ICAF 2011 Structural Integrity: Influence of Efficiency and Green Imperatives

ICAF 2011 Structural Integrity: Inﬂuence of Efﬁciency and Green Imperatives Jerzy Komorowski (Editor) ICAF 2011 Stru...

Author: Jerzy Komorowski

26 downloads 1321 Views 44MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

ICAF 2011 Structural Integrity: Inﬂuence of Efﬁciency and Green Imperatives

Jerzy Komorowski (Editor)

ICAF 2011 Structural Integrity: Inﬂuence of Efﬁciency and Green Imperatives Proceedings of the 26th Symposium of the International Committee on Aeronautical Fatigue, Montreal, Canada, 1-3 June 2011

ABC

Editor Jerzy Komorowski National Research Council Canada Institute for Aerospace Research 1200 Montreal Road K1A 0R6 Ottawa Ontario Canada Telephone: 613-993-0141 Fax: 613-952-7214 E-mail: [email protected]

ISBN 978-94-007-1663-6

e-ISBN 978-94-007-1664-3

DOI 10.1007/978-94-007-1664-3 Library of Congress Control Number: 2011926989 c 2011 Springer-Verlag Berlin Heidelberg This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, speciﬁcally the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microﬁlm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer. Violations are liable to prosecution under the German Copyright Law. The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a speciﬁc statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. Typeset & Cover Design: Scientiﬁc Publishing Services Pvt. Ltd., Chennai, India. Printed on acid-free paper 987654321 springer.com

International Committee on Aeronautical Fatigue

ICAF General Secretary Dr. Anders Blom Defence & Security, Systems and Technology Swedish Defence Research Agency (FOI) SE-164 90 Stockholm, Sweden [email protected]

National Delegates Australia Phil Jackson Helicopter and Transport Aircraft (Project) Structural Integrity Defence Science and Technology Organization Mail 506 Lorimer Street Fishermens Bend VIC 3207, Australia [email protected]

Canada Jerzy P. Komorowski Institute for Aerospace Research National Research Council Canada 1200 Montreal Road, Bldg. M-3 Ottawa, Ontario K1A 0R6, Canada [email protected]

Finland Dr. Aslak Siljander VTT Technical Research Centre of Finland Machines and Vehicles P.O. Box 1000 FI-02044 VTT, Finland [email protected]

France Dr. Thierry Ansart Structures Division Toulouse Aeronautical Test Center (CEAT) BP 93123, 47 rue Saint Jean 31131 Balma, France [email protected]

Germany Dr. Claudio Dalle Donne EADS Innovation Works 81663 München, Germany [email protected]

Israel Abraham Brot Israel Aerospace Industries Dept. 4444, Engineering Div. Ben-Gurion Airport 70100, Israel [email protected]

Italy Prof. Luigi Lazzeri Universita di Pisa Department of Aerospace Engineering Via G. Caruso, 8 56122 Pisa, Italy [email protected]

VI

International Committee on Aeronautical Fatigue

Japan

The Netherlands

Prof. Nobuo Takeda University of Tokyo Dept. of Aeronautics and Astronautics 7-3-1 Hongo, Bunkyo-ku Tokyo 113-8656, Japan [email protected]

Marcel J. Bos National Aerospace Laboratory (NLR) P.O. Box 153 NL-8300 AD Emmeloord, The Netherlands [email protected]

Poland Dr. Antoni Niepokólczycki NET Institute Institute of Aviation Al. Krakowska 110/114 02-256 Warsaw, Poland [email protected]

Sweden Hans Ansell Saab Aerosystems Saab AB SE-581 88 Linköping, Sweden [email protected]

Switzerland Dr. Michel Guillaume RUAG Aerospace P.O. Box 301 CH-6032 Emmen, Switzerland [email protected]

United Kingdom Dr. Steve Reed Technical Advisor Structural Integrity Physical Sciences Department Defence Science and Technology Laboratory Desk 236, Room 102, Bldg 5 (Porton Down) Salibury, Wilts SP4 0JQ, UK [email protected]

United States of America Dr. Ravinder Chona Structural Sciences Center US Air Force Research Laboratory (AFRL / RBSM) Wright-Patterson Air Force Base Ohio, USA [email protected]

Preface to ICAF 26th Symposium Proceedings

The International Committee on Aeronautical Fatigue Conference and Symposium are coming back to Canada after a long trip around the world, to member countries spread over 5 continents. Twenty four years ago the 14th ICAF Symposium was organized in Ottawa, Ontario by then Canadian National Delegate Mr. David Simpson. ICAF has since visited two major aerospace clusters - Seattle (1999) and Toulouse (2001). It is with great pride that I have the pleasure of inviting everyone to the third major aerospace cluster, Montréal! The city is home to many well recognized aerospace companies: Bombardier Aerospace, Pratt and Whitney Canada, Bell Helicopter Textron Canada and CAE as well as many important lower tier suppliers. The nearly 40,000 Montréal aerospace employees often point out that within 30km of downtown Montréal one can design and build, from the ground up, a complete aircraft – something not possible in Seattle or Toulouse. I would expect that Conference and Symposium attendees will not only enjoy the technical presentations and tours but also take the time to appreciate Montréal as a unique multicultural city that represents much of the best that Canada has to offer. ICAF was formed in 1951 in response to the growing concerns regarding fatigue problems in metal aircraft structures. The stated aims of ICAF are to exchange information concerning aircraft structural fatigue and to encourage contacts between people active in this field. To this end a Conference and Symposium are organised every two years for attendance by representatives of industry, universities and institutes, regulatory agencies and operators throughout the world. The two-day Conference consists of reviews of aeronautical fatigue activities presented by the National Delegates of ICAF member nations. It is followed by the three-day Symposium that consists of specialist papers presented by authors with backgrounds and expertise in design, manufacturing, airworthiness regulations, operations and research. Much has changed since the first symposium and ICAF has had to adapt to these changes. The organization has now 14 member nations. While keeping the established and well recognized ICAF acronym, the full name of the organization has changed to International Committee on Aeronautical Fatigue and Structural Integrity in recognition of the expanded scope of ICAF interests as well as the technological changes in the industry. The Conference and Symposium theme “Structural Integrity: Influence of Efficiency and Green Imperatives” is in keeping with one of the new concerns of aerospace. It is no longer only safety, weight and cost but also environmental sustainability of the industry that matters. Each ICAF symposium starts with a lecture honoring the memory of Dr. Fredrick J. Plantema, the founder of the International Committee on Aeronautical Fatigue. Dr. Plantema took the initiative of forming the ICAF in 1951 with the

VIII

Preface to ICAF 26th Symposium Proceedings

stated objectives of forming closer cooperation with various institutes carrying out non-classified work. This year Professor Graham Clark, currently at Aeronautical Design School of Aerospace of RMIT University in Melbourne Australia, joined a long list of distinguished Plantema lecturers. Professor Clark has been involved with ICAF since 1990 and was Australian National Delegate to ICAF from 2002 to 2010. ICAF has been primarily focused on the structural integrity of fixed wing aircraft, with rotorcraft and engine related issues having been given less space. In a departure from this tradition I have sought help of Dr. Bogdan Krasnowski of Bell Helicopter in organizing a dedicated helicopter session. It is my sincere hope that this innovation will become a permanent feature of ICAF Symposia as the fixed wing and helicopter communities have a lot to share. With the Bell/Boeing V-22 and Bell/Augusta BA609 the distinction between the two aircraft types has become blurred. Organization of a major international conference is not possible without a team effort. I have been particularly privileged to have the help of Mr. David Simpson a long time Canadian National Delegate and the previous Secretary General of ICAF. Mr. Simpson and Mr. Pierre Lamoureux of the National Research Council Canada have formed the core of the ICAF2011 Management Office. Without their initiative, enthusiasm, experience and diligence ICAF2011 would not be possible. Mr. Simpson has also chaired the Technical Committee consisting of following individuals: N. Bellinger, National Research Council of Canada, E. Burczak, Bell Helicopter Textron (Canada), J. Dubuc, L-3 MAS Canada, R. Fews, Concordia University, J. Gaerke, Department of National Defence, L. Kok, Bombardier Aerospace, P. Lortie, Bombardier Aerospace, P.V. Straznicky, Carleton University. I am greatly indebted to these individuals for their invaluable assistance in shaping the Symposium technical program. Financial support for ICAF 2011 has been received from various organizations. Bombardier Aerospace, Bell Helicopter Textron Canada, L3 Military Aircraft Support, Fatigue Technology Inc., MTS, IMA Dresden and Consortium for Aerospace Innovation in Quebec have all stepped forward as Symposium Partners. The contribution by Tourisme Montréal is also gratefully acknowledged. Finally I would like to thank all other individuals, organizations, Secretary General and National Delegates for their contribution to the success of ICAF 2011. The papers included in this volume should be of interest to the broad aerospace community concerned with structural integrity and to the newcomers to the field. A particular expression of gratitude is in order to the authors and their organizations. Without their contribution the proceedings would have never been published. Jerzy Komorowski Conference Chair Institute for Aerospace Research, National Research Council of Canada

Table of Contents

Plantema Memorial Lecture ∗

Fleet Recovery and Life Extension – Some Lessons Learned . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Graham Clark

1

Airworthiness and Other Considerations ∗

27

Analysis of Requirements on Fatigue and Damage Tolerance for Civil Transport Airplanes . . . . . . . . . . . . . . . . . . . . . . B.G. Nesterenko, G.I. Nesterenko

39

Material Selection and Detailed Design – Requirements and Responsibilities of an Accredited and Qualiﬁed Test Laboratory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . R. Best, Th. Fleischer, R. Franke, S. Goldbach, J. Gruner, S. Reichard, J. Ridzewski

61

Sticks and Stones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Steve Swift

Advanced Materials and Innovative Structural Concepts ∗

Bombardier Aerospace FSW Demonstrator . . . . . . . . . . . . . . . . . L.J.J. Kok, Ken Poston, Gary Moore

73

∗

Evaluation of Fatigue Crack Growth Behavior in FSW Joint by Experiment, Analysis and Elasto-Plastic FEM . . . . . . T. Okada, K. Kuwayama, M. Asakawa, T. Nakamura, S. Machida, S. Fujita, H. Terada

∗

Oral presentation.

83

X

Table of Contents

∗

Recent Development on Bonded Structures . . . . . . . . . . . . . . . . . G. Delgrange, J.C. Ehrstr¨ om

93

∗

Recent Advancements in Thin-Walled Hybrid Structural Technologies for Damage Tolerant Aircraft Fuselage Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 R.C. Alderliesten, C.D. Rans, Th. Beumler, R. Benedictus ∗

Fatigue Life Assessment for Composite Structure . . . . . . . . . . . 119 Andrew Makeev, Yuri Nikishkov ∗

Potential of CFRP Used for Light Weight Structures: Some Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 Th. Fleischer, J. Ridzewski, M. Sachse, F. Schirmacher ∗

An Implementation of an Accelerated Testing Methodology to Obtain Static, Creep and Fatigue Master Curves of a T300/913 Unidirectional Composite Material . . . . 145 Yuval Freed, Sven Rzepka Improvement of Vibration Damping and Flexural Fatigue Property Incorporating Nanoclay into Glass/Epoxy Composite . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155 A. Kabir, S.V. Hoa Formation of a Metal Coating by Means of Friction Stir Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 D. Koca´ nda, A. G´ orka, D. Zasada Inﬂuence of the Carbon Nanotube Type, Loading and Chemical Functionalization on the Fatigue Resistance of Aluminum Lap Joints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 Iosif D. Rosca, Roham Mactabi, Suong V. Hoa Fatigue Damage Behavior of Glass/Epoxy Composites Using Carbon Nanotubes as Sensors . . . . . . . . . . . . . . . . . . . . . . . . . 189 H. Hena-Zamal, S.V. Hoa DSTO – NLR Collaborative Programme on Fatigue Properties of β-Annealed Ti-6Al-4V: Preliminary Results . . . . 199 E. Amsterdam, A. Shekhter, S.A. Barter, M. McDonald, R.J.H. Wanhill Damage Tolerance Demonstration of Flange Joint for Aircraft Engine Composite Fan Case . . . . . . . . . . . . . . . . . . . . . . . . 207 Y. Ueda, H. Kuroki, T. Murooka, A. Tanaka, K. Miyazawa, I. Okumura, Y. Shigenari, K. Oikawa, H. Morita ∗

Oral presentation.

Table of Contents

XI

Fatigue Crack Growth and Life Prediction Methods ∗

The Formation/Nucleation of Fatigue Cracks in Aircraft Structural Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219 David W. Hoeppner ∗

Modelling of Continuing Damage for Damage Tolerance Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231 Yan Bombardier, Min Liao, Guillaume Renaud ∗

The Critical Importance of Correctly Characterising Fatigue Crack Growth Rates in the Threshold Regime . . . . . . . 249 K.F. Walker, S.A. Barter ∗

The Relationships between Crack Closure, Specimen Compliance and ‘Eﬀective’ Fatigue Crack Growth Rate . . . . . . 265 D.L. Ball, J.K. Donald, M.A. James, R.J. Bucci ∗

Experimental and Numerical Study of Stress and Strain Field around the Rivet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277 W. Wronicz, J. Kaniowski, B. Korzeniewski, E. Gadalinska ∗

Fatigue Analyses of Riveted Lap-Splice Joints in a Narrow-Body Aircraft . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289 J.C. Newman Jr., R. Ramakrishnan Crack Growth Rate Curves: Which Part Dominates Life Prediction and When? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303 C. Wallbrink, P. Jackson, W. Hu Critical Distance for Fatigue Life Prediction in Aerospace Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315 Yoichi Yamashita, Yusuke Ueda, Hiroshi Kuroki A Uniﬁed Variable-Amplitude Model for Crack Initiation and Crack Propagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325 A.B. Chattopadhyay, G. Glinka Development of an Eﬃcient Methodology and Tool to Determine Stress Intensity Correction Factors for Complex Aircraft Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335 Guillaume Renaud, Min Liao, Yan Bombardier Improved SIF Calculation in Riveted Panel Type Structures Using Numerical Simulation . . . . . . . . . . . . . . . . . . . . . . 347 S.C. Mellings, J.M.W. Baynham, T.J. Curtin ∗

Oral presentation.

XII

Table of Contents

An 1D-Beam Approach for Both Stress Analysis and Fatigue Life Prediction of Bonded Joints . . . . . . . . . . . . . . . . . . . . . 359 E. Paroissien, A. Da Veiga, A. Laborde Cyclic Stress-Strain and Strain-Life Properties of Aerospace Metallic Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375 S.K. Walker, A.C. Quilter Crack Propagation Calculation for Aluminium Aircraft Structures Considering the Inﬂuence of Load Sequences . . . . . 389 R. Buchholz Analysis of Fatigue Crack Growth under Random Load Sequences Derived from Military In-ﬂight Load Data . . . . . . . . 399 C. Mattrand, J.-M. Bourinet, D. Th´eret Statistical Analysis of Fatigue Crack Growth Based on the Unigrow Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415 S. Mikheevskiy, S. Bogdanov, G. Glinka Fatigue Life Estimation of Structures Subjected to Vibratory Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427 M. Fressinet, F. Fuchs, P. Madelpech A Structural Defect Expansion Model Based on Physical Correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443 S. Ito, S. Sugimoto, T. Okada

Structural Health and Structural Loads Monitoring ∗

Link between Flight Maneuvers and Fatigue . . . . . . . . . . . . . . . . 453 Juha Jylh¨ a, Marja Ruotsalainen, Tuomo Salonen, Harri Janhunen, Ilkka Ven¨ al¨ ainen, Aslak Siljander, Ari Visa ∗

Health and Usage Monitoring of Unmanned Aerial Vehicles Using Fiber-Optic Sensors . . . . . . . . . . . . . . . . . . . . . . . . . . 465 I. Kressel, A. Handelman, Y. Botsev, J. Balter, P. Gud’s, M. Tur, S. Gali, A.C.R. Pillai, M.H. Prasad, A.K. Yadav, N. Gupta, S. Sathya, R. Sundaram ∗

Airframe Loads and Usage Monitoring of the CH-47D “Chinook” Helicopter of the Royal Netherlands Air Force . . . 473 A. Oldersma, M.J. Bos

∗

Oral presentation.

Table of Contents

XIII

Damage Detection System for Automated Hot Spot Monitoring Based on Diﬀerent Technologies Used in Component Testing for Shot Peening Validation . . . . . . . . . . . . . 495 C. Stolz, M. Neumair, L. Benassi Memorization and Detection of an Arrested Crack in Foam-Core Sandwich Structures Using Embedded Metal Wires and Fiber-Optic Sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 507 Shu Minakuchi, Nobuo Takeda, Yasuo Hirose Structural Load Monitoring Systems for Military Aircraft in the Polish Armed Forces with Examples of Selected Activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 519 M. Kurdelski, A. Leski, S. Klimaszewski, M. Stefaniuk

Full Scale Fatigue Testing of Aircraft and Aircraft Components ∗

A320 ESG Full Scale Fatigue Test - Lessons Learned . . . . . . . . 529 G. Hilfer, N. R¨ oßler, C. Peters, C. Herrmann ∗

New Connection Strap Concepts for A320 Wheel Well Area Tested during the Airbus A320 Extended Service Goal Full Scale Fatigue Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 539 Nikola Cenic, Bernd Zapf, Till Haberle ∗

Full-Scale Static and Fatigue Testing of Composite Fuselage Section . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 551 Rushabh Kothari ∗

Durability and Damage Tolerance Evaluation of VaRTM Composite Wing Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 561 Yuichiro Aoki, Yoshiyasu Hirano, Sunao Sugimoto, Yutaka Iwahori, Yosuke Nagao, Takeshi Ohnuki Development of Load Spectrum for Full Scale Fatigue Test of a Trainer Aircraft . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573 Andrzej Leski, Piotr Reymer, Marcin Kurdelski

Inservice Experience, Life Extension and Management of Ageing Fleets ∗

ATR Life Extension Project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 585 Maurizio Cajani, Roberto Ciotola, J´er´emy David, Jaco Salvi

∗

Oral presentation.

XIV

Table of Contents

∗

Beneﬁts of Using a Risk Process in ASIP – The CF-18 Experience . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 601 Yves Beauvais ∗

Integrated Probabilistic Analysis of Damage Tolerance and Risk for Airframe Structural Locations . . . . . . . . . . . . . . . . . . 615 W. Hu, R.F. Torregosa Aircraft Joints and Corrosion Control . . . . . . . . . . . . . . . . . . . . . . . 625 Ung Hing Tiong, Graham Clark A Case Study of Nose Landing Gear Failure Caused by Fatigue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 635 V.Y. Guertsman Test Method for Determining the Eﬀect of Chromate Primers on Fatigue Crack Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . 645 Yongwon Lee, Sarah E. Galyon Dorman, Matthew J. Hammond Life Extension: Fatigue Lifetime Updating of the French Xingu Fleet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 655 P. Madelpech, D. Th´eret, M. Fressinet, J. Despujols, B. Andr´e Fatigue Life of Cold Expanded Fastener Holes at Short Edge Margins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 671 G.M. Valli`eres, D.L. DuQuesnay Environmentally Assisted Cracking in Advanced Aerospace Aluminums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 685 E.M. Arnold, J.J. Schubbe, P.J. Moran, R. Bayles An Overview of Fretting Aspects Relating to Aero-Engine Dovetail Attachment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 697 Raghu V. Prakash, K. Anandavel, P. Balasubramani Fatigue Analysis of the Compressor Blades with V- Notches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 721 Lucjan Witek

Fatigue Life Enhancement Methods and Repair Solutions ∗

Fatigue and Residual Strength Performance of Bonded Repairs to Metallic Fuselage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 735 John G. Bakuckas Jr., Bud Westerman ∗

Oral presentation.

Table of Contents

XV

∗

Life Extension Techniques for Aircraft Structures – Extending Durability and Promoting Damage Tolerance through Bonded Crack Retarders . . . . . . . . . . . . . . . . . . . . . . . . . . . . 753 P.E. Irving, X. Zhang, J. Doucet, D. Figueroa-Gordon, M. Boscolo M. Heinimann, G. Shepherd, M.E. Fitzpatrick, D. Liljedahl ∗

Damage Tolerance of Adhesive Bonded Stiﬀened Panels: Experimental and Analytical Investigation of the Fatigue Crack Propagation Underneath the Stringers . . . . . . . . . . . . . . . . 771 Ivan Meneghin, Gianluca Molinari, Goran Ivetic, Enrico Troiani ∗

Development of a New Fiber Metal Laminate Variant Optimized for Cold Expansion and Riveting of Holes . . . . . . . . 785 David Backman, Thomas Sears, Eann A. Patterson ∗

Applying the Damage Tolerance Approach to Expanded Bushing and Rivetless Nut Plate Installations . . . . . . . . . . . . . . . 797 Len Reid ∗

Fatigue Life Improvement of Metallic Aerospace Structures via Crenellations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 811 M.V. Uz, Y.J. Chen, N. Huber Fatigue Life Improvement Using In-situ Robotic Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 827 Zahi Hajjar, Benoit Leblanc Investigations into the Fatigue Enhancement Provided by the Hole Cold Expansion Process Using Accurate 3D FEA Simulations and Fatigue Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 839 S.J. Houghton, S.K. Campbell, A.D. James Characterisation of Fatigue and Crack Propagation in Laser Shock Peened Open Hole 7075-T73 Aluminium Specimens . . . 855 G. Ivetic, I. Meneghin, E. Troiani, G. Molinari, A. Lanciotti, V. Ristori, J.L. Oca˜ na, M. Morales, J.A. Porro, C. Polese, A.M. Venter

Helicopter Fatigue and Damage Tolerance ∗

Damage Tolerance of Titanium Alloy Rotorcraft Components: Advantages and Challenges . . . . . . . . . . . . . . . . . . . . 867 X. Li, B.R. Krasnowski, W.P. Green ∗

NH90 Qualiﬁcation According to Damage Tolerance . . . . . . . . 877 Alain Struzik ∗

Oral presentation.

XVI

Table of Contents

∗

Damage Tolerance for Composite Parts . . . . . . . . . . . . . . . . . . . . . 899 Rupert Pfaller ∗

Towards Weight Savings for Damage Tolerant Masts and Driveshafts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 915 W.P. Green, B.R. Krasnowski, X. Li ∗

Challenges in Damage Tolerance Approach for Dynamic Loaded Rotorcraft Components – From Risk Assessment to Optimal Inspection Planning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 927 Jack Zhao, David Adams ∗

Improvements in Fatigue Evaluations of Helicopter Transmissions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 959 Ugo Mariani, Rosanna Molinaro, Sergio Sartori, Giuseppe Gasparini, Carlo Gorla Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 971

∗

Oral presentation.

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Fleet Recovery and Life Extension – Some Lessons Learned Graham Clark Innovation Professor, Aerospace Design, RMIT University, Melbourne, Australia

Abstract. Extending the life of an existing fleet which still has acceptable operational capability can be enormously attractive in economic terms. Ideally, such extension programs will be planned and managed (via an ASI program), although many are urgent “recovery” programs required when substantial problems are discovered. This paper discusses examples of planned and unplanned programs, highlighting the differences in approach required. Regulatory systems usually demand that we preserve the prevailing “acceptable level of safety” during fleet life extension, although inevitably, progressive life extension must lead to enhanced risk of unforeseen events which are absent from our structural integrity models. We cannot remove this risk, but we can mitigate it. Paradoxically this requires additional (and potentially unwelcome) investment in broad investigative strategies such as teardowns and damage enhancement test programs. This paper will provide examples of a management program that was successful precisely because it contained such strategies. The paper argues that we may underestimate the extent to which organisational issues may bring an additional (and perhaps more important) threat to the safety of old aircraft. Two examples are provided in which complacency and a perception that the fleet is nearing end-of-service promoted drawing down of maintenance/safety resourcing, leading to maintenance underperformance, increased risk, accidents, and loss of life. These issues will be particularly evident where we have poor corporate culture, weak organisational structure, progressive deferral of fleet withdrawal dates and increased operational demand. The examples suggest that our structural safety models are in themselves of limited value if these broader system/organisational risks are neglected.

1 Introduction Cost, and the value of Life Extension and Recovery (LEx) programs The economic impact of replacing a major fleet – even in terms of the relatively small Australian Defence Force this will amount to tens of billions of dollars – is so large that planning to extend the life of current aircraft which are still offering a reasonable level of capability is usually seen as worthwhile. The benefit may not be just economic – usually the new fleet is an aircraft which is still being *

Oral presentation.

2

G. Clark

developed, and deferring an acquisition for a few years is likely to result in the new fleet, when it arrives, to be a more mature and more capable system rather than early production models; this is particularly true in the development of onboard systems which may be at a low level of development in the early models. The increase in cost of new acquisitions is very substantial; it has been noted [1] that “since the 1950’s aircraft payload has increased by a factor of five, performance/range increased by a factor of four, but cost increase is times one thousand, and the main reason is SYSTEMS costs”. Usually, and ideally, the deliberation about whether or not to extend life, and the subsequent decision-making, occurs in a planned environment, and a suitable trade-off between declining capability and cost can be reached. However, more often than not, this process is disrupted by one of two issues: (i) it is fairly usual to see delays in the “new” fleet’s development – this leads to continual rewriting of the Planned Withdrawal date (PWD) for the existing aircraft, each time a further delay is announced. (ii) there are many cases where the existing fleet experiences an accident, a significant maintenance issue involving unacceptable increases in cost-ofownership. These life-limiting events usually result in demand for fleet recovery involving desperate efforts to achieve a minimal level of capability with a much reduced fleet. The lead time in any fleet withdrawal and replacement is usually so long that if there is a need to retire a fleet very rapidly, and there is no means of providing the lost capability, there result will be costly short-term purchase (or lease) of alternative aircraft. This paper discusses two cases – both of which involved planned life extensions, but in one case such planning disappeared when there was an accident requiring urgent fleet recovery. Both programs successfully provided economic benefit and preserved capability, but they required very different approaches; the structured example used large-scale testing backed by laboratory-level support, while the recovery program required development of innovative tools “on the run”. Interestingly, both led to significant advances in understanding of structural lifing issues. Maintaining an acceptable level of safety in Life Extension Given the substantial economic benefits of extending the life of a fleet which can still provide capability, what factors limit this process? In the absence of accidents or major surprises such as discovery of structural problems, individual structural integrity issues do not usually present us with a “hard” limit – even where safe-life presents a “retire by” date, RAAF can usually plan for and implement, transitions to inspection-based approaches. The examples given later outline the need to do this on a “whole-of-aircraft” approach, even though individual structural issues may initially attract attention. The key life-limiting features are usually resolvable by introducing inspections, for example, although such activity will often result in an increase in cost which will need to be sustained throughout the extended life, and it may be more economical (in terms of maintenance and reliability) to introduce a terminating repair or replacement for some issues.

Fleet Recovery and Life Extension – Some Lessons Learned

3

In contrast, there is often a need for systems upgrades to maintain the fleet capability throughout the proposed life extension, and this presents the operator with a major cost step; such a systems upgrade – which is usually extremely costly relative to the structural issues – can become the economic driver for a desired service life limit. In all of the discussion preparatory to life extension, maintenance of an acceptable level of safety is usually taken to be a “given”, and in terms of life extension, our target is usually maintenance of the current level of structural safety. It is worth asking whether these are realistic, and the following sections will attempt to illustrate some of the key issues. Damage and crack initiators: Take, for simplicity a fleet managed on a safe-life basis. Figure 1, which is discussed in more detail in [2] illustrates the conventional representation of a (log) normal distribution of fleet fatigue lives. In this, cracking will develop readily from small, numerous materials or manufacturing features in any location which has suitably high stress. The initiators are assumed to be numerous throughout the structure, as shown in the notional “inherent” distributions in Fig 2. Fatigue life is dominated by variability in geometry, design and local stressing, and by applying a suitable scatter factor1 fatigue life is restricted to a notional safe life which is a fraction of the mean life. That mean life is demonstrated by full-scale testing under conditions which are expected to match as closely as possible service conditions. The testing is also expected to reveal unidentified critical locations, an assumption which relies on the test article being fully representative of the fleet. However, it has long been recognised that the promulgated fatigue lives of fleets can be very optimistic. In many cases, this reflects the existence of other features, design oversights, incorrect design/usage data, or damage – perhaps manufacturing features or geometrical anomalies, or unaccounted-for material inhomogeneities – which can generate cracking much earlier in life. What is important here is to recognise that this additional class of damage, here called “rare”, some of which carries the likelihood of being in a location with sufficient stress to allow significant growth, may occur in only one or two locations and perhaps in only a few aircraft in the fleet, and we cannot therefore rely on finding it in a full scale test. It is hard to avoid mention in this context of the “rogue flaw”; the author takes the view that the term has sometimes been used to label an initiator or feature as something which we could not reasonably have been expected to foresee. Unfortunately this might be used to conclude that we cannot be expected to foresee other “rogue flaws”. An alternative view would be that such features are usually failures of process, whether that be design, maintenance, or one of the many other facets of a complex systems and its management, and such a view might be more productive in terms of encouraging the broadest possible effort to minimise risk.

1

In Damage Tolerance, the equivalent would be assuming conservative starter crack sizes and inspection intervals.

4

G. Clark

Fig. 1 The (log) normal distribution curve usually considered representative of fleet fatigue life reflects the concept that many initiators exist, allowing cracks to develop readily at high stress locations; a scatter factor can be used to protect against all but the most extreme cases in the distribution. However, the same scatter factor also protects against the rare discrete damage which occurs sporadically in some fleet aircraft.

Fig. 2 Notional set of crack initiating features (as equivalent crack size) indicating the need to sum the extreme cases (upper tails of distributions) of very numerous initiating sites such as inclusions, and the rare or extreme discrete features such as mechanical damage which may occur only once or twice in a fleet.

Is this additional population of rare damage important? In reality, of course, the scatter factor we adopt for the fleet safe-life also covers the fleet against many of these cases (the “unknown unknowns”) although one issue of concern is that the potential existence of this ‘hidden” population should not be forgotten in situations where there is an attempt to refine the scatter factor based on full scale testing. One obvious reason for the limited discussion of this issue is that it is clearly difficult to

Fleet Recovery and Life Extension – Some Lessons Learned

5

introduce any additional, uncertain limits on fleet life service life, based simply on “unknown unknowns”. Such conservatism would need to be justified. What is clear, though, is that progressive life extension must attract an enhanced exposure to risk of unforeseen events – however much we rely on our scatter factors (or inspections) this increased risk will arise from the “unknown unknowns”. Corrosion, for example, can be one such source, causing time-based changes in distributions as illustrated in Fig 2, and corrosion at unexpected locations should feature more highly in a life-extended fleet. Indeed the whole process of breakdown in protective systems –paints, sealants – would be expected to provide many additional problems only later in life. Managing increasing risk from unforeseen events: How can we monitor this enhanced risk and mitigate it? Firstly, we need to observe that aviation accidents are sufficiently rare that simply waiting for accidents (monitoring fleet health on the basis of losses) is not an acceptable approach - risk mitigation requires a proactive, not a purely reactive approach! The answer is that there are some tools which can be applied; (i)

(ii) (iii)

one obvious one is using the life extension program to introduce enhanced corrosion management, for example by increased scope of inspection, improved broad-area application of corrosion protection and improved maintenance of protection systems; this will help defer the onset of widespread corrosion later in life and will contain fleet whole-of-life costs and minimise the risk of losing capability through surprise discovery of major damage. However, we need to acknowledge that we cannot cover all eventualities by, for example, increasing inspection - some of the areas needing to be examined are simply not readily accessible. More aggressive exploratory programs such as additional teardowns – an approach adopted extensively in Australia Introduction of damage enhancement testing to promote growth of small cracks already present in the fleet. An example of a damage enhancement test program provided later highlights the value of these proactive programs in mitigating the risk of problems emerging from “unknown unknowns”.

Feedback to improved design: An additional benefit flows from the investigations which can be used to support life extension. One such benefit is that access to inservice airframes and systems means that damage enhancement testing, while it cannot guarantee discovery of “rare” fleet damage, can identify issues that may be systemic (manufacturing issues, for example). In addition to supporting management of the issue for the in-service fleet, this has the potential to provide valuable feedback to support improved design and manufacturing. Some examples come to mind – experience with SCC in aging aircraft fleets proved clear impetus for the4 development of SCCresistant materials for new design and manufacture, and also led to the development of innovative methods of restoring SCC-resistance to existing components (eg. Retrogression and Re-Aging). Future teardowns and damage enhancement tests will allow “field” evaluation of approaches currently being used to produce fatigue –resistant

6

G. Clark

holes, and more complex joint designs, and will also provide valuable feedback on the performance of composites and adhesively bonded joints. Impact on overall LEx planning: A practical, political difficulty is that these approaches involve additional investment; damage enhancement testing and teardowns, in particular, are significant cost items, and arguing for such additional investment, particularly in the context of a life extension proposal focussed on economy, can present a challenge. Nevertheless it is vital to mount an effective argument to ensure that such proactive activities are included in programs such as Aging Aircraft Audits to minimise structural integrity “surprises” from the kind of sources described earlier.

2 Investment to Mitigate the Risk of Surprises Case Study: Teardown and damage enhancement: RAAF F-111 The RAAF operated the F-111 fleet until late in 2010, some ten years after the USAF retired its aircraft; to allow this required substantial investment in activities (the Sole Operator Program, SOP) designed to equip Australia with the tools and knowledge which would allow operation to a planned withdrawal date of 2020. This SOP program included developing and using a full structural model allowing loads/stress analysis, undertaking management of stress corrosion cracking in an SCC sensitive fuselage structure, and addressing a number of issues relating to wing reinforcement. The program also allowed development of innovative rework approaches [3] based on structural optimisation approaches which culminated in reshaping of critical regions in the high strength steel wing structure (Fig 3), reducing stresses dramatically, and extending inspection to much more manageable intervals.

Fig. 3 Structural optimisation rework developed and applied to F-111 wing pivot fiting, showing a major reduction in local stress (summarised in [3]).

Fleet Recovery and Life Extension – Some Lessons Learned

7

A key factor in the SOP was the funding of structural teardown of aircraft to identify and document the numerous areas in which corrosion and fatigue degradation might become difficult as the fleet continued in service. More importantly, it transpired, was the conduct of a Wing Damage Enhancement Test (WDET) in which a service wing – which was presumed to contain small cracks that would be difficult to find – was subjected to additional fatigue loading to grow the damage to the stage where a teardown would allow it to be detected, documented and analysed. This test was expected to provide core data for future fatigue evaluations. The test was not a fatigue test which aimed for an exact replication of service conditions, but focussed instead on rapid testing, and the loading was not fully representative of RAAF flight loads.

Fig. 4 Manufacturing damage leading to fatigue cracking in F-111 wing structure, and a region of high stress in the Australian wing configuration.

In 2002 the test wing failed, revealing numerous sites (Fig 4) with manufacturing damage – damage which had never been revealed or suspected in earlier tests. It also highlighted some high stress regions which showed that the basis for fatigue certification was unreliable - the RAAF wing geometry differed from that in the original test and earlier testing had not addressed this difference. An additional issue complicating fatigue evaluation was that the holes and fasteners relied on achieving a specified level of interference fit, and many lay outside specification. Recovery from this situation required extensive effort –the older model wings were retired, and when the replacement (used) wings were introduced, and were found to also feature build-quality issues, a two-pronged approach was needed: (a) a new full scale fatigue test was used to address the overall fatigue life certification of the wing, and (b) a major NDE development was started to allow examination and inspection-based management of the manufacturing damage issue, based on full inspection to identify anomalies in holes. It is interesting to note that this two-pronged approach reflects the two sources of failure discussed earlier – dealing with the overall certification issue based on a full scale, representative test (ie. using a test article declared to be as representative as possible), and a program to deal with the variable and scattered manufacturing damage throughout the fleet.

8

G. Clark

So, how valuable was the investment in the WDET test, and the SOP program? The testing allowed detection of a significant fleet issue which had never been suspected, and which, if the RAAF had continued its fleet operations to 2020, may well have led to loss of aircraft, and an immediate loss of the capability that fleet provided. The same issue may also have become a problem for the USAF, had they decided to extend their fleet life. On a practical level, the Australian result allowed a managed withdrawal of a fleet whose costs were high, and which, as demonstrated, was likely to provide surprises. The proactive testing in this case provides an excellent example of the value of this investigative approach. The message arising from this is that while extending fleet operations is undoubtedly of economic benefit, the risk of surprise events, and accidents, will rise, and this translates to unreliability in terms of fleet capability. To mitigate this risk requires additional investment in the form of damage enhancement testing, or teardown. While such additional investment might be unattractive in terms of life extension focussed on economy, the F-111 case shows that it is vital to argue that such reinvestment is justified as part of a structural integrity management program.

3 Life Extension Programs, Structured or Otherwise A planned Service Life Extension Program: RAAF’s P3-C fleet SLAP Program: The Australian Defence Force (ADF) performs its maritime surveillance role with a fleet of P3-C aircraft; supporting a proposed major avionics update would require extending the PWD to 2015, implying a safe-life extension by approximately 50%. Several limitations were apparent in the basis of the - safe life approach then in use (adopted from the USN); in particular, both the aircraft usage spectrum and key structural elements had changed since the original fatigue test conducted back in 1961. In addition, there were shortfalls in estimating some flight and ground loads and the way they were applied during testing. This led to concern that any inspection program based upon the findings of the original test would not accurately target the critical locations. In addition, the true economic life ie. the point at which generalised cracking became common (whether WFD or not) was not known. Australia joined a P-3 Service Life Assessment Program (SLAP) proposed by the USN, along with Canada and The Netherlands. The objective of the program was to substantiate the 15,000 hour design life of the structure with an 85th percentile mission profile usage in accordance with USN philosophy [4]. In this program, conducted between 1999 and 2006 [5], Australia performed flight loads measurement, wing teardown and an empennage full scale fatigue test on a retired structure [6]. The wing teardown (Fig. 5) on a high-time fleet wing revealed valuable data on the extent of cracking (relatively little) and corrosion (relatively widespread).

Fleet Recovery and Life Extension – Some Lessons Learned

9

Fig. 5 RAAF P-3C Empennage fatigue test. US-based wing test (from [7], and in-service wing teardown.

The P-3C SLAP involved the use of an 85th percentile USN test spectrum to substantiate the 15,000 hour design life of the aircraft [7]. This spectrum was found to be more damaging than the RAAF usage (Fig 6) and so the results from SLAP had to be adjusted so the findings could be applied to the RAAF P-3C fleet. Research Outcomes: The full-scale empennage fatigue test provided an opportunity for exploration of one issue which affects lightly-loaded aircraft structure, namely the difficulty of running a fatigue test long enough to achieve meaningful results in the low-load (and consequently high-scatter) regime. The test included a phase of augmented loading, which normally would raise the prospect of an unrepresentative fatigue result because of local complex effects involving overload, yielding and retardation; after substantial discussion – essential when such tests involve a major full scale test article – the augmented spectrum was clipped at a level which was able to minimise these potential effects. After two lifetimes of baseline loading (30,000 SFH), a further two lifetimes of augmented loading was applied, the second lifetime of testing including newly-introduced damage (sawcuts) to provide experimental data to support crack growth calculations. The test successfully revealed the key fatigue critical locations at the base of both the fin and horizontal tail, one location being adjacent to an existing inspection point established analytically under the existing aircraft management regime. The P3 ASI management program that was then in place, therefore, would not have kept the aircraft safe.

10

G. Clark

Secondly, the results also allowed a comparison of different lifing approaches using the same test data [6, 8]; in particular, it was intended to allow exploration of the UK DEF STAN 00-970 Safe S-N [9] concept of variable scatter factor associated with different structural features. This aspect of the test concluded that inspection thresholds calculated using FAR 25.571 and DEF STAN, while not identical, were fairly close; but, there was a difference in inspection intervals, attributed largely to the different factors called up by the two approaches. DSTO was responsible for the interpretation, for RAAF, of all the tests results and, translating those results back to ADF usage and proposing a new inspectionbased structural management plan for the aircraft. The analysis approach for this interpretation was a generally conventional test interpretation, see Fig 7, incorporating teardown data, as it became available, and from this, presenting options for inspections and structural replacement to the fleet managers.

Fig. 6 P3 spectra (left) RAAF, Lockheed –Martin wing test and RAAF wing spectra (right) DSTO empennage test baseline and augmented spectra.

The plan also included the determination of more modern elements such as the onset of Widespread Fatigue Damage (WFD) and the Limit of Validity (LOV) of the in-service maintenance program. In this part of the program, RAAF was fortunate to have access to a large quantity of in-service inspection data that had been generated by the USN. With the lead fleet, the USN had begun inspecting their

Fleet Recovery and Life Extension – Some Lessons Learned

11

aircraft at the critical locations identified in the fatigue tests and quickly began finding cracks. The data was of sufficient quantity to enable probabilistic assessments to be made to refine the test based inspection thresholds. In this case RAAF, with its relatively younger fleet, was able to take advantage of the time afforded between the conduct of the SLAP and the thresholds for structural inspections and replacements to both plan an orderly and optimised on-going maintenance program and manage the drawdown of aircraft to meet the introduction into service of the replacement capability.

Fig. 7 Fleet lifing parameters for RAAF P3-C.

The test concluded that [10] “testing the structure to eventual failure provided a significantly extended structural clearance than did the equivalent conventional test that was halted after a pre-defined service life multiple”. In that sense, the economic penalty of extra testing was offset by the value of increased confidence in structural integrity of the life-extended fleet, in part as a result of use of a carefully thought-out engineering approach to the augmented-load test which allowed test life to be contained. This program clearly demonstrated the value of collaboration to allow a wholeof-aircraft SLAP, with major fatigue test distributed worldwide, and extensive data sharing, combined analysis and test interpretation providing exceptional robustness for the management strategy. The results of the full scale testing undertaken by the SLAP participants allowed development of inspection and lifing proposals and individual Aircraft Tracking which would permit RAAF to implement management of the fleet to FAR 25.571.

12

G. Clark

An unplanned Service Life Extension Program - fleet recovery post-accident Until 1990 the structural integrity of the RAAF fleet of jet trainer aircraft – the Aermacchi MB326H – had been founded on a “safe life” approach, with aircraft retirement based on achieving a proportion of the life demonstrated in a full-scale fatigue test conducted by Aermacchi in 1974/75 [11] which demonstrated first failure in the lower spar of the centre section, and a second failure location in the wing spars. In the early 1980s, the RAAF conducted a Life-Of-Type EXtension (LOTEX) program to extend the fatigue life of the fleet to 1992. Various modifications were developed by RAAF and carried out by an Australian contractor. Wing spars were replaced, and the original centre sections were replaced by improved centre sections with an (estimated) improved safe life, although, crucially, no fatigue test was undertaken so RAAF had no knowledge of the second most critical fatigue location in the wings.

Fig. 8 Fracture surface from failed Macchi wing spar, showing manufacturing damage which promoted cracking, and manufaturing damage in other holes.

In the late 1980s the RAAF decided to extend the planned withdrawal date of its MB326H fleet to the year 2000, to align it with the introduction of a new leadin-fighter fleet. At this stage, the fleet had been managed for many years on a safe-life basis with some locations managed on the basis of safety-by-inspection. In-flight failure: In 1990, the wing of an aircraft separated during an air combat manoeuvre, at a service life substantially below the safe life currently in force for the fleet. The fleet was immediately grounded, and recovery of parts from the sea

Fleet Recovery and Life Extension – Some Lessons Learned

13

identified a severe fatigue cracking problem in the lower wing spar cap (Fig 8) and, after substantial fractographic analysis - complicated by the extensive stable tearing fracture - to determine the crack growth rate, inspection methods were developed to inspect fleet wing spars in this region and to permit the resumption of normal service operations. Initial Fleet Life Amendment: However, after examining additional recovered parts, many more build quality issues were identified. The situation was made far more complex when significant service cracking was also identified in undamaged holes in the non-failed wing from the accident aircraft. Of course, in a safe-lifed fleet, where the philosophy relies on not allowing the development of significant cracks, such a situation should not arise, and it was discovered that there had been deficiencies in interpreting the results of the original rudimentary fatigue test to develop the service fleet life. The combination of damaged holes and this lifing error required a re-assessment of the safe life of the aircraft, the lifing methodology and fatigue life management tools. This led to a major effort involving the teardown and assessment of five additional high-life wings using fractographic tools, and projection of the lives of these wings. Pooled results from a number of wings led to application [12] of a reduced, conservative, life limit in 1991 which was approximately one-half that originally promulgated. Fig 9(a) shows the fleet usage accumulated at the time, and the impact of this life reduction led to a dramatic reduction in available aircraft – from 69 to 11. This reduced capability meant that the fleet would not reach its planned withdrawal date. Recovery Actions: The RAAF and DSTO collaborated to find a means of managing the wing spar structural integrity problem, and the first action was to extend the teardown program; soon the database included data on eleven wings, and over 1000 holes of which about 100 had fatigue cracking. The situation was equivalent to a fatigue test in which there is one failure and ten run-outs, and a key issue was how to project the service data to achieve a useful life prediction. The method developed was based on the observations of Goldsmith and others [13-16] who observed that an exponential growth law of the form: a = a0eβN

(1)

where crack length a at service N (cycles or hours) develops from an initial size a0 at a rate influenced by loads, material and geometry related factor β. This provides a remarkably good representation of variable amplitude crack growth in service, as illustrated in Fig 9b, which shows crack depth vs service data for several cracks at four different hole locations. Two cracks shown were measured in detail fractographically, and for the rest, crack growth estimates could be made by joining the measured initial and final defect sizes. In this figure, it is interesting to note that four families of curves can be identified, associated with the four hole locations.

14

G. Clark

(a)

(b) Fig. 9 (a) fleet usage distribution at the time of the accident (b) results of fractographic analysis [12] using log-lin crack growth relationship to develop predicted lives from the observed cracking.

The log-linear formulation is consistent with a model by Frost and Dugdale [17, 18] who developed an earlier representation by Head [19]. This simple relation was used to project forward the observed crack scenarios to allow development of predicted lives for the wings torn down. As a result of the predicted lives, a revised fleet safe-life was developed for the wings, supplemented by an order for new wing sets. Centre Section Cracking and life: Having determined a suitable wing spar recovery approach, attention was then focussed on other potentially critical areas. The steel booms of the centre section had been managed by inspecting specific bolt holes known to have high stress. However because of changes in boom design at LOTEX, several booms were subjected to component fatigue testing for comparative purposes and to determine whether or not the bolt holes could be cold-worked to extend boom life. Here the lack of a suitable full scale test to identify the second-most-critical area was very evident. The tests revealed a complex set of

Fleet Recovery and Life Extension – Some Lessons Learned

15

different failure mechanisms, illustrated in Fig. 10. These modes included manufacturing metallurgical issues, fatigue from corrosion pitting in the bore of the wing attachment lug, crack growth from bolt holes, and also cracking from a complex geometrical region in which small screw holes came close to penetrating the bolt holes. Resolving the issue of criticality order for these different failure modes became paramount, and analysis was based on further development of the loglinear relationship.

Fig. 10 Cracking observed in centre-scetion spar booms (a) lug failure (b) cracking from corrosion pits in lug bore (c) cracking initiating at a forging/intermetallic (d) cracks from flange holes (e) complex crack configurations involving bolt holes and near-intersecting, corroded screw holes.

The exponent parameter beta describes the severity of the “cracking” response of the material and geometry to loading, and this exponent was used to evaluate the criticality of the many holes in the spar, as illustrated in Fig 11, where the beta factor is shown for various locations and configurations along the boom. The results confirmed that bolt holes near the boom ends had a higher beta and were therefore were more prone to cracking. The important result however was that two locations involving screw hole/bolt hole intersection had a high beta value, meaning that there was no prospect of extending the lives of the bolt holes since that would leave these regions vulnerable. As a result of these anlyses, inspections were introduced (magnetic rubber and eddy current) for the flange bolt holes and lugs. The scerew hole corrosion was removed and protection introduced, and the screw hole/bolt hole intersections were monitored using the same approach, although two of these locations were deemed life-limiting.

16

G. Clark

Fig. 11 The severity of cracking at various locations and details along the centre-section booms, indicated by the beta growth exponent from the log-linear growth model. The results confirmed the severity of cracks at the end bolt hole region, but also showed the severity of cracking at screw hole/bolt hole intersections. From [12].

This procedure provided the RAAF with a workable, well-substantiated and acceptable centre section structural integrity management strategy Corrosion Issues: The teardowns, which by now included twelve wings and two fuselages, allowed assessment of the condition of the whole aircraft. Two new corrosion issues were identified: (a) corrosion of the magnesium alloy centre sections spacer blocks; this was not structurally critical, and could be rectified, and (b) the discovery of stress-corrosion cracking in primary structure (spars) in the tailplane. The tailplane SCC illustrated the complexity of dealing with such corrosion damage in structural terms; the SCC often progressed along a cylindrical surface inside the spar as shown in Fig. 12. Detection of this profile of cracking, in unknown locations, presented a major problem, and it was decided that it was prudent to replace the spars with a corrosion-resistant 7xxx alloy. However, the replacement time demanded assessment of whether the spars were safe to fly, and a series of actions was implemented to mitigate risk. These included liberal use of Corrosion Inhibiting Compounds to prevent or limit further development of the cracks, full scale tests on some tailplanes to determine failure loads in bending and torsion. In addition, inspections were introduced insofar as they could be accomplished.

Fleet Recovery and Life Extension – Some Lessons Learned

17

Fig. 12 Stress corrosion cracking observed in tailplane spars.

Overall outcome: The program was ultimately successful; training could be continued, and the capability of the Macchi fleet was preserved until the fleet was replaced by a more modern aircraft. Several major lessons were learned: (i) The fleet service life was based on the results of a full-scale fatigue test which had become increasingly unrepresentative as changes in aircraft configuration were introduced to extend the fleet service life; a full scale fatigue test at LOTEX would have revealed the safe-life problems and probably the build quality issues. (ii) A teardown program to support the final life extension proposed would have revealed the build quality issues. (iii) Environmental degradation changed the critical crack location. (iv) The recovery program ultimately relied on development – and implementation – of new analytical tools, a process which ideally would have been followed in a more structured and robust manner. The life extension/recovery program, while successful, must be seen as fragile – the program moved from one emerging issue to another, for some five years. At any stage, these emerging problems, which were in fact addressed by innovation and effort, had the potential to derail the whole effort, which would have led to enormous outlay to acquire and alternative training capability.

4 A Bigger Threat to Safety – Organisational Issues? Acceptable level of safety First we need to note that the concept of an acceptable level of safety, which is, after all, our goal, is itself complex – what is acceptable? And how useful is the

18

G. Clark

concept? Recently the FAA Transport Airplane Directorate published a Risk Methodology Handbook, [20] which included the illustration shown below in Fig 13, showing the current level of risk associated with various activities; in this commercial aviation provides approximately 10-8 risk of individual fatality per hour. Is this an “acceptable level” of risk? The same figure shows several other risk levels which greatly exceed this, suggesting that political or economic issues, and perhaps the extent to which those interests deem fatalities sufficiently newsworthy to achieve public prominence, may play a large role. It seems likely that the alleged cause of the losses (eg. terrorism, mechanical failure) are parameters which influence the result. For military aviation, risk of aircraft loss is substantially higher – a few decades ago, Australia lost approximately one-third of a fighter fleet over its service life, representing approximately 10-6 losses per hour, although with ejection systems, individual fatality rate is substantially lower than this. The reasons are again complex – the military role is perhaps more complex, more variable, and demanding of equipment and planning.

Fig. 13 Data (illustrative only) for a range of risks. For aviation, data is from NTSB 2002-6. From [20].

Fleet Recovery and Life Extension – Some Lessons Learned

19

To illustrate the variable levels of acceptability, it is interesting to note Fig 14 [21-23] which provides another risk representation, in this case associated with an event (not specifically aviation); para. 136 of [22] states “Where societal concerns arise because of the risk of multiple fatalities occurring in one event from a single major industrial activity, HSE2 proposes that the risk of an accident causing the death of fifty people or more in a single event should be less than one in five thousand per annum”. Using these figures suggests an accident involving total loss of a civil transport aircraft would be seen as “intolerable” only if it occurred 10-3 times per annum, and we would be successful in achieving a “negligible” risk if such an accident occurred 10-6 times per annum – perhaps a worthy goal, but are we anywhere near that? Naturally, such comparisons are made more complex by the different populations of events being assessed, but it is clear that judging public acceptability of risk in aviation is difficult, and we should perhaps avoid assuming that current levels of risk are indeed acceptable – a small number of highly publicised losses could change public perception very dramatically; indeed in the earlier days of aviation, civil aviation safety in Australia was greatly influenced by such losses. One obvious conclusion is that assuming the current level of losses as acceptable is something we should avoid if we are to avoid complacency. There are of course many contributors to the overall level of aviation risk, and ASI is only one of them. In an attempt to put that balance into perspective, the last section of the paper raises the issue of non-structural factors affecting safety of old fleets.

Fig. 14 Frequency / number risk diagram, after [23]. 2

(The UK Health and Safety Executive).

20

G. Clark

System threats to safety of an old fleet More challenging, however, is another issue – however laudable our focus on managing structural risks through life, we cannot rely solely on this process. The results – fleet safety – are threatened by organisational issues such as financial cutbacks which affect maintenance and cloud the chain of responsibility. Older fleets in particular are vulnerable through diversion of resources to other programs, and the possibility of progressive decay of maintenance support systems such as documentation. These issues, particularly when compounded by the uncertainty of progressive deferral of fleet withdrawal dates, and increased operational demand, can, and do, lead to “maintenance debt”, accidents, and loss of life. Two examples of this are described briefly below. What is at issue here is that our structural safety models are in themselves of limited value if the system risks are neglected. Taking the larger view raises the question of prioritisation of effort when safety risks – losses – may be dominated by non-structural issues such as crew or maintenance performance. Example: Sea King accident 2005 In 2005, a RAN Sea King helicopter crashed in a remote location in Indonesia with the loss of nine lives [24]. The Board of Inquiry report [25] noted that the primary cause of the accident was flight control system failure caused by separation of components; the separation was “the result of a series of errors and noncompliances with Maintenance Regulations which ultimately led to…deficient fitment of the split-pin and nut that secured the pivot bolt”. Importantly, the BoI stated that the accident was “not an isolated random event caused by the actions of a few maintenance personnel. Rather it was the result of a complex interaction of individual and systemic failing”. Limited staff resources, personnel seeing little future opportunity working on that type, the favouring of newer aircraft types, declining spares resourcing, declining documentation control, and a pervasive culture of maintenance work-around and non-compliance. One key observation was that the performance of maintenance on this aircraft had already been observed to be deficient – many breakdowns in performance of maintenance had already been observed, but recovery of maintenance performance to a more appropriate level did not occur, in part because of declining interest in the older aircraft, reluctance to invest in facilities, staff and training, and the gradual predominance of a culture of ‘work-arounds” to cope with limited resourcing (in terms of staff, facilities, tools, spares, documentation). In this case, the environment – an old aircraft approaching the end of its service life – clearly contributed to organisational issues in an already under-resourced maintenance environment, and was a major factor leading to loss of life. The issues are detailed in ref [25] and have also been summarised using an alternative “AcciMap” approach [26].

Fleet Recovery and Life Extension – Some Lessons Learned

21

Example: Nimrod accident There are many similarities between the conclusions of the RAN Sea King BoI and those of a review [27] of the safety issues underlying the crash of a RAF Nimrod in 2006. RAF Nimrod XV230 was lost on 2 September 2006 on a mission over Afghanistan; a catastrophic mid-air fire led to the total loss of the aircraft and the death of all 14 service personnel on board. While a ‘Safety Case’, to identify, assess, and mitigate potentially catastrophic hazards before they could cause an accident, had been mandated for military aircraft and other military platforms by regulations introduced in September 2002, the Review concluded “the Nimrod Safety Case was a lamentable job from start to finish. It was riddled with errors. It missed the key dangers. Its production is a story of incompetence, complacency, and cynicism”. More significantly for the present discussion the Review [25] stated, “The Nimrod Safety Case process was fatally undermined by a general malaise: a widespread assumption by those involved that the Nimrod was ‘safe anyway’ (because it had successfully flown for 30 years) and the task of drawing up the Safety Case became essentially a paperwork and ‘tickbox’ exercise”. In this instance, as in the RAN Sea King earlier warning signs of fuel system problems had been ignored by some parts of the safety system. A Nimrod report in 1998 [28] had warned of “the conflict between ever-reducing resources and ... increasing demands; whether they be operational, financial, legislative, or merely those symptomatic of keeping an old aircraft flying. The pressures…ensue from reducing resources place additional burdens on a ‘can do’ organisation such as the Nimrod Force and call for highly attentive management, closely attuned to the incipient threat to safe standards, if airworthiness is to be safeguarded.”. The Haddon-Cave [27] review concurred: “The Nimrod fleet of aircraft was going to require more (not less) care, resources and vigilance and a strengthening (not weakening) of the airworthiness regime and culture if these ‘legacy’ aircraft were going to continue to operate safely until their extended Out-of-Service date.” One of the issues discussed earlier as likely to lead to increased risk was repeated changes (ie. deferral) of Planned Withdrawal date. The Review [27] noted in regard to this issue: “the MR2 Out-of-Service Date was continually being put back led to planning, spares, sourcing and long-term investment problems”. In this instance the continual extension of life to match the introduction of new equipment was a clear factor contributing to the fleet problem, and this was compounded by a “muddy” system of airworthiness/safety responsibility. A similarity exists to the state of the RAAF Boeing 707 fleet in the early 2000’s; repeated deferral of the PWD for the fleet meant that the management system for the fleet – particularly any elements of proactive maintenance – was degrading, to the extent that with each PWD deferral, it became harder and harder to restore a functional maintenance program. The fleet was effectively being managed on an individual aircraft basis, day to day, records were becoming progressively less complete, and the retention of a fleet capability was threatened continually by each inspection.

22

G. Clark

In summary, the tools and procedures we have developed for managing structural integrity are highly effective, but the question remains…as engineers and scientists focussed on safety management, do we do enough to engage with other safety-related issues (such as the maintenance of old fleets) which in the broader picture, may have far more impact on overall fleet safety?

5 Conclusions Overall, fleet life extension has the potential to be extremely cost-effective, by delaying costly fleet replacement. However, several issues demand attention: •

•

•

• • • •

When we consider the nature of the damage or features which can cause failure, we need to include the class of initiators (including damage) which are not widely represented in the fleet, and which are not necessarily present in “representative” test articles. Using the descriptor “rogue flaw” misses the point that such initiators are failures of process, involving complacency, or omission in design, maintenance, and overall systems management. The presence of this population of discrete, individually unique, damage suggests that fleet life extension will be accompanied by an increased risk of accident/failure, since our models do not address “unknown unknowns” and these risks inevitably increase with increasing service time. There are methods to mitigate these risks; they involve proactive investigative programs such as teardown, coupled with more aggressive programs such as damage enhancement testing, and proactive management of deterioration such as corrosion. Even though such on-going investment in ASIP is of course something that should be part of a “normal” fleet management approach, adding these costly programs will require significant reinvestment in an old fleet. They should be made an integral part of the LEx proposal, and that may require strong argument on safety grounds. One example was provided of a successful program in which damage enhancement testing identified, safely, some serious issues with an old fleet in Australian service. Innovative risk mitigation programs in a LEx program can provide valuable information to guide future design and manufacturing. Life extension programs will naturally function most effectively in a planned environment, and can include significant research and development which will provide improved capability in Structural Integrity. An example of a successful life extension program showed that it can provide an opportunity to consider recent improvements to ASIP and ongoing airworthiness processes by civil and military authorities; these include limits of validity of maintenance programs (due to limits of underpinning data), initiatives to improve assessments by using probabilistic techniques (eg. by developing distribution functions for crack growth to support further analysis).

Fleet Recovery and Life Extension – Some Lessons Learned

•

• •

23

Unfortunately, life extension often occurs in an unplanned manner, as a result of an accident, or other failure. Such extensions can be successful, and can be extremely cost-effective (avoiding the cost of an “emergency” acquisition of alternative capability) but an example provided shows the difficulties which are encountered in such “reactive” recovery-based programs, and highlights the fragility of the outcomes. The concept of an “acceptable” level of risk is complex, because of a strong tendency to equate it to “current” level of risk. Finally, the paper has raised the issue of just where our major safety risks lie, noting that there is evidence that progressive deferral of retirement dates for old fleets can lead to declining maintenance effort and effectiveness. The impact of such issues - some recent accidents involving such issues have caused heavy casualties - may be to introduce risks very much greater than those posed by structural integrity matters, and this perhaps places a demand on ASI specialists to contribute to the broader issue of fleet risk.

Acknowledgements While the opinions presented in this paper are the author’s own, the examples used to illustrate the paper involved outstanding work by many people, principally in the Defence Science and Technology Organisation (DSTO) of the Australian Department of Defence, in which the author was Research Leader Structural Integrity, and in the Australian Defence Force’s Directorate-General of Technical Airworthiness (DGTA-ADF).

References [1] Johnston, C.: Commander, Naval Air Systems Command, JACG Principals Panel. In: 7th Joint DoD/FAA/NASA Conference on Aging Aircraft, New Orleans, Louisiana, USA (September 2003) [2] Clark, G., Jackson, P.: Structural integrity and damage type in military aircraft. Fatigue and Fracture of Engineering Materials and Structures 33(11), 752–764 (2010) [3] Heller, M., Burchill, M., Wescott, R., Waldman, W., Kaye, R., Evans, R., McDonald, M.: Airframe Life Extension by Optimised Shape Reworking –Overview of DSTO Developments. In: Bos, M. (ed.) Proceedings of the 25th ICAF Symposium Bridging the Gap between Theory and Operational Practice, pp. 279–299. Springer, Heidelberg (2009) [4] Iyyer, N.P., Phan, N.: Durability Issues and Management of Aging P-3C Aircraft. In:11th International Conference on Fracture, Turin, Italy (2005) [5] Teunisse, B., Mongru, D., Jackson, P., Matricciani, E., Hartley, D.: P-3C service life assessment program - Australian test interpretation report for the USN wing/fuselage/landing gear test articles. DSTO Report, Melbourne, Australia (2006)

24

G. Clark

[6] Hartley, D., Jackson, P., Matricciani, E., Teunisse, B., Phillips, M.: P-3C service life assessment program Australian test interpretation report for the empennage test articles. DSTO Report, Melbourne, Australia (2005) [7] Iyyer, N., Sarkar, S., Merrill, R., Phan, N.:Managing aging aircraft using risk assessment models-lessons learned from P3-C fleet. In: 24th ICAF Symposium, Naples, Italy, May 16-18 (2007) [8] Jackson, P., Cardrick, A.W.: The Challenge of Conducting a Meaningful Fatigue Test on a Transport Aircraft Empennage. In: Proc. 22nd Symposium of the International Committee on Aeronautical Fatigue, Lucerne (May 2003) [9] Ministry of Defence. Design and Airworthiness Requirements for Service Aircraft. Defence Standard 00-970 (2), Part 1, Section 3, Leaflet 35 (December 1999) [10] Jackson, P., Mongru, D., Hartley, D.: Durability and Damage Tolerance Substantiation of a Transport Aircraft Metal Tailplane Structure. In: Australia 24th ICAF Symposium Napoli, Italy, May 16-18 (2007) [11] Final report of Fatigue Test on Wing and Centre Section Structures MB- 326H Aircraft. Aermacchi report No. 1945 (September 1975) [12] Clark, G., Jost, G.S., Young, G.D.: Recovery of the RAAF MB326H Fleet; the Tale of an Aging Trainer Fleet. In: Poole, P., Cook, R. (eds.) Proceedings of the 19th ICAF Symposium Fatigue in New and Ageing Aircraft, pp. 39–58. EMAS, Warley (1997) [13] Clark, G., Barter, S.A., Goldsmith, N.T.: Influence of initial defect conditions on structural fatigue in RAAF aircraft. In: Blom, A. (ed.) Durability and Structural Reliability of Airframes, vol. I, pp. 281–304. EMAS, Warley (1993) [14] Goldsmith, N.T., Clark, G.: Analysis and interpretation of aircraft component defects using quantitative fractography. In: Bernard, S.M., Susil, P.K. (eds.) Quantitative methods in fractography, STP, vol. 1085, pp. 52–68. American Society for Testing and Materials, Philadelphia (1990) [15] Goldsmith, N.T.: Fractographic examinations relevant to the F+W Mirage fatigue test. Dept. Defence Aeronuatical Research Laboratory Materials Tech. Memo 371 (August 1978) [16] Anderson, B.E., Goldsmith, N.T.: Prediction of crack propagation in Mirage wing fatigue test spar. Aeronuatical Research Laboratory Structures Note 448 (1978) [17] Frost, N.E., Marsh, K.J., Pook, L.P.: Metal fatigue. Clarendon Press, Oxford (1974) [18] Frost, N.E., Dugdale, D.S.: The propagation of fatigue cracks in test specimens. J. Mech. Phys. Solids 6, 92–110 (1958) [19] Head, A.: The growth of fatigue cracks. Phil. Mag. 44(7), 925–938 (1953) [20] FAA Transport Airplane Risk Assessment Methodology Handbook, Federal Aviation Administration, Transport Airplane Directorate, document ANM-100 draft 22-12-10 [21] Health and Safety Executive (HSE), UK Government, Report: The Tolerability of Risk from Nuclear Power Stations (1992) [22] Health and Safety Executive (HSE), UK Government, Report: Reducing Risks, Protecting People (“R2P2”) (2001) [23] Knott, J.F.: The integrity and durability of structures and machines. In: Proc. 9th International Conference on Engineering Structural Integrity Assessment, October 15 - 19, vol. 51180, pp. 1–21. ESIA Publication, Beijing (2007) [24] Athiniotis, N.A., Lombardo, D., Clark, G.: Simulation, assessment and technical conclusions from a major accident investigation. Engineering Failure Analysis 17, 353–360 (2010) [25] Royal Australian Navy. Nias Island Sea King Accident Board of Inquiry Report, released June 21 (2007)

Fleet Recovery and Life Extension – Some Lessons Learned

25

[26] Debrincat, J., Bil, C., Clark, G.: Assessing organisational factors in aircraft accidents: methodologies and limitations. In: Proc. 27th Congress of the International Council of the Aeronautical Sciences, Nice, France, September 19- 24 (2010) [27] Haddon-Cave, C.: An independent review into the broader issues surrounding the loss of the RAF Nimrod MR2 Aircraft XV230 in Afghanistan in 2006, HC, vol. 1025. HM Stationery Office, London (2009) [28] Nimrod Airworthiness Review Team report 1998; quoted in [27] p. 359

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Sticks and Stones

(Could the Words of Aeronautical Fatigue Hurt Us?) Steve Swift Steve Swift Pty Ltd, Canberra, Australia

Abstract. This paper appeals for care with the language of aeronautical fatigue. Communication is as important as calculation. Bad words undo good engineering. It gives examples (including ten troublesome words), suggests causes (including pride and insecurity), and offers help (including plain English and Simplified Technical English).

1 Introduction ‘Sticks and stones will break my bones but words will never hurt me!’ is an old saying I often heard at school. But we know from everyday life that this is not true—words can hurt. What about the words of aeronautical fatigue? Could they hurt us too? 2 The Hurt Words have hurt us in the past. Here are just three examples: Tenerife 1977 After two Boeing 747s crashed into each other, killing 583, the Dutch report said ‘misunderstanding has arisen from normal but ambiguous terminology’ [1]. The Spanish report recommended the ‘use of standard, concise and unequivocal aeronautical language’ [2]. The crash led to a major international effort to improve communication between pilots, and between pilots and air traffic controllers. University of New South Wales 1976 This was a personal experience. While studying aeronautical engineering, for my thesis I designed a wing with another student who stressed it. By telephone (this was before fax and email) he told me the length I needed for a part of the spar. He told me the semi-span. I thought he meant full span. So my part was too short. Thankfully the wing broke on test, not in flight, so it only hurt my pride. Ansett 2000 This was the incident that started my thinking for this paper. In 1997, a revision to Boeing’s Maintenance Planning Data for the 767 said at the front: *

Oral presentation.

28

S. Swift Revised Airworthiness Limitation section to reflect the reassessment of the ‘50-Series’ Supplemental Structural Inspection Program.

Ansett, then Australia’s second-largest airline, thought ‘50-series’ meant that these fatigue inspections did not start until 50,000 flights, the 767’s design service goal. So they put the revision aside. But ‘50-series’ was Boeing jargon for something else. The inspections should have started at 25,000 flights, a point which some of Ansett’s 767s had already passed. Ansett remained unaware for three years, until Christmas 2000. The immediate groundings for inspections and repairs stranded thousands of passengers. Ansett did not recover and closed in 2001. See [3].

3 The Words Here are ten words that are troublesome and could hurt us: • • • • • •

initiation failure on-condition condition monitoring safe life fail-safe

• • • • • •

damage tolerance ageing aircraft structural integrity detectable supplemental inspection documents service history based inspections

Initiation ASTM E1823–10 defines crack initiation as ‘the onset of crack propagation from a pre-existing macroscopic crack’—leaving us to then define onset [4]. For AFGROW, ‘initiation is defined by the strain-life data used in a given prediction’. For its ‘sample data’, it is ‘a 2.5 mm (0.01 inch) crack’ [5], a common size ‘engineers have arbitrarily set’ [6]. Unfortunately, engineers do not always say which size they set, or even the ‘level of observation’ [7]. It could be a visual check, load drop-off, or separation. While initiation works generally (‘notches initiate cracks’), spatially (‘the crack initiated here’) and metallurgically (‘initiation involves slip bands’), it is not definitive chronologically (‘the crack initiated then’), because it is not the ‘singular event’ that many think it is [8]. It is a multi-stage process. Better reflections of that are Schijve’s ‘initiation period’ [9] and Hoeppner’s nucleation [10]. Fortunately, civil rules do not use initiation to design aircraft or their inspections. Despite that, I have seen attempts to use initiation instead of critical for setting the inspection threshold. This error is unconservative if: •

critical precedes initiation (for strong brittle alloys as Hoeppner warns [11])

•

critical follows initiation and the compressive zone (for cold-expansion [12]).

Sticks and Stones

29

Failure Failure is hard to define because, like initiation, it is usually a multi-stage process rather than a singular event. Where in the process do we draw the line—is it initiation, ultimate strength, limit strength or separation? Safe lives for the wing of the Embraer Bandeirante once differed widely because some authorities thought its fatigue test failed at ultimate strength, and others at separation—much later. Bandeirante wings flew more hours less safely in some countries. It is even more serious if international differences cause operators to distrust all safe lives. MSG-3 has a good definition of failure: ‘the inability of an item to perform within previously specified limits’ [13]. Let’s remember to specify those limits. On-condition MSG-2 defined on-condition as ‘repetitive inspections, or tests to determine the condition of...portions of structure’ [14]. However, many thought the word implied doing nothing, and hoping that damage will be obvious before it is dangerous. Another problem was the word sounding like condition monitoring (the next word in this list). So, it was good when MSG-3 replaced on-condition by its maintenance tasks to ‘eliminate the confusion associated with the various interpretations of Condition Monitoring (CM), On-Condition (OC), Hardtime (HT) and the difficulties encountered when attempting to determine what maintenance was being accomplished on an item that carried one of the process labels’ [13]. Condition monitoring MSG-2 defined condition monitoring: For items that have neither hard time limits nor on-condition maintenance as their primary maintenance process. Condition monitoring is accomplished by appropriate means available to an operator for finding and resolving problem areas. These means range from notices of unusual problems to special analysis of unit performance. No specific monitoring system is implied for any given unit.

It could mean doing nothing, as many thought on-condition meant. It was just as confusing. So, MSG-3 replaced it by its maintenance tasks too. Safe life Safe life has many meanings. Some say it is a design property—a single load path. But we know there are single load paths that are damage tolerant and there are multiple load paths that have safe lives. Some say safe-life is a method of analysis—S-N curves and Miner’s rule. But we know there are safe lives that are set by fracture mechanics and tests. Some say safe life is an assumption—of perfect material and manufacturing quality. But we know that the well-used S-N curves in FAA AC 23-13A [15]

30

S. Swift

came from fatigue tests of military fighters built in the Second World War—was their quality perfect? FAR 25.571(c) defines safe life as ‘service life without detectable cracks’. It is vague because it does not specify the inspection method—and perverse because it discourages a good one. MIL-STD-1530C (3.29) is better: ‘that number of events such as flights, landings, or flight hours, during which there is a low probability that the strength will degrade below its design ultimate value due to fatigue cracking’. (AC 91-82 [16] copies MIL-STD-1530C but is not a rule.) Unlike on-condition and condition monitoring, MSG-3 kept safe life. But the emphasis is the associated maintenance task—discard—‘the removal from service of an item at a specified life limit’. It is clear what to do. (I thought it was clear what to do with a safe life until an aircraft manufacturer told me they fatigue-tested a fuselage to find its safe life, but then did not publish it. They did not see a need for a maintenance task. So operators exceed it, unaware of the risk.) Some prefer safety by retirement (SBR). But discard is simpler, shorter and, being in an international standard, more established. Fail-safe Fail-safe was a worthy goal—a structure can fail and stay safe; damage will be obvious before it is dangerous; there is no need for a maintenance task. But even though the practice rarely met the promise, the word fail-safe still fostered complacency. In the 1970s, it hindered authorities wanting to mandate maintenance for fail-safe jets, such as the Boeing 707. It took until 1978, when the UK CAA’s Airworthiness Note 89 required ‘structural audits’ [17]. In 1981, the FAA followed with AC 91-56 [18]. The results were the now familiar Supplemental Inspection Documents (SIDs)—more on those soon. The complacency still hinders us. In 2007, Cessna issued a good SID for the 441 Conquest turboprop. In 2011, some operators still question the need for the SID’s inspections and limit of validity because of overconfidence in the 441’s failsafety. Fail-safe is no longer in FAR 25.571. Some lament losing the requirement for structure to survive large damage. Perhaps MSG-3’s emphasis on the maintenance task could help us again. Could we restore fail-safety to damage tolerance by restricting the inspection task to operating crew normal duties? Damage tolerance At ICAF 2005, Eastin and I argued that damage tolerance is not a design property or fracture mechanics, but ‘a method...for assuring safe inspection intervals’ [19]. Then, at ICAF 2007, Gallagher seemed to contradict us [20]. He said: ‘contrary to popular belief, the damage tolerance design approach was not created to support the development of an inspection program’. He said it was a design objective. Both are correct in context. Eastin and I spoke for a civil rule, FAR 25.571(a)(3):

Sticks and Stones

31

Based on the (damage tolerance) evaluations required by this section, inspections or other procedures must be established…

Gallagher spoke for a military rule, MIL-STD-1530C (3.8): Damage tolerance is the attribute of a structure that permits it to retain its required residual strength for a period of unrepaired usage after the structure has sustained specific levels of fatigue, corrosion, accidental, and/or discrete source damage.

So, civil and military damage tolerance rules differ, as Eastin explains more fully in [21]. He concludes that ‘the use of the same words for different things can lead to confusion and needless debate. This has been the case with the words damage tolerance and fail-safe’. I have seen ex-military engineers get so confused evaluating damage tolerance for civil aircraft that they ignore inspectability, so damage could be dangerous before it is detectable. Like fail-safe, the words foster complacency, so some think they do not need to inspect or repair damage-tolerant aircraft. And again, damage tolerance is not a maintenance task. Some prefer safety by inspection (SBI). But again, simpler, shorter and more established is MSG-3’s task: inspection. Ageing aircraft When is an aircraft an ageing aircraft? The FAA’s Aging Airplane Safety Rule says 15 years [22]. Australia’s Civil Aviation Safety Authority (CASA) says it is ‘the time they leave the production line’ [23]. Does definition matter? Not if we don’t need the words. ‘Ageing aircraft’ is just a new name for an old activity—continuing airworthiness. It is a fad. Unfortunately, some think it really is new. For example, a recent recommendation that an authority should have an Ageing Aircraft section overlooks that ageing aircraft are already core business for its Continuing Airworthiness Section. We would be wise to avoid such wasteful duplication and risky division. The fad is passing. Ageing Aircraft conferences in the US and Australia are now Aircraft Airworthiness and Sustainment conferences—a good start. Structural Integrity ICAF 2011’s theme is ‘Structural Integrity: Influence of Efficiency and Green Imperatives’. But what do we mean by structural integrity? By definition and common usage, rarely is it wholeness and purity. Mostly, practically, all we mean is strength. If so, do we really need more jargon that is neither clearer nor shorter? If we also mean stiffness—if rules like FAR 25.629 (d) require us to—we can say so. But, if not (and we rarely do), let it be clear. Aircraft manufacturers and airworthiness authorities like structural integrity because it is a euphemism. Service bulletins and airworthiness directives warn more of ‘reduced structural integrity’ than of ‘failure’. The first causes less alarm, but the second causes more action.

32

S. Swift

Detectable ‘Damage tolerance evaluations’ to civil rules (like FAR 25.571) must assess ‘detectability’. But at ICAF 2007, Gallagher preferred ‘missability’. Asking about ‘missability’ instead of ‘detectability’ helps the optimistic among the inspectors to see the ‘glass half-empty’ instead of ‘half-full’, to answer for a probability of detection (POD) of 90% instead of 10%. Safety must consider psychology. SlD Airworthiness Limitations (in the ALS) and Supplemental Inspections (in the SID) are the same inspections for the same safety. The only difference is their time of creation relative to type certification: the ALS before, the SID after. Yet operators obey the ALS more than the SID. It is the different words as well as different laws—many think ‘supplemental’ means ‘optional’. We are getting less safety than we had hoped from SIDs like those for Fairchild Metros [24] and Cessna piston twins [25]. Good lessons from ICAF are underused. Service history based inspections In AC 91-82 [16], the FAA’s service history based inspections differ from damage tolerance based inspections by not needing even a crude check of three basics of damage tolerance that Eastin and I argued are essential to assure safety [19]: • • •

detectable (or missable) dangerous (or critical) duration (or interval).

Service history, if available, can help us estimate them—but it cannot replace them. The dichotomy is false.

4 The Reasons We might use bad words—or misuse good words—because we are: • • •

unaware—we do not see a problem unprepared—we do not see a solution uncaring—we do not want a solution.

The first one is obvious. The second I discuss in the next section. But the third— why would anyone not want their writing to be shorter, simpler and clearer? Is it schooling—did your teacher specify the minimum length of your essays? If so, did you learn verbosity? Is it engineering—do we like our words complex like our numbers? Is it pride—is complexity more likely to impress? Is it insecurity—would you still be the expert if others understood you?

Sticks and Stones

33

Is it laziness? Paradoxically, simple writing is harder. Eagleson says ‘it involves thinking—but if your thinking is muddled, your writing will be muddled too’ [26]. I do not judge—I am guilty of all.

5 The Help The solution is ergonomics, the ‘science of work’, according to the International Ergonomics Association (IEA) [27]. Ergonomics Ergonomics is more than office furniture and cockpit controls. The IEA defines three branches: • • •

physical ergonomics—‘human anatomical, anthropometric, physiological and biomechanical characteristics as they relate to physical activity’ (includes office furniture and cockpit controls) cognitive ergonomics—‘mental processes, such as perception, memory, reasoning, and motor response, as they affect interactions among humans and other elements of a system’ (includes communication) organisational ergonomics—‘optimization of sociotechnical systems, including their organisational structures, policies, and processes’.

In cognitive ergonomics, ICAO’s SHELL model has two interfaces [28]: • •

liveware-software interface—written communication liveware-liveware interface—spoken communication.

For the first, ICAO warns that ‘delays and errors may occur while seeking vital information from confusing, misleading or excessively cluttered documentation and charts’. We once liked complexity in the cockpit (the more, the better!), but then we matured. We learned that increasing complexity increases error. We now strive for simplicity and clarity. Likewise, we should strive for simpler, clearer communication for aeronautical fatigue. To this end, three aids are Plain English, Simplified Technical English and Safety Management Systems. Plain English Eagleson describes Plain English as: The opposite of gobbledegook and of confusing and incomprehensible language. Plain English is clear, straightforward expression, using only as many words as are necessary. It is language that avoids obscurity, inflated vocabulary and convoluted sentence construction. It is not baby talk, nor is it a simplified version of the English language. Writers of plain English let their audience concentrate on the message instead of being distracted by complicated language. They make sure that their audience understands the message easily. [26]

34

S. Swift

Gowers adds [29]: ‘When experts are writing only for their fellow experts they may achieve their aim of conveying to their readers exactly what they intend to convey by the use of language which the rest of us find obscure or even quite unintelligible.’ ... ‘This is not to say such writing always is clear, even to the intended readers. Its obscurities are sometimes due, not to the requirements of the subject matter, but to muddled thinking or mere pretentiousness; and for obscurities of this kind, as for any other misuses of language, the writers are not to be forgiven merely because they are writing for fellow experts.’

Both business and government support plain English. Citibank adopted it in 1973—a move followed by many companies. The USA passed its first law in 1975 and its latest, the Plain Writing Act, in 2010. The Canadian Legislative Drafting Provisions adopted plain English in 1976. One lawyer, Asprey, in Plain Language for Lawyers, argues it is good for justice and for business [30]. But many lawyers resist for reasons similar to engineers. In aviation, if plain English improves clarity, it improves safety. Writing Plain English How do you write plain English? Three helpful books are: • • •

Writing in Plain English by Eagleson (Australia) [26] The Complete Plain Words by Gowers (England) [29] The Elements of Style by Strunk (USA) [31].

More recent and for aviation is: Writing User-Friendly Documents, A Handbook for FAA Drafters [32]. Two of its applications to fatigue are: • •

AC 23-13A, Fatigue, Fail-safe, and Damage Tolerance Evaluation of Metallic Structure for Normal, Utility, Acrobatic, and Commuter Category Airplanes [15] FAR 39, Airworthiness Directives.

Simplified Technical English Simplified Technical English (STE) is a controlled language for technical documentation. It started for European aviation but other countries and industries now use it. Standard ASD-STE-100 has ‘writing rules’ and ‘approved words’ [33]. Sometimes there is tension between STE and plain English. For example, where plain English would suggest ‘inspect’ as a verb, STE would want ‘do an inspection’ because it approves the noun but not the verb. But the irritations are only minor in a standard worth ICAF’s attention. Format Plain English and Simplified Technical English also concern format. Sadly, we arrange many of our fatigue and corrosion maintenance instructions like an old

Sticks and Stones

35

cluttered cockpit. Gallagher says the ‘root causes’ for ‘inspection misses’ include ‘confusing...documents and instructions’ [20]. We keep adding more clutter. It is easy to miss or misread an instruction—just as Ansett did. Safety Management Systems Since 2009, ICAO has required ‘that an operator implement a safety management system...that...identifies safety hazards...and aims to make continuous improvement’ [34]. For a lot longer—since 1951—ICAF has been doing exactly that for technical safety hazards. A modern safety management system should remind us and help us to watch our words as well, as we ‘exchange information concerning aeronautical fatigue’ (ICAF’s aim).

6 Conclusion The sentiment of ‘sticks and stones’ is wrong. The words of aeronautical fatigue could hurt us. Communication is as important as calculation. Let us not allow bad words undo our good engineering.

7 Recommendation Here are three: • • •

Be aware—be a critic, especially of your own words Prepare—be a wordsmith, even if you find it hard, like I do Care—be the professional you are—in all things.

8 Quote We spent over fifty years on the hardware, which is now pretty reliable. Now it’s time to work with people. —Don Engen, FAA Administrator, 1986

References [1] Final Report and Comments of the Netherlands Aviation Safety Board of the Investigation into the Accident with the Collision of KLM Flight 4805, Boeing 747-206B, PH-BUF and Pan American Flight 1736, Boeing 747-121, N736PA at Tenerife Airport, Spain on March 27 (1977) [2] Collision of KLM Boeing 747 PH-BUF and Pan Am Boeing 747 N737PA at Los Rodeos (Tenerife) on Recommendation 3.2, p. 60, Ministry of Transport and Communication, Spain (March 27, 1977)

36

S. Swift

[3] ATSB, Investigation into Ansett Australia maintenance safety deficiencies and the control of continuing airworthiness of Class A aircraft, Australian Transport Safety Bureau, Canberra, Australia (2002) [4] ASTM, E1823–10, Standard Terminology Relating to Fatigue and Fracture Testing, ASTM International, West Conshohocken, PA, USA (2010) [5] Harter, J.: In a private email. January 20 . LexTech Inc., USA (2011) [6] Grosskreutz, J.: The Mechanisms of Metal Fatigue (II) (1971) [7] Hoeppner, D.: Corrosion Fatigue Considerations in Material Selections and Engineering Design. In: Corrosion Fatigue. NACE-2, pp. 3–11. NACE, Houston (1972) [8] Hoeppner, D.: Future Technology Requirements (Related to Defects and Quantitative Material Behavior to Aid in Implementation of Damage Tolerance for Design of Engine Structures). In: AGARD meeting, Mierlo, The Netherlands (1988) [9] Schijve, J.: Fatigue of Structures and Materials. Springer, The Netherlands (2009) [10] Hoeppner, D.: From Safe Life to Holistic Structural Integrity Design. In: NRC IAR Structures Workshop, Ottawa, Canada (2002) [11] Hoeppner, D.: Parameters that Input to Application of Damage Tolerance Concepts to Critical Engine Components. In: Conference Proceedings AGARD-CP 393, Damage Tolerance Concepts for Critical Engine Components, pp. 1–4. NATO-AGARD, France (1985) [12] Reese, et al.: Bridging the Gap between Theory and Operational Practice. In: ICAF 2009, p. 1275. Springer, Heidelberg (2009) [13] ATA, MSG-3, Operator/Manufacturer Scheduled Maintenance Development. Air Transport Association of America, Washington DC, USA (2007) [14] FAA, AC 120-17A, Maintenance Control by Reliability Methods, Appendix 1, USA (1978) [15] FAA, AC 23-13A, Fatigue, Fail-safe, and Damage Tolerance Evaluation of Metallic Structure for Normal, Utility, Acrobatic, and Commuter Category Airplanes, USA (2005) [16] FAA, AC 91-82, Fatigue Management Programs for Airplanes with Demonstrated Risk of Catastrophic Failure Due to Fatigue, USA (2008) [17] CAA, Airworthiness Notice No. 89, Continuing Structural Integrity of Transport Aeroplanes, Issue 1, Civil Aviation Authority, Redhill, England (1978) [18] FAA, AC 91-56, Supplemental Structural Inspection Program for Large Transport Category Airplanes, USA (1981) [19] Eastin, R., Swift, S.: Rough Diamond. In: Proceedings of the 23rd ICAF Symposium, Hamburg, Germany, p. 43 (2005) [20] Gallagher, J.: A Review of Philosophies, Processes, Methods and Approaches that Protect In-Service Aircraft from the Scourge of Fatigue Failures. In: Proceedings of the 24th ICAF Symposium, Naples, Italy, p. 1 (2007) [21] Eastin, R.: Contrasting FAA and USAF Damage Tolerance Requirements.In: ASIP, Memphis, Tennessee, USA (2005) [22] FAA, Aging Airplane Safety; Final Rule, February 2. Federal Register, USA (2005) [23] ACTS, Presentation to the Ageing Aircraft Advisory Group #2 Meeting, Canberra, Aviation Concepts and Training Services Pty Ltd, Australia (2010) [24] FAA, DOT/FAA/AR-00/18, Development of Supplemental Inspection Report for the Fairchild Metro SA226 and SA227 Airplane, USA (2000) [25] FAA, DOT/FAA/AR-98/66, Supplemental Inspection Document Development Program for the Cessna Model 402, USA (1999)

Sticks and Stones

37

[26] Eagleson, R.: Writing in Plain English. Australian Government Publishing Service, Canberra (1990) [27] IEA International Ergonomics Association (2011), http://www.iea.cc [28] ICAO, Human Factors Training Manual, Doc 9683, 1edn., Montreal, Canada (1998) [29] Gowers, E.: The Complete Plain Words. Penguin Books, Harmondsworth (1987) [30] Asprey, M.: Plain Language for Lawyers. Federation Press, Leichhardt (2003) [31] Strunk, W.: The Elements of Style, New York, USA (1918) [32] FAA, Writing User-Friendly Documents, A Handbook for FAA Drafters, prepared for the FAA by the Plain English Network, USA (2000) [33] ASD, ASD-STE-100, Simplified Technical English, Aerospace and Defence Industries Association of Europe, Brussels, Belgium (2010) [34] ICAO, Annex 6, Operation of Aircraft, section 3.3.4, Montreal, Canada (2001)

26th ICAF Symposium – Montreal, 1-3 June 2011 Analysis of Requirements on Fatigue and Damage Tolerance for Civil Transport Airplanes B.G. Nesterenko and G.I. Nesterenko Central Aerohydrodynamic Institute, Russia

Abstract. The basic step to ensure safe operation of aircraft structure is to comply with regulatory requirements to structural fatigue and damage tolerance. This paper gives the analysis of the requirements stated in the FAA Regulations and Advisory Circulars (USA), in the former USSR Aircraft Airworthiness Regulations and Russian Aviation Rules. Principal requirements are given on fail safety, damage tolerance and prevention of widespread fatigue damages in operated aircrafts. Some results on multiple site fatigue damages of full-scale aircraft structure study are shown. The concept of step-by-step prolongation of Russian aircraft service lives is considered. Emphasized is the harmonization of Russian rules with USA and European regulations. The values are given of design goals, fatigue test results of full-scale structures and in-service number of flights for different airplane types. The required values of static and cyclic crack resistance of structural materials are substantiated and compared with current values of these characteristics of contemporary aluminum alloys.

1 Introduction After well-known accidents with two British jet passenger airplanes “Comet” in 1954 caused by fatigue damages in pressurized fuselages, safe life and fail-safe concepts were developed to prevent aircraft structural failure due to fatigue cracks. The “safe life” is based on the statement that airplane safety is defined by the absence of fatigue cracks in the structure during its service life. The “fail-safe” concept considers a structure that would be designed capable to sustain significant regulated load after partial or complete fracture of its one primary element. Great Britain had accepted the “safe life” concept. The USA aviation experts had admitted the “fail-safe” concept providing structural safety from catastrophic failure due to fatigue cracks or other damages. The earlier USSR Requirements had been influenced by the British ones. Hence the first USSR Airworthiness Requirements for Civil Aircraft contained the only concept of aircraft safety in longterm operation, i.e. “safe life”. Russian Requirements had been seriously changed after the 1972 accident with passenger turboprop An-10A airplane due to multiple site fatigue cracks in the wing lower surface. And the damage tolerance concept has been introduced into Airworthiness Requirements as equivalent to safe life concept.

40

B.G. Nesterenko and G.I. Nesterenko

Based on the results of comprehensive test-analytical studies performed in 1970s by TsAGI in collaboration with Antonov, Tupolev, Ilyushin aircraft design companies the Russian philosophy on damage tolerance in aircraft structures had been developed [1]. This philosophy contained a number of principal differences from the damage tolerance requirements of the USA Regulations FAR25.571 of the 1970s [2]. The analysis of Russian and US Regulations to ensure transport airplane structural damage tolerance is outlined below as well as the methods to ensure and verify structural damage tolerance.

2 Principles of Russian Regulatory Rules First of all in accordance with above-mentioned Russian idea “to ensure acceptable damage tolerance characteristics satisfying the requirements considered design of the structure should be performed basing on the definite criteria of tolerable damage sizes and their propagation period” [1]. Specific damage tolerance parameters were defined from the requirement that the calculated number of flight hours per one accident of the airplane with design goal of 50000 hours would be of 108 hours [1]. Recommended (regulated) damages are presented in Fig. 1 [1], [44].

Fig. 1 Regulated aircraft damages used in its damage tolerance analysis.

In the stiffened structures those are two-bay skin cracks with broken central stiffener (stringer in the wing and fuselage, frame in the fuselage). Wing structure should carry limit load under simultaneous failure of spar chord and the panel skin crack of one inter-stringer distance size, as well as the spar web crack equal to the half of the web height.

Analysis of Requirements on Fatigue and Damage Tolerance

41

In the fuselage cut-out zones it is reasonable to consider the cut-out edge failure and adjacent skin crack of one inter-stringer (inter-frame) distance size to be a tolerable damage. The structure having such damages should maintain carrying capability under simultaneous action of operating overpressure and external limit load Р limit. The structure having regulated damages should maintain the limit load. Basing on in-service maintenance program analysis and experience from operation of passenger aircraft some the requirements to damage growth duration and structural inspection intervals were formulated [1], [3], [44]. To provide the economical efficiency of the advanced design airplanes structures it is reasonable to give the opportunity to make seldom inspections with intervals about 1/4-1/5 of service life value. That corresponds to the airframe repair intervals in the factory, i.e. 3000-5000 flights [1], [3]. Safety factor due to crack growth duration scatter in the wing structure should be assumed > 2 [1], [3].It is reasonable to design the joints having poor inspectability according to safe life concept [1]. The specific recommendations to provide structural safety in case of multiple site damages (MSD) are given in Ref.[1]. In case of MSD in one of the wing build-up panels the entire wing structure must maintain the required residual strength while the full failure of that panel and the other panels intact. In case of MSD of several panels, none of the panels should be completely broken. One of the main tasks to conduct the full-scale fatigue tests of aircraft structures is to find out the probability of airframe MSD during operation. As the MSD occurrence depends on local stress concentration which is sometimes difficult to calculate precisely, the practical MSD characteristics are stated experimentally when the required number of flights for the structure is reached by the test program loading. According to recommendations of Ref. [1] the number of flights for the wing and the fuselage by the test program loading is determined from the conditions to find out the potential MSD. It is taken to be equal of three times the aircraft service life (3DSG.). After such number of flights the wing and the fuselage structures must maintain carrying capacity under the limit load [1]. According to the methodology developed in the Soviet Union the extent of full-scale laboratory structural tests is 3-5 times larger as the required economic service life [22]. Enhanced damage tolerance of the structures could be provided by application of advanced materials with high crack resistance. To meet the damage tolerance criteria mentioned above the required static and fatigue crack resistance characteristics had been defined to improve aluminum alloys [3],[4]. To determine the required static crack resistance wide experimental data were generalized considering stress intensity factors Kapp, that had been received from tension tests of plates with the width W=700-1200 mm without eliminating the buckling near the crack [4]. Analyzing these data a conclusion was done that material fracture toughness for the airframe skin could be increased at least up to the Kapp = 140 -150 MPa√m due to increasing metal pureness and applying appropriate thermal and mechanical treatment [4]. To providing the required residual strength values for the wing structures having two-bay skin crack under the broken stringer some specific experiments have been conducted on the stiffened panels made of D16T alloy. The test panels made

42

B.G. Nesterenko and G.I. Nesterenko

of the improved skin material with fracture toughness Kapp =140-150 MPa√m failed at gross stresses σfr =270-280 MPa [3]. Such level of the allowables gives high weight efficiency of the wing structure. The required values of cyclic crack resistance were defined basing on crack growth analysis for the wing skin with the broken stringer. The typical values were assumed in these calculations for the relative stringer area to the skin Fstr/Fskin=0,5-0,75, reliably detectable skin crack lengths were 2a = 25 mm, inspection interval T =3000-5000 flights, equivalent cyclic stresses σequiv=140-145 MPa. It was found out that in order to meet the requirements stated, the improved Al-alloys with high crack resistance should be used. Specifically, for the stress intensity factors ΔK=31 MPa√m and ΔK=62 MPa√m the crack growth rates are approximately 0.002 mm/cycle and 0.02 mm/cycle, correspondingly [3].The above methodology of the aircraft structure damage tolerance is used in major up to now.

3 Principles of the USA Regulatory Rules Evolution of structural damage tolerance principles for civil transports in the USA Airworthiness Regulations is considered below. Some significant differences from corresponding Russian approach are noted. Also the differences in terminology should be mentioned while comparing these regulations. The US FAR Regulations contain the terms “fail-safe” and “damage tolerance”. Fail-safety is the ability of the structure to maintain its required residual strength during its operation period without any repair after the failure or the partial failure of a principal structural element. Damage tolerance is the attribute of the structure that permits it to maintain its required residual strength during the period of its operation after the structure has sustained a given level of fatigue, corrosion, accidental or discrete source of damage [5]. Thus the fail safety is based on the apparent detection of large damages. Damage tolerance is based on establishing the structural inspection threshold and intervals in operation in order to detect potential acceptable damages by means of non-destructive inspection methods. The Russian concept of “operational survivability” covers both “fail-safe” and “damage tolerance” concepts [1], [10]. According to the fail-safe concept (the first survivability criterion) the residual strength should be guaranteed at tolerable regulated damages [1]. Damage tolerance concept (the second survivability criterion) requires a crack growth duration from the reliably detectable till regulated sizes such as to enable seldom airframe inspections [1], [4]. The widespread fatigue damage (WFD) in FAR includes multiple site fatigue damage (MSD) and multiple element damage (MED) [5]. WFD is characterized by the simultaneous presence of cracks in many structural elements that are of significant size and density of location, thereafter the structure would not be able to satisfy the requirements of damage tolerance (i.e. to maintain its required residual strength after partial structural failure). MSD is a special case (or the source) of WFD characterized by simultaneous presence of cracks in one and the same structural element (i.e. fatigue cracks that

Analysis of Requirements on Fatigue and Damage Tolerance

43

may link with or without other damages and lead to a loss of required residual strength). MED is a special case of widespread fatigue damage characterized by simultaneous presence of fatigue cracks in adjacent structural elements. Used in Russian official terminology is only the WFD term. But while the analysis of the structure several cases must be considered, and those cases correspond to MSD and MED cases of FAR terminology [1]. For example, while the wing residual strength analysis two types of damages are considered: multiple damages of one separate panel and the absence of cracks in remaining panels; onesite damages of several panels (the presence of one crack in several panels) [1]. During the last 46 years the US Airworthiness Regulations for civil transports have been improved several times [6]. It was caused by several airplane accidents resulting from principal structural damages in the lifetime, as well as due to requirements developed by the USAF The contribution of the USAF into implementing damage tolerance concepts should be recognized [7]. These requirements for military aircraft differ in details but not essentially [7]. In 1956 a new section CAR 4b.270 has been introduced into US Civil Air Regulations that gave an option to certify the airplane either by safe life concept or by fail-safe [2]. There was no regulated damage size for the fail-safe analysis. Thus in practice the aviation experts determined the damage size by their subjective criteria requiring the ease of structural failure detection and fast repair before the failure of remaining structure [6]. It is noted in Ref.[6] that most airplanes of the recent generation, e.g. Boeing 777 and Airbus A380, are designed so that fuselage structure is capable to carry the required load at the presence of two-bay crack (longitudinal or circumferential). But in the latest draft rules developed by General Structures Harmonization Working Group [8] it is suggested to consider one-bay crack instead of two-bay crack. The term “fail-safe” is excluded in this draft and replaced by the term “structural damage capability” (SDC). SDC is defined as an attribute of the structure which permits it to retain its required residual strength in the presence of large damage. Thus it is the partial failure between the elements containing the damages, i.e. one-bay crack. According to the author of Ref. [2] such replacement is stipulated by the following. The Working group developed the guidelines on WFD onset in 11 aging airplane fleets. This group assumed that not all the airplanes of these 11 fleets could sustain two-bay skin cracks with broken central stiffener. The author of Ref. [2] thinks it is inadmissible to give up the two-bay criterion for new airplanes having today available materials. It should be noted that the criterion of two-bay skin crack with one broken central stiffener in the Russian methodology was proposed and accepted in the early 1970s [1], it is still valid and recommended in the draft of МОС 25.571 [9]. The accidents of AVRO 748 in Argentina and Boeing 707 in Zambia due to fatigue cracks witnessed some deficiency of fail-safe concept. Hence a concept of damage tolerance [2] has been introduced into FAR 25.571 in 1978. The failsafe and the damage tolerance were considered as equivalent [2]. In Amendment 72 to FAR 25.571, issued in 1990, reference to fail-safe was removed to show that damage tolerance and fail-safe were not necessarily synonymous [2]. Current

44

B.G. Nesterenko and G.I. Nesterenko

Regulations completely cover fail-safe concept [6]. For instance, it is required for the fail-safe structure to define inspection threshold basing on crack growth analysis, i.e. to provide its damage tolerance [5]. Admitted in regulations the damage tolerance concept appeared to be the significant progress, as it resulted in implementing directed inspections to detect damage before it led to decrease of structural strength below allowable limits [6]. Introduction of damage tolerance concept became a stimulus on giving an importance to reliability of flaw detection, essentially for nondestructive inspection methods [7]. After the accident with Aloha Boeing 737 in 1988 at the Hawaii due to MSD cracks in the pressurized fuselage an Amendment 96 to FAR 25.571 was issued in 1998 and supported by Advisory Circular АС 25.571-1C [5]. These documents were recommended by FAA Technical Oversight Group for Aging Aircraft [2]. Three of the most important changes were introduced: 1) requirement to include manufacturing defects as a damage source; 2) requirement to establish inspection threshold based on crack growth for single load path structures as well as for the fail-safe multiple load path structures and crack arresting fail safe structure, for which it can not be demonstrate that load path failure, partial failure or crack arrest will be detected and repaired during normal maintenance, inspection or airplane operation prior to failure of the remaining structure; 3) requirement to demonstrate the sufficient full-scale fatigue test evidence that WFD will not occur within the design service goal of the airplane by teardown inspections following the completion of fatigue testing. Adequate verification by full-scale fatigue tests includes fatigue testing up to two and more design goals followed by inspection and analysis [5]. It should be noted that the requirement to prevent structural failure due to multiple damages have been formulated in Russian “operational survivability” [1] more than 20 years earlier than in FAA [5]. To demonstrate that multiple fatigue damages would not occur in operation it was recommended to perform a full-scale fatigue test in extent of three service goals [1]. The new generation airplanes, such as Airbus А380 [11] and Boeing 777 have been tested up to three design goals. The damage tolerance characteristic can be shown analytically by using reliable or conservative methods, and supported by test evidence [5]. Characteristics of crack growth should be determined for each detail design point. These data combined with the results of analyses and tests on residual strength are the basis for the inspection program. The following data are required for crack growth calculations: applied loading spectrum; assumed initial crack size and its shape; stress concentration and applicable stress intensity solutions; actual crack growth scenario and algorithms; crack growth rate parameters. Crack growth calculation is based on the approaches of linear fracture mechanics. Applied stress intensity solutions could be taken from public domain or may need to be calculated [8]. There is a number of recommendations on development methods and application of stress intensity factors.

Analysis of Requirements on Fatigue and Damage Tolerance

45

In January, 2011 a new Advisory Circular АС 25.571-1D was issued. It includes an important change – the introduction of a principal concept of Limit of Validity (LOV) – the period of time up to which it should be demonstrated by test evidence, analysis, and if available, service experience, that WFD will not occur in the airplane structure. Established LOV in effect, is the operational life of the airplane consistent with accomplished evaluations and maintenance actions aimed to prevent WFD. To opinion of the authors, an important approach coupling LOV and the full-scale fatigue tests is that test duration of 3 times the LOV without WFD occurrence eliminates the need for WFD-related mandatory maintenance requirements. And in case the test duration equals 2 times the LOV, it may result in required inspections and\or modifications prior to the LOV. Such an approach is close to traditional Russian practice in aviation [45].

4 Full Scale Structure Tests According to the US Regulations of 1978 on damage tolerance no full-scale fatigue testing was required for the airplane certification. Thus the US and European aircraft manufacturers developed difference concepts in order to certify their products [6]. The Boeing company in the beginning had conducted the full-scale fatigue tests in the extent of one service life in order to improve the airframe design [6]. Full-scale fatigue tests of Boeing 727 and 747 were conducted to one service life. Only hydro-fatigue test have been performed for Boeing 707 up to 2.5 design goals. In 1987 the rear fuselage of Boeing 737 was taken from service and fatigue tested in order to find out MSD effect on its damage tolerance [7]. After 59’000 in-service flights the fuselage was cyclically loaded up to 70’000 pressurizations [12]. MSD have initiated and resulted in two-bay crack followed by the safe fuselage depressurization due to flapping [7]. Boeing 747 fuselage was tested after 20’000 in-service flights to additional 20’000 pressurizations. While the test some MSD cracks have linked to one 150 mm long crack by the end of the tests [7]. Based on these tests inspection intervals were defined: for Boeing 737 – 12 years with assumed 3000 flights per year; for Boeing 747 – 7 years with assumed 1500 flights per year [7]. To opinion of the author of Ref. [7] such fatigue tests are not enough to guarantee the fleet safety, and the carefully done inspections are essential. Though fatigue testing of new models gives useful data, industry has reasonable objections against establishing service life basing on such tests [7]. Full-scale tests of Boeing 737 were not conducted, but then Boeing extended fatigue full-scale testing up to two service lives for Boeing 757 and 767 [12], and up to three service lives for Boeing 777. Airbus considered full-scale tests as certification tests with simulation of two service lives at least [6]. Figs.2, 3 and 4 present the comparison of design goals, number of in-service flights and fatigue tests of pressurized Boeing fuselages [12], full-scale structures of Douglas [13], Airbus [6] and Russian [26] airplanes. Fatigue testing of Tu-204, Tu-334, An-148 and SSJ-100 are still in progress [26].

46

B.G. Nesterenko and G.I. Nesterenko

Fig. 2 Service lives of Russian airplanes.

Fig. 3 Service lives of Boeing airplane fuselages.

Analysis of Requirements on Fatigue and Damage Tolerance

47

Fig. 4 Service lives of Airbus and Douglas aircraft structures.

All the Russian airplane types had been always full-scale tested (up to current days) for certification of their service lives. Today a number of Russian longoperated (aging) airplanes have exceeded their design goals (Fig. 2). In contrast to common practice all the full-scale structural tests of Russian airplanes were performed till WFD formation or structure failure. Next structures were been tested for residual strength (teardown) followed by crack growth analysis using fractography methods [14], [15].

5 Investigation of MSD Multiple site cracks studies were started after the accident with Aloha Boeing 737 in 1988 due to MSD cracks in the longitudinal lap joint of the pressurized fuselage skin [2], [6]. Most experimental results were taken from panels and specimens tests simulating longitudinal skin lap joints [16], [17]. Damage tolerance calculations for such elements having MSD assume the presence of one leading (large) crack starting at one of the holes and small cracks at other [6], [7], [17], [18]. In fact the scenario of multiple site cracking in full-scale structures is more complicated. The problem of MSD cracks was studied in Russia mainly on the results of fullscale structure tests. Table 1 presents some experimental criteria of residual strength for Russian airplane structures damaged by multiple fatigue cracks [19], [20]. Here:

σ app fr. netto - net stresses calculated with regard to decreased section area

of load bearing elements due to the holes and initial crack lengths;

σ Cfr.netto

-

48

B.G. Nesterenko and G.I. Nesterenko

stresses taking into account stable crack growth; Kfr - stress intensity factor at the structural failure. Analysis of experimental data (Table 1) has shown that residual strength of structures with multiple site damages is affected by a large number of factors such as structural elements features, stresses due to element bending, material plasticity, mutual location of cracks, holes, stable crack growth at a single static loading [14]. Table 1 Residual strength of full-scale aircraft structures with widespread fatigue damages.

Type of damaged primary structural element, airplane 1 Skin and stringers near stringer joint of An-10A lower wing surface panels Skin, stringers and spar of An-10A lower wing surface near stiffening patch edge Skin and stringers of Tu-134A lower wing surface panels near stiffening patch edge Skin and stringers of Il-62 integral stiffened lower wing surface panels near stringer holes for fuel flow Spars, joint profiles of Il -62 upper wing surface panels Joint profiles of Il-62 upper wing surface panels Stringer and circumferential skin joint patch of Tu -154B pressurized fuselage Il-86 pressurized fuselage skin near 3-row rivet seam of longitudinal skin joint Il-86 pressurized fuselage skin near 3-row rivet seam of longitudinal skin joint An-24 pressurized fuselage skin near 2-row rivet seam of longitudinal skin joint Tu-134A pressurized fuselage skin between two frames and two stringers (19 through-thickness notches) Tu-134A pressurized fuselage skin between two frames and two stringers (19 through-thickness notches) Il-86 strap joining pressurized fuselage with pressure bulkhead Skin and stringer of Tu-104 lower wing surface panels near stiffening patch edges Strap joining skin sheets of lower wing surface panels Wing turning assembly

! app fr .net ! 0.2

! Cfr.net ! 0.2

K fr

K fr

K app

KI C

3

4

5

6

0.8

1.0

0.5

-

0.9

1.0

0.6

-

0.9

1.0

0.5

-

D16Т

0.7

0.83

1.0

-

D16Т D16Т D16АТV D16Т

0.3 0.7

0.7 1.0

0.5 0.75

-

0.75

0.88

1.0

-

D16АТV

0.57

1.0

0.5

-

D16АТV

0.63

1.05

0.9

-

D16АТV

0.48

0.85

0.7

-

D16АТV

0.9

0.9

1.0

-

Material 2 D16АТNV D16Т D16АТNV D16Т D16АТV D16Т

D16АТV

0.85

0.85

1.0

-

D16АТV V95АТ1V V95 Т1 V95АТ1V V93 Т1

0.16

0.17

0.45

-

0.45

0.46

1.0

-

0.4 0.4

0.41 0.4

0.4 -

1.0 1.0

Growth of MSD cracks studies used fatigue test results of full-scale structures and experience of structures operation [15]. Figs.5-6 illustrate the estimation of MSD crack growth durations based on the analysis of cracks in the wing structures of operated airplanes [15]. Crack growth duration curves are given for the probabilities of р = 0.5; 0.05 и 0.001 [43]. Analysis of experiments has shown that the ratio of MSD crack growth time ΔT (from initial sizes 0.1 - 0.5 mm till critical sizes) to gross life Т0+ΔT (life till crack initiation plus crack growth time) depends on structural features of load bearing elements and it lies in the range of 0.03-0.8 [15]. This ratio is about 0.5 for the typical wing structural elements. From specially performed tests on fracture of longitudinal lap joints in fuselage skin due to MSD (Fig.7) the ratio ΔT = 0,35 was obtained [41],[42]. It should be noted T0 + ΔT that the longitudinal skin lap joint is loaded by tension and bending.

Analysis of Requirements on Fatigue and Damage Tolerance

49

Fig. 5 Relative crack growth duration for MSD cracks in lower wing surface stringers (data from operation).

Fig. 6 Relative crack growth duration for MSD cracks in joint of upper wing surface panels (data from operation).

50

B.G. Nesterenko and G.I. Nesterenko

Fig. 7 Flat panel fracture at tension of longitudinal fuselage skin lap joint, damaged by MSD.

6 Application of Safe Life Concept Current regulations require to design the structure according to damage tolerance concept unless it leads to complications that an effective damage tolerant structure cannot be achieved due to limitations of geometry, inspectability or good design practice. Under these circumstances a design that complies with fatigue evaluation (safe-life) requirements is used. A typical example of the structure that cannot be designed by damage tolerance concept is the landing gear and its attachments [5]. Russian experience has demonstrated that it is reasonable to use the “safe life” concept in design of joints with poor inspectability [1]. Regular airframe zones – longitudinal wing and fuselage joints – define the mass of load bearing structure. It is assumed that fatigue resistance of longitudinal joints uniquely defines the service life of the whole wing structure. Onset of fatigue cracks in longitudinal joints has, in general, a widespread character and it indicates that structural service life is exhausted [21].

7 Stages of Establishing Airplane Service Life in Russian Practice Particular difference of the Russian methodology on providing safe service life of the airplane structure in contrast to common practice methodology is the continuous extension of operation limits – that means to assign currently “approved” service life incrementally, i.e. by stages [22], [25]. When the leading airplanes of the fleet are approaching the actual “approved” service life of the structure it is required to consider the further operation of the given fleet and to assign the next

Analysis of Requirements on Fatigue and Damage Tolerance

51

specified life limit, i.e. the next approved service life. Definition of specified lives is in the limits of design goal. Such step-by-step service life extension is caused by difficulties in prediction of operational conditions for decades, structural fatigue and damage tolerance characteristics, possible changes in the structure resulting from operation, technology change etc. [23]. Special feature of safe operation of aged airplanes is that they are operated beyond design goals according to the current regulations and damage tolerance requirements. In compliance with accepted methodology the service life and the lifetime are extended step-by-step in the individual way [26]. Damage tolerance was not considered for the structure while the design stage of old airplanes. Therefore utilized is some “natural” damage tolerance inherent to every structural type to some degree [23]. To ensure safe operation of old airplanes it is required to formalize not only the baseline Approval for fleet operation, but also an individual Approval on operation for each plane of the fleet with obligatory inspection of this board before service life extension in accordance with its specially developed program [23].

8 Harmonization of the US, European and Russian Regulations In 1990s it became necessary to harmonize domestic Russian Regulations with current common world-wide regulation requirements on aircraft structural strength and, in particular, on aircraft structural fatigue and damage tolerance [24]. Harmonization of European and US regulations on the problem considered was based on practically total approval in the Joint Aircraft Regulations (JAR) of Europe the general requirements of paragraph 25.571 from Federal Aircraft Regulations (FAR) of the USA [6]. It meant the recognition of great and “pioneer” US contribution into the development and practical approval of the methods on solving this problem [24]. Similar to the earlier harmonization of FAR 25.571 and JAR 25.571 requirements, it was decided to base the Russian Aviation Rules (AP 25.571) on the FAA regulations. Section 4.9 of the 3rd edition of Airworthiness Regulations for civil USSR aircraft (NLGS-3, 1984) was “ideologically” close to the requirements of FAR 25.571 [24]. Specific research on aviation strength standards traditionally was carried out by TsAGI experts in collaboration with the experts of the leading aviation design bureaus [24]. In present time the harmonization project “Methodical materials in defining compliance of Aviation Rules AP 25.571 (2004) on fatigue and damage tolerance. МОС 25.571” with the joint FAA/JAA circular АС 25.571-1Х [8] is in process now [9]. Realization of aircraft structure damage tolerance concept in solving the problem of ensuring and maintaining airworthiness of operated flying vehicles is outlined in the book [25].

52

B.G. Nesterenko and G.I. Nesterenko

9 Improvement of Aluminum Alloys Desired characteristics of static strength, fatigue, damage tolerance and weight efficiency of contemporary aircraft structures is provided generally by using improved aluminum (Al) alloys. In 1970s in Russian to reach the increased damage tolerance of airplanes the required values of fracture toughness KCapp were defined as equal to 140-145 MPa√m and crack growth rates equal to 0.002 mm/kcycle at ΔK=31MPa√m for Al-alloys [3],[4]. The substantiations of these values of material crack resistance are presented in Refs. [3], [27], [28]. Crack resistance properties of mass-production Al-alloys produced in 1970s were much lower than the required characteristics [3]. But by the end of the 1970s the Aviation Industry of the USSR has formed and put in action the System of Aircraft Material Quality. The main principle of this system was an integrated centralized development, implementation and standardization on manufacturing, application and operation of materials in the products [29]. As a result the alloys 1163Т, 1161Т, 1163АVТ and other with advanced crack resistance properties have been developed [3],[4]. Tables 2 – 7 present properties of improved Al-alloys applied in the structures of Russian and foreign airplanes. The following characteristics are given: KCapp – apparent stress intensity factor determined on the standardized specimens; v31 = d(a)/dN - crack growth rate at delta stress intensity factor ΔK=31MPa√m and cycle ratio R=0; N133- average fatigue life of standard specimens with central hole under cyclic gross stresses σmax=133 MPa at R=0; σb- ultimate strength; σ 0,2 – yield strength; δ – elongation at fracture. Material properties presented in Tables 2 – 7 are gathered based on Refs. [30][41]. Figs.7-8 illustrate the improvement by Alcoa company of strength and crack resistance in Al-alloys applied in the structures of different airplanes [36], [37]. Table 2 Material properties of wing lower surface skin. Plates. Alloys 1163T

1163 T7

TU-204

IL96-300

σB , MP a σ0.2 , MPa

460 340

500 390

δ, % N133 , cycles da/dN, mm/kcycle ΔK=31, R=0 Kapp, MPa √ m W =1200mm

20 205000

14 200000

12 275000

2,5

3

175

163

Material characteristics

2324-T39 C 433 -T351 Airplanes Boeing 767, А340 А380 500 490 460 380

C433-T39

2024-T351

500 460

Boeing 747, А310 490 390

15 250000

12 -

15 115000

2,5

1,1

2,1

2

148

153

163

135

Boeing 777

Analysis of Requirements on Fatigue and Damage Tolerance

53

Table 3 Material properties of wing upper surface skin. Plates. Alloys Material characteristics

V95ochT2

V96 c-3pchT12

IL96-300, TU-204

VIAM

δ, % N1 3 3, cycles da/dN, mm/kcycle ΔK =31, R=0

540 460 10

635 595 10

170000 3,25

Kapp , MPa √ m W=1200mm

175

σB , MPa σ0.2, MPa

7150-T651

7055-T7751

Bo eing 757/767, A310 Boeing 777, A380 >580 550 >7

620 595 7

320000

-

300000

5

4,3

3,5

70

85

90

Table 4 Material properties of wing lower surface skin. Sheets. Alloys Material characteristics

1163ATB

1163 RDTV

2524-Т3 (t=3,8mm)

6013-Т6

1370Т1

1441 RT1

IL96 -300

TU-204

Bo eing 777

А380

VIAM

VIAM

riveted σB, MPa σ0.2, MPa δ, % N1 3 3, cycles da/dN, mm/kcycle ΔK =31, R=0 Kapp , MPa √ m W=1200mm

welded

430 315 24 100000

460 340 23 115000

450 345 19 168000

400 365 13 85000

440 350 10 105000

420 340 13 85000

2

2

1,7

2,5

2,5

2,7

120

130

140

110

100

100

Table 5 Material properties of fuselage upper surface skin. Sheets. Alloys Material characteristic s

σB, MPa σ0.2 , MPa δ, % N133 , cycles da/dN, mm/kcycle ΔK=31, R=0 Kapp , MPa √ m W=1200mm

1163ATB

1163 RD TV

2524-Т3 (t=1,6mm) 2524-Т3 (t=3,8 mm)

IL96 -300

TU-204

А380

Boeing 777

430 315 24 100000

460 340 23 115000

430 325 20 100000

450 345 19 168000

2

2

2

1,7

120

130

132

140

Airplanes

54

B.G. Nesterenko and G.I. Nesterenko

Table 6 Material properties of wing lower surface stringers. Extrusions. Alloys 1163

V95ochT2

TU-204, SSJ-100

IL-96-300

Bo eing 777

AL COA

σB , MP a

570

570

590

>540

σ0.2 , MPa

440

510

430

>490

δ, %

7

7

10

8

N1 3 3 , cycles

270000

190000

-

-

KI C , MP a √m

37

31

Material characteristics

2224-Т3511

Al-Li 2099-Т8967

Airplanes

63

Table 7 Forging alloys. Load-bearing frames, ribs, spars, fittings. Alloys Material characteristics

1933Т123

7085Т7452 Airplanes

"Mria", AN-148, Yak-130, SSJ-100

Боинг 787, А380

σB , MP a

510

500

σ0.2 , MPa

460

460

δ, %

8

9

N1 3 3 , cycles

140

-

da/dN, mm/kcycle ΔK =31, R=0

2.6

2.7

KI C , MP a √m

37

36

Analysis of Requirements on Fatigue and Damage Tolerance

55

Fig. 8 Improvement of Al-alloys, ALCOA data yield.

Fig. 9 Improvement of Al-alloys for Boeing aircraft.

10 Conclusion Resuming the evolution and development of main concepts on providing required damage tolerance of airplane structures of the USA, Europe and Russia the following could be concluded:

56

• • • • •

•

• • • • •

B.G. Nesterenko and G.I. Nesterenko

Russian Rules AP 25.571 on fatigue and damage tolerance of transports are harmonized with US regulations FAR 25.571 and European regulations JAR 25.571. In Russian practice the main principle on providing structural safety in the scope of fatigue strength is the “operational survivability”. It incorporates fail-safe and damage tolerance concepts. According to Russian circulars it is recommended to use the “safe-life” concept for landing gear, longitudinal fuselage joints, as well as for longitudinal and lateral wing panel joints. The experience of airplanes operational has shown the necessity to satisfy damage tolerance criteria by providing the required residual strength of the structure tha contains “two-bay” skin crack with broken central stiffener. For the fail-safe structures the inspection starting point may be defined either by fatigue or by slow crack growth criterion from initial manufacturing flaw. Inspection threshold fore non fail-safe structures should be defined only by slow crack growth criterion. To verify zero probability of WFD occurrence during the aircraft design goal it is required to conduct the fatigue tests of the aircraft full-scale structure for two and more design goals followed by teardown, disassembly and flaw analyses. One of the particular features in assigning the service life for Russian airplane is the step-by-step and individual operation time extension. Required fatigue and damage tolerance characteristics of aircraft structure could be reached by using advanced Al-alloys with high static and fatigue crack resistance. Duration of MSD crack growth and residual strength of the structure with such cracks depend on many factors such as element structural features, material properties, stress state, plasticity etc. Damage tolerance characteristics of aircraft structures should be calculated analytically using reliable conservative methods and verified by tests. Current damage tolerance analysis is performed basing on linear fracture mechanics approaches and methods.

References [1] Nesterenko, G.I.: Damage tolerance of aircraft structures. In: Proceedings of Kiev Institute of Civil Aviation Engineers. Strength, reliability and life of aircraft structures, Kiev, vol. (2), pp. 60–70 (1976) (in Russian) [2] Swift, T.: Fail-safe design requirements and features, regulatory requirements. In: AIAA/ICAS International Air and Space Symposium and Exposition: The Next 100 Years, Dayton, Ohio, July 14-17, p. 23 (2003) [3] Nesterenko, G.I.: Requirements to ensuring operational survivability of passenger and transport airframes at design stage. In: Proceedings of Scientific-Technical Conference “Complex Ensuring Aircraft Structural Service Life”, TsAGI, pp. 199–211 (1984)

Analysis of Requirements on Fatigue and Damage Tolerance

57

[4] Nesterenko, G.I.: Analysis of damage tolerance parameters of aircraft structures based on fracture mechanics. Journal: “Physical-Chemical Material Mechanics” AN USSR (1), 12–20 (1983) (in Russian) [5] AC 25.571-1C Damage Tolerance and Fatigue Evaluation of Structure. Department of Transportation. Federal Aviation Administration, 4/29/98 [6] Schmidt, H.-J.: Damage tolerance concept, methods and experiments utilized in the structure of modern transports for compliance with FAA/JAA regulations. Thesis for Candidate of Technical Sciences, IMASH RAN, Moscow (2002) (in Russian) [7] Goranson, U.G.: Damage Tolerance, Facts and Fiction. Presentation at International Conference on Damage Tolerance of Aircraft Structures, September 25. Delft Technical University, Delft (2007) [8] Damage Tolerance and Fatigue Evaluation of Structures. FAR/JAR § 35.571. General Structures Harmonization Working Group Report, FAA (2002) [9] Nesterenko G.I., Dubinsky V.S., Trunin Yu. P., Ya, S. V.: Methodical materials to identify compliance with requirements of Aviation Rules AP25.571 (2004) in Fatigue and Damage Tolerance.25.571-1A, Draft, TsAGI (2008) (in Russian) [10] Airworthiness regulations for USSR civil airplanes, Amendment, ch. 2-4,2 edn. (December 25, 1976) (in Russian) [11] Yudi, A., Peter, B., Thomas, N.: Test program for the A380 major fatigue test. In: Proceedings of the 23rd ICAF Symposium, Hamburg, Germany, June 8-10, vol. 1, pp. 353–364 (2005) [12] Rao, V.S., McGuir Jac, F.: Boeing structural design technology improvements. In: Proceedings of the FAA-NASA Sixth International Conference on the Continued Airworthiness of Aircraft Structures, Atlantic City, New Jersey, June 27-28, pp. 75–81 (1995) [13] Hoggard Amos, W.: Design, analysis and testing of durable aircraft structures. Presented to International Symposium Experimental Facilities and Aircraft Certification, August 22-25 (1995) [14] Nesterenko, G.I.: Multiple site fatigue damages of aircraft structures. In: AGARD Conference Proceedings 568 Widespread Fatigue Damage in Military Aircraft. Papers presented at the 80th Meeting of the AGARD Structures and Materials Panel, held in Rotterdam, the Netherlands, May 10-11, p.11-1 –11-8 (1995) [15] Nesterenko, G.I.: Fatigue and damage tolerance of aging aircraft structures. In: Proceeding of the FAA-NASA Symposium on the Continued Airworthiness of Aircraft Structures. Atlanta, Georgia, August 28-30, pp. 279–299 (1996) [16] Fawaz, S.A., Schijve, J., de Koning, A.U.: Fatigue crack growth in riveted joints. In: Proceedings of the 19th ICAF Symposium, Edinburgh, Scotland, June 16-20, pp. 553–574 (1997) [17] Gruber, M.L., Wilicius, K.E., Worden: Investigation of fuselage structure subject to widespread fatigue damage. In: Proceedings of the FAA-NASA Symposium on the Continued Airworthiness of Aircraft Structures, Atlanta, Georgia, August 28-30, pp. 439–459 (1996) [18] Swift, T.: Damage tolerance capability. Journal of Fatigue 16, 75–94 (1994) [19] Nesterenko, G.I.: Fatigue and damage tolerance of aging aircraft structures. Trudy TsAGI, 2631, p.67–75 (1998) (in Russian) [20] Nesterenko Grigory, I.: Ensuring damage tolerance of aging aircraft structures. In: Proceedings of the Second Joint NASA/FAA/DoD Conference on Aging Aircraft. Langley Research Center, Hampton, Virginia, pp. 163–178 (January 1999)

58

B.G. Nesterenko and G.I. Nesterenko

[21] Vorobyev, Z., Ol’kin, B.I., Stebenev, V.N., Rodchenko, T.C.: Fatigue strength of structural elements. Mashinostroyeniye Moscow, 240 (1990) (in Russian) [22] Selikhov, A.F., Leybov, V.G., Nesterenko, G.I., Raikher, V.L.: Methodology and experience of ensuring structural safety of aging aircraft. Trudy TsAGI, 2631, p.21–29 (1998) (in Russian) [23] Dubinsky V.S., Nesterenko G.I., Raikher V.L., Stuchalkin Yu.A.: Maintaining airworthiness of structures of certified aircraft by service life conditions Trudy TsAGI, 2631, p.73–75 (in Russian) [24] Raikher, V.L., Dubinsky, V.S., Nesterenko, G.I.: The features of aircraft structure fatigue resistance certification and airworthiness maintenance in contemporary conditions. In: Proceedings of International Symposium Experimental Facilities and Aircraft Certification, Zhukovsky, Russia, August 22-25, pp. 235–245 (1995) [25] Arepyev A.N., Gromov M.S., Shapkin V.S.: Problems of aircraft structural damage tolerance. Vozdushniy Transport, Moscow, p. 422 (2002) (in Russian) [26] Dubinsky V.S.: About service life state of domestic transport and passenger airplanes. Trudy TsAGI, 2683 (2009) (in Russian) [27] Nesterenko, G.I.: Requirements to properties of advanced structural materials for airframe. Journal Technology of Light Alloys 2, 43–51 (1995) [28] Nesterenko, G.I.: Analytical characteristics of advanced structural materials for airframe. Journal Technique of Air Transport 3-4, 1–9 (1995) (in Russian) [29] Kachanov Ye, B., Berenson, V.F.: Issuing passports for aircraft materials. Journal Technology of Light Alloys 2, 19–21 (1995) [30] Fridlyander, I.N.: Al-alloys in flying vehicles for the periods of 1970-2000 and 20012015. Journal Technology of Light Alloys 4, 12–17 (2002) [31] Fridlyander, I.N., Sadkov, V.V., Sandler, V.S., Fedorenko, T.P.: Semiproduct properties for high-tech Al-Li alloy 1441. Journal Technology of Light Alloys 4, 24–27 (2002) [32] Senatorova, O.G., Sukhih, A., Yu., S.V.V., Goloviznina, G.V., Matviyenko, S.V.: Evolution and prospects of applying high-strength Al-alloys for rolled semiproducts. Journal Technology of Light Alloys 4, 28–33 (2002) [33] Tkachenko, Y.N., Latoushkina, L.V., Val’kov, V.Y., Shomin, V.A.: Effect of homogenization modes on structure and properties of ingots and extruded semiproducts of 1933 alloy. Journal Technology of Light Alloys 4, 34–37 (2002) [34] Kolobnev, N.I., Khokhlatova, L.B., Ovsyannikov, B.V., Ivanovsky, N.P.: Adoption of semiproduct manufacture from corrosive resistant welding alloy of 1370 series with Al-Mg-Si-Cu system. Journal Technology of Light Alloys 4, 44–47 (2002) [35] United States Patent, 4,294,625 (October 13,1981) [36] Airliner Boeing Structural Design and Technology Improvements (April-June 1996) [37] Sawtell, R., Liu, J., Newman, J., Bray, G., Sakharutov, A.: Advanced aluminum alloys and products for aerospace application. ALCOA. Presented to Technical Seminar for Yakovlev Design Bureau (January 2009) [38] Basov V.N., Nesterenko B.G., Nesterenko G.I.: Fracture of high-strength Al-alloys. Journal ”Flight” TsAGI Mashinostroyeniye, Moscow, 87–92 (2008) (in Russian) [39] Nesterenko, B.G., Nesterenko, G.I., Basov, V.N.: Fracture behaviour of skin materials of civil airplane structures. In: Proceedings of the 25th Symposium of the International Committee on Aeronautical Fatigue (ICAF), May 27-29, pp. 661–683. Springer, Rotterdam (2009)

Analysis of Requirements on Fatigue and Damage Tolerance

59

[40] Dmitriyev, V.G., Zamula, G.N., Konovalov, V.V., Nesterenko, G.I.: Priority directions in improving materials for advanced aircraft structures. Journal Technology of light alloys 1, 3–8 (2003) [41] Nesterenko, B.G., Nesterenko, G.I., Schmidt, H.-J.: Fatigue and damage tolerance of fuselage skin longitudinal joints. In: Proceedings of the 24th Symposium of the ICAF, Naples, Italy, May16-18, vol. 11, pp. 1006–1018 (2007) [42] Nesterenko, G.I., Kozlov, A.G., Nesterenko, B.G., Stoyda, Y.M.: Damage tolerance of riveted joints in fuselage skin. Journal Problems of engineering and reliability of machines 4, 111–115 (2008) [43] Ya, S. V.: Analysis of fatigue crack growth parameters in aircraft structural elements by operation data. Trudy TsAGI, 1671, p. 17–27 (1975) (in Russian) [44] Nesterenko, G.I.: Service life and damage tolerance of aircraft structures. Journal Problems of engineering and reliability of machines 1, 106–118 (2005) (in Russian) [45] 25.571-1D Damage Tolerance and Fatigue Evaluation of Structure. Department of Transportation. Federal Aviation Administration (January 13, 2011)

26th ICAF Symposium – Montreal, 1-3 June 2011 Material Selection and Detailed Design – Requirements and Responsibilities of an Accredited and Qualified Test Laboratory R. Best, Th. Fleischer, R. Franke, S. Goldbach, J. Gruner, S. Reichard, and J. Ridzewski, IMA Materialforschung und Anwendungstechnik GmbH Dresden, Germany

Abstract. For the characterisation and accreditation of materials the aeronautical industry accesses the expertise and the support of independent test laboratories. The following paper outlines the requirements and responsibilities of a test laboratory. Those are referring to the performance of destructive and nondestructive tests as well as to the corresponding quality management system. The field of testing is generally handling the requests of the test definitions and the optimisation of test methods and their influencing terms and conditions. The field of quality management system attends to the basics and the necessity of the certifying and accreditation process of a test laboratory. For the execution of characterisation tests of materials (metals, composites or hybrids) it is important that the test laboratory can fulfil the requirements of the testing expertly and is able to perform enhancements and counsels to the test concepts as well as to the test rigs in front of the background of its accreditations and qualifications.

1 Introduction The implementation of new materials and design concepts for new types of aircraft structures requires a large number of material tests starting from coupon tests through different levels of component tests (shell, barrel) up to full scale testing of complete aircraft structures accompanied by measurements and non-destructive inspections. IMA Dresden have been involved in a variety of structural tests of development and certification programs of the international aerospace industry over the past years. Our test activities cover all static and fatigue mechanical structural tests, starting with material tests at coupon level, furthermore, different levels of structural and component testing up to testing of complete aircraft structures (major tests). Being involved in full scale fatigue test projects as well as in performing barrel tests, curved and flat panel tests, coupon and other structural tests requires not only a wide-spread basis of engineering services and know-how but also the appropriate qualification and accreditation.

62

R. Best et al.

The test pyramid for all mechanical structural tests (static and fatigue) consider material tests at coupon level, different levels of structural and component testing up to testing of complete aircraft structures (major tests), Figure 1.

Fig. 1 Test pyramid.

2 Static and Fatigue Testing Standard testing At the beginning of a material qualification, investigations in form of material tests take place. Often the tests are standardized in international / national standards and more detailed in in-house test specifications. For the standardised tests not only a precise performance, analysis and documentation is essential. A precondition for the laboratory performing such tests is having facilities capable of meeting all applicable temperature and humidity requirements as well as cleanliness and structuring. Further requirements in the course of testing have to be met also, as there are:

checking clearness and condition of all test specimens, appropriate treatment and storage of specimens, dimensional check and measurement with adequate and calibrated measuring equipment, correct clamping and alignment of specimens and test rigs, compliance with climatic requirements such as temperature and humidity, test execution with all required test parameters e. g. test speed, method of Young’s modulus determination,

Material Selection and Detailed Design

63

adequate strain measurement devices, determination and analysis of characteristics acc. to the test standard applied, documentation of all relevant data.

Even for standardised test methods a variety of errors are possible. Starting with the measurement of specimen dimensions with inappropriate measuring equipment, further specimen clamping without adequate alignment, to the point of an evaluation at improper measuring characteristics such as for the determination of the modulus; the elimination of these errors has to be ensured by means of the quality management system of the testing laboratory, such as:

very well educated and experienced personnel who are able to review the test results and to identify meanderings, test equipment which is calibrated and aligned acc. to national / international standards, participation in internal and external round robin tests, tailored clamping devices which provide a maximum of clamping load by a simultaneous minimum of notch factor.

Fig. 2 Examples of standard tests for metals and FRP.

Material tests under climatic conditions Material tests which are to be performed under climatic conditions other than standard climate challenge test institutes consistently. The specific test conditions have to be provided. It must be ensured, that the specimens are subject to the actually required climate. Particular attention has to be turned to the utilised details of the equipment. As example some requirements which need to be met:

low temperature tests down to -196 °C, high temperature tests up to +1000 °C, high humidity and temperature tests, operation temperature of utilised equipment, freezing, melt and flash points of used liquids, cables or auxiliary materials.

64

R. Best et al.

Development In addition, even standards have to be improved from time to time. Here test institutes also have the responsibility to suggest improvements and to develop test equipment further as can be necessary. An example for such further development is the compression test on fibre reinforced plastics. A large number of test standards exist for this test, e. g. DIN pr. EN 2850, ISO 14126, AITM 1-0008, ASTM D 3410. All standards vary regarding the way of load introduction (shear-, end- or combined loading), the test rig to be applied, the specimen dimensions as well as the evaluation method. The challenge lies in the way of load introduction without causing failure in the tab area. Additionally, it is essential to exclude external bending influences. Resulting from those requirements IMA Dresden have developed a compression test device with hydraulic clamping (Patent No. DE 103 44 544 B3) which nowadays is in global use, Figure 3.

Fig. 3 Compression test on fibre reinforced plastics.

The developed compression test device allows testing of various specimen geometries at low and high temperatures due to a simplified handling of specimens. The suitability for aviation related tests has been validated in several interlaboratory comparisons as well as permanent tests. Zwick GmbH & Co. KG, a member company of Zwick Roell AG, is now responsible for the worldwide distribution of the Hydraulic Composite Compression Fixture (HCCF) [5]. This example briefly shows a successful development meeting the responsibility of a test institute to put its know-how effectively into practice. Specific tests according to customer requirements A multitude of tasks assigned by customers cannot be covered by existing standards. In that field of testing the laboratory has the responsibility to perform the tests as well as to counsel the customer. Hereby, the following issues need to be considered:

Material Selection and Detailed Design

65

exact specification of all test parameters, customisation of test rigs (load introduction, clamping, anti buckling devices, deformation measuring) and specimens, provision of calculations and simulations.

Examples of customised tests are displayed in the following figures.

Fig. 4 Bolted joint; patch test.

Fatigue tests Accurate fatigue life prediction becomes more and more important, as composites are nowadays used for critical structural components. Diverse loading patterns are applied for the determination of fatigue data as there is:

constant amplitude loading, two-stage fatigue loading, multiple-stage fatigue loading, random or spectrum fatigue loading with variable amplitudes and mean stress.

Due to the rather specific characteristics of fatigue tests nearly every test performance needs a great amount of customisation in order to achieve the desired results. One important point is the alignment of the test rig. The test laboratory has to make sure that the specimens do not receive any bending influence especially in compression loading patterns. For the most test set-ups additional anti buckling devices are needed and have to be validated by the test laboratory. The second point is the specimen geometry. Especially the load transfer from the clamping to the testing range is a challenge. In order to receive valid failures the specimen geometry often has to be adapted and the tests calculated and simulated.

66

R. Best et al.

3 Non-destructive Testing The implementation of new materials for aviation business requires a large number of tests and validations. With all of the performed tests static as well as fatigue properties are studied. For the safety of the materials respectively structures the damage tolerance behaviour is important. Considerably more complex than detecting the crack propagation of metallic materials is the prediction of flaws in fibre reinforced composites which gain more and more importance for aircraft industry. Cracks in metallic materials generally show continuous growth; the durability of the constructional element can be estimated by their location and dimensions. Damages in fibre reinforced plastics such as CFRP and GFRP, e. g. delaminations or fibre breakage, in the first instance often stay undetected but can lead to fast failure of the structure in certain circumstances. The application of carbon fibre reinforced plastic materials is only one important aspect for new design, inspection and testing demands [2]. The behaviour of bonded structure parts and the material properties after impact damage or under cyclic conditions are of important interest also – starting with simple structure parts up to whole CFRP panels with complex joining of diverse materials. For the fulfilment of demanding requirements NDT is applied to aerospace material. The concept of NDT is focused on the ability of how quantitatively all the possible damages can be detected according to the test pyramid. The damage tolerance approach (DTA) requires that all possible damages must be detected before they reach a critical status under maintenance inspection schedule. Inspection requirements The quantitative detection of all potential damages requires the determination of an adequate monitoring period for structures – a capable criterion for detectable damages has to be found. Various aspects are essential for the determination:

the inspection starting point (initial inspection), the inspection time, the repeat inspection interval [1].

Fig. 5 Inspection interval [1].

Material Selection and Detailed Design

67

Therefore, a diversity of inspection methods needs to be applied under testing conditions, subject to the kind of material and the areas of interest (e. g. impacted skin areas; bonded stiffener elements; structure cuts; joining areas):

Manual pulse-echo-technique, Ultrasonic phased array technique, e. g. Figure 6 and Figure 7 Ultrasonic through transmission, Ultrasonic immersion technique, Eddy current inspection (conductivity measurement, layer thickness measurement, surface crack detection, multi-frequency technology, Figure 8) Optical measurement (e. g. ARAMIS, PONTOS), Figure 9.

Fig. 6 Ultrasonic phased array scan of impacted Glare skin.

Fig. 7 Inspection of CFRP curved panel.

The bore hole inspection with multiple frequency technique will be used for determining the damage location, the damage depth into the bore hole and the length into the material up to 2 mm. It is qualified for aluminium and can reduce the repair effort for in-service repairs, Figure 8.

68

R. Best et al.

Fig. 8 Bore hole inspection with multiple frequencies.

The principle of an inspection based on ARAMIS method is:

3-D displacement of the surface, deformation of the surface (out of plane), strains in X,Y direction, shear strain angle, 3D – surface, information about single points and lines, cuts in several lines and different steps, diagrams, overlay pictures, data export.

Fig. 9 ARAMIS inspection of curved fuselage panel.

Material Selection and Detailed Design

69

4 Certification and Accreditation Process The general conditions at manufacturers, suppliers and end users are largely determined by the introduction and realization of modern quality philosophies reflecting themselves in terms of quality, quality assurance, quality management, audit, certification, accreditation of laboratories of DIN EN ISO/IEC17025, DIN ISO 9001, EN 9100 or NADCAP. All these terms are meant to justify the claim of trustworthiness in connection with the term quality and, therewith, to create the preconditions of lasting marketability. Quality management The term quality management QM subsumes planning, steering, organization and further development of quality assuring actions that are suitable to assure sufficient and homogeneous composition of products by proving compliance with the requirements of safety, reliability and usability. Quality management is governed by the linkage of a companies’ quality policy, the kind and extend of work-, test- and documentation instructions which must verifiably be known and by regulations for corrective actions that must be applied on the basis of information about actual or potential errors, that means by the quality of the company and the quality of products and services [3], [4]. Basis of a companies’ quality management (QM) system is the quality manual. Its preparation increases transparency in the company, reveals existing shortfalls in planning and operational bottlenecks, regulates authorities and areas of competence in the companies´ structure; certain operational procedures such as processing of orders or complex test activities are formalized and become comprehensible. Procedures and guidelines for uniform assessment of design and extent of QMsystems are described in DIN EN ISO 9000:2005. This standard describes the elementary basics for quality management systems. It serves a general understanding and does not contain requirements to QMS. The current version from 2005 is issued in three languages. Standard DIN ISO 9001:2003 (new draft 2009) is one that makes requirements to a QMS. It requires documented procedures for the following QM elements:

Document control, Control of QM records, Internal audits, Control of defective products, Corrective actions, Preventive measures.

The EN 9100 standard series is based on the general quality management standards according to EN ISO 9001:2000. Including quality assurance models for construction, development, production, assembly and maintenance it provides the frame for comprehensive quality management systems for suppliers of the aerospace industry.

70

R. Best et al.

General requirements to the competence of test- and calibration laboratories are defined in DIN EN ISO/IEC17025:2005. Since 2000 this standard replaces DIN EN 45001 which was valid until then. By means of these guidelines qualified and independent institutions (certification bodies) may check the appropriate QM system (audit) independent from the industrial sector, certify that a company fulfils the requirements of the appropriate level of QM and in that sense qualifies for ISO/IEC-standard. Accreditation and certification Accreditation is the acknowledgement of competence of a test laboratory. During an audit the validation and neutral assessment of the effectiveness of a quality management system or its parts is made. Certification is the attestation of the conformity with a corresponding standard (such as ISO 9001). The certification itself takes place in four steps. The certificate is issued for 3 years if it is re-acknowledged in yearly review audits [4]. In events of damage companies must be able to produce proof of having done everything to their best available technology. Increasing safety requirements oblige more and more companies to keep record of quality assuring measures for years and to the general responsibility of documentation. The great diversity of requirements of different customers oftentimes causes new, very complex quality management requirements such as the application of special company-internal standards and procedures. Responding to the requirements of the market for standardization and simplification of the certification process quality assurance must nowadays increasingly be demonstrated already during the certification process. Such certifications and accreditations etc. are issued by various institutions.

5 Summary and Future Prospects Being able to perform wide-spread testing requires a lot more than just being able to perform certain tests. This became apparent in a lot of certification and accreditation processes in the past. The diversity of materials to test, e. g. metallic or nonmetallic, and the complexity of the combination of different material types as well as the necessity of the right kind of test considering every demand leads to numerous options in performing destructive tests with static and fatigue loading as well as non-destructive tests. IMA Dresden as an independent accredited test laboratory with a unique position between research and industry offer those complex services for development, qualification and certification testing, developing test concepts and also accepting the responsibility for design and procurement of test equipment. The requirements of aviation business demand continuous improvements. IMA Dresden, therefore, permanently continue certification processes necessary to ensure the provision of high-quality data for aerospace industries for a reliable material selection and detailed design.

Material Selection and Detailed Design

71

References [1] Heida, J.H., Grooteman, F.P.: Airframe Inspection Reliability using Field Inspection Data. In: Proceedings of Airframe Inspection Reliability under Field/Depot. Conditions, Brussels (1998) [2] Franke, R., Ridzewski, J., Goldbach, S.: Werkstoffe, Fügeverfahren und Testmethoden in der Passagierflugzeugentwicklung. Regionaltag der sächsischen Luft- und Raumfahrt 2 (November 2009) [3] Mengedoht, F.-W., Grossmann, A., et al .: Flexibles Qualitätsmanagement. Flexibilisierung von Qualitätsmanagementsystemen nach DIN EN ISO 9000 im Zusammenhang mit simultaner Aufgabenbearbeitung im gesamten Produktionsentstehungsprozess in kleinen und mittleren Unternehmen (KMU) Forschungshefte Forschungskuratorium Maschinenbau e.V., Band 222, pp. S1–S520 (1997) [4] Atzmüller, H.: Ein Gerüst für hohe Ansprüche. Qualitätsmanagement: vierter und letzter Teil; PC Magazin, Heft 3, Seite 50 -51 (1997) [5] http://www.zwick.com/en/news/news-detail/article/ zwick-launches-an-innovative-solution-for-compositestesting.html

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Bombardier Aerospace FSW Demonstrator L.J.J. Kok1, Ken Poston2, and Gary Moore2 1

Bombardier Aerospace, Toronto, Canada 2 Bombardier Aerospace, Belfast, UK

Abstract. The development of Friction Stir Joining as a new class of fastening technologies promises to open new design possibilities. The ability to join large components into a unitized structure or super panels offers the potential of reducing the number of manufacturing processes; leading to reduced build times, all the while achieving weight savings; a natural for reducing aircraft production and operational CO2 footprint. Over five years ago a mutli-site Friction Stir Welding project was launched within Bombardier Aerospace to examine these potentials. Fundamental to the project was the advancement of the technology readiness level from then current levels to near production levels presently on our full family of aircraft products; business, regional and mainline feeder. To that end, butt weld and lap weld were principally investigated.

1 Introduction The ability to join not only homogeneous material joints, but also heterogeneous material joints opens up the design currently only achievable with mechanical fastening or bonding. This concept of unitized build is of course different than that on an Integral Aircraft Structures machined out of monolithic plate materials or fabricated out of integrally stiffened extrusions. [1] As with any novel technology, before any commitment to production is made, it is prudent to have in hand technical knowledge of the candidate process to be exploited. In that vain, Bombardier has undertaken a collaborative FSW project to develop the expertise to exploit the process in-house. To that end a multi-year project leveraged between various Bombardier sites, and industrial collaborators has led to a joint development of capability to industrialize butt and lap joints in an aggressive loading application such as a regional aircraft fuselage. Current Industry Challenges Current aircraft structures design relies on the integration of plurality of materials, structural concepts and cost models to achieve a optimum design that can be coupled with a mission suite to deliver cost and performance targets as per the customer’s expectations. After some 100 years of aircraft construction we are seeing *

Oral presentation.

74

L.J.J. Kok, K. Poston, and G. Moore

definite classes of materials and structures combinations. Whereas in the early days of aircraft numerous small parts were integrated; be it by bonding as seen in design by de Havilland and Fokker [2-6], or mechanical fasteners using rivets and threaded fasteners, as seen in other design in the industry. Part consolidation through machining led to more integral structures as per Douglas and Boeing design in the early 1950’s. As the industry matured the cost of assembly became a more important consideration and as such the goal to reducing parts, labour and material cost have become drivers for the industry. [7] On the material sides, metallics as a general class of materials still provide a cost effective solution, but work is always on going at reducing this side of the cost equation. Mechanical fastening costs have dropped through automation but there is still a prevalence of manual labour involved, even metal bonding is making a resurgence aircraft applications. [8,9] Integral structures are moving forward to due the increasing availability of lower cost machining solutions as opposed to assembling from smaller, simpler detail parts; thus supplanting increasing labour costs. This process works well for applications where a single material is used for the whole component. In recent years composite structures have made great strides but material cost and manufacturing yield lessons have yet to be learned and still remain a challenge in current cost structures. This of course has had the effect of driving further work on optimizing metallic structures. The design problem of joining different alloys has typically led to fastening solutions. If we revisit older designs, we see application of welding in Russian aerospace products but Western design have also relied on it in past; Sud Aviation Caravelle, Handley-Page Herald and Budd Conestoga are notable examples. [10,11] In all some 67 welding techniques are have been documented. [12] Consider current aircraft designs, where typically stringers are of 7xxx series extrusions and skin is of a 2xxx series aluminum, neither alloy particular easy to weld in a homogenous fashion let alone a heterogeneous weld; technology used to bring forth with FSW allows design parodies moving away from conventional fastening techniques, costs and design assemblies.

2 The Friction Stir Weld In a friction stir welded butt weld, as described in the patent [13], a cylindricalshouldered tool, with a profiled threaded/unthreaded probe (nib or pin) is rotated at a constant speed; plunged into the work piece, Figure 1; and fed at a constant traverse rate into the joint line between two pieces of sheet or plate material, which are butted together. The parts have to be clamped rigidly onto a backing bar in a manner that prevents the abutting joint faces from being forced apart. The length of the pin is slightly less than the weld depth required and the tool shoulder should be in intimate contact with the work surface. The pin then moves against the work piece, or vice versa.

Bombardier Aerospace FSW Demonstrator

75

Fig. 1 Friction Stir Weld Process on a butt weld.

Frictional heat is generated between the wear-resistant welding tool shoulder and nib, and the material of the work pieces. This heat, along with the heat generated by the mechanical mixing process and the adiabatic heat within the material, cause the stirred materials to soften without reaching the melting point (hence cited a solid-state process), allowing the traversing of the tool along the weld line in a plasticised tubular shaft of metal. As the pin is moved in the direction of welding, the leading edge of the pin probe, assisted by a special pin profile, forces plasticised material to the trailing edge of the pin probe, while applying a substantial forging force to consolidate the weld metal. The welding of the material is facilitated by severe plastic deformation in the solid state, involving dynamic recrystallization of the base material, thus leaving a solid phase bond between the two parts. [14,15] Developing the Weld Parameters and Pin Tool Using design of experiment techniques, a pin tool that achieved adequate weld strength was developed for a heterogeneous lap weld. The important observation being that FSW parameters for strength did not produce desired fatigue lives; as inferred from the weld morphology in Fig. 2. Down hooking at the interface between the two work pieces on the retreating side of the weld led to substantially inferior fatigue lives, whereas the presence of up hooking led to much improved fatigue properties.

76

L.J.J. Kok, K. Poston, and G. Moore

Fig. 2 FSW Lap Weld DOE Morphology.

Further iterations on this basic pin tool focused on improved weld morphology and fatigue strength, and finally tool wear. As shown in the Figure 3 below, coupon and detail specimen tests confirmed that the weld had the desired fatigue life required for the application. This gave imputes to advance to the next level of sub-component tests and full-scale panels manufacturing.

Fig. 3 Fatigue Life Curves of Test Coupons.

Bombardier Aerospace FSW Demonstrator

77

Component Testing A series of component tests were designed to not only explore the fatigue life of the weld; but also to evaluate crack initiation, crack growth and residual strength. The stiffened panel residual strength test exhibited failure much the same as seen in monolithic machined structures as opposed to those seen in bonded or riveted skin stringer panels.

Fig. 4 Sub Component Fatigue and Residual Strength.

During the course of the test program a small number of rogue cracks were exhibited on the specimens well into the fatigue tests. To this end an investigation to track the source of these observations were made. The availability of the X and Y force on the tool, spindle RPM and traverse speed from the welding head allowed a post weld analysis to be conducted. [16] The challenge for the investigation centred around the fact that the data had been down sampled by a factor of 6 over that which would normally allow for a reasonable evaluation using the FSW Analysis Tool (FSW-AT) program [17]. Hence, the data may suffer not only from lack of information, but also from noise due to aliasing. A Fourier analysis done on the X and Y forces, there was a consistent dominant peak located at about 21 Hz in the frequency spectra of the feedback forces. It was assumed that the aliasing peak was not related to the dynamics of the material flow; because no phase lag between the X force and the Y force was observed; a tell tale sign in accordance with the material flow. Due to the down sampling, the latest evaluation approaches are not applicable to the underlying weld data. Therefore, the stability criterion based method described in [18] was used for evaluating the weld quality.

78

L.J.J. Kok, K. Poston, and G. Moore

Fig. 5 Weld Parameter Tracking.

The discrete Fourier transforms (DFTs) of the X and Y forces. DFTs were scaled by the amplitude of the corresponding dominant peak. To check the stability of the Y force in phase space, the following procedures were performed for each weld data. From the Y force data, 3-second-long timeseries segments were consecutively extracted. Every two adjacent segments had no overlap. The stability of the Y force oscillation was quantified by the Poincaré map method [18], yielding a stability number calculated using Eq. 1.

stability number ∀

! A

(1)

where σ is the standard deviation of Poincaré map of the bottom-to-top piercing, and A is the amplitude of a spindle frequency oscillation. Since the feedback data did not capture the spindle frequency oscillation, the amplitude of the dominant peak located at 21 Hz in the frequency spectrum of the Y force was assigned to A instead. According to the empirical study, a stability number in the range 0 to 1.0 usually corresponds to a defect-free weld as shown in the upper right image of Fig.5, and a stability number greater than 1.0 indicates the presence of volumetric defect, as in the lower right image. The analysis presented in graphics as on the lower left that identified an area of potential bad weld as indicated by the arrows and verified on the test specimen panels above it. Thus the power of the procedure was demonstrated even when presented with extensively down sampled data.

Bombardier Aerospace FSW Demonstrator

79

Full Scale DADT Testing As a final phase of the fatigue and damage tolerance assessment of FSW lap joint, full scale component fuselage panels were manufactured and subject to full scale fatigue and damage tolerance assessment; Fig. 6. The skin stringer panels, as designed have successfully completed two lifetimes of regional aircraft fuselage spectrum loading and residual strength testing. The first lifetime of 80,000 pressure, bending and torque cycles was applied on the as manufactured panels, with only some blend outs as could be expected in a production rework setting. The second lifetime focussed in on monitoring deliberately initiated flaws to assess their affects on the integrity of the structure. The final set of assessments involved extending predetermined damage in order to facilitate multiple bay residual strength tests. These tests showed nothing remarkable with respect to rapid crack extension or loading capability.

Fig. 6 FSW Demonstrator Fuselage Test.

3 Conclusions As demonstrated, the Bombardier FSW demonstrator program developed a set of weld parameters for lap weld joints of heterogeneous material to achieve regional aircraft loading and fatigue lifetimes. A viable NDE of welding parameters was

80

L.J.J. Kok, K. Poston, and G. Moore

demonstrated from processing data much as an SPC process would. Full-scale lap weld sub-component and component test validated design rules and confirmed the absence of any hereto-unexpected failure modes. From the encouraging results to date it is fully expected that a decision to commit FSW lap welds in the near future can be made with little technical risk.

References [1] Pettit, R.G., Wang, J.J., Toh, C.: Validated Feasibility Study of Integrally Stiffened Metallic Fuselage Panels for Reducing Manufacturing Costs. NASA/CR-2000209343, Langley (2004) [2] Pettit, R.G., Wang, J.J., Toh, C.:Papers from the Structural Adhesive Bonding Conference Presented, NASA Marshall Space Flight Center, Clearinghouse for Federal Scientific and Technical Information, March 15-15 (1966) [3] Potter, D.L., et al.: Primary Adhesive Bonded Structure Technology (PABST) Design Handbook for Adhesive Bonding, Report AFFDL-TR-79- 3129, Douglas Aircraft Co., Air Force Flight Development Laboratory (FBA) Air Force Systems Command, WPAFB (1979) [4] Schliekelmann, R.J.: Adhesive Bonding in the Fokker-VFW F-28 Fellowship, National Technical Information Service, Springfield, Virginia (1973) [5] Higgins, A.: Adhesive Bonding of Aircraft Structures. Int. J. Adhesion and Adhesives 20(5), 367–376 (2000) [6] Bishop, John, A.: The History of Redux and the Redux Bonding Process. Int. J. Adhesion and Adhesives 17(4), 287–301 (1997) [7] Munroe, J., Wilkins, K., Gruber, M.: Integral Airframe Structures (IAS): Validated Feasibility Study of Integrally Stiffened Metallic Fuselage Panels for Reducing Manufacturing Costs. NASA/CR-2000-209337,Langley, p. 138 (2000) [8] Ioannou, M., Kok, L., McNeil, N., Fielding, T.: New Material and Fatigue Resistant Aircraft Design. In: Simpson, D. (ed.) Proceedings of the 14th ICAF Symposium, Ottawa, Canada, pp. 127–147. EMAS, United Kingdom (1987) [9] Kok, L.: In: Vermeeren, C. (ed.) Around Glare: a New Aircraft Material in Context, pp. 99–118. Kluwer, Netherlands (2001) [10] Kok, L.: Resistance Spot Welding in the Caravelle Aircraft. Engineering and Aerospace Technology 34(3), 86–87 (1962) [11] Gardner, N.K.: Recent advances in the application of resistance welding to airframe construction. Production Engineers Journal 36(4), 238–252 (1957) [12] Jeffrus, L.F.: Welding: Principles and Applications, 5th edn. Thomson Delmar Learning, p. 8 (2004) [13] Thomas, W.M., Nicholas, E.D., Needham, J.C., Murch, M.G., Temple-Smith, P., Dawes, C.: Friction-stir butt welding, GB Patent No. 9125978.8, International patent application No. PCT/GB92/02203 (1991) [14] Kallee, S.W.: Friction Stir Welding at TWI. The Welding Institute (TWI) (2006), http://www.twi.co.uk/content/fswintro.html [15] Kallee, S.W.: Friction Stir Welding and Processing. In: Rajiv, S., Mishra, M.W. (eds.) ASM International, Materials Park, OH 44073-0002 (2007)

Bombardier Aerospace FSW Demonstrator

81

[16] Kallee, S.W.: Model 493.20 Chassis Product Information, MTS Systems Corporation, Eden Prairie, MI 55344-22908 (2004) [17] Arbegast, W.J.: Using Process Forces as a Statistical Process Control Tool for Friction Stir Welds. In: Jata, K.V., et al. (eds.) Friction Stir Welding and Processing III. TMS (The Minerals, Metals and Materials Society) (2005) [18] Boldsaikhan, E., Corwin, E.M., Logar, A.M., McGough, J., Arbegast, W.J.: Phase Space Analysis of Friction Stir Weld Quality. In: Mishra, R.S. (ed.) Friction Stir Welding and Processing IV, TMS (2007)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Evaluation of Fatigue Crack Growth Behavior in FSW Joint by Experiment, Analysis and Elasto-Plastic FEM T. Okada1, K. Kuwayama2, M. Asakawa3, T. Nakamura1, S. Machida1, S. Fujita2, and H. Terada4 1 2

Japan Aerospace Exploration Agency, Tokyo, Japan Graduate Student, Waseda University, Tokyo, Japan 3 Waseda University, Tokyo, Japan 4 Japan Aerospace Foundation, Tokyo, Japan

Abstract. Elasto-plastic FEM is used to examine the crack opening stress of a 2024-T3 Aluminum alloy sheet and a friction stir welded (FSW) panel. To investigate the effect of plastic deformation around the crack on the crack opening stress, the effects of the following two parameters, the distance between the weld line and the center of the crack starter, and the magnitude of the tensile residual stress on the crack opening stress, are evaluated. The da/dN-ΔK curves and a-N curves for the FSW plate are obtained numerically using an experimental da/dNΔK curve for the base material and the calculated crack opening stress. In addition, the da/dN-ΔK curves where ΔK is evaluated by correction factor and its a-N curves are obtained analytically. Comparison of these numerical results with the results of empirical tests shows while that the FEM result conforms to experimental data, the calculations based on the correction factor show poorer correspondence. These results demonstrate that FEM can reasonably predict the da/dN-ΔK curves and a-N curves for an FSW panel.

1 Introduction Fatigue failure of a structure consists of several stages in which different factors play important roles, and several concepts are suggested [1, 2]. Aircraft primary structures are required to be damage tolerant, and one of the key issue for this requirement is to be able to predict the fatigue crack growth behavior of the structure from discontinuities such as a scratch introduced while drilling a hole. In order to apply the friction stir welding (FSW) process to aircraft structures, much research is focusing on the relationship between the fatigue crack growth properties of a panel and the residual stress in and around the weld line [3-7]. Irving et al. evaluated the fatigue crack growth behavior of an FSW panel in the case of a crack growing within and across the weld line [5]. They also evaluted the crack trajectory when an open hole exists near a fatigue crack growing within the weld line. Lemmen et al. evaluted fatigue crack growth rates and growth direction for an FSW panel in which the weld line was at an angle to the loading direction and showed the relationship between the yield strength and residual stress [6]. At ICAF2009, the authors presented fatigue crack growth test results of an 2024-T3 *

Oral presentation.

84

T. Okada et al.

aluminum alloy FSW panel and showed that the fatigue crack grew perpendicular to the fatigue loading irrespective of the angle between the loading direction and the weld line [7]. An analytical investigation using a correction factor was also conducted and compared to the empirical result. In this paper the following two parameters, distance between the weld line and the center of the crack starter, and the magnitude of the tensile residual stress, are analyzed using the elasto-plastic Finite Element Method (FEM) to estimate their effects on crack opening stress. The analysis considers the effects of residual stresses and external stresses on plastic deformation around the crack and on the crack opening stress. The numerical results are compared with the earlier empirical results. An analytical procedure incorporating a correction factor is also applied using other residual stress distributions and compared with the empirical and FEM results.

2 Experimental Data As explained in our previous paper, the authors carried out fatigue crack growth tests of base material and FSW panels using MT specimens in accordance with ASTM E647-00, and evaluated the crack opening stress by the compliance method during the crack growth test. The specimen geometry is shown in Fig. 1. The distance between the weld line center and the center of the initial notch, λ, are 10 or 40 mm.

Geometry t h

of

initial

Fig. 1 Specimen geometry.

3 Analytical Procedure Several approaches exist to predict fatigue crack growth behavior in a residual stress field. One procedure is based on the crack closure concept, which evaluates the crack opening stress or the stress intensity factor for the crack opening stress

Evaluation of Fatigue Crack Growth Behavior in FSW Joint

85

[8]. Another approach uses a correction factor for the stress intensity factor to take into account the effect of the residual stress field. In our other paper [7], the correction factor proposed by Terada [9] is employed with the residual stress distribution function proposed by Terada [9] and Tada and Paris [10]. The functions for the residual stress distribution are as follows : σy(ξ) = σ0 ( 1 - ξ 2 ) exp( -0.5 ξ 2 ) σy(ξ) = σ0 ( 1 - ξ 2 ) / ( 1 + ξ 4 )

(1) (in reference (2) (in reference

[9]) [10])

Here, σ0 and ξ are the peak residual stress and the normarized coordinate, respectively. However, the obtained da/dn - ΔK curve did not coincide well with the experimental data because these functions overestimate the compressive residual stress around the welding line. The residual stress distribution function proposed by Tada [11] was found to be more accurate for the residual stress distribution of our specimen and is used for this evaluation. σy(ξ) = σ0 ( 1 - ξ 2 ) / ( 1 + ξ 2 ) 2

(3) (in reference

[11])

4 Numerical Procedure A commercial FEM code, ABAQUS, was used for this analysis. From the experiment, the crack direction was found to be perpendicular to the loading direction for all specimens. Therefore, only half of the specimen symmetric to the crack growth path is modeled for the analysis. A two-dimensional analysis is carried out and the plane stress condition is assumed. The analytic model for an FSW panel with λ = 10mm is shown in Fig. 2. For smaller crack lengths a 0.025mm fine mesh

Fig. 2 Finite element model.

86

T. Okada et al.

is used in the crack region, while a 0.05mm mesh is used for longer crack lengths. The mesh sizes on and around the weld line and at the crack region are same for other panels. The multiliner elasto-plastic relationship was measured by experiment and the obtained stress–strain relationship is shown in Fig. 3. Isotropic hardening and the von-Mises yield criterion are assumed for the specimen.

Fig. 3 Stress – strain relationship of 2024-T3 used for FEM analysis.

The analysis evaluates the effects of residual and external stresses on plastic deformation around the crack and on the crack opening stress. LaRue and Daniewicz evaluated the crack opening stress of a plate with pre-existing residual stress [12]. They simulated the crack growth of one finite element length by releasing the crack tip node on each load cycle. The crack opening and crack closure were achieved by changing the constraints of nodes on the crack surface. In our analysis, the longitudinal node at the crack tip is constrained at first. After the maximum load is applied, the constraint is released. Crack closure is modeled by the contact between the crack surface and a rigid body introduced as a contactant. The distribution of residual stress from the welding process is simulated by temperature differences on the welding line. This residual stress is applied before evaluating the crack opening stress.

5 Results and Discussion Comparison with experimental data Figure 4 shows the FEM-computed crack opening stresses and corresponding empirical data. The mesh size on and around the cracked region affects the convergence of the crack opening stress [13]. In this case, a fine 0.025mm mesh is

Evaluation of Fatigue Crack Growth Behavior in FSW Joint

87

employed at the crack region. As shown in Fig. 4, except for smaller crack lengths (less than 10mm), the calculated crack opening stress is comparable to the experimental result.

Fig. 4 Crack opening stress for base material.

For the base material the crack opening stress depends on the stress ratio, and several equations that correlate the stress ratio and the ratio of stress intensity range, U, are proposed. The equation proposed by Schijve[14] is as follows: U = 0.55 + 0.35 R + 0.1 R2

(4)

U = σop / σmax

(5)

The calculated crack opening stress using this equation is 32.56MPa. This value is similar to both the experimental and FEM results. The numerical crack opening stresses of the FSW panels are plotted in Fig. 5. The FEM results indicate that the calculated crack opening stress of an FSW panel starts decreasing when the crack tip reaches the welding line and the minimum crack opening stress is achieved when the crack tip is within the welding line. After the crack tip passes through the welding line, the crack opening stress asymptotically reaches the opening stress of the base material. Additionally, the figure shows that the minimum crack opening stress when the crack tip is within the welding line decreases with increasing distance between the weld line and the center of the notch.

88

T. Okada et al.

Fig. 5 Crack opening stress obtained by elasto-plastic FEM.

The da/dN-ΔK curves for the FSW panel are numerically obtained by using the experimentally determined da/dN-ΔK curve for base material and the calculated crack opening stress for the FSW panel. The da/dN-ΔK curves for λ=10 and 40mm are shown in Fig. 6. The experimental data indicate that the crack lengths for each side can differ due to the effect of residual stress. Then, the crack length from the center of the specimen for each side is used to evaluate the stress intensity factor range and the correction factor proposed by Ishida is employed [15]. Figure 7 shows the a-N curves obtained by integrating the calculated da/dN-ΔK curves for λ=10 and 40mm. The empirical results are also plotted in the figure.

(a) λ=10mm

(b) λ=40mm Fig. 6 da/dN-ΔK curve.

Evaluation of Fatigue Crack Growth Behavior in FSW Joint

(a) λ=10mm

89

(b) λ=40mm Fig. 7 a-N curve.

As shown in these figures, the curves based on the elasto-plastic FEM results coincide well with the experimental data, irrespective of the distance between the center of the weld line and the center of the initial notch. On the other hand, while the da/dN-ΔK curves where ΔK is evaluated by the correction factor match the experimental data relatively well, the calculated a-N curves have some discrepancy. The employed residual stress distribution is closer to the experimental data and indicates lower compressive residual stress around the welding line compared to the distributions used in our previous evaluation. The maximum compressive residual stress for ref. [11] is about half of that for ref. [10]; however, it still overestimates the compressive stress beside the weld line. This would underestimate the crack growth rate where the crack tip is around the weld line as shown in fig. 6, and is a small source of discrepancy between the calculated crack growth rate and the experimental one. The difference becomes more pronounced for the a-N curves. It is much more apparent for λ =40mm. These results indicate that elasto-plastic FEM can adequately estimate the crack opening stress of an FSW panel. By using the calculated crack opening stress, the da/dN-ΔK curves and the a-N curves for an FSW panel can also be reasonably predicted. Effect of peak residual stress amplitude Next, the magnitude of the peak residual stress is set to 295, 250 or 115Mpa and its effect on the crack opening stress is also assessed. Figure 8 shows the crack opening stress for λ = 10mm. As shown in the figure, the peak crack opening stress within the weld line for each case decreases by 13.0, 12.5, and 9.5 Mpa, respectively. This indicates that the crack opening stress for an FSW panel decreases nonlinearly and its change gets dull with the increase of the peak residual stress.

90

T. Okada et al.

Fig. 8 Effect of peak residual stress on crack opening stress.

6 Conclusions The fatigue crack growth behavior of an FSW panel is evaluated by elasto-plastic FEM and an analytical procedure using a correction factor. These are compared with the experimental data. The results indicate that the procedure using the correction factor exhibits some differences with experiment, especially for the a-N curve, because the introduced residual stress distribution overestimates the residual compressive stress. On the other hand, elasto-plastic FEM is able to adequately evaluate the crack opening stress of an FSW panel. It is also considered that the da/dN-ΔK curves and the a-N curves for an FSW panel ca n b e r e aso n ab l y p r ed ic ted using the calculated crack opening stress.

References [1] Hoeppner, D.W.: Damage Tolerance Concepts for Critical Engine Components. Advisory Group for Aerospace Research and Development, Neuilly-sur-Seine, 4/1–4/16(1985) [2] Bellinger, N.C., Shi, G., Prost-Domasky, S., Brooks, C.L.: Proceedings of the ICAF 2007, pp. 126–146 (2007) [3] Bussu, G., Irving, P.E.: Int. J. of Fatigue 25,77–88 (2003) [4] Biallas, G., Dalle-Donne, C., Juricic, C.: Advances in Mechanical Behaviour, Plasticity and Damage, pp. 115–120. Elsevire Science, Kidlington (2000) [5] Irving, P., Ma, Y.E., Zhang, X., Servetti, G., Williams, S., Moore, G., Santos, dos, J., Pacchione, M.: In: Proceedings of the ICAF 2009, pp. 387–405 (2009) [6] Lemmen, H.J.K., Alderliesten, R.C., Benedictus, R.: Proceedings of the ICAF 2009, pp. 619–642 (2009)

Evaluation of Fatigue Crack Growth Behavior in FSW Joint

91

[7] Okada, T., Kuwayama, K., Fujita, S., Asakawa, M., Nakamura, T., Machida, S.: Proceedings of the ICAF, pp. 899–908 (2009) [8] Newman Jr., J.C.: NASA TM 81941 (1981) [9] Terada, H.: Role of Fracture Mechanics in Modern Technology, pp. 899–910. Elsevier Science Publishers B.V, Amsterdam (1987) [10] Tada, H., Paris, P.C.: Int. Journ. of Fracture 21, 279–284 (1983) [11] Tada, H.: The Stress Analysis of Cracks Handbook, 3rd edn., p. 531. ASME Press (2000) [12] LaRue, J.E., Daniewicz, S.R.: Int. J. of Fatigue 29, 508–515 (2007) [13] Kuwayama, K.: Master thesis, Waseda University (2000) [14] Schijve, J.: Eng. Fract. Mech. 14, 461–465 (1981) [15] Ishida, M.: Trans. of the ASME 33, 674–675 (1966)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Recent Development on Bonded Structures G. Delgrange1,2 and J.C. Ehrström1 1

Alcan Centre de Recherche de Voreppe, Voreppe, France 2 Delft University of technology, Delft, The Netherlands

Abstract. Adhesive bonding is a technology that offers significant advantages for the construction of Aluminium aircraft structures. It is used industrially on several commercial aircraft. This assembly technique combines the stiffness of junctions allowing higher compression loads with a multiple load path configuration with very efficient load transfer between the elements, leading to high damage tolerance. Used together with new Aluminium Copper Lithium alloys (AIRWARETM technology), weight savings over 20% on typical wing components are obtained. An overview of tests and modelling activities performed by Alcan internally and in partnership with public laboratories to demonstrate the quantitative impact of bonding on damage tolerance and compression stability is given. These include large stiffened panels tested in compression and fatigue crack growth, tests on laminates, and the support to the development of a debond prediction model at TUDelft. An assessment of the technology is proposed from an industrial point of view. Manufacturing issues, in particular non-destructive quality assurance of adhesive bonds and out of autoclave bonding appear to be the main priorities to address in order to widen the use of this efficient technology that could represent a major advantage for future metallic aircraft to be developed around 2020.

1 Introduction Bonding is a technology that has demonstrated a very high efficiency for metallic structures. From the preliminary trials of Fokker and what became British Aerospace after World War II, until the application of hybrid material such as GLARE on the latest aircraft like A380, the technique has been developed and mastered [1, 2]. Several aircraft are currently flying with bonded joints, but the technology suffers somewhat from a lack of confidence from the designers. The assessment of the bond quality after curing seems to be the current main industrial issue [3], now that durability tests of bonded joints between Aluminium components after 30 years have been published [1]. Alcan, in collaboration with public partners (NLR and TU Delft), has undertaken an effort to further evaluate and demonstrate the structural benefits of *

Oral presentation.

94

G. Delgrange and J.C. Ehrström

bonded joints, with several tests combining adhesive bonding and new Aluminum Copper Lithium alloys AIRWARETM. Large stiffened panels in 2195-T8 alloy have been manufactured with top-hat and J stringers, with both riveted and bonded joints. These panels having an optimised geometry for a given load flow have been tested at NLR in order to evaluate the effect of stringer geometry and joining technique. Large integral and bonded panels in 2027-T3 have been produced and tested with two purposes: -

evaluate the debonded area that could occur around the crack tip when the crack progresses below a bonded stringer show the crack retardation effect in comparison with the same geometry with an integral design and, in addition, demonstrate the effect of crack retardation features that were proposed under the name of “crenellations” [4]

As delamination is a major aspect of bonded structures behaviour, a model was developed with TU Delft, based on energy release rate, to predict delamination. This model, stemming from a model for Fibre Metal Laminates is applied to bonded doublers. Further developments could be used to predict debonding in compression as well. Finally, the manufacturing advantages in terms of forming and size of subcomponents as well as the remaining issues like non destructive testing are highlighted. Even more than for design, only a close collaboration between Original Equipment Manufacturer, Aluminium producers and possibly certification societies can lead to a more widespread application of bonding.

2 Improvement of Load Carrying Capability in Compression The following tests were performed to evaluate the effect of bonding on the buckling load of stiffened panels. The test campaign aims at demonstrating the improvement that an advanced top wing structure can offer versus a conventional top wing using J riveted stringers in 7449-T79, riveted to a 7449-T79 skin. AIRWARETM 2195-T8 is used, offering a 5% density benefit with the same static properties and improved fatigue and damage tolerance properties. Top-hats are used in comparison to J stringer. Both geometries are tested in the riveted and bonded configuration. Experimental details J-shape stringer reinforced panel were tested in compression at NLR. One panel was using riveted attachment and the other one was bonded. The same test was performed with top-hat stringers. Both geometries were optimized to reach a 1300 KN collapse load. As a result, the panel with top-hat stringers is 15% lighter than the panel with J-shape stringers.

Recent Development on Bonded Structures

95

The panels were designed as follows: Stringers and skin are made in 2195 T8 alloy 3 stringers per panel, 134 mm stringer pitch 1100 mm long 536 mm wide (4 x 134 mm) Anti-buckling guided for the two lateral free flanges, both ends of the panel fully clamped. See Figure 1. Aerospace standard were used for the surface treatment and the bonding (BR127 primer, AF-123-2K adhesive cured at 120°C and 1.5 bar).

Compression

Epoxy support

Anti-buckling guide

LVDT

Compression platen

Fig. 1 General picture of the test rig.

Results of compression tests The comparison of the 4 panels is presented on Figure 2.

96

G. Delgrange and J.C. Ehrström

Fig. 2 Comparison of riveted and bonded panels loaded in compression.

The maximum collapse load measured was 1266 for the riveted J-stringer stiffening and 1439 for the bonded J stringer stiffening.. Bonding increased the collapse load by 13.5%. For the top hat, the benefit is increased up to 16% between riveted and bonded.

3 Damage Tolerance Tests Bonding is an efficient solution to improve the damage tolerance of a structure. For example the effect of replacing a bulk skin by a laminate (without any composite intermediate layer) is known from the first publication of Schijve [5], followed by others [6]. This is particularly true when the panel is loaded under spectrum. Also, when bonding is used in preference to traditional techniques for stringer attachment, the damage tolerance of a reinforced structure is increased. This can be explained by the fact that the crack propagates under the stringer without propagating in it (as it would for integral stringers). Then the stringer bridges the crack and slows it down. This is the same phenomenon with riveted attachment, but due to the bond, a delamination will occur preferably to the failure of the stringers and the structure will remain efficient. This principle is also used with bonded straps. To illustrate this principle, two panels in 2027-T3 aluminium were prepared with a crenellated skin and 5 bonded stringers, and compared to integrally machined panels with the same geometry. The mass of the panels was the same. The geometry of the panel (see Figure 3) was defined using optimisation done by simulation. The principal dimensions and characteristics of the panels are: -

5 stringers per panel, the stringer pitch is 160 mm. 720 mm wide panels. 1000 mm long (the tested length is 700 mm). Equivalent thickness is 8.5 mm

Recent Development on Bonded Structures

-

97

Stiffening ratio is 0.23 (stringer section / total section). Total section of each panel is 6125 mm².

Fig. 3 Mean section of the bonded stringers + complete panel.

Aerospace standard were used for the surface treatment and the bonding. Tests were performed at constant amplitude. Figure 4 plots the crack lengths (a1 = left, a2 = right) versus the number of cycles for the two bonded panels. These results are compared to those obtained on integrally machined panels with crenellation. The integral panel with smooth skin is also plotted for comparison. The arrows 1 and 2 correspond respectively to the improvement brought by the crenellation (integral panels with smooth or crenellated skin) and the improvement brought by the bonding (crenellations, integral or bonded stringers). In this particular geometry, this corresponded to: -

10 000 cycles gained with the crenellations 31 000 cycles gained with the bonding

Crack length (mm)

250

1

200

2

150

Integral smooth a1 Integral smooth a2 Integr. cren. a1 Integr. cren. a2 Bonded cren. A5, a1 Bonded cren. A5, a2 Bonded cren. A6, a1 Bonded cren. A6, a2

100 50 0 0

20000

40000

60000

80000

100000

Number of cycles

Fig. 4 Comparison of integrally machined and bonded stringers, with of without crenellations.

98

G. Delgrange and J.C. Ehrström

Crack initiation in stringer occurred after crack passes the stringer (Figure 5) which validates the assumption that the stringer bridges the crack.

Fig. 5 Propagation of the crack under the stringer and initiation of a new crack.

4 Modeling In order to be able to establish the criteria by which bonded solution may be designed in preference to conventional joining technique, the ability to predict the behaviour of such structures is needed. Two different methods can be considered to assess the failure of a bonded interface: analytical solutions and finite element analysis. They are both base on the stress energy release rate (SERR) approach. The SERR is used as a parameter of a failure criterion on a similar way it is done with the stress intensity factor (critical value, Paris law …). Those static and fatigue failure criteria have to be established experimentally. A methodology is proposed in [7]. Analytical approach For the analytical method, the value of the SERR is estimated from the stress in the layers at the delaminated interface. This method for instance has been used to assess the fatigue crack propagation behaviour in Glare sheet (see [8]). Finite element analysis For the finite element analysis, the value of the SERR is computed with the virtual crack closure technique (VCCT) [9]. A study was held in collaboration with TUDelft to study the accuracy of the prediction models. Data that were published at the previous ICAF conference [10] were used as an application example for the finite element method. A test campaign had been performed to investigate the influence of several design solutions for stiffened fuselage panels.

Recent Development on Bonded Structures

99

Fig. 6 Application example for the comparison with FEA – from [10].

Comparison prediction/measurement Several configurations have been experimentally tested by Meneghin, Pacchione, and Vermeer [10] to check the influence parameters like panel alloy, doubler geometry or prepreg reinforcement. Only a few comparisons will be reported here as summarized in Table 1. Table 1 Cases from [10] selected for comparison with FEA.

Case 1

Skin 2524-T3 (1.6mm)

doubler 2524-T3(0.7mm)

comment Reference

1b

2524-T3 (1.6mm)

none

Influence doubler

2

2024-T3 (1.4mm)

2024-T3(0.8mm)

Reference

2b

2024-T3 (1.4mm)

2024-T3(0.6mm)+3 glass pp layers

Influence of hybrid reinforcement

of

the

Figure Figure 7 ; Figure 8 Figure 7 ; Figure 8 Figure 9 ; Figure 10 Figure 9 ; Figure 10

100

G. Delgrange and J.C. Ehrström

Fig. 7 Experimental curves for cases 1 and 1b – from [10].

Fig. 8 Prediction curves for cases 1 and 1b.

Fig. 9 Experimental curves for cases 2 and 2b – from [10].

Recent Development on Bonded Structures

101

Fig. 10 Prediction curves for cases 2 and 2b.

5 Manufacturing Bonding can be applied in three main types of applications: laminates, bonded reinforcement (including repair) and bulk component joining. Advantages: There are numerous reasons to use bonding in preference to other technologies for instance: - Laminating can be an answer to complex shape forming problem (it is also known as self-forming technique). Thinner sheets are easier to form, the curing being done in the mould. - In combination with the laminating technique, splice joint concepts can allow joining two sheets to make a wider one. The material can then be provided to the desired dimension, without being restricted by the industrial constraints (width of the rolling machine for instance). - Bonding can be used to tailor the structure. This is actually the root of the composite materials. For example hybrid material such as Glare is aluminium sheets tailored with glass fibres. Also, solutions such as bonded straps proved to be an efficient local reinforcement. - Riveting is a complex manufacturing process that becomes very complex when the part to joint are made of different materials (drilling through fibre and metal). Bonding can be a solution to such problems. Reliability: One of the main challenges for industry is to find a solution to guarantee that once the bond is cured, there is no defect. Characterisation: Today few adhesive datasheet contains the required characterisation to predict the damage growth of the adhesive for fatigue loading for instance (and to a certain limit, is there any fracture model that can act as a reference?). It

102

G. Delgrange and J.C. Ehrström

means that the damage growth of an adhesive has to be fully characterised. The amount of testing required may be huge (influence of the temperature, moisture, etc…). This problem is rather limited since the adhesives that are currently used are well known but it may limit the application of new products. Equipment: Currently an autoclave is needed to bond strongly enough for primary application. The spread of composite material in the aerospace industry is such that the required equipment is quite common. But the high production rates, that will be required for the new short range aircrafts, will make the use of autoclave undesirable. An out-of-autoclave process that can give a strong bond might be necessary. Recyclability: A metallic structure that is riveted, welded or integral can be easily melted at the end of its life in a low environmental impact process. The use of reinforcement with glass fibre in a polymeric resin might make the final product much more difficult to recycle.

6 Discussion The first advantage of bonding shown is the increased resistance to buckling when loaded in compression. When a panel reinforced by stringers is loaded in compression, there is a load transfer between the panel and the stringers. When the stringers are riveted, the load transfer is punctual, allowing some freedom. A bonded attachment makes the load transfer continuous along the stringer and thus is much stiffer. Combined with stringer shape (top-hat) and new alloys a weight saving of about 25% is achievable versus a conventional structure having J riveted stringers, with stringers and skin in 7449-T79: -

bonding increases the buckling load by 15% which typically can be translated into approximately 5% weight saving top-hats offer 15% weight saving 2195-T8 offers 5% weight saving versus 7449-T79 thanks to reduced density (at the same strength and improved fatigue and damage tolerance properties).

It as also been demonstrated that bonding is a good solution to increase the damage tolerance of a structure. A bonded stringer will be protected from the propagation of the crack in it (when compared to integral reinforcement). And also it will bridge the crack and slow down the crack. This benefit can be combined with the benefit brought by the crenellation. During the test, it has been demonstrated: - 10 000 cycles gained with the crenellations which as been estimated equivalent to a 14% weight saving. - 31 000 cycles gained with the bonding. This corresponds to a weight benefit between 13% and 29% depending on the critical crack length we consider. When compared to riveted solutions, the main advantage of bonding (in addition to the ability to delaminate) is the absence of stress concentration point, and thus

Recent Development on Bonded Structures

103

of early crack initiation. However, a difference in thermal coefficient between bonded materials leads to internal stresses after curing and cooling. This is observed in the case of aluminium bonded to glass fibre for example. The models that were developed to predict the delamination for complex type of loading are promising - Experimental curves 1 and 1b show a difference only in the zone very close to the doubler. This is exactly what the prediction show as well. The measurement shows that the crack was in the doubler zone at around 8000 cycles. The prediction gives 5000 cycles. Logically, once the doubler is broker, the two set of test show the same trend. This is noticeable on both measurement and prediction. - For case 2 and 2b, in the prediction plot there is much less difference between the two cases before the failure than there is actually in the experiment. It is to be noted in that in the experiment the doublers that were bonded under the stringers were different for panel 2 and 2b (respectively 0.8 mm and 0.6mm). This was not taken into account in the model and can explain the difference we observe. - If we focus on 2b, the comparison between the prediction and the measurement gives the same kind of conclusions we had for the others cases: before the crack reached the doubler, the prediction is good (around 2000 cycles). Under the doubler, the time period is underestimated by the model (4000 cycles for the model, 6000 for the measurement) but then both cases reach the stringer at T=12000 cycles.

7 Conclusions An overview of Alcan testing and modelling contribution in the domain of adhesive bonding of aircraft structure is proposed. The following conclusions can be drawn at this stage: -

-

-

the technology is not new in the industry: it has been developed by airframers since 1945 and more than 30 years old joints have shown a good performance structural advantages are confirmed by the present tests: high joint stiffness increases buckling resistance of stiffened panels, while maintaining a multiple load path. The bridging effect of stringers is even higher than with riveting, and it is demonstrated that very little delamination ahead of the crack tip occurs. The stringers remained intact while the crack propagates. Modelling of fatigue crack growth and delamination based on strain energy release rate is promising. Further developments should include the prediction of delamination in compression.

Further collaboration with public labs and OEM is required to increase the use of bonding that would enable the manufacture of very weight efficient structures at the 2020 horizon.

104

G. Delgrange and J.C. Ehrström

Acknowledgments Part of this work was performed at Delft University of Technology. The Authors would like to thanks R. Alderliesten for his knowledgeable assistance and for supervising the modelling work. The authors would also like to thanks NLR which performed the bonding process of the panels and some testing as well.

References [1] Higgins, A.: Int. J. of Adhesion and Adhesives 20, 367 (2000) [2] Vlot, A.: Glare, History of The Development of a New Aircraft Material. Kluwer Academic Publishers, Dordrecht (2001) [3] Furfari, D.: Bridging the Gap between Theory and Operational Practice. In: Bos, M. (ed.) Proceedings of the 24st ICAF Symposium, vol. I, pp. 1305–1319. Springer, Rotterdam [4] Ehrström, J.-C., Muzzolini, R., Arsène, S., Van der Veen, S.: In: Dalle, C. (ed.) Proceedings of the 23rd ICAF Symposium, vol. I, pp. 79–90. DGLR Publ., Hamburg (2005) [5] Schijve, J., van Lipzig, H.T.M., van Gestel, G.F.J., Hoeymakers, A.H.W.: Fatigue properties of adhesive-bonded laminated sheet material of aluminium alloys. Eng. Frac. Mech. 12(4), 561 (1979) [6] Morkovkine, A.V., Petoukhov, Y.V.: Techno Metall. (4) (2003) ISBN 620.191.33 [7] Delgrange, G.: Bridging the Gap between Theory and Operational Practice. In: Bos, M. (ed.) Proceedings of the 24st ICAF Symposium, vol. I, pp. 1305–1319. Springer, Rotterdam (2009) [8] Alderliesten, R.C.: Analytical prediction model for fatigue crack propagation and delamination growth in Glare. Internatioanl Journal of Fatigue 29(4), 628 (2005) [9] Krueger, R.: Virtual crack closure technique: History, approach and applications. Applied Mechanics Reviews 57(2), 109 (2004) [10] Meneghin, I., Pacchione, M., Vermeer, P.: Bridging the Gap between Theory and Operational Practice. In: Bos, M. (ed.) Proceedings of the 24st ICAF Symposium, vol. I, pp. 1305–1319. Springer, Rotterdam (2009)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Recent Advancements in Thin-Walled Hybrid Structural Technologies for Damage Tolerant Aircraft Fuselage Applications R.C. Alderliesten1, C.D. Rans1, Th. Beumler2, and R. Benedictus1 1

Delft University of Technology, Delft, The Netherlands 2 Airbus, Hamburg, Germany

Abstract. This paper presents and discusses some of the recent advancements that have been achieved in the development of the FML concept for thin single-aisle fuselage structures. It is shown that the inherent damage tolerant nature of this concept can be further exploited for these structural applications.

1 Introduction The concept of Fibre Metal Laminates (FMLs) has been developed at Delft University of Technology (TUDelft) as a robust and damage tolerant material concept for primary aircraft structures. In a joint effort between Airbus, TU Delft and Dutch partners the concept has been successfully developed towards major skin material application on the Airbus A380 fuselage, and to impact resistant leading edges of the horizontal and vertical tail planes. A key characteristic of the FML concept is the high resistance against (fatigue) cracking and impact, and a significant residual strength in case of (fatigue or impact) damage. This damage tolerance enables designing structures with high allowable stresses, leading to significant weight reduction at a high safety level. Compared to the A380, single aisle aircraft imply different requirements to the hybrid concept. Key parameters to be considered are the significantly higher Design Service Goals (DSG) and inspection intervals of short range aircraft, in conjunction with minimum thickness related design criteria due to the relatively low mechanical- and differential pressure loads. Figure 1 provides an overview about those design criteria, which are mainly load- and stress dependent. For the shown design and material combination, the majority of the investigated fuselage skin panels is obviously designed by Fatigue and Damage Tolerance (DT) criteria, i.e. to meet the target inspection interval (low crack propagation rates), the target Large Damage Capability (LDC, to be provided at limit load) and option to design permanent riveted repair (the influence of bonded repairs on thin walled structures is an interesting topic as well, but not discussed in this paper). *

Oral presentation.

106

R.C. Alderliesten et al.

Fig. 1 Typical design criteria for single aisle pressurized fuselage shell.

Less load- but more material dependent is a set of design criteria, which might be summarized as contributors to the robustness of the structure, beyond the items discussed above. These criteria consider a high utilization of the aircraft because of a level of structural damage capability, which does not lead to the immediate grounding of the vehicle in order to perform a repair. Items to be considered are the easy detectability of any kind of damages (i.e. dents, lightning strikes); limited strength degradation due to damage, in order to provide a comfortable period before required repair; and the possibility of simple, permanent repairs. Unfortunately these design criteria can not be captured as easy as structural weight dependent to load-, design- and material related design values. However, DT and structural robustness is the domain for which FMLs have been invented. It is therefore worth to investigate the specific challenges of thin-walled fuselage structures made of FMLs.

2 Thin-Walled Design Options Trent setters as the B737 and the A320 fly minimum skin thickness in the fuselage skin “pockets” in the order of 0.9mm to 1.2mm, at hoop stress levels in the order of 80 MPa to 100 Mpa during cruise, i.e. at load case 1g plus the differential pressure (fatigue case). As indicated in Figure 1, most of the fuselage skins can be designed by hoop stress. While that is dominating the front fuselage, almost equal tensile fatigue stress levels have to be considered in flight direction in the upper rear fuselage. Significant static compression cases in the upper front- and lower rear fuselage have to be managed by the stringer, which are preferably bonded to the skin in order to provide large load carrying (skin) width for the compression cases and a high tensile residual strength in the LDC case. Metal foils made of either 2024 alloy or 7475 alloy are available in 0.3mm and 0.4mm thickness, clad or non-clad. Larger foil thickness can be manufactured, but its application is restricted by the self-forming technique and the fatigue properties of the bonded splice. 0.2mm foils can be manufactured as well but are a challenge for handling. Most likely applicable thin FMLs are therefore:

Recent Advancements in Thin-Walled Hybrid Structural Technologies

(1) (2) (3)

107

FML3-2/1-0.3 (t = 0.85mm, ρ = 2.54kg/dm3 if made with 2024) FML4-2/1-0.4 (t = 1.18mm, ρ = 2.52kg/dm3 if made with 2024) FML5-2/1-0.4 (t = 1.30mm; ρ = 2.46kg/dm3 if made with 2024)

All three solutions consist of 2 aluminium foils, but with different fibre volumes being sandwiched. The 2/1 lay-up principle is illustrated in Figure 2. An option to increase the thickness of the lay-up is the inclusion of an interlaminar doubler, as indicated on the left side of this figure.

Interlaminar doubler

Fig. 2 FML in 2/1 lay-up with interlaminar doubler.

Interlaminar doubler would be required locally in combination with FML3-2/10.3 in order to provide enough thickness for fastener installation. Taking into account a countersunk depth of <1.0mm for a 4.0mm flush head fastener, the other two options can fulfil the countersunk criterion without local reinforcement, but local load transfer between skin and fastener, e.g. at the skin/frame joint will nevertheless require higher skin thickness. Two minimum thickness fuselage panel design principles have been developed and justified by panel tests, supported by elementary- and coupon tests. Details will be reviewed in the following sections.

3 Impact Resistance Keeping structural integrity under control in case of impacts is an important maintenance item. For this purpose aircraft manufacturers provide Structural Repair Manuals (SRM), which define the required actions in case of a finding, i.e. of a detected plastic deformation. Here one distinguishes between dents with and without cracks. In the former case, repairs are required at once or within a pre-defined number of flights; small deformations do not require a repair at all. Note that in any case it is required to identify possible damages in back-up structures in order to assess the consequence of damage on strength correctly. Unfortunately, large number of variables must be considered for the simulation of impacts on in-service aircraft, i.e. the location of damage in relation to stiffeners, the shape of the impacting part, its angle of attack and impact velocity. The following information is limited to impacts in the centre of a “far field” skin between frames and stringers, introduced with a gas gun and 15mm diameter impact head.

108

R.C. Alderliesten et al.

In frame of an extensive coupon impact test program, performed with various materials of different thickness at TUDelft, 2024 has been compared with Standard-GLARE4 and 1441-FML4, all with a nominal thickness of 1.2mm. Some results are summarized in Figure 3. The failure modes considered are • • •

Dent without crack Superficial crack (on non-impacted side) or through crack <0.3mm Through crack >0.3mm

Fig. 3 Damages related to impact energy.

Concerning the 2024 benchmark, first cracks appear on the impact surface at a plastic deformation above 5mm. The corresponding impact energy is in the order of 17 J to 22 J for this type of specimens. Standard-GLARE4 showed first cracks above 18 J, first cracks on the front side of 1441-FML4 occurred above 15 J. The fibre bridging in the laminate obviously balances the low impact resistance of the 1441 aluminium lithium alloy to some extent. Summarizing, and related to the coupon results, a repair of 1.2mm Standard-GLARE is not required earlier than in monolithic 2024 of the same thickness. FML made with aluminium lithium is more sensitive to impact damages and would require repairs at lower impact energies. This information is important for the selection of a technology for a future single aisle structure. Additional impact investigations have been performed on 2/1 lay-ups in a thickness range of 0.9-1.05mm. Figure 4 indicates the different energies required to create similar permanent deformations in both, a small coupon and a stiffened panel. Obviously, the detailed failure mode is dependent on the plastic deformation, not on the impact energy. First cracks appear in a 1.2mm stiffened 2024 panel at about four times the impact energy level observed in the coupon, at similar dent depth. A similar dent depth is achieved in 0.9mm Standard-GLARE at approximately 45 Joule impact energy, see Figure 4. As expected, stiffened FML panels with skin thicknesses below 1.0mm are more sensitive to impact cracking than 1.2mm monolithic 2024; their application would potentially require earlier repairs.

Recent Advancements in Thin-Walled Hybrid Structural Technologies

109

8 2024-T35 - reference (1.2mm) Glare 3-2/1-0.3 (0.9mm) 1441-FML 3-2/1-0.35 (1.05mm) Glare 3-2/1-0.3 1441-FML 3-2/1-0.35

dent depth [mm]

7

6

5

4

3

2 10

20

30 Impact energy [J]

40

50

Fig. 4 Dent depths in coupons and on stiffened panels.

Similar impact comparisons as shown in Figure 4 have been conducted with 1.2mm 2024- and Standard-GLARE4 panels, calibrated with full scale impact results. First cracks in GLARE4 have been detected above 100 J. The lower the structural stiffness – the lower the portion of energy transferred into plastic deformation.

4 Application of Waffle Plate Concepts The thin FML fuselage panel concept has been investigated with a large fuselage panel tested in bi-axial loading including internal pressurization. The geometry and dimensions were selected to be representative for a single-aisle aircraft fuselage. To investigate multiple advanced material solutions in the same tests, the panel was designed symmetrically to allow direct comparison between 1441-FML and GLARE, between overlap splices and welded aluminium sheets, and between different fibres. The panel utilizes the concept illustrated in Figure 2, which implies interlaminar application of a 0.4mm thin waffle plate to provide sufficient thickness underneath stringers and frames. The panel was manufactured according to the state of the art industrial FML environment in accordance with Airbus requirements. In preparation of the tests, the determination of the most appropriate load introduction was supported by detailed finite elements analysis to assure that specified stress levels were obtained in skin, stringers and frame. The tests consisted of a fatigue test simulating one DSG, after which artificial impact damages were created in the panel to be subsequently tested for another half DSG. Finally, saw-cuts were created after which another half DSG was applied. Figure 5 illustrates the observed crack growth in both 1441-FML and GLARE.

110

R.C. Alderliesten et al.

80

Standard-Glare 1441-FML

70 60

a [mm]

50 40 30 20 10 0 45

47.5

50

52.5 N [kflights]

55

57.5

60

Fig. 5 Comparison between crack growth in 1441-FML and GLARE.

5 Evaluation of FML4-2/1-0.4 1.2mm minimum thickness panels have been designed and manufactured in both, Standard-GLARE and 1441-FML. Key feature is an alternate circumferential bonded splice concept as shown in Figure 6. Because of the maximum aluminium foil width of 1500mm, preferably 1300mm (lower price), and the required 2/1 layup, the alternate circumferential splice at any frame position is the obvious solution. At any second frame position the outer and respectively the inner sheet is spliced. Frames and clips in the test panels have been standard A320 parts. The bonded stringers have been made of aluminium 2196. Tasks and advantages of the circumferential splices are: • • • • •

providing enough thickness to accommodate countersunk heads of clip rivets providing damage arrest capability (LDC, to be confirmed) providing a location of increased thickness / decreased stress to accommodate fastener for riveted repairs providing the option to manufacture wide panels / avoid riveted longitudinal joints (maximum size assumed by manufacturing specialists: 180° shell) provide less complicated circumferential joint design

Recent Advancements in Thin-Walled Hybrid Structural Technologies

111

Fig. 6 Design principle GLARE4A-2/1-0.4 with circumferential splice.

The disadvantages of the circumferential splices are: • •

more complex riveted longitudinal splice design required at edges application in double curved locations might require post-stretching of aluminium foils

The crack propagation capability of GLARE3-2/1-0.3 is discussed in the previous chapter, comfortable inspection intervals have been justified by test. At similar applied loads, the crack propagation rates are even lower in the thicker FML. The justification of LDC against hoop stress required two tests because of the alternating different splice types. It was the target to justify a skin damage above broken frame at 1.15∆p, as demanded by FAR 25.571(b). Predictions using GLARE4 R-curves from [1] lead to the engineering guess, that a crack above a broken frame should become unstable at approximately 2a=500mm. It was decided to first introduce a saw cut in aircraft length direction as shown in Figure 6, i.e. an unstable crack would run towards the location of the inner sheet splices. Indeed, below 2a = 500mm, no unstable crack growth could be observed. At a certain stage, the saw cut was extended to 2a=533mm (equals one frame bay) and internal pressure was applied again. At 0.91∆p the crack extended unstable to 2a=592 mm and the internal pressure collapsed because of test system errors. However, in a second trial the differential pressure was increased to 715 mbar, the crack extended towards the internal splices and was arrested as expected. The two-bay crack was repaired and the same procedure was repeated successfully with the same panel for the outer sheet splices. In both cases, the crack was arrested at or in front of the bonded aluminium sheet overlap, see Figure 7 and Figure 8. The LDC standard of the A380 GLARE panels has been demonstrated with 1.2mm thin FML shells, offering more than 13% weight saving (for the skin, only) compared with a monolithic 2024 design.

112

R.C. Alderliesten et al.

Fig. 7 Crack arrest at internal splice.

Fig. 8 Crack arrest at external splice.

6 Generic Evaluation oF 1441-FMLS The application of the 1441 alloy in the FML concept was evaluated by testing 1441-FML coupons under static and fatigue conditions. An important result of the tensile tests was that the mechanical properties of the 1441-FML are in general better compared to the Standard GLARE. The 1441-FML stiffness was observed to be 5% higher and the yield strength even 20% higher than GLARE.

Recent Advancements in Thin-Walled Hybrid Structural Technologies

113

Although the required stiffness improvement can be achieved with the 1441 alloy, the implications with respect to damage tolerance had to be considered. For this reason fatigue crack growth tests on small coupons and large flat panels were performed and compared with the data and predictions for Standard GLARE. An example of the correlation between both FMLs is given in Figure 9. 120 MPa

155

100 MPa

80 MPa

135

a(mm)

115

95

75

55

Standard Glare 1441-FML

35 0

50000

100000

150000

200000

250000

300000

350000

Cycles

Fig. 9 Comparison between constant amplitude fatigue crack growth for Standard GLARE and 1441-FML3-3/2-0.4 at R = 0.1.

The case illustrated in this figure consists of cracks emanating from a large saw-cut (2a0=75mm). Key observation in this figure is that at lower maximum stress levels (i.e. below 100 MPa) the crack growth for both Standard GLARE and 1441-FML are similar. However, at higher stress levels, the crack growth resistance of the 1441-FML appeared to be lower compared to Standard GLARE. With the coupon tests it was further observed that not only at high crack growth rates, but also at the very low crack growth rate regime, the crack resistance of the 1441 alloy is less than 2024.

7 Reparability of Thin-Walled Panels The skin surfaces of the GLARE4-2/1 panels provide a constant thickness of 1.2mm between the splices. No pad-up was considered between skin and stringer. The measured hoop stress in the “pocket” centre is 90 Mpa at 622mbar (average between internal and external surface), related to the modulus of StandardGLARE4A-2/1-0.4. It must be the target to provide a permanent riveted repair for a typical Single Aisle DSG, let say 60000 flights. Various repair design solutions are under experimental validation at TUDelft, all using 1441-FML4A-2/1-0.4.

114

R.C. Alderliesten et al.

Diversifications are external and internal repair patches (flush repair), 2024 patches and 1441-FML patches, three and four rivet rows, NAS1097 fastener and Hi-Loks. It is expected to finish the investigations at the end of the year 2011. A typical test specimen design is shown in Figure 10. It is expected that the repair solution with an external 2024 patch, jointed with 4 mm Hi-Loks, provides the minimum fatigue life. A crack initiation prediction for this repair design is shown in Figure 11 for 100 MSD scenarios and included all relevant scatter and safety factors. The crack initiation life is already calibrated by first coupon test results. With 50% probability, cracks nucleate after 16000 cycles (= flights, factored result). Figure 12 shows the entire strength justification for one of the scenarios, which is closest to the average. The lead crack is detectable by NDI after 32000 flights (detectable crack length 5mm from the hole edge, diagram starts from hole centre). At this point the residual strength is still significant above ultimate load capability and inspections would be not required. This is also true at 60000 flights, the considered DSG for the here discussed example.

Fig. 10 Typical repair joint coupon specimen.

However, almost all cracks in the mating aluminium foil of the skin would have linked up. A recommended inspection threshold would be 42000 flights, when the average crack would be detectable. According to these predictions, a simple external patch made of 1.2mm 2024 with 3 rivet rows and 4.0mm Hi-Loks is not expected to qualify for a permanent repair covering 60000 flights. Note that during the entire range of investigation there is no crack initiation in the second aluminium layer. Crack propagation rates and residual strength of this joint will be validated until the end of 2011. Predictions made for repair solutions with 4 rivet rows and FML patches estimate permanent repairs without inspections.

Recent Advancements in Thin-Walled Hybrid Structural Technologies

115

Fig. 11 Crack initiation prediction for FML skin with external 2024 patch.

Fig. 12 F&DT strength justification for FML skin with external 2024 patch.

8 Further Exploiting the Hybrid Concept Several studies have been performed in the past on the application of high stiffness fibres in the FML to obtain FML stiffness similar to the stiffness of monolithic aluminium. Key observation in these studies is that the limited strain to failure reduces the residual strength of FMLs. To exploit the composite nature of the FML concept a study has been performed by TU Delft to investigate whether local application of high stiffness fibres would sufficiently reduce the stiffness

116

R.C. Alderliesten et al.

mismatch between FML skin and aluminium stringers, while still improving damage tolerance (crack growth and residual strength). The investigation was performed on flat panels and stiffened panels, for which the concept is illustrated in Figure 13.

3 mm 3 mm

Fig. 13 Concept to apply high stiffness fibres underneath the stiffeners.

Compared to the GLARE panels, the fatigue initiation life was increased by 20%, while the crack growth life from 2a=75 mm to 250 mm almost doubled. The residual strength at that final crack length was about 22% higher compared to the same panel without the high stiffness fibre straps.

9 Conclusions A civil passenger transport aircraft fuselage is mainly designed by fatigue and damage tolerance, which makes the consideration of FML structures obvious. GLARE skin panels with a minimum thickness of 1.6 mm are successfully applied on the A380, demonstrating high structural performance and robustness in terms of high damage resistance and relaxed maintenance tasks required. In the footprints of Bombardier, an extensive investigation performed by TUDelft, Airbus and VIAM demonstrated similar characteristics as validated with the A380 GLARE panels for thin FML structures, i.e. made with 0.85 mm GLARE3 and 1.2 mm GLARE4A. As usual, the combination of particular design features and joining techniques (e.g. the bonded stringer) with the characteristics of FML materials is required to create and validate an efficient structure, which satisfies all Airworthiness Requirements at minimum structural weight and minimum maintenance requirements. In the present case, for minimum thickness fuselage panels, the weight saving potential of stiffened FML panels (skin + frames + clips + stringer) is in the order of 12% to 15% compared with conventional 2024 structures of the 1965 to 1985 design generation. Because the electrical conductivity of FML is almost similar to the one of monolithic aluminium, the structural weight equals the system weight. Note that the fire resistance of FML structures (>15 minutes burn

Recent Advancements in Thin-Walled Hybrid Structural Technologies

117

through resistance justified in frame of A380 development) provides additional weight saving opportunities for the lower fuselage panels against FAR 26.856 (b). Investigations Single Aisle shaped fuselage structures have been performed with the so-called Standard-GLARE, i.e. FMLs containing 2024 aluminium foils, as well as with FMLs containing 1441 aluminium foils. The latter provide a density advantage of 6.5 % for the metal volume fraction, at almost similar material properties. Interesting enough, the disadvantages of aluminium lithium alloys compared with 2024, i.e. lower ductility, lower fracture toughness and lower strain to failure, can be compensated by the glass fibres in a FML

References [1] de Vries, T.J.: Blunt and sharp notch behaviour of Glare laminates, PhD dissertation, Delft University of Technology (2001) [2] Lucas, C.: Static, fatigue and damage tolerance behaviour of the Fibre Metal Laminate GLARE based on the new aluminium lithium alloy: 1441-T11, Thesis, Delft University of Technology (2010)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Fatigue Life Assessment for Composite Structure Andrew Makeev and Yuri Nikishkov Georgia Institute of Technology, Atlanta, GA (404)894-8166 [email protected]

Abstract. This work presents some of the most recent advances in the technologies which could enable accurate assessment of useful life for composite aircraft fatigue-critical, flight-critical components and structure. Such technology advances include: (1) nondestructive subsurface measurement shift from just detection of defects to three-dimensional measurement of defect location and size; (2) material characterization methods ability to generate 3D material allowables at minimum time and cost; and (3) fatigue structural analysis techniques ability to capture multiple damage modes and their interaction. The authors summarize their recent results in all three subjects.

1 Background Currently, composite designs adopt metal design philosophy and use the same factor of safety of 1.5 to determine the ultimate design load from the limit load even though composite parts are inherently more susceptible to variations in manufacturing processes than metal parts. In addition to material variation in the resin content, bulk factor, and fiber alignment, part fabrication process variations such as operator skill, tooling setup, humidity fluctuation, equipment control, etc. are common causes that contribute to variation in part quality. Consequently, the increased sensitivity of composite part quality to material and process variations lowers production yields. In order to increase production yields, heavy burden is placed on composite manufacturing communities to understand and control their processes. Production yields of greater than 90% remain a “hit-and-miss” target. The effects of manufacturing process parameters on structural strength, durability and damage tolerance are not well understood. In particular, the effects of inadequate design method and manufacturing process used to produce carbon/epoxy and glass/epoxy composite aircraft fatigue-critical, flight-critical components manifest themselves as defects such as wrinkles and porosity/voids, and such defects impact the performance and the service life of these components. Manufacturing defects can severely deteriorate the matrix-dominated properties resulting in degraded strength and fatigue structural performance of composites. *

Oral presentation.

120

A. Makeev and Y. Nikishkov

Although it might not be practical to eliminate all the defects in a composite part, it is possible to avoid assumptions of the worst-case scenario and address improved part durability and damage tolerance once the defects and their effects are captured. Advanced structural methods that account for manufacturing defects in composite parts are needed to enable accurate assessment of their capability and useful life and enhance current design and maintenance practices. This work presents some of most recent advances in the technologies which could enable accurate assessment of useful life for composite aircraft fatiguecritical, flight-critical components and structure. Such technology advances include: (1) nondestructive subsurface measurement shift from just detection of defects to accurate three-dimensional measurement of defect location and size; (2) ability of material characterization methods to generate 3D material allowables at minimum time and cost; and (3) fatigue structural analysis techniques ability to capture multiple damage modes and their interaction. The authors summarize their recent results in all three subjects.

2 Subsurface Measurement Computed tomography (CT) is a proven nondestructive evaluation (NDE) technology enabling three-dimensional measurement of manufacturing defects including wrinkles and porosity/voids [1]. Figure 1 shows the operation basics for a modern industrial CT system. The system uses three major components: an x-ray tube, x-ray detectors, and a rotational stage. New generation micro-focus x-ray tubes and amorphous silicon flat panel area detectors offer micron-scale resolution which cannot be matched by the other available NDE methods. A CT scan typically includes a series of 2D x-ray images of the object rotating 360 degrees (complete rotation) or 180 degrees (half rotation). CT systems acquire between 120 and 3600 digital images, the image size 3 to 10 megapixels, depending on the desired resolution. Once the scan is complete, CT reconstruction algorithms are used to generate the 3D volumetric information. Due to recent advancement in the fast CT reconstruction software and computer hardware, the reconstruction process can be accelerated to a few minutes. It is possible to manipulate the volume in real time, e.g., slice anywhere inside the object, after the CT reconstruction. A recent feasibility assessment demonstrated the ability to detect wrinkles and porosity/voids in composites with a Micro-CT system. North Star Imaging (NSI) M5000CT industrial CT system with a 225 KV Microfocus X-ray tube and Varian 4030E series Flat Panel detector was utilized. Figure 2 shows examples of CT volume slices with multiple manufacturing defects and structural damage in glass/epoxy and carbon/epoxy tape composite laminates.

Fatigue Life Assessment for Composite Structure

121

Fig. 1 Major components and operation principle for a modern industrial CT system [1].

Wrinkles and Voids

Matrix Cracks and Delaminations

De laminations

Matrix Cracks

Fig. 2 Micro-focus CT volume slices show the level of subsurface detail resolution including wrinkles and voids in 30-ply thick glass/epoxy tape laminate; and matrix cracks and delaminations in 16-ply thick carbon/epoxy tape laminate.

122

A. Makeev and Y. Nikishkov

Figure 3 shows CT reconstruction results for a Glass/Epoxy laminate structural detail representative of a helicopter tail rotor flexbeam up to 1.5 inches thick.

Fig. 3 A CT volume slice shows wrinkles and voids in a Glass/Epoxy laminate structural detail representative of a helicopter tail rotor tapered flexbeam section.

Figure 4 shows a Glass/Epoxy main rotor blade spar section near the root end. Although the wrinkles are located at the surface, wrinkle measurement conducted by visual inspection and aided by rudimentary measurement tools such as a ruler or a caliper could result in unacceptable measurement variation and affect the objectivity at making disposition decision of the affected part. Figure 4 shows that CT data provide high spatial resolution and high clarity wrinkle images essential for repeatable and reproducible in-plane and out-of-plane characterization [1]. Figures 5 and 6 illustrate CT reconstruction results for another composite laminate main rotor blade spar structure. Wrinkles and voids around a molded-in metallic section are clearly shown although large difference in the densities between the metal and the composite causes scattering artifact during the reconstruction. Although the industrial micro-CT system used in the feasibility assessment demonstrated that the micro-CT technology is well-suited for characterization of wrinkles and porosity/voids in composite structures, the system configuration is not suitable for the inspection of long composite parts such as wing spars, rotor blades, yokes, flexbeams, etc. It is not practical to manipulate such parts in the CT scanner without high risk of deforming them. The parts must be stationary and the x-ray tube and detectors must move around the parts. Such arrangement is similar to the medical CT systems.

Fatigue Life Assessment for Composite Structure

123

Fiberglass rotor spar structure

Surface Wrinkle ~0.04” Deep

1

2

Surface Wrinkle ~0.12” Deep

Fig. 4 CT data show the material structure and surface wrinkles for a fiberglass main rotor spar root section.

A successful CT system for efficient and accurate measurement of manufacturing defects in large and thick composite parts must combine the industrial microCT resolution ability and the medical CT ability to scan large objects. And the measurement and characterization of defects must be fully-automated. The defect measurements must also be converted into finite element-based failure models to assess the effects of the defects on structural performance. Coupling accurate measurements and rigorous failure models will improve overly conservative part rejection criteria and enable lower scrap rates in the flight-critical, fatigue-critical composite parts.

124

A. Makeev and Y. Nikishkov

Fig. 5 Main rotor blade spar structure: wrinkles and voids around a molded-in metallic section.

Fig. 6 Voids in the main rotor blade spar structure.

CT can become a powerful diagnostic tool and enable the ability to predict part capability and remaining useful life. The quality of micro-focus CT detail reconstruction allows for the ability to automate the defect interpretation. The automated recognition of the defects is essential for use of the defect data in the durability and damage tolerance models based on accurate three-dimensional geometric characterization. Manual defect measurement is time consuming and also could result in unacceptable variability affecting part disposition decisions as mentioned earlier.

Fatigue Life Assessment for Composite Structure

125

3 Material Characterization Accurate three-dimensional stress-strain constitutive properties are essential for understanding of complex deformation and failure mechanisms for materials with highly anisotropic mechanical properties. Among such materials, glass-fiber and carbon-fiber reinforced polymer-matrix composites play a critical role in advanced structural designs [2]. A large number of different methods and specimen types currently required to generate three-dimensional allowables for structural design slow down the material characterization [3]. Also, some of the material constitutive properties are never measured due to prohibitive cost of the specimens used for the material characterization. Recent work [4, 5] shows that a simple short-beam shear (SBS) test, coupled with the Digital Image Correlation full-field deformation measurement and simple stress analysis, is well-suited for measurement of 3D constitutive properties for composite materials, and that can enable a major shift toward accurate 3D material characterization. The SBS test methodology introduced three fundamental contributions to the experimental mechanics. First, tensile, compressive, and shear stress-strain relations in the plane of loading are measured in a single experiment. Figure 7 shows multiple standard methods that can be reduced to a single test and specimen configuration. It is simple to machine short-beam coupons in the 0-deg. and 90-deg. material directions; and load in the 1-2 (in-plane), 1-3 (interlaminar), and 2-3 material planes and measure 3D constitutive properties.

A single test method?

Fig. 7 A single method to generate 3D material constitutive properties for composites enables affordable material characterization.

126

A. Makeev and Y. Nikishkov

Second, a counter-intuitive feasibility of closed-form stress and modulus models, normally applicable to long beams, is demonstrated for short-beam coupons. Linear axial strain distributions through the specimen thickness observed in the coupons enable simple stress/modulus approximations. Figure 8 shows a linear axial strain distribution for a unidirectional glass/epoxy coupon loaded in the 1-3 material plane. The strain distributions are measured using the Digital Image Correlation (DIC) technique. Accurate stress-strain curves, strength, and modulus data are generated for multiple glass/epoxy and carbon/epoxy materials [4-6].

ε11 = − ky − b

Fig. 8 A linear axial strain distribution through the specimen thickness enables simple stress and modulus approximations.

And third, the test method is viable for measurement of stress-strain relations at various load rates including static, fatigue [7-9], and impact load conditions. The most recent results verified simplicity of the SBS test based constitutive property approximations at impact load rates. Figure 9 shows a high-strain rate test setup.

Fig. 9 Short-beam shear test setup for impact load rates: Instron Dynatup load frame and Photron SA1 high-speed (up to 200,000 frames/second) camera system.

Fatigue Life Assessment for Composite Structure

127

It is worth noting that SBS test methodology, micro-focus computed tomography measurement and automated transition of the subsurface measurement into a structural model able to predict failure response could enable accurate assessment of the effects of manufacturing defects on material-scale properties.

4 Structural Analysis This section presents a few selected results of recently developed fatigue structural analysis methodology and models able to predict initiation and progression of ply- cracks and delaminations in composites. The structural analysis is based on three-dimensional solid finite element (FE) techniques combined with failure initiation criteria and damage propagation algorithms that capture multiple damage modes and their interaction including effects of defects. The following results are summarized: (a) fatigue delamination of thick tensile composite articles with ply-waviness defects; (b) analysis of porosity defects; and (c) fatigue damage in open-hole tensile articles that includes matrix cracks and delaminations and their interaction. A priori model predictions are correlated with tests. Fatigue Analysis of Thick Composite Articles with Ply-Waviness (Wrinkles) A 88-ply [(+453/02)3/+454/02/+454/02]S IM7/8552 Carbon/Epoxy tape laminate with wavy plies subject to tensile fatigue load is considered first. Three articles illustrated in Figure 10 were loaded at constant amplitude 0.1 load ratio and at 10 Hz frequency, to delamination failure onset. The ply-waviness geometry is obtained from digital images. Nonlinear in-ply and interlaminar stress-strain relations account for micromechanical damage. 1

2

3

Fig. 10 Articles with ply-waviness defects. Zero-degree plies look lighter in the digital images.

128

A. Makeev and Y. Nikishkov

Table I lists the peak loads and compares cycles to delamination onset predicted using the Hashin failure criterion [10] and a modified LaRC04 fracturetoughness-based failure criterion [11], and test measurements. Table 1 Predicted and measured cycles to failure for articles with ply-waviness. Test

Peak Load

Hashin [10]

Fracture [11]

Test Data

710,000 cycles

750,000 cycles

1

9875 N (2200 lbs) 3,800,000 cycles

2

8674 N (1950 lbs)

700 cycles

600 cycles

<1,000 cycles

3

6894 N (1550 lbs)

225,000 cycles

93,000 cycles

100,000 cycles

Test data at 750000 cycles

Predictions, no ply-terminations, Predictions, with ply1,000,000 cycles terminations, 750,000 cycles

Fig. 11 Comparison of fatigue delamination patterns.

The DIC technique was used to monitor peak strains at 500-cycle intervals during fatigue loading. Figure 11 shows measured and predicted fatigue delamination patterns for article 1 at the specified number of cycles. Explicit modeling of plyterminations present in the +45-degree ply-groups at the failure-critical locations was required to capture the ply-interface damage locations [8]. Analysis of Structure with Porosoty Defects Test data show that porosity/voids at critical locations may significantly reduce strength and fatigue life of composite laminates. In particular, when lower curing pressure was used to reduce wrinkles in the thick IM7/8552 carbon/epoxy composite tensile test articles, they delaminated at much lower than theoretically predicted loads when porosity was ignored in the structural model. In this work, the technical approach to account for the porosity/voids and the combinations of manufacturing defects in the failure models is based on the following workflow: (1) measure ply-orientation defects (wrinkles); (2) detect

Fatigue Life Assessment for Composite Structure

129

porosity/voids shapes and locations; (3) calculate local stress fields and the stress concentrations associated with the porosity geometric shapes; (4) build a FE mesh that accounts for the wrinkles; (5) combine local stress field due to the porosity geometric shapes and the stresses in the model with the ply-orientation defects; and (6) use static/fatigue failure criteria to predict the load/cycles to structural damage. A 104-ply [(+453/04)3/+454/04/+454/02]S IM7/8552 Carbon/epoxy tape laminate with wrinkles and porosity is considered in this section. Material orientations of wavy plies are defined in the FE model using edge detection analysis of the digital images and interpolation of the wrinkle curves tangent orientations. Porosity locations and sizes are also found by edge detection analysis; and the stress concentration field due to porosity is included in post-processing. The test data show a 12.9 kN (2900 lbs) failure initiation. The failure load prediction for the FE model of the laminate with wrinkles and no porosity predicts is 34.7 kN (7800 lbs); and the porosity included in the failure model as ellipsoidal surface voids reduces the failure load to 13.3 kN (3000 lbs). Figure 12 shows the results of defect detection algorithm. Figure 13 demonstrates measured failure locations (on top of the 4th wrinkle) and the simulation results with stress concentration from porosity (on top and below the 4th wrinkle). Voids appear as black areas in +45-degree plies.

8

7

6

5

4

3

2

8

7

6

5

4

3

2

Fig. 12 Imperfection detection results for wrinkles (left) and porosity (right).

130

A. Makeev and Y. Nikishkov

Failure locations

7 9 8 7 6

5 4

6

5

4

3

3 2 1

Fig. 13 Test data (left) and simulation results (right) show similar locations for delamination failure.

As shown in the Subsurface Measurement Section, 3D material orientations and porosity defects could be measured based on CT volume data reconstruction. Fatigue Analysis of Open Hole Tensile Coupons This sub-section presents fatigue failure model predictions and subsequent test correlations for open-hole tensile articles subject to cyclic loads that result in matrix cracks and delaminations and their interaction in three dimensions. The first 16-ply quazi-isotropic IM7/8552 Carbon/Epoxy tape open-hole tensile (OHT) laminate was subject to constant amplitude load to 1,000,000 cycles at 10 Hz frequency, 22.2 kN (5,000 lbs) peak load and 0.1 load ratio. The laminate dimensions are 38.1 × 190.5 × 2.642 mm (1.5 × 7.5 × 0.104 in); and the holediameter is 6.35 mm (0.25 in). The fatigue structural analysis methodology and models, including FE mesh and fatigue simulation algorithm, is described in detail in Ref. [12]. Sub-modeling is used to allow sufficient mesh size for convergence of interlaminar stresses. To allow matrix damage propagation in fiber directions the sub-model is assembled from the fiber-aligned regular meshes representing laminate plies that are connected using mesh constraints. The progressive fatigue failure algorithm determines the elements failed during fatigue cycles, and recalculates stresses due to element damage. Delamination is simulated by matrix failure of thin (10% ply thickness) solid element layers between the plies. A modified three-dimensional fracture toughness-based criterion [7, 11] accounts for the interlaminar tensile and shear failure modes in both 1-3 and 2-3 material planes.

Fatigue Life Assessment for Composite Structure

131

Delaminations Test (CT Data)

Fatigue Model

Matrix cracks Failure cycles encoded by colors from blue (low cycles) to red (high cycles)

Fig. 14 Damage patterns predicted by the fatigue failure model matches the test results for the IM7/8552 Carbon Epoxy OHT laminates.

Figure 14 compares predictions of matrix cracks and delaminations shown as failed elements, and the CT data of the specimen obtained at 1,000,000 cycles of fatigue loading. The Figure shows excellent correlation of the locations of the major matrix cracks in the surface and sub-surface plies and delaminations between plies. The fatigue damage progression algorithm that included multiple damage modes was able to conservatively predict largest crack lengths and delaminations within crack measurement tolerance. The simulations that included only a single failure mode were not able to obtain the conservative predictions. Table 2 Comparison of largest crack lengths and delaminations for OHT articles at various loads and cycles. *Different quasi-isotropic layup.

Peak load, kN (lbs) 21.4 (4800) 22.2 (5000)* 23.1 (5200) 24.0 (5400) 24.9 (5600)

Cycles 3,000,000 1,000,000 400,000 200,000 100,000

Simulation Max crack Max length, delam, mm (in) mm (in) 6.3 (0.25) 2.0 (0.08) 5.6 (0.22) 2.5 (0.1) 6.3 (0.25) 1.8 (0.07) 6.3 (0.25) 1.8 (0.07) 6.3 (0.25) 1.8 (0.07)

Test Max crack Max length, delam, mm (in) mm (in) 8.1 (0.32) 1.0 (0.04) 5.3 (0.21) 2.5 (0.1) 7.4 (0.29) 1.5 (0.06) 6.9 (0.27) 1.3 (0.05) 6.6 (0.26) 0.8 (0.03)

132

A. Makeev and Y. Nikishkov Matrix cracks in surface -45° layer

Max Crack Length

Max Crack Length

rd Delamination in 1st subsurface -45/90 interface Delamination in 3 subsurface 45/0 interface and matrix cracks in 3rd subsurface 45° layer and matrix cracks in surface 45° layer

Fig. 15 Comparison of matrix cracks (top) and delaminations (bottom) in the test specimen CT scan and simulation.

To further verify the methodology, the progressive fatigue failure model predictons were compared with CT data for multiple OHT articles tested at various fatigue loads and cycles to crack and delamination sizes. Table II shows the comparison for predicted cracks and delaminations. Figure 15 shows a typical comparison of the simulations with the CT data of the OHT article subjected to 24 kN (5400 lbs) maximum load and tested to 200,000 cycles. The FE simulations show good qualitative and quantitative agreement with the tests. The simulations were able to predict the locations and sizes of matrix cracks and delaminations and their development.

5 Comments The main objective of this work is to inform the aerospace engineering community about advanced structural methods and prognostics that could account for manufacturing issues in the fatigue-critical, flight-critical composite parts and enable accurate assessment of their capability and useful life. Composite aircraft structures must undergo a fundamental shift from fleet statistics to accurate assessment of condition for individual parts in order to enable both safe and economical usage and maintenance. Technologies to measure defects and understand their effects on fatigue performance could potentially enable that shift. Our goal is to make such technologies the industry standard practice for structural diagnostics in the existing aircrafts and the emerging composite aircraft platforms. It is critical to enable: (a) accurate three-dimensional nondestructive

Fatigue Life Assessment for Composite Structure

133

measurement of manufacturing defects location and size; and (b) defect characterization based on fatigue structural models that automatically take the subsurface measurement data and account for multiple failure modes and their interaction to predict remaining useful life for the inspected parts. A close relation among the technology elements presented in this work is essential for success. For example, the SBS test methodology could minimize the number of test methods required for material characterization; and the structural analysis techniques applied at the coupon-level could enable a heavy reliance on analysis (virtual tests) to capture laminate-scale strength and fatigue behavior. Such relation could reduce costly experimental iterations to qualify every new composite material which delay insertion of advanced materials that address performance and operations efficiency requirements for aircraft structural designs, and limit the design space of material configurations. One of the critical topics in the composite structures for the aerospace industry is to enable the ability to design and build a fatigue-critical, flight-critical part right the first time (achieve at least 80% yield). Currently the first-time yield is estimated about 20% to 30%, and 70% to 80% in production. The designers do not have the appropriate tools to achieve such goal quickly (weeks.) One problem is when the new part is designed it is sized for static in-plane properties but the interlaminar structural behavior is not well-understood. We know that fatigue failure in a composite laminate starts with interlaminar damage. Manufacturing imperfections also contribute to fatigue failure initiation. About 40% of the budget to develop a new part is spent before starting the fatigue qualification when it is too late in the process and the design is patched later to fix the fatigue issues when the part is not developed right. First-ply failures in composites typically do not affect their residual capability and useful life, and damage progression to significant (detectable) size is required for life assessment. A comprehensive fatigue structural analysis methodology that captures multi-stage failure modes and their interaction in composites, and predicts initiation and progression of structural damage to detectable size without a priori assumptions of the initial damage or the damage path is required. The structural analysis methodology being developed by the authors attempts to satisfy these requirements to successfully predict life of aircraft composite parts. The structural analysis methodology is based on 3D solid FEM that simulate the initiation and progression of structural damage to detectable size. No a priori assumptions of the initial damage or the damage path are required. The models account for micromechanical damage through nonlinear interlaminar stress-strain constitutive relations. Stress-based and fracture mechanics-based failure criteria are used to predict initiation of ply-cracks and delaminations as well as their progression. The element failure simulations are consistent with the failure criteria: material stiffness loss in the transverse tension and shear directions simulates the matrix-dominated failures. The failure models must be supported by test evidence. Successful verification efforts started at laminate-level and element-level must be continued and expanded to full-scale parts.

134

A. Makeev and Y. Nikishkov

Acknowledgements This work is sponsored by the US Office of Naval Research, NAVAIR, and the National Rotorcraft Technology Center, U.S. Army Aviation and Missile Research, Development and Engineering Center (ARMDEC.) Such support is gratefully acknowledged. The views and conclusions contained in this article should not be interpreted as representing the official policies, either expressed or implied, of the U.S. Government.

References [1] Makeev, A., Nikishkov, Y., Carpentier, P., Lee, E., Noel, J.: Manufacturing Issues and Measurement Techniques for Assessment of the Effects on Structural Performance of Composite Parts. In: Proceedings of the AHS 66th Annual Forum, Phoenix, AZ (2010) [2] Dobyns, A., Rousseau, C.Q., Minguet, P.: In: Kelly, A., Zweben, C. (eds.) Comprehensive Composite Materials, 6th edn., pp. 223–242. Elsevier Ltd., Amsterdam (2000) [3] Department, U.S.: of Defense, Military Handbook - MIL-HDBK-17-1F: Composite Materials Handbook, vol. 1 (2002) [4] Makeev, A., He, Y., Carpentier, A.P., Shonkwiler, B.: A Method for Measurement of Three-Dimensional Constitutive Properties for Composite Materials. submitted for publication in the J. Comp. Mat (2010) [5] Carpentier, P., Makeev, A.: Frac. and Dam. Mech. IX. Key Eng. Mat., 401–404 (2010) [6] Makeev, A., Ignatius, C., He, Y., Shonkwiler, B.: J. Comp. Mat. 43(25), 3091–3105 (2009) [7] Nikishkov, Y., Makeev, A., Cline, C., Beasley, J., Fay, R.: Finite Element-Based Damage Tolerance Methods for Aircraft Composites. In: Proceedings of the 36th ERF, Paris, France (2010) [8] Nikishkov, Y., Makeev, A.: Fatigue Damage Simulation in Composites. In: Proceedings of the AHS 66th Forum, Phoenix, AZ (2010) [9] Makeev, A., Nikishkov, Y., Seon, G., Lee, E.: Fatigue Structural Substantiation for Thick Composites. In: Proceedings of the 17th ICCM, Scotland, UK (2009) [10] Hashin, Z.: J. App. Mech. 47, 329–334 (1980) [11] Davila, C.G., Camanho, P.P., Rose, C.A.: J. Comp. Mat. 39, 323–345 (2005) [12] Nikishkov, Y., Makeev, A., Seon, G.: Simulation of Damage in Composites based on Solid Finite Elements. Accepted for publication in the Journal of American Helicopter Society (2010)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Potential of CFRP Used for Light Weight Structures: Some Experimental Results Th. Fleischer, J. Ridzewski, M. Sachse, and F. Schirmacher IMA Materialforschung und Anwendungstechnik GmbH, Dresden, Germany

Abstract. This paper focuses on mechanical tests of materials and structures. Those tests are essential to assess structural integrity of aircrafts. Material tests are performed in order to obtain material properties used for evaluation of structural integrity. That means, it is essential to obtain correct results in order to minimize uncertainties during the design process of aircraft structures. This paper shows IMA Dresdens approaches to obtain several properties for fibre reinforced materials. Testing methods to determine fatigue strength, especially for unidirectional laminates, in-plane shear modulus, compressive strength and inter-fibre failure for tension are presented. The second topic is structural testing related to curved fuselage panel testing. A new approach to increase capabilities of that technology is presented.

1 Introduction Mechanical testing still plays an important role during developing of aircraft structures. On one hand you have to perform tests in order to determine material properties that are later on used to assess the capabilities of structures. The need for those tests is increasing, as the variety of materials available increases, especially in the field of fibre reinforced plastics. Due to different technologies for obtaining different properties of those materials, such as compressive or in plane shear strength, inter-fibre tension strength or fatigue strength, the variety of test methods is also quite large. IMA Dresden developed different methods for testing, hardware as well as procedures that shall be shown in the following paragraphs. But not only is the material testing essential to make best use of the properties offered by fibre reinforced plastics. If the current design of structures made of those materials changes from “black metal” towards solutions taking into account the non-isotropic properties more consequent, new series for structures will be needed in order to support the development. IMA Dresden is going forward to have the test technologies ready in terms of fuselage panel testing. The new approach for that special kind of structure testing will be shown, too. To make use of the potential of fibre reinforced plastics it is necessary to obtain material properties as exact as possible. Current methods are very conservative and maybe not efficient enough for detailed investigations. This paper shows some ideas and hints to improve that situation. *

Oral presentation.

136

Th. Fleischer et al.

2 Material Testing – Fibre Reinforced Plastics Challenges of fibre reinforced plastic testing Obtaining mechanical properties of fibre reinforced plastics, especially of unidirectional laminates, has several difficulties to fight with. That is why there are no holistic approaches and design methods known as a standard in this field. In general, the damage mode of fibre reinforced plastics is totally different from the one of metals, as can be seen in the Figure 1. In the end, the damage state of a laminate can be described with: • Debonding • Number and size of interlaminar cracks • Delaminations • Local failure of fibres due to tension or compression. This variety of damages to be triggered by testing on the one hand and the anisotropic behaviour of the material on the other hand are the reason that several testing technologies have been developed in the past. IMA Dresden developed some improvements that shall be shown in the following.

Homogenious Material

Fibre-reinforced Plastic Cyclic loading

Numerous damages within the matrix and at the fibre matrix interface Crack initiation Interlaminar stresses in the crack area, inter-fibre failure, local strain increase Creation of fibre breakage and delaminations Crack growth Increasing number of failures Æ increasing stresses of load carrying UD-layers Failure of remaining area

Failure of ramianing fibres of those layers

Fig. 1 General failure modes.

Potential of CFRP Used for Light Weight Structures: Some Experimental Results

137

Obtaining fatigue strength values for unidirectional laminates Unidirectional laminates are tested to obtain material properties for a single layer within a “real” laminate. Challenging for such tests is the strong anisotropic behaviour of such a material. So far, there is no specimen geometry or clamping interface available that provides: • • •

uniaxial stress conditions evenly distributed stresses failure of the specimen within the designated area

That means, so far only a rare number of valid test results is available, even in the literature. Very often the specimen fails within or nearby the clamping interface and thus making the test result invalid (see Figure 2). That means, test results reflect more the boundary conditions of testing than the actual material properties.

Fig. 2 Examples for invalid failed specimens.

For that reason an optimized specimen geometry has been developed. The general geometry is twice waisted. It has been optimized by the use of numerical analysis in order to ensure even stress distribution with the stress maximum being in the area of interest.

Fig. 3 General UD specimen geometry.

138

Th. Fleischer et al.

This specimen geometry allows testing of UD laminates with different R-ratios. Various test series showed valid specimen failure (see Figure 4) for all specimens and therefore valid testing results.

Fig. 4 Valid failed specimen.

Results in general show, that current assumptions regarding fatigue behaviour of fibre reinforced plastics are still conservative. Compression testing of fibre reinforced plastics The challenge for compression testing of fibre reinforced plastics is the load introduction in the specimen. There are several standards available, such as ISO 14126 or ASTM D 695. There are several test fixtures available to test according to those standards. Those devices are often complex in use or allow small specimen dimensions only. But specimen geometry is maybe more important than for tests with metal specimens. Due to fibre layout there are structures within the material with a size of several millimetres, as Figure 5 shows.

Fig. 5 Typical fibre layout.

IMA Dresden recommends the usage of hydraulic clamping devices and a special specimen preparation process. This ensures effective testing and a low number of invalid test results. This can be seen in Figure 6, at a specimen with 0° fibre orientation and one with 90° fibre orientation, each failed validly.

Potential of CFRP Used for Light Weight Structures: Some Experimental Results

139

Fig. 6 Hydraulic clamping fixture and failed specimens.

Determination of in-plane shear strength of fibre reinforced plastics As the in-plane shear strength of laminates can only be obtained experimentally, it is essential to have an appropriate testing technology at hand. Main objectives for such a technology are: • • • • •

pure, homogeneous shear stress conditions within the specimen obtaining of shear modulus and shear strength possible fast and easy specimen preparation and test conduct easy calculation of demanded values out of measurements possibility for further investigations

And indeed, there are several test methods described in standards available: • • • • •

Shear test with shear frame according to DIN 5399-2 (1982) Rail shear method o Two-Rail Shear Test Method according to ASTM D 4255 (1983) o Three-Rail Shear Test Method according to ASTM D 4255 (1983) Torsion test at tube specimens according to ASTM D 5379 (1993) V-Notched Beam Method according to ASTM D 5379 (Iosipescu) (1993) Tension test at ±45°-laminates according to DIN EN ISO 14129 (1995)

Each of those methods has its specific advantages and disadvantages. Furthermore, they do not fulfil the above mentioned requirements to the same extent. Therefore, a new method was developed as a derivate of the proven ones. The Two-Rail Shear Method and the V-Notched Beam Method are combined to the VNotched Rail Shear Method.

140

Th. Fleischer et al.

Two Rail Shear Method

Iosipescu Method

74 - 76

6x

149 - 151

45

mm r

15

V-Notched Rail Shear Method Fig. 7 Evolution of V-Notched Rail Shear Method.

That Method has been evaluated quantitatively by comparison with others and qualitatively by comparison between FE-Analysis and Digital Image Correlation. The results are shown in the following figures.

Shear modulus of biaxial reinforced laminates GFRP-BX 4500

CFRP-BX 5,5%

4,2%

3,9% 0,0%

4000 -1,5% 3500 a P M3000 / s lu 2500 u d o 2000 M r a e 1500 h S 1000

-3,6%

-3,0% -10,5%

-10,7%

-20,0%

-22,8% -40,0%

-60,0%

-80,0%

500 0

-100,0% V-Notched Rail Shear Tension test at Test +/ -45 ° Laminates

V-Notched Beam Method - Iosipescu

Torsiontest at tube specimen

Shear frame

Test Method

Fig. 8 Derived shear moduli from different test methods, compared with the classic laminate theory.

Potential of CFRP Used for Light Weight Structures: Some Experimental Results

141

Fig. 9 Comparison of shear strain εxy between FE (left) and ARAMIS® (right).

3 Structure Testing – Curved Fuselage Panels Current available possibilities of fuselage panel testing CFRP is the chosen material for the fuselage of two wide body airplanes. So far, the potential of that material is not used up to its full extent. The possibility to tailor the material stress-related could lead to changes in fuselage design. Especially cut outs could become interesting due to their load redirecting character. Being confident, that fuselage panel testing is a major keystone of structure development, IMA Dresden developed its capabilities towards panels with large cut outs. In the past test rigs for curved fuselage panels aimed for evenly distributed stresses within a certain area of interest within the specimen. Key to performing such tests is the ability of the concept to simulate the pressure difference on the fuselage between inside and outside during high level cruising. The roots of the full-scale panel tests with pressurisation go back to the middle of the last century. These began with two mirrored panels clamped at the longitudinal boundary which were pressurised cyclically. It seems this way was not practicable, because the next step was a pressurisation of the entire fuselage inside the water tank. In parallel several static and fatigue tests had been performed for complementary single loads like tension and shear at plane as well as curved panels. The first step on the development path of technologies for testing single curved fuselage panels had been the test method for superposing hoop stress and longitudinal tension stress. This condition applies for a single crown panel of the forward fuselage of a commercial airplane. In later development steps the option to apply in-plane shear was added and the possibility to apply longitudinal compression implemented [1].

142

Th. Fleischer et al.

Fig. 10 Development of curved fuselage panel test technology.

Loading possibilities grew quite complex, including combinations of: • • •

constant longitudinal tension and/ or compression internal pressurisation using either water or air constant in-plane shear

This test philosophy aims on testing of smaller features within a panel and their local effects or for validation in terms of panel material and fastening technologies. Also repairs of a certain extent are widely researched with that test concept [2], [3]. Current technology improvement A new technology has been developed and an appropriate test rig is currently put into service. It adds the following points to the already available features of such test rigs: • • •

larger panel geometries in order to test significant cut outs and their surrounding structures within the scope of a panel test parabolic distributed shear at the curved interface linear distributed longitudinal forces at the curved interface

A CAD model can be seen in Figure 11. The loading of the panel is realized with the panel being part of a closed cross section. The “tube”, consisting of the panel itself and a reusable complementary structure is loaded between two stiff plates by moving them relatively to each other. The movement is controlled comparable to a hexapod. This principle allows a broad variety in loading resulting in the above mentioned possibilities. The test rig is designed modularly. That means that various specimen geometries can be adapted easily. Different radii, lengths and angles are possible, depending on aircraft geometry and test aim needs. Higher loaded structures for instance would require a smaller panel. This has to be considered when planning tests for structures designed for a certain fuselage area. The parameters for loading in general are the following:

Potential of CFRP Used for Light Weight Structures: Some Experimental Results

• • • • •

Longitudinal Force Torque Shear Force Bending Moment Over-pressure

143

±5.000 kN ±4.000 kNm ±1.500 kN ±10.000 kNm 0,15 MPa

Fig. 11 Test rig design.

References [1] Bode, M.D., Sippel, W., Bakuckas, G.: In: Bros, M. (ed.) Proceedings of the 25th ICAF Symposium Bridging the Gap between Theory and Operational Practice, ICAF 2009, pp. 109–121. Springer Science+Business Media B.V, Amsterdam (2009) [2] Bakuckas, G., McIver, K., Hsu, C.: In: Bros, M. (ed.) Proceedings of the 25th ICAF Symposium Bridging the Gap between Theory and Operational Practice, ICAF 2009, pp. 407–425. Springer Science+Business Media B.V, Amsterdam (2009) [3] Best, R., et al.: In: Bros, M. (ed.) Proceedings of the 25th ICAF Symposium Bridging the Gap between Theory and Operational Practice, ICAF 2009, pp. 3–14. Springer Science+Business Media B.V, Amsterdam (2009)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

An Implementation of an Accelerated Testing Methodology to Obtain Static, Creep and Fatigue Master Curves of a T300/913 Unidirectional Composite Material Yuval Freed1 and Sven Rzepka2 1

Israel Aerospace Industries, Ben-Gurion International Airport, Israel [email protected] 2 Fraunhofer ENAS, Chemnitz, Germany [email protected]

Abstract. This study focuses on the prediction of long term residual strength, creep and fatigue characteristics of T300/913 unidirectional composite material. Both tensile and three point bending specimens were tested. The relation between time and temperature was established, and residual strength, creep and fatigue master-curves were determined. These curves were compared to test results at different elevated temperatures, applied strain rates and cyclic loading frequencies. From the results presented in this paper, it can be readily concluded that the timetemperature superposition principle indeed holds for creep, residual strength and fatigue behavior of the T300/913 unidirectional tape.

1 Introduction With the increase use of unidirectional composite materials as primary structures in advanced light-weight aerospace products, the ability to predict long term behavior of composite materials becomes essential. Since these products are designed to a target life-goal of several dozens of years, it is needed to establish an accelerating testing methodology that can replace a long term real-time testing. Prof. Miyano and his co-workers introduced an accelerating testing methodology for the determination of long term behaviour of fiber reinforced composite materials [1-5]. Their methodology was based upon a well-known timetemperature superposition principle for polymers, which was originally developed to obtain non-destructive material properties, and is summarized in an ASTM standard [6]. Recent studies showed that this principle holds for failure parameters of certain types of composite materials as well. The outcome of this methodology is a set of master-curves, in which the long term behavior of the composite material is described in terms of applied loads, number of cycles to failure, load frequencies, and operational temperatures. This is an efficient and systematic approach to life prediction, since only one fatigue master curve is needed to *

Oral presentation.

146

Y. Freed and S. Rzepka

predict the lifetime of the composite material at any load frequency or elevated temperature. Since this simple coupon testing is performed in short time durations, it significantly reduces the overall testing time (by approximately 99%), as compared to real-time testing along the product design life. Due to the acceleration in this testing, a significant reduction of energy consumption is achieved as well as reduced environmental impact. This meets the ICAF 2011 theme of influence of efficiency and green imperatives. This study focuses on the prediction of long term residual strength, creep and fatigue characteristics of T300/913 unidirectional composite material. Both three point bending and tensile specimens were tested. The relation between time and temperature was established, and residual strength, creep and fatigue mastercurves were determined. These curves were compared to test results at different elevated temperatures, applied strain rates and cyclic loading frequencies. From the results presented here, it can readily be concluded that the time-temperature superposition principle indeed holds for creep, residual strength and fatigue behavior of the T300/913 unidirectional tape.

2 Accelerated Testing Methodoly The accelerating testing methodology is based upon determination of a relation between the temperature and the testing time periods. This relation is established by performing viscoelastic testing at several elevated temperature states. This relation, usually referred to as a 'time-temperature superposition principle' (TTSP), holds for creep, residual strength, and fatigue behavior of the composite material. With additional sets of simple constant strain rate and fatigue coupon tests, the degradation of the mechanical properties of the composite material upon applied cyclic loading over long term periods in standard operational temperatures can be determined. This procedure is schematically described in Fig. 1.

VISCOELASTIC TESTING AT SEVERAL ELEVATED TEMPERATURES

CONSTANT STRAIN RATE TESTING AT SEVERAL ELEVATED TEMPERATURES

FATIGUE TESTING AT SEVERAL TEMPERATURES

TIME-TEMPRATURE SHIFT FUNCTIONS

resin

RESIDUAL STRENGTH MASTER CURVE

CREEP MASTER CURVE

FATIGUE MASTER CURVE

Composite material

Fig. 1 Schematic description of the accelerating testing methodology.

An Implementation of an Accelerated Testing Methodology

147

Three hypotheses are assumed in the scheme of the accelerated testing methodology: 1.

It is assumed that the time-temperature superposition principle holds for all strength. In other words, once the time-temperature shift functions are determined from a set of viscoelastic coupon tests of the matrix resin (see Fig. 1), the static residual strength master curve can be obtained using the same time-temperature relation.

2.

A linear cumulative damage law for monotonic load is assumed. Static loading is considered as a set of discrete monotonically increased creep loading for discrete time periods. The damage upon static loading can be calculated as a linear superposition of the discrete damage upon different levels of creep loading. By exploiting this linear relation between the creep and residual strength master curves, the creep master curve can be established directly from the static residual strength master curve. This assumption was demonstrated theoretically and experimentally by Christensen and Miyano [8].

3.

A linear dependence of the fatigue strength upon the stress ratio R=σmin/σmax is assumed. This assumption has no physical meaning; however, it is convenient for durability design. Fatigue strength mastercurves can be determined by performing fatigue tests at a single frequency and several elevated temperature, combined with the timetemperature shift functions, the creep master curve (R=1) and the static residual strength master curve (R=0, Nf = 1/2).

It may be noted that the physical foundations of this methodology are beyond the scope of this paper. The reader is directed to Refs. [1-5] for further information, clarifications, and discussions. The suggested accelerated testing methodology has some advantages. It provides a practical acceleration of certain tests, which is required for the design process of new products. It is an efficient and systematic approach, and its outcome is a single set of fatigue strength master-curves to predict the lifetime of composite materials. In other words, while a conventional S-N curve provides information on the lifetime of the composite material as a function of the applied load and the number of cycles to failure, but is valid only for a specific temperature and load frequency, the fatigue strength master-curve is presented as a function of applied load, number of cycles, temperature, and load frequency. However, this method also has some limitations. It is not valid for PEEK resins and if the fibers show time and temperature dependence on the mechanical properties as well. In contrast to this, PAN-based carbon fibers are considered as good candidate for this methodology. The aforementioned methodology is used in this study with stress ratios of 0 ≤ R ≤ 1.

148

Y. Freed and S. Rzepka

3 Experimental Procedure Unidirectional carbon fiber tape pre-impregnated (Hexcel T300/913) composite material was investigated in this study. Its typical B-Basis mechanical properties at room temperature are listed in Table I. This unidirectional tape is designed to carry loads for prolonged times at temperatures up to 80ºC. Table 1 B-Basis mechanical properties of Hexcel T300/913 unidirectional tape (at room temperature).

Property Ultimate tensile strength Ultimate flexural strength Tensile modulus Flexural modulus Transverse modulus Transverse modulus of the fiber Ultimate elongation Fiber volume fraction Curing Temperature Glass Temperature

Value (B-Basis) 1.79 GPa 1.84 GPa 127 GPa 114 GPa 58 GPa 40 GPa 1.8% 0.6 135ºC 130ºC

The complete test matrix is presented in Table II. To evaluate the viscoelastic behaviour of the resin, three-point bending tests were conducted on transverse 90º specimens according to Ref. [7]. The material properties of the resin were backcalculated according to the rule of mixture ∗ Vm 1 1 ⎧⎪1 + V y 1 ⎫⎪ ∗ = ∗⎨ − ⎬, V y = 0.516 Vf E m V y ⎪⎩ ET E fT ⎪⎭

(1)

Where ET and EfT denote the transverse moduli of the composite and the fiber, Em represents the modulus of the resin while Vf and Vm are the volume fractions of the fiber and the resin matrix, respectively. Constant strain rate (CSR) and fatigue tests were performed on 0º-direction specimens. These tests were exploited to predict the fatigue lifetime for any arbitrary combinations of time to failure, temperature, frequency and stress ratio. Additional tests were performed to validate the predicted results. The test matrix is associated with a total of 429 test specimens.

An Implementation of an Accelerated Testing Methodology

149

Table 2 Test matrix.

Viscoelastic DMA (90ºdirection)

Deflection rate V(mm/min)

Frequency f (Hz)

Stress ratio R

Temperature T (ºC)

Remarks

-

0.5Hz – 50Hz (5 steps per decade)

-

-40 – 260 (in steps of 5ºC)

To obtain TTSF

Constant Strain Rate (0º-direction)

1

-

Constant Strain Rate (0º-direction)

0.5, 10

-

Constant Strain Rate (0º-direction)

1

-

Constant Strain Rate (0º-direction)

0.01, 100

-

Flexural fatigue test (0º-direction)

-

2

Flexural fatigue test (0º-direction)

-

0.1, 50

Flexural creep test (0º-direction)

-

-

To obtain tensile 25, 50, 80, 100, 120 residual strength master curve To validate tensile residual 25 strength master curve To obtain flexural residual 25, 50, 75, 100, strength master 125, 150 curves To validate flexural 25, 50, 75, 100, residual 125, 150 strength master curve To obtain fatigue strength 0.05 25, 75, 125 master curve (and validation) To validate fatigue 0.05, 0.5 25, 75, 125 strength master curve To validate 25, 75, 100 creep master curve

4 Results and Discussion The normalized relaxation modulus master curve and its corresponding timetemperature shift functions are presented in Figs. 2 and 3 for a reference temperature T0 of 25ºC. The relaxation modulus master-curve can easily be reproduced with respect to any reference temperature T, simply by shifting the curve by

log aT 0 (T ) = log t − log t ′

(2)

where t is the actual testing time ('short time' = several minutes), and t` denotes the 'reduced time'; that is, a time scale that represents the long term period. The parameter aT0 is the time-temperature shift factor and it represents the amount of shifting (in logarithmic scale) between the actual testing time and the long term

150

Y. Freed and S. Rzepka

'reduced time'. It is obtained by measuring the creep compliance or the relaxation modulus at various temperatures. Note that aT0 is a function of the temperature, and may be calculated as

log aT 0 = −

⎛1 ΔH 1 ×⎜ − ⎜ 2.303 R ⎝ T Tref

⎞ ⎟ ⎟ ⎠

(3)

where ΔH is the activation energy below or above the glass temperature Tg and R = 8.314×10-3 kJ/(K·mol) is the gas constant. T ref = 25°C

1.20

8000

1.00

7000 6000 ET [MPa]

ET/ET,MAX

0.80 0.60 0.40

5000

T=25°C

4000

T=60°C T=80°C

3000 2000

0.20

T=40°C

1000

T=100°C T=120°C T=140°C T=160°C T=180°C

0.00 0 -2.00 -1.00 0.00 1.00 2.00 3.00 -10.00

0.00

10.00

20.00

30.00

log (t'[min])

log (t[min])

Fig. 2 Normalized relaxation modulus master-curve.

0.00 20

log(a

)

-5.00

25

30

35

40

T

-10.00 -15.00 -20.00 -25.00 -30.00 1/T 10- 4 [1/°K]

Fig. 3 Time-temperature shift function.

Next, the master-curve of the flexural residual strength was obtained. It was determined by means of the time-temperature shift functions combined with constant strain rate test results at various elevated temperatures. The flexural residual strength master-curve is presented in Fig. 4. Nice agreement between the timetemperature superposition principle and the test results used for validation is achieved.

T = 23 °C T = 100 °C T = 125 °C T = 150 °C

2,5

2,0

1,5

1,0

0,5

Flexure (CSR) strength smax (t,T) (GPa)

Flexure (CSR) strength σmax (t,T) (GPa)

An Implementation of an Accelerated Testing Methodology

0,0 -1

0

1

2

151

T = 175 °C 2,50 T = 200 °C 2,25 T = 225 °C T = 250 °C 2,00

T = 23 °C T = 100 °C T = 125 °C T = 150 °C

1,75

T = 175 °C T = 200 °C T = 225 °C T = 250 °C

1,50 1,25 1,00 0,75 0,50 0,25 0,00 4 -25

3

-20

-15

log time [min]

-10

-5

0

5

10

log time [min]

Fig. 4 Flexural residual strength master-curve.

The normalized tensile residual strength master-curve for several reference temperatures is presented in Fig. 5. As expected, the residual strength of the composite material decreases upon increasing the temperature. Recall that the T300/913 unidirectional tape is designed to carry loads for prolonged times at temperatures up to 80ºC. To emphasis this directive, it may be observed that for the temperature of 80ºC, the residual strength of the composite material drops rapidly after 105 minutes, which are equivalent to two months. For typical room temperature of 25ºC on the other hand, the residual strength drops rapidly after more than 1015 minutes; this is equivalent to 109 years. 1.2

σs/ σMAX

1 0.8 0.6

T = 25°C

0.4

T = -40°C

T = 80°C

0.2 0 0

5

10

15

20

25

30

log (t's[min])

Fig. 5 Normalized tensile residual strength master-curve and a demonstration of the ability of the composite materials to carry load upon prolonged times at different temperatures.

152

Y. Freed and S. Rzepka

Fig. 6 Fatigue master-curve.

Fig. 6 shows the fatigue master-curve. The data was obtained by exposing the specimen to cyclic loads of different magnitudes and at several temperatures. In addition, the result of the CSR tests has been involved as the limiting case of specimen failing right within the first loading cycle (denoted as 100 in Fig. 6). Applying the visco-elastic shift function, the three fatigue curves are arranged along the CSR master-curve. In total, the set of fatigue data creates an array of graphs, which provide estimates to the number of cycles to failure (Nf) in the range of 100...104 for arbitrary temperatures. Note that this master-curve can be reproduced with respect to any reference temperature T, simply by shifting the curve as described in Eq. (2).

5 Summary and Conclusions In this study, the long term residual strength, creep, and fatigue characteristics of T300/913 unidirectional composite material are determined. Both, three point bending and tensile, specimens were tested. The relation between time and temperature was established and residual strength, creep, and fatigue master-curves were determined. These curves were compared to test results at different elevated temperatures, applied strain rates, and cyclic loading frequencies. From the results presented here, it can readily be concluded that the timetemperature superposition principle indeed holds for creep, residual strength, and fatigue behavior of the T300/913 unidirectional tape. Our next goal is to assess the effect of water absorption on the residual strength and fatigue behavior of the unidirectional T300/913 tape. A similar methodology will assist us to determine the long term response of the composite material upon any temperature and moisture environmental conditions.

An Implementation of an Accelerated Testing Methodology

153

Acknowledgement This study was carried out in the framework of the European research project "Clean-Sky". The authors wish to thank the "Clean-Sky" project for its financial support.

References [1] [2] [3] [4] [5] [6]

Miyano, Y., Nakada, M., McMurray, M.K., Muki, R.: J. Comp. Mater. 31, 619 (1997) Miyano, Y., Nakada, M., Kudo, M., Muki, R.: Adv. Comp. Mater. 8, 235 (1999) Miyano, Y., Nakada, M., Muki, R.: Mech. Time-Depend Mater. 3,141 (1999) Miyano, Y., Nakada, M., Sekine, N.: Comp. B 35, 497 (2004) Miyano, Y., Nakada, M., Sekine, N.: J. Comp. Mater. 39, 5 (2005) ASTM D2990-01, Standard Test Methods for Tensile, Compressive, and Flexural Creep and Creep-Rupture of Plastics [7] ASTM D-5023, Standard Test Method for Plastics: Dynamic Mechanical Properties:In Flexture (Three-Point Bending) [8] Christensen, R., Miyano, Y.: Int. J. Fract. 143, 35 (2007)

26th ICAF Symposium – Montreal, 1-3 June 2011 Improvement of Vibration Damping and Flexural Fatigue Property Incorporating Nanoclay into Glass/Epoxy Composite A. Kabir and S.V. Hoa Concordia Center for Composites Department of Mechanical and Industrial Engineering, Concordia University Centre de recherche en Plasturgie et Composites (CREPEC) Montreal, Quebec H3G 1M8, Canada [email protected] [email protected]

Abstract. This study demonstrates that nanoclay fillers can improve vibration damping property and fatigue behavior of conventional long fiber reinforced composites. Nanoclay was dispersed in epoxy resin by a solvent-free high-speed mixing method using a high speed homogenizer. This modified resin was used to manufacture S-glass/epoxy composite laminates by hand lay-up and autoclave curing. The dynamic properties of the samples were tested using a dynamic mechanical analyzer DMA 983. A maximum of 11.1% improvement in the flexural storage modulus and 16.9% in the loss modulus were achieved for adding up to 2 wt.% nanoclay. To see the damping effect of nanoclay at higher frequency and amplitude, a log decrement test was carried out where a maximum of 55.5% increase in damping ratio was observed. The effect of damping improvement on fatigue life was also investigated. A fixed amplitude flexural fatigue test was performed using an MTS machine. Maximum of 66% and 133% improvement in flexural fatigue life at respectively 1 and 2 wt.% nanoclay incorporation were achieved. Nanoclay increases the fracture resistance significantly which was characterized by optical microscope.

1 Introduction Recently, nano reinforcement of polymers to form nanocomposites has attracted the attention of researchers for their potential in property development. Organically modified nanoclay, nano-fiber or carbon nanotubes having ultra-high strength and stiffness became very popular during last decade for nano reinforcement of polymers. Nano-particles at very low concentration (<5 wt.%) and well dispersed in polymer resins often impart superior mechanical, thermal, barrier and electro-magnetic properties. Damping is a very important factor related to the study of dynamic behavior especially related to vibration for fiber reinforced composite structures. This property reduces noise and vibration of the structures

156

A. Kabir and S.V. Hoa

thus increases service life. Depending on the sources of energy dissipation, there are several mechanisms of damping in fiber-reinforced composites. One of these sources of damping is fiber and matrix interface [1,2]. Depending on the nature and properties of interphase, different mechanical properties including damping are affected. In composite structures, energy can also be dissipated due to the slippage and friction between fiber and matrix or delamination which is again related to interface [1,2]. So by increasing the amount of interfacial surface area within a given composite system, its damping property can be improved. Nano particles such as nano layered silicate or nanoclay having thickness around 1 nm and lateral dimensions in the order of few microns [3], have very high aspect ratio and specific surface area (around 657 m2/g)[4]. So even at very low concentration, these nanoclays can create a huge network of interfacial surface area when well dispersed in a polymer resin systems. It can be expected that adding nano-clay in a polymer matrix can improve the ability of energy dissipation under dynamic loading thus enhancing the damping property. A number of works have been carried out to predict the damping property of nanoclay reinforced polymer. In some recent works, it is pointed out that organically surface treated nanoclay has a significant enhancement on the damping property of composite structures [5-8]. A number of researchers have used the clay modified resin system to further reinforce with long fiber and investigated the effect of nanofillers on damping property. It was reported that incorporation of organoclay in glass/epoxy composites significantly enhances the vibration damping property [9-12]. It is worth mentioning, if vibration damping or dynamic properties are improved, the flexural fatigue life of the structure can be improved. Moreover, nanoclay increases the fracture toughness of polymer [8] which also helps increase the fracture resistance. A number of recent works have reported that nanoclay significantly improves fracture behavior of nano-filler modified neat polymer composites [13-16] and nano-filler incorporated conventional composites [17,18] at low concentration of nano-filler and low stress level.

2 Experimental Procedure Materials and sample fabrication In this study, Epoxy EPON 828 has been used as polymer resin and EPICURE 3046 as hardener. Both supplied by Hexion Specialty Chemicals. Organoclay Nanomer I.30E treated with long chain primary amine intercalant supplied by Nanocor Inc. is used as nanoclay additives. Unidirectional S-glass fiber manufactured by AGY World Headquarters and supplied by Aerospace Composites Products Inc. is used for further reinforcement of clay modified resin.

Improvement of Vibration Damping and Flexural Fatigue Property

EPO N 828

157

C la y I .3 0 E

S tirr in g w ith h ig h sp e e d h o m o g e n iz e r E P IC U R E 3 0 4 6

M ix in g

D egas

H an d L ay-u p

S -G la s s p r e f o rm

V a c u u m B a g g in g

A u to c la v e C u r in g

Fig. 1 Flow chart for sample manufacturing.

A high speed stirring method using a high speed homogenizer is used to disperse the clay in resin. The high speed homogenizer consists of an internal rotor and external stator assembly which is called a rotor-stator generator. The rotor is connected with a variable speed electric motor and can be operated at a minimum speed of 6,500 rpm to a maximum of 24,000 rpm. The rotor is designed in such a way that it acts as a centrifugal pump under operation. The differential speed between the rotor and stator produces extremely high shear force which subsequently generates a very high turbulent energy in the shear gap. This energy disperses the clay in the resin. Epoxy resin EPON 828 was first preheated to 45 ºC to reduce the viscosity. The necessary amount (up to 2 wt.%) of organoclay Nanomer I.30E was added to the resin. The clay is mixed into the resin by hand with a spatula. After that, the high speed homogenizer is used to disperse the clay up to a maximum rotational speed of 20,000 rpm for 20 minutes. The temperature of the suspension was closely monitored using a thermometer and kept below 100ºC throughout the process to avoid self-polymerization. Hand lay-up and autoclave molding process were used to fabricate samples. Figure 1 shows the flow chart of sample manufacturing procedure. A steel plate of dimensions 60cm 60cm 1.5cm was used as the tool to fabricate a flat plate laminate. 16 layers of fibers were laid-up on this plate then a vacuum bag was prepared and the assembly was subsequently cured in an autoclave. The samples were cured at 120°C for 2 hours and post cured at 140°C for another 2 hours. For convenience, a flat Cross-ply [02/902/02/902]s laminate was first manufactured and then the desired size experimental specimens were cut from the laminate using a diamond cutter.

158

A. Kabir and S.V. Hoa

DMA analysis Dynamic mechanical or viscoelastic properties of material are analyzed by DMA technique. DMA can simultaneously measure elastic or storage modulus and the energy dissipating or loss modulus of the material as a function of several parameters such as temperature, time, and frequency. A Du Pont 983 DMA coupled with TA 2100 thermal analyzer was used to measure the storage modulus, loss modulus and tanδ. All samples had the width between 8 mm to 10 mm and thickness between 1.25 mm to 2 mm. The dynamic properties were determined using fixed frequency method of 1Hz and amplitude of 0.2 mm. The samples were heated from room temperature to 180°C at a ramp (heating rate) of 2°C/min. Log decrement test The log decrement test is a method to measure damping ratio experimentally of an under-damped system. In an under-damped system, the amplitude of vibration exponentially decays over time and the natural log of amplitudes of any two successive peaks is called the logarithmic decrement or log decrement. So the log decrement, usually denoted by δ, can be defined by (1) Where A0 is the amplitude of the first peak and An is the amplitude of the peak after n period. From this log decrement, the damping ratio ζ can be calculated using the following formula (2)

The length and width of the samples are 20 cm and 2.54 cm respectively and the thickness is between 1.25 mm and 1.35 mm. Figure 2 shows the log decrement test set-up. One end of the sample was clamped on a rigid support with sufficient clamping force at distance of 20 cm from the free end. An accelerometer (Bruel & Kjaer 4393) was attached on the tip of the sample using wax glue to measure the amplitudes of vibration. The tip of cantilever composite beam is then excited with initial amplitude of 20 mm approximately and let to vibrate freely until the amplitude decays. These amplitude signals were then amplified by a Kistler 504E dual mode amplifier and the amplitude versus time graph was plotted using an Agilent 54624A oscilloscope. From the free decay curve, logarithmic decrement δ, and damping ratio ζ was calculated using equations 1 and 2 respectively.

Improvement of Vibration Damping and Flexural Fatigue Property

159

Fig. 2 Experimental set-up for log decrement test.

Fig. 3 Experimental set-up for flexural fatigue test.

Flexural fatigue test A fixed amplitude flexural fatigue test was designed using an MTS machine to measure the fatigue life of glass/epoxy composite laminate with and without nanoclay. Sample manufacturing procedure and dimensions were the same as for the log decrement test samples. Figure 3 shows the schematic illustration of experimental set-up for the flexural fatigue test. One end of the sample was clamped with lower grip of MTS machine and the other end was clamped to a fixed and rigid support outside of the machine at a distance of 15 cm. To monitor the damage inside the laminate sample, a strain gage (Vishay CEA-06-125UW-350) was attached to the laminate at a distance of 6 cm from the fixed end. The MTS machine was programmed to deflect the sample to maximum amplitude of 5° at frequency of 4 Hz. For each sample 300,000 fatigue cycles were applied. Strain on the sample was recorded after every 25,000 cycles of fatigue.

160

A. Kabir and S.V. Hoa

3 Results and Discussion DMA analysis The temperature dependent curves of storage modulus and loss modulus for the glass/epoxy laminate (with and without nanoclay) are shown in figure 4 as determined by flexural dynamic mechanical analysis. The temperature dependent curve of storage modulus in figure 4(a) shows that nanoclay significantly improves the storage modulus of the cross-ply laminate in the glassy region and the maximum improvement is observed at room temperature. At room temperature, 1 wt.% and 2 wt.% incorporation of nanoclay increased the storage modulus by 8.4% and 11.1% respectively. Nanoclay also has a positive effect on loss modulus in the glassy region shown in figure 4(b). The improvements in the loss modulus are 8.3% and 16.9% at room temperature for the same clay contents. But in the rubbery region, the loss moduli of laminates with nanoclay are slightly reduced compared to the laminates without nanoclay. After Tg, energy dissipation ability due to friction is decreased because of relative motion between the molecules of the resin system. So the loss modulus falls sharply but due to the presence of glass fiber, the laminate still maintains some stiffness. This could be a possible reason for this reduction of loss modulus in rubbery region. It is also observed from the peak of loss modulus that, the glass transition temperature Tg is not affected by the addition of nanoclay.

Fig. 4 Temperature dependent curves with different nanoclay contents; (a) Storage Modulus and (b) Loss Modulus.

Log decrement test The log decrement curves with different nanoclay contents are shown in figure 5. It can be observed clearly that the amplitude of the laminates with nanoclay decays faster than the laminate without nanoclay and the damping effect is higher at higher nanoclay content. After 4.5 seconds, the amplitudes of laminate with 1

Improvement of Vibration Damping and Flexural Fatigue Property

161

wt.% and 2 wt.% nanoclay have been reduced by 64.6% and 66.4% respectively. The log decrement value and the damping ratios of laminate with different nanoclay contents are given in Table I which indicates that at 1 wt.% incorporation of nanoclay can improve the damping ratio by 41.7% and at 2 wt.% nanoclay content the improvement is 55.5%. The log decrement test results are even more impressive than DMA results. This indicates that nanoclay acts more positively at higher frequency and amplitude of vibration.

Fig. 5 Log decrement curves with different nanoclay contents. Table 1 Log decrement values and damping ratios of laminates with different nanoclay contents.

Laminate Without Nanoclay With 1 wt.% clay With 2 wt.% clay

Logarithmic decrement, δ (eqn. 1) 0.036 0.051 0.056

Damping ratio, ζ (eqn. 2) 0.00573 0.00812 0.00891

162

A. Kabir and S.V. Hoa

Flexural fatigue test Figure 6 shows the strain versus number of cycles curves with scatter in the strain values. Test results indicate that the maximum strain values remain consistent up to a certain fatigue cycles. Then there is a sudden increase in the strain values. As fatigue damage starts accumulating within the laminate, it loses its stiffness and the laminate tends to bend more. Thus the jump in maximum strain is observed.

Fig. 6 Strain versus no. of cycle curves; (a) No clay, (b) 1 wt.% clay and (c) 2 wt.% clay.

The number of fatigue cycles at which this change in strain value occurs can be considered as a measure of the flexural fatigue life. The results clearly indicate that the presence of nanoclay significantly improves fatigue life of glass/epoxy laminate. The maximum strain value of glass/epoxy laminate without nanoclay is quite consistent up to an average of 100,000 fatigue cycles. Then the strain increases rapidly. That means fatigue damage starts accumulating within the laminates after approximately 100,000 fatigue cycles. On the other hand, the strain value starts changing after an average of 166,000 fatigue cycles for laminates with 1 wt.% nanoclay and the rate of increase in the strain value is slower compared to the laminate without nanoclay because of improved fatigue resistance. The

Improvement of Vibration Damping and Flexural Fatigue Property

163

laminates with 1 wt.% nanoclay have 66% improved flexural fatigue life. For the laminate with 2 wt.% nanoclay, a slight change in the strain value is observed after an average of 233,000 fatigue cycles. At 2 wt.% incorporation of nanoclay, the flexural fatigue life of the laminates was increased by 133% compared to conventional glass/epoxy composites. The curves in Figure 6 also indicate that the scatter in the results is more in the area where the strain values start increasing. This is possibly due to the uncontrolled nature of failure in composite materials. But all the curves have similar trend.

Fig. 7 Optical micro-graph of the cross section of the laminate after 200,000 fatigue cycles; (a) No clay, (b) 1 wt.% clay and (c) 2 wt.% clay.

The increase in the strain value occurred due to the accumulation of microcracks and delaminations inside the laminate. Figure 7a to 7c shows the optical microscopic image of the cross section of the laminate with different nanoclay contents after 200,000 fatigue cycles. After 200,000 fatigue cycles, the laminate without nanoclay has the height concentration of cracks (shown by thin arrow) and delaminations (shown by thick arrow) within 3rd and 4th layer (transversely oriented fiber) and the delaminations are taken place parallel to the crack propagation directions, those are clearly observable in figure 7a. The laminate with 1 wt.%

164

A. Kabir and S.V. Hoa

nanoclay also has micro-cracks concentration within the same plane shown in figure 7b but the extent of damage is less than the laminate without nanoclay. Whereas the laminate with 2 wt.% nanoclay has very few cracks as shown in figure 7c. This indicates that nanoclay has a very significant effect on the fracture resistance of polymer composites under cyclic loading. This improvement in fracture resistance might be attributed from the improvement in vibration damping combined with improved fracture toughness and improved strength of resin system due to nanoclay incorporation.

4 Conclusion In this study, vibration damping and flexural fatigue behavior of glass/epoxy/nanoclay composite were studied. Organically modified nanoclay was dispersed in epoxy resin by high speed mixing method using a high speed homogenizer. To measure the vibration damping property, DMA analysis and log decrement tests were carried out. DMA analysis indicates that nanoclay incorporation significantly improves both storage modulus and loss modulus. That means, nanoclay enhances the energy dissipation capability of glass/epoxy composite material and at the same time increases its stiffness. A maximum of 11.1% improvement in storage modulus and 16.9% in loss modulus was obtained at 2 wt.% nanoclay incorporation of nanoclay loading. Also the glass transition temperature was not affected due to nanoclay incorporation. In the log decrement test, a maximum of 55.5% improvement in damping ratio was observed at 2 wt.% nanoclay incorporation. Thus it can be said that nanoclay has a very positive effect on vibration damping property and the improvement is more significant at higher frequency and amplitude. A low amplitude flexural fatigue test was designed so that fatigue could be attributed from vibration and free from thermal effect. Test results show that nanoclay significantly improves flexural fatigue life. Flexural fatigue life of the laminate was increased by 66% and 133% at only 1 and 2 wt.% nanoclay incorporation respectively. This improvement in the fatigue life is most likely to be attributed from the improvement in vibration damping coupled with the improvement in fracture toughness and strength of resin system due to the addition of nanoclay. The optical microscopic images of cross-section of fatigued samples indicate that nanoclay addition significantly reduce micro-crack formation within the laminate thus improving its fatigue life.

References [1] [2] [3] [4]

Chandra, R., Singh, S.P., Gupta, K.: J. Sound Vibrat. 262(3), 475–496 (2003) Chandra, R., Singh, S.P., Gupta, K.: Compos. Struct. 46(1), 41–51 (1999) Ray, S.S., Okamoto, M.: Prog. Polym. Sci. 28(11), 1539–1641 (2003) Helmy, A.K., Ferreiro, E.A., De Bussetti, S.G.: J. Colloid Interface Sci. 210(1), 167–171 (1999) [5] Chen, C., Curliss, D.: Nanotechnology 14(6), 643–648 (2003) [6] Mohan, T.P., Kumar, M.R., Velmurugan, R.: J. Mater. Sci. 41(18), 5915–5925 (2006)

Improvement of Vibration Damping and Flexural Fatigue Property

165

[7] Sarathi, R., Sahu, R., Danikas, M.G.: J. Electr. Eng. 60(6), 358–361 (2009) [8] Ngo, T.D.: In: Understanding the effect of adding nanoclays into epoxies, PhD thesis, Concordia University, Montreal (2007) [9] Haque, A., Shamsuzzoha, M.: J. Compos. Mater. 37(20), 1821–1837 (2003) [10] Avila, A.F., Donadon, L.V., Duarte, H.V.: Compos. Struct. 83(3), 324–333 (2008) [11] Chandradass, J., Kumar, M.R., Velmurugan, R.: Mater. Lett. 61(22), 4385–4388 (2007) [12] Velmurugan, R., Jeyaprakash, P., Balaganesan, G.: Damping study of hybrid nano composites by low velocity impact. In: Proceedings of International Conference on Aerospace Science and Technology, Bangalore, India, June 26-28 (2008) [13] Varadharajan, B.R.: In: Fatigue behavior of α-Zirconium Phosphate/epoxy nanocomposites, Master’s thesis, Texas A&M University (2005) [14] Song, M., Yao, K.J.: Mater. Sci. Technol. 20(8), 989–993 (2004) [15] Blackman, B.R.K., Kinloch, A.J., Sohn Lee, J., Taylor, A.C., Agarwal, R., Schueneman, G., Sprenger, S.: J. Mater. Sci. 42(16), 7049–7051 (2007) [16] Wetzel, B., Rosso, P., Haupert, F., And Friedrich, K.: Eng. Fract. Mech. 73(16), 2375–2398 (2006) [17] Chisholm, N., Mahfuz, H., Rangari, V.K., Ashfaq, A., Jeelani, S.: Compos. Struct. 67(1), 115–124 (2005) [18] Grimmer, C.S., Dharan, C.K.H.: J. Mater. Sci. 43(13), 4487–4492 (2008)

26th ICAF Symposium – Montreal, 1-3 June 2011 Formation of a Metal Coating by Means of Friction Stir Processing D. Kocańda, A. Górka, and D. Zasada Military University of Technology, Warsaw, Poland

Abstract. Some aspects of a metal coating formation of structural components by means of friction stir processing are considered here. The paper provides a simplified description of a thermal field induced either by the process of friction bonding of materials (FSW) or by modifying surface layers of components by means of friction stir processing (FSP). Additionally, a theoretical analysis of mass transport in a high gradient thermal field initiated in micro-areas of materials subjected to FSP and FSW processes are included as well. Regarding the practical aspect of this research the analysis focuses on a Fe-Ni and Fe-Cu diffusion systems in a structural materials. Verification of the presented considerations was carried out for adhesive nickel or copper coatings on S235JR and S355JR low alloyed steels.

1 Introduction Since last years, new manufacturing technologies for materials of specialist applications are dominated by the technologies that make use of a friction phenomenon as an energy inducer of processes of bonding and modifying surface layers of various structural materials. These technologies are represented by friction stir welding (FSW) and friction stir processing (FSP). Interest in these ones was determined by their applications in aircraft and military equipment production as well as the industry that works for the needs of universe exploration [1, 2]. New technologies make it possible to bond various materials which are difficult to be bonded by means of traditional methods without losing strength and fatigue properties of those materials as well as to create multi-layer coatings characterized by reduced internal stresses [1]. Variable thermal conditions that run below recrystallization temperature favour diffusion processes at short distances for the majority of structural materials. These processes do not lead to structural transformations in large areas of a layer and do not cause any drastic changes in the strength of this layer. Results of practical researches conducted on new bonded materials and published by numerous authors dominated also theoretical studies in the field of modelling and simulation of structures created using the FSP process.

168

D. Kocańda, A. Górka, and D. Zasada

The presented study focuses on description of a thermal field induced in a FSW or in a FSP processes as well as on a theoretical analysis of mass transport in a high gradient thermal field initiated in micro-areas of materials subjected to FSP processes. These descriptions do not give considerations to the influence of numerous material-related and technological factors which determine the effectiveness of the friction process itself. Therefore, they can be treated as simplified descriptions. However, even in the simplified forms the descriptions are precise enough to be well used for widely understood technological research. After taking into consideration thermal parameters variability of matrix and coating materials and their dependence on space-time coordinates, the descriptions may serve for more precise research including test on diffusive changes as the effect of a thermodiffusive mass transport in micro-area of a modified surface layer [3]. A complex system of the potentials of chemical activity of the system components will significantly affect the coating structure. If the character of friction welding, that is a significant thermal gradient initiated in micro-areas, as well as a small transformation zone and a very short duration of the process are taken into consideration, adverse conditions for maintaining structural homogeneity of micro-coatings in the friction pair can be presumed. For that purpose, it is essential to trace the behaviour of particular components of the process in the model system and, thereby, predict a probable stoichiometric composition of the anticorrosive coating being created.

2 Model of a Surface Heat Source in a FSP Process In basic FSP processes various shaping tools are used depending on the material being processes, thickness of the layer being modified as well as predicted basic process parameters [4,5]. In the model considered here a cylindrical shape of the tool of a base radius Rc (Fig. 1) was taken. The base process parameters were a constant pressure force Fn, a rotational speed ω, and a linear velocity VL of the processing tool. In the model of process, the time of thermal field initiation is not accounted for in the description of this field but it constitutes a starting point of a surface heat source formation. In practice, it means that the model of thermal field does not encompass the process of a surface heat source formation. In solving boundary conditions for a particular thermal field, a surface heat source accounts for the mentioned time as a system variable of some kind that models the surface heat source. Taking into account the tool shape it was assumed that the geometry of surface heat source power is modeled by an approximation model of annular normal distribution (Fig. 2) that depends on the conditions and parameters of the FSP process. However, a friction force Fn caused by a rotary motion of the tool has the biggest contribution to the formation of surface heat source geometry.

Formation of a Metal Coating by Means of Friction Stir Processing

169 An aproximation model of a heat source

Fn

! VL

VL

RC

RO

Fig. 1 Model of a FSP process.

RC

Fig. 2 Geometric and an approximation models of a surface heat source at the initial stage of a FSP process.

It is a consequence of significant differences in linear velocities which result from rotary and translational motions of the tool. For the simplified friction pair model, the heat resulted from a material deformation in the layer are not accounted for. Therefore, the surface heat source power distribution q(V) for a rotarytranslational motion of the tool may be expressed by the form Eqn. (1):

q (V ) = μ ⋅ Fn ⋅ (2π ⋅ R ⋅ ω + V L )

(1)

where μ is friction coefficient, VL - a linear velocity of translational motions of the tool along basic symmetry axes for the assumed FSP model, Fn is a normal pressure force of a tool and R – a current radius of a processing tool (0 < R < Rc ). It is a description of a so-called geometric model of a heat source, which precisely represents the participation of a rotary motion of a tool in the formation of power distribution in a surface heat source. A geometric shape of the heat field model is presented in Fig. 2. This figure also illustrates power distribution in a surface heat source at the initial stage of the FSP process. In the proposed approximation model of heat source, it was assumed that the initiation time of the FSP process on the material surface deformed only the geometry of surface heat power distribution through normal and surface heat flux in the processing tool and in the base material. On the basis of the results published in [4, 6- 9] as well as own research, the approximation model of a surface heat source may be described by means of the relation Eqn. 2: ⎡ 2 ⋅ (R − Rc )2 ⎤ ⎥ I ( R ) = I o ⋅ exp ⎢− 2 ⎢ ⎥ R o ⎣ ⎦

(2)

170

D. Kocańda, A. Górka, and D. Zasada

where I o =

2⋅Q ⎡ ⎛ 2 ⋅ Rc 2 ⎞⎟ π ⋅ Ro 2 ⋅ ⎢exp⎜ − + 2⋅π ⎢ ⎜ 2 ⎟ R o ⎠ ⎣ ⎝

⎛ R ⋅ ⎜⎜ c ⎜ Ro ⎝

⎞ ⎟ ⋅ erfc⎛⎜ − ⎜ ⎟⎟ ⎝ ⎠

2 ⋅ Rc Ro

⎞⎤ ⎟⎥ ⎟⎥ ⎠⎦

Symbol Q means a total power of a surface heat source, R0 – a Gaussian distribution radius (notation as in Fig. 2), Rc – a radius of annular normal power density distribution (Fig. 2), I (R) – intensity of a surface thermal field in a function of distance R. Equation (2) reflects actual conditions of the discussed FSP process relatively accurately. At the same time it constitutes a boundary conditions for the thermal field model. In this model parameters Ro, Rc, Q are constant. In practice, they are determined for particular process conditions, dimensions and shapes of used processing tools, thermal properties of a coating material, a base and a tool etc. In the present work, the said parameters will be defined as invariants of the process.

3 Description of a Thermal Field in an FSP Process Energy balance equation for the discussed model of FSP process (Fig. 1) and for variable thermal parameters of a processed material is given by Eqn. 3: ∂ ∂ xo

⎡ ∂ ∂T ⎤ ⎥+ ⎢λ (T ) ⋅ ∂ xo ⎦ ∂ y ⎣ o

⎡ ∂ ∂T ⎤ ⎥+ ⎢λ (T ) ⋅ ∂ yo ⎦ ∂ z o ⎣

⎡ ∂T ⎡ ∂T ⎤ ∂T ⎤ − VL ⋅ ⎥ ⎥ = c(T ) ⎢ ⎢λ (T ) ⋅ ∂ x o ⎦ (3) ∂ zo ⎦ ⎣ ∂t ⎣

with the following initial boundary conditions: Tp = T ( xo = ±∞, yo, zo, t ) = T ( xo, y = ±∞, zo, t ) = T ( xo , yo, zo = ±∞, t ) = T ( xo, yo, zo, t = ∞)

I ( R ) = −λ (T ) ⋅

∂T ( x o , y o , z o => 0, t ) ∂z o

c(T ) = c p (T ) ⋅ ρ (T )

where: (x0, y0, z0, t) - space-time coordinates of a fixed reference system, Tp - an initial process temperature, I(R) - intensity of a thermal field defined by Eqn. (2), VL - a linear velocity of a processing tool, λ(T) - a thermal conduction coefficient of a processed material, cp(T) - a thermal capacity of a material at a constant pressure, ρ(T) - mass density of a material at a constant pressure, t - time in the process is counted from the moment of a surface heat source formation. A solution of the above heat condition equation is presented only for constant material parameters in the form of Eqn. 4:

Formation of a Metal Coating by Means of Friction Stir Processing

171

⎧ 2⎤ ⎫ ⎡ ⎪ k ⋅ ⎢ y 2 + ⎛⎜ x + α ⋅ ⎛⎜1 + ξ 2 ⎞⎟ ⎞⎟ ⎥ ⎪ ⎠ ⎠ ⎥ ⎪⎪ ⎝ ⎪⎪ k ⋅ z 2 ⎝ ⎢ 2 ⋅ Tmax 2⋅ k ⋅a⋅t 1 ⎦ dξ T ( x , y, z , t ) = T p + ⋅ ⋅ exp ⎨− − ⎣ ∫ ⎬ 2 2 2 π 1+ ξ ⎪ ξ ⎪ 0 1+ξ ⎪ ⎪ ⎪⎩ ⎪⎭

Tmax =

Io ⋅ π 2⋅λ⋅ k

λ = λ (T = const.)

k=

1

x

Ro 2

y

a = a(T = const.)

} = ∓ ( R ± Rc )

α=

(4)

VL 4⋅a ⋅ k

where: (x, y, z, t) - space-time coordinates of a moving system connected with a maximum temperature of the thermal field Tmax (t→∝), α - a correction constant of a moving reference system, a - a temperature equalization coefficient for T=const, I0 - defined by Eqn. (2), R, Rc, Ro –notations as in Figs 1 and 2. The above equations are a basic thermal field description which constitutes a foundation for further verification research on thermal effects observed in a model of FSP process. In this model Rc, ω, VL are independent process parameters, that is process invariants. Whereas dependant parameters that considerably influence process effects are as follows: -

-

-

a normal distribution, Ro, accounted for in an approximation model of a friction pair as a variable dependant mostly on a base, a coating and a tool materials, a tool shape as well as independent process parameters, relative time t, related to a stationary thermal field model, dependant on a surface heat source shape in the first place and indirectly on the type and thickness of a surface layer being formed as well as process conditions (e.g. cooling conditions in a friction pair, types of used materials etc.), Vm a resolved linear velocity of the tool and the base material in the process zone.

The above mentioned variable parameters of a FSP process are determined at the stage of a thermal field initiation. Considering both their complexity and their practical character in the study, they will not be subjected to an in-depth analysis. The assumed purpose and domain of the study do not require such an analysis. In the study, focus was placed on temperature distributions along basic symmetry axes for the assumed FSP model under fixed process conditions. Figure 3 illustrate sample spatial temperature distributions of a moving heat source at different depths of the heat affected zone in a steel base for friction process parameters. Depth of the layer and field temperature were described by relative units related to the field.

172

D. Kocańda, A. Górka, and D. Zasada

a)

b)

c)

Fig. 3 Spatial temperature distributions in a steel base for the model parameters, respectively: a) Ro=Rc, to=const, VL=const, z=1, b) Ro=Rc, to=const, VL=const, z=0.1, c) Rc=2Ro, to=const, VL=const, z=0.1, z – coordinate of a depth.

These distributions are different, especially in the zone of maximum field values. In the case of tools with larger diameters Rc (Rc >> Ro), the differences are even more distinct (Fig. 4c). This fact determines variable thermal conditions in the layer. Under actual process conditions, it may lead to the existence of different conditions of coating and base structures formation. In the case of typical constructional materials and assumed temperatures of an FSP process (Tmax=900 1050o C), high temperature mass intermix determines diffusion bonding of the materials. This is a primary qualitative feature of the discussed process. At the same time, variable thermal conditions in the zone of thin coatings created in an FSP process may lead to a so-called undulation of a separation line (a variable thickness of a coating being created) Fig. 4. In the case of thick coatings and variable direction of resultant vector Vm, it is a positive property. It may lead to the preferential directions of mass intermix which are shown in Fig. 5. However, in the case of thin layers, it may lead to their discontinuity. Simulation of thermal effects on a coating-base boundary is not that unequivocal. Under actual conditions, it should be expected that a separation boundary undulation occurs as a result of variable thermal conditions of the field, that is temperature and duration of the said conditions at a particular field level. In the case of a boundary between a thin coating and a base, the undulation will look like the one which is shown in Fig. 4a. A boundary separation shape depends on thermal parameters of a coating material to a large extent. In the case of thicker coatings with comparable thermal parameters of a base, coating materials and tool material, the undulation will be of a different nature (Fig. 4b). The effects will significantly influence processes of layers formation on particular bases. Temperature distribution inside the base supports the assumption (Fig. 4c). Variable thermal conditions in a coating being created as well as inside a base may considerably affect layer properties observed once the process has been completed.

Formation of a Metal Coating by Means of Friction Stir Processing

a)

173

Coating Base

b)

c)

Fig. 4 An exemplified thermal field lines observed in a steel base after a cut, a) simulation of thermal effect observed on a boundary between a thin coating and a base at a large participation of a cooling heat flux in a coating material, b) simulation of thermal effect observed on a coating-base boundary for comparable thermal parameters of a coating, a base and a tool materials, c) simulation of thermal effect observed in a base material at its greater depths, directly under a coating being formed.

a)

b)

c)

Fig. 5 Practical realization of coating formation by means of FSP: (a) cross section of the surface layer along the direction of tool motion, (b) cross section of the surface layer, the case of dextrorotatory tool motion relative to its translational movement, (c) cross section of the surface layer, the case of levorotatory tool motion.

174

D. Kocańda, A. Górka, and D. Zasada

4 Mass Transport in Micro-areas of a Modelled Friction Pair Steel- nickel and steel-copper were used as exemplified components of a binary diffusion system. In binary systems adopted for analysis, on the separation boundaries of layers being formed, in short time intervals, the resultant mass flux is a sum of non-coupled effects of concentration and thermal inputs [10, 11]. Under the condition of a terminal thermodynamic equilibrium, resultant mass fluxes in such a model lead to the formation of characteristic concentration distributions. From among the mentioned concentration distributions, there can be distinguished ones that are created as a result of either interaction or discrimination of noncoupled components [10, 11]. On the basis of previous research [3, 4 ,7, 8] it can be predicted that in the FSP model a diffusive character of the separation boundary will induce mass transport in both directions of the vector grad(T). For better understanding of diffusion phenomena, that take place in a thermal field, it is essential to investigate the analytical form of the concentration distribution line of a saturating element in the conditions of intermediate as well at relative - "frozen" thermodynamic equilibrium. In the case of correlated thermal and concentration flows in an open diffusion system, it is possible to introduce to the analytical description a parameter that defines changes in the concentration of the solution components under thermodiffusion. This parameter is described by the Soret coefficient [2, 3]. The analytical solution of the complex problem of mass transport in a semi-infinite body along the line of the highest temperatures of the field can be presented in the following form [11, 12]:

c( z, t ) =

⎡ ( z + D ⋅ h ⋅ t )2 ⎤ Q ⋅ h ⎡ z − D ⋅ h⋅t ⎤ ⎥+ ⋅ exp ⎢− ⋅ exp(− h ⋅ z ) ⋅ erfc ⎢ ⎥ 4⋅ D ⋅t 2 ⎢ ⎥ π ⋅ D⋅t ⎣ 2⋅ D⋅t ⎦ ⎣ ⎦ Q

(5)

where: Q = co ⋅ d h=s⋅g s is the Soret coefficient, constant for a narrow temperature range, g - gradient of temperature, constant for a narrow temperature range, co – relative mass concentration of an element in diffusion system, d – thickness of coating level of a saturating element, t – time, z – coordinate of layer depth. An exemplified comparison of the tracer concentration distribution, for the variants of identical directions of thermal and concentration diffusion inputs (unidirectional grad(T) and grad(N) vectors), indicates that for the thermodiffusive model the mass transport rate is many times higher (Fig. 6). This rate should guarantee a complete diffusion bonding of materials in a formed layer. For that purpose, a verification research was undertaken, despite of the FSP process guaranteed low temperatures.

Formation of a Metal Coating by Means of Friction Stir Processing

175

Fig. 6 Exemplified distributions of the tracer concentration for a thermal diffusion and a concentration diffusion characterized by identical directions of thermo-dynamic inputs [3].

5 Verification Research In verification tests on the FSP process, a tool of a radius Rc= 20mm and a flat face was used ( Fig. 1). Copper and nickel coatings were spread over a 355 steel base with the use of galvanic methods (a condition for an adhesive bonding of the coating and the base). The research was carried out with the use of a standard milling machine at a rotational speed of the tool of ω=890[rev/min] and a feed speed of approximately VL= 0.1[m/min]. The results of tests concerning surface layers being formed are presented in Figs 7- 10. In Fig. 7a, there can be observed a copper coating on a 355 steel base after the FSP process. After this process, the geometry of the coating-base separation line changed from a rectilinear to an undulated one and showed a significant similarity to the line predicted in the simulated process presented in Fig 4a. The mentioned layer, which originally had an adhesive bonding with the base, showed a diffusive character of the bond along the whole width of the tool. The band structure of the layer is noteworthy. This structure explicitly documents the phenomena that accompany the FSP process in the formation cycle. It is a new quality in the processing technology. A similar situation takes place in the case of a nickel coating (Fig. 7b), except that the undulated line has a somewhat different character than it would follow from Fig. 4b. It can result from great similarity of the cooling fluxes either in the coating, the base or the processing tool. Moreover, in the layers being formed, there can be observed a diffusive character of the separation lines of bonded materials (Fig. 8). This fact, together with the changes in the thermal field illustrated in Fig. 4c, document the complex process of layer formation in the FSP process. Therefore, it is necessary to allow for diametrically different conditions of the layer structure formation at various depths. Thereby, when considering complex diffusion process, a greater homogeneity of the basic components of materials bonded in the FSP process should be expected.

176

D. Kocańda, A. Górka, and D. Zasada

a)

Ni coating

b) Steel base

Processing tool diameter

Fig. 7 Actual images after the completion of a FSP process on 355 steel base: a) Cu coating, b) Ni coating.

a)

b)

c)

Fig. 8 Structure of the coating layers of the area marked by a rectangle in a) Figure 7a b) Figure 7b after the completion of a FSP process.

For that reason, there was conducted an analysis of surface distributions of particular components of bonded materials. In the case of Cu coating deposited on a 355 steel base, primary focus was given to the surface distribution of copper and iron in the layer (Fig. 9). In the case of Ni coating, surface distributions of iron, chromium, nickel and manganese were subject to analysis (Fig. 10). In this research, the phenomena connected with structural transformations that take place as a result of mass transfer were not taken into account. The mentioned phenomena are observed in standard structural tests in the form of distinct bands in the structures of layers being formed, which are noticeable in Fig. 8.

Formation of a Metal Coating by Means of Friction Stir Processing

a)

177

c)

b)

Fig. 9 Results of spectral analysis of copper coating on 355 steel base after the FSP process: a) an image of the area being analysed, b) surface distribution of Cu in the coating, c) surface distribution of Fe in the coating.

a)

b)

d)

c)

e)

Fig. 10 Results of spectral analysis of nickel coating on 355 steel base after the FSP process: a) an image of the area being analysed, b) surface distribution of Fe in the coating, c) surface distribution of Cr in the coating, d) surface distribution of Ni in the coating, e) surface distribution of Mn in the coating.

On the basis of the research results presented above, it can be stated that main coating and base components of the formed layer are uniformly distributed (mixed) as a result of the FSP process. Observed diffusion processes aid the support this process.

6 Conclusion In the study, main focus was given to the theoretical aspect of the FSP process simulation as well as on diffusion mass transport in the micro-areas of a layer

178

D. Kocańda, A. Górka, and D. Zasada

formed in the FSP. The proposed description of the thermal field from an analytical point of view is of practical significance in the initial analysis of phenomena initiated in the FSP process. It should be noted that this description does not give consideration to the initial time of the formation of a stable thermal field due to the complexity of phenomena induced at this initial phase. Thanks to such an assumption, the field description has a simpler form. Regarding the practical aspect of the study, the analysis of mass transport in the FSP process was focused on steel-nickel and steel-copper binary diffusion systems. Verification study confirmed the correctness of the presented thermal field description which was particularly noticeable in the case of simulation of diffusion separation line between the Cu or the Ni surface coating on a 355 steel base by means of the FSP process. On the basis of a simplified model analysis and the results of verification study, it is possible to plan technological research connected with the use of the FSP process to form surface layers on the base of constructional materials. Acknowledgement. The research was financially supported by the Polish Ministry of Science and High Education.

References [1] Kocańda, D., Górka, A.: Biul, WAT 2, 395 (2010) (in Polish) [2] Schmidtand, H.J., Schmidt-Brandecker, B.: In: Proceedings of the 2nd Int. Conf. Material and component performance under variable amplitude loading, Darmstadt, Germany, pp. 67–87 (2009) [3] Kocańda, D., Górka, A.: J. Strength of Materials (2010) (in Press) [4] Schmidt, H.B., Hattel, J.H.: Elsevier Scripta Materialia 58, 332 (2008) [5] Hattingh, D.G., Blignault, C., van Niekerk, T.I., James, M.N.: Elsevier Journal of Materials Processing Technology 203, 46–57 (2008) [6] Górka, A.: Laser technology VI: Applications. In: Proecedings of the SPIE, Szczecin, Poland, vol. 4238, pp. 163–173 (1999) (in Polish) [7] Schmidt, H., Hattel, J., Wert, J.: Int. J. Modeling and Simulation in Materials Science and Engineering 12, 143–157 (2004) [8] Nandan, R., Roy, G.G., Debroy.: Metallurgical and in Materials Science and Engineering (April 2006) [9] Kar, A., Langlais, M.D.: J. Optical and Quantum Electronics. 27, 1165 (1995) [10] Krupkowski, A.: Fundamental Problems of Theory of Metallurgical Process. PWN Warszawa, Kraków (1974) (in Polish) [11] Buda, M.J.: Suplement Biul. WAT 1, 329 (1980) (in Polish). [12] Górka, A.: Biul. WAT 9, 89 (1993) (in Polish)

26th ICAF Symposium – Montreal, 1-3 June 2011 Influence of the Carbon Nanotube Type, Loading and Chemical Functionalization on the Fatigue Resistance of Aluminum Lap Joints Iosif D. Rosca, Roham Mactabi, and Suong V. Hoa Concordia University, Mechanical and Industrial Engineering, Concordia Center for Composites, Center for Research in Polymers and Composites (CREPEC), Montreal, Canada

Abstract. Different types of carbon nanotubes (CNTs) with and without attached functional groups and with different loadings were used to prepare electrically conductive epoxy adhesive. By incorporating only 1 wt% of pristine single-wall CNTs the resistance of single lap joints (SLJs) was reduced by over 10 orders of magnitude compared to the neat resin. The type, loading and functionalization have little effect on the apparent shear strength of SLJs as the joints displayed mostly adhesive failure. Nanotube loading and functionalization have a strong influence on the fatigue life of the SLJs. Over 4 wt% of multiwall CNTs the fatigue life is only half of that of the neat resin. Repeated functionalization gradually decreases the length of nanotube pullouts and significantly increases the fatigue life. The tradeoff of improved fatigue life is a higher electrical resistance.

1 Introduction Structural adhesives are extensively used to build lightweight structures in aerospace and automotive industries. Electrical continuity and electrostatic dissipation capabilities are usually requested for these structures. Since all of these adhesives are electrical insulators, the structures must be grounded by time intensive operations like silver brazing or strapping. Currently, carbon nanotubes (CNTs) are intensively investigated as efficient fillers for electrically conductive composites [1-2]. Also, CNTs were found to positively influence fatigue life and fatigue crack propagation rates when dispersed in epoxy matrices [3-6]. However, there are no studies on the fatigue behavior of adhesives joints based on CNTs. The aim of the present paper is to investigate the influence of the CNT type (single and multiwall), loading and chemical functionalization on the fatigue resistance and electrical conductivity of aluminum lap joints.

2 Materials and Methods Materials Industrial grade multiwall carbon nanotubes (MWCNTs) with a measured average length and diameter of 3.3 μm and 11.5 nm respectively [1] were purchased from

180

I.D. Rosca, R. Mactabi, and S.V. Hoa

NanoLab Inc. Single wall nanotubes (SWCNTs) were provided by Nikkiso Co. with a measured average length and bundle-diameter of 5.4 μm and 15.2 nm respectively [2]. The epoxy resin Epon 862 and the curing agent Epikure W were purchased from Hexion Specialty Chemicals. Benzoylperoxide (BPO) was purchased from Sigma Aldrich and used without further purification. Aluminum (2024 T3 alloy) plates were purchased from McMaster Carr. Single lap joint (SLJ) preparation CNTs were dispersed in the epoxy resin by three-roll milling as described in our previous work [1]. Briefly, the resin, the curing agent (26.4 wt%) and a determined quantity of CNTs were weighed and hand mixed to form batches of 10 g. The batch was three-roll-milled for several times on a laboratory scale three rollmill (EXAKT 80E, EXAKT Technologies, Inc.) at a gradually smaller gap between the rolls. Next, the mixture was degassed in a vacuum oven at 90 °C for 30 min. The aluminum plates cut to dimensions as shown in Fig. 1 were degreased in acetone and etched in chromic acid solution for 30 min at 65 °C. The adhesive was applied in less than one hour after the surface preparation. SLJs were prepared by applying a thin layer of conductive adhesive, on each adherent. Next, the two adherents were tightened together using C clamps and cured in an oven at 175 °C for 4 hours. Bondline thickness of 0.2 mm was controlled by adding glass beads. CNT functionalization Typically, 1 g of CNTs, 10 g of benzoylperoxide and 500 mL of toluene were loaded in a round bottomed flask. The mixture was stirred on an oil bath at 102 °C for 4 hours. The functionalized CNTs were separated on a Nylon filter membrane of 45 μm pore size and washed twice with toluene and dried at 120 °C for 12 hours. In order to increase the degree of functionalization the above procedure was repeated up to 8 times (8xBPO).

Fig. 1 Single-lap joint for fatigue testing; dimensions in mm.

Measurements The resistance of the bonded joint was measured by four-wire method using a current source (Keithley 6220 DC) and a nanovoltmeter (Keithley 218A).

Influence of the Carbon Nanotube Type, Loading and Chemical Functionalization

181

The apparent shear strength of the simple lap joints was measured on an MTS 100kN testing machine at 1.3 mm/min strain rate. In order to expedite the fatigue tests, the maximum load was 60% of the average shear strength and the load ratio was 0.1. The tests were carried out on an MTS 100kN testing machine at 5 Hz. The Raman spectra were recorded on an inVia Renishaw spectrometer equipped with a 514.5 nm argon-ion laser. The thermogravimetric analysis (TGA) was run on a TA Instruments Q50 in nitrogen atmosphere at 10 °C/min.

3 Results and Discussion Fig. 2 presents the apparent shear strength of the SLJs prepared with adhesive containing different pristine or functionalized CNTs at different loadings. The CNT type, loading and functionalization have no significant influence on the shear strength of the SLJs as the joints displayed mostly adhesive failure. The CNT type, loading and functionalization have a strong influence on the electrical resistance of the SLJ (Fig. 3). Lap joints prepared with neat resin have resistances of 5x1011 Ω. By adding 1 wt% of SWCNTs the bond resistance is reduced by more the 10 orders of magnitude to 26.5 Ω. At the same loading the SWCNTs results in lower resistances than MWCNTs because the former are longer [2] and display a more uniform structure i.e. narrow diameter distribution, similar morphology and less waviness (Fig. 4 b). Longer CNTs with uniform structure will generate a more conductive network throughout the polymer matrix [1, 2]. 30

Neat resin MWCNT SWCNT

20 15

106 104

10

102

5 0

MWCNT SWCNT

Resistance, Ω

Shear Strenght, MPa

25

108

0%

1%

2%

4% 8x-1%8x-2% 1% 8x-1%

CNT loading, wt%

Fig. 2 Shear strength of the SLJs.

100

1%

2%

4% 8x-1% 8x-2% 1% 8x-1%

CNT loading, wt%

Fig. 3 Electrical resistance of the SLJs.

182

I.D. Rosca, R. Mactabi, and S.V. Hoa

Fig. 4 TEM micrographs of the as received MWCNTs (a) and SWCNTs (b).

While CNTs seems to have less influence on the shear strength they have a strong influence on the fatigue life of the SLJs (Fig. 5). During the fatigue tests SLJs showed mostly adhesive failure. We have observed also crazing, especially for the adhesives containing SWCNTs. Crazing appears as a thin whitened line normal to the load direction (encircled region in Fig. 6). According to Zhang et al [7] crazing contributes to energy dissipation, but the fatigue life of SLJ containing CNTs did not show a clear improvement over the neat resin with one exemption of that of the highly functionalized SWCNTs (8xBPO-SWCNT in Fig. 5).

Neat resin MWCNT SWCNT

Number of cycles to failure

12500 10000 7500 5000 2500 0 0%

1%

2%

4% 8x-1% 8x-2% 1% 8x-1%

CNT loading, wt%

Fig. 5 Fatigue life of the considered SLJs.

Influence of the Carbon Nanotube Type, Loading and Chemical Functionalization

183

Fig. 6 SEM micrograph of a fractured SLJ.

Nanotube loadings over 1 wt% increase the brittleness of the adhesive [8-10]. For brittle adhesives increase in crack growth rate was observed [11], which explains the reduced fatigue life at high MWCNT loadings (> 4 wt%). Recently, Gojny et al. [12] and Mirjalili et al. [13] evidenced that CNT bridging may inhibit crack initiation and reduce crack propagation by toughening the polymer matrix. SEM investigation of the fractured samples evidenced CNT-bridging (Fig 7a, c), but there was no clear improvement in the fatigue life compared to that of the neat resin (Fig. 5). As it can be observed from Fig 7a and c the CNT pullouts are quite long (several micrometer long) that means low interfacial strength. As the CNTs have smooth surface, the van der Waals forces alone seems to be insufficient to efficiently anchor the nanotube into the matrix. In a recent review article Bose et al. [14] showed that covalent functioanalization of the CNT surface improves the mechanical properties of the composites by increasing the interfacial strength between the CNT and the polymer matrix. Increased interfacial strength implies shorter CNT pullouts. Indeed, functionalized CNTs display much shorter pullouts than the as received CNTs (Fig. 7 b, d and e). However, the fatigue life is clearly improved only for the highly functionalized SWCNTs (8x BPO). TGA analysis and Raman spectroscopy evidenced that MWCNTs and SWCNTs behave totally different in the functionalization process. Due to their smaller diameter and higher curvature, SWCNTs are more reactive than MWCNTs. This was confirmed by TGA analysis. After 8 times of repeated functionalization SWCNTs contain more than four times more functional groups than MWCNTs (Fig. 8). More functional groups lead to an increased interfacial strength. As the number of functionalization cycles increases the CNT pullouts get smaller and more CNT ruptures are observed (Fig 7 d and e).

184

I.D. Rosca, R. Mactabi, and S.V. Hoa

Fig. 7 SEM micrographs of the fractured joints at 1 wt% CNT loading; a - as received MWCNTs, b - 8xBPO functionalized MWCNTs, c-As received SWCNTs, d - 4xBPO SWCNTs, e - 8xBPO SWCNTs. (scale bar 1 μm).

100

100

AR 8.6%

Weight, %

12.4%

80 24.9%

60

36.5%

50 300 400 500 Temperature, °C

a.

6.5 % 7.9 %

90

4x 8x

4x

85

8x

200

AR

95

2x

70

40 100

1x

Weight, %

90

600

700

100

200

300 400 500 Temperature, °C

600

b.

Fig. 8 TGA plots of the functionalized CNTs; a - SWCNTs; b – MWCNTs.

700

Influence of the Carbon Nanotube Type, Loading and Chemical Functionalization G-1592 cm-1

G-1592 cm-1

1200

1300

1400 1500 1600 Raman shift, cm-1

a.

cm-1

Intensity, a.u.

Intensity, a.u.

D-1343

D-1343 cm-1

185

AG/D 6.02 5.58 3.72 4.48 50.5

8x 4x 2x 1x AR

1700

1800 1000

AG/D 1.23

1250

8x

1.19

4x

1.11

AR

1500 1750 Raman shift, cm-1

2000

b.

Fig. 9 Raman spectra of the functionalized CNTs; a - SWCNTs; b – MWCNTs.

Even after 8 times functionalization of the MWCNTs the fatigue life of the joint is still not visibly improved, because the functional group content is relatively low (7.9 %) probably insufficient to significantly increase the interfacial shear strength. The area ratio of the G and D peaks (AG/D) of the Raman spectrum is commonly used to assess the extent of the CNT functionalization. In the case of the MWCNTs (Fig. 9b) this ratio increases unexpectedly after consecutive functionalizations. This can be explained by the fact that MWCNTs contains important amounts of amorphous carbon (AG/D close to the unity) which is partly rendered soluble by the reaction with BPO and eliminated during the washing and filtration processes. The influence of functionalization on the SWCNTs is twofold. First, after subsequent funtionalization the bundle diameter decreases (Fig. 10) leading to an increased interfacial surface, which in turn contributes to a higher interfacial strength. Secondly, high degree of functionalization (36.5 % Fig. 8a) results in very short pullouts and more CNTs ruptures proving an efficient load transfer (Fig. 7e). The combined effects of small bundle-diameter and short CNT-pullout increases significantly the fatigue life of the SLJ based on 8xBPO SWCNTs compared to that of the neat resin (Fig. 5). The tradeoff of improved fatigue resistance is a significantly increased electrical resistance from 26.5 Ω of the as received SWCNTS to 100 kΩ of the 8xBPO SWCNTs.

186

I.D. Rosca, R. Mactabi, and S.V. Hoa

Fig. 10 SEM micrographs of functionalized SWCNTs (scale bar 1 μm).

4 Conclusions Adding small amounts of MWCNTs or SWCNTs the electrical conductivity of structural adhesives is increased by up to 10 orders of magnitude. SWCNTs showed lower electrical resistances as they are longer and have more uniform structure than MWCNTs. The resin system used to prepare the SLJs displayed mostly adhesive failure in quasi static as well as in fatigue testing. The type, loading and functionalization have little influence on the apparent shear strength of the SLJs. The fatigue life depends strongly on the CNT loading and functionalization. At high MWCNT loadings (> 4 %) the fatigue life of the SLJ is only half of that of the neat resin. One-time functionalization is not enough to generate significant amount of functional groups. Only after 8 times of functionalization SWCNTs displayed a significantly improved fatigue life compared to that of the neat resin and nonfunctionalized CNTs. The tradeoff of improved fatigue life is a significantly increased electrical resistance.

References [1] Rosca, I.D., Hoa, S.V.: Carbon 47, 1958 (2009) [2] Rosca, I.D., Hoa, S.V.: In: Proceedings ASC 24th Tech. Conf., CD ROM., Newark, DE (2009)

Influence of the Carbon Nanotube Type, Loading and Chemical Functionalization [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14]

187

Grimmer, C., Dharan, C.: J. Mater. Sci. 43(13), 4487 (2008) Zhang, W., Picu, R.C., Koratkar, N.: Appl. Phys. Lett. 91,193109 (2007) Zhang, W. Srivastava, I. Zhu, Y. Picu, C. Koratkar, N.: Small 5 (12),1403 (2009) Rafiee, M., Rafiee, J., Wang, Z., Song, H., Yu, Z., Koratkar, N.: ACS Nano. 3(12), 3884 (2009) Zhang, W., Srivastava, I., Zhu, Y.F., Picu, C.R., Koratkar, N.A.: Small 5(12),1403 (2009) Breton, Y., Desarmot, G., Salvetat, J.P., Delpeux, S., Sinturel, C., Beguin, S., Bonnamy, S.: Carbon 42, 1027 (2004) Li, X.F., Lau, K.T., Yin, Y.S.: Composites Science and Technology 68, 2876 (2008) Yeh, M.K., Hsieh, T.H., Tai, N.H.: Materials Science and Engineering A 483-484, 289 (2008) Azari, S., Papini, M., Schroeder, J.A., Spelt, J.K.: Engineering Fracture Mechanics 77, 395 (2010) Gojny, F.H., Wichmann, M.H.G., Fiedler, B., Schulte, K.: Composite Science and Technology 65, 2300 (2005) Mirjalili, V., Hubert, P.: Composite Science and Technology 70, 1537 (2010) Bose, S., Khare, R.A., Moldenaers, P.: Polymer 51, 975 (2010)

26th ICAF Symposium – Montreal, 1-3 June 2011 Fatigue Damage Behavior of Glass/Epoxy Composites Using Carbon Nanotubes as Sensors H. Hena-Zamal and S.V. Hoa Concordia Center for Composites Department of Mechanical and Industrial Engineering, Concordia University Center for Research in Polymers and Composites (CREPEC) Montreal, Quebec, H3G 1M8, Canada [email protected], [email protected]

Abstract. Carbon nanotubes have been used as sensors in glass fiber/epoxy laminates for monitoring damage under fatigue loading conditions. Carbon nanotubes are embedded in epoxy resin. The modified resin is infused into glass fibers to make composites. Silver paste electrodes are deposited on the surface of the laminate. Changes in electrical resistance along the plane and through thickness of the laminate are measured. The changes in resistance are compared with the strain changes measured by strain gages. It is found that the changes in electrical resistance are more sensitive to the degradation in the composite laminate as compared to strain gauges. This technique may be used to provide in-situ monitoring of damages in composites.

1 Introduction During the last decades, fiber reinforced polymer (FRP) composite materials are extensively used in many industrial applications. For proper utilization of composite materials and for the safety of the structure, the detection and monitoring of damages, crucially fatigue damage of composite parts, have become important issues. A number of non-destructive damage detection techniques such as, ultrasonic, eddy current, radiography, dye penetrants, magnetic particles, fiber optical sensor are available. However, most of them are not suitable for in situ damage monitoring because of the requirement of special laboratory set ups. For carbon/epoxy composites, monitoring the change of electrical resistance during loading cycles is one attractive method as it allows in-situ monitoring of fatigue damage with great simplicity, low cost and without compromising the properties of host materials. For glass fiber reinforced polymer composites where both the matrix and reinforcement are non-conductive, this method fails, however, for obvious reasons. Carbon nanotubes (CNTs) which possess outstanding thermal and electrical properties combined with high specific stiffness and strength, and very large aspect ratios (l/d) can provide a good solution for this problem. CNTs are incorporated into the epoxy matrix to make it conductive. This facilitates the application of electrical resistance measurement as a damage sensing technique. Multi walled carbon nanotubes (MWCNT) are used since they offer the highest potential for enhancement of electrical conductivity in polymer composites. As a

190

H. Hena-Zamal and S.V. Hoa

result, even at very low concentration, MWCNT can create an conductive network in interfacial surface area when well dispersed in polymer resin system [1-3]. In some recent research works [4-31], it is established that electrical resistance measurement method is capable of monitoring damages in composite materials effectively. The electrical resistance is very sensitive to the formation of micro damage in a structure which changes in a regular pattern making it relatively easy to detect and characterize the extent of damages in the composite structures. Among numerous works on this subject, [14-25] investigated the effectiveness of electrical resistance method with Carbon Fiber Reinforced Polymer (CFRP) composites with or without the embedded CNTs. On the other hand, the studies in [26-31] were dedicated to the monitoring of fatigue damage behavior of Glass Fiber Reinforced Polymer (GFRP) using the electrical resistance method exploiting the embedded CNT networks. Boger et al. [26] investigated the damage behavior of CNTs and carbon black incorporated [0o,+45o,90o,-45o,+45o,90o,-45o,0o] GFRP laminates produced by vacuum assisted resin transfer molding (VARTM) using the electrical resistance method. They found this method very promising for damage monitoring and supposed it to be more sensitive than other damage sensing techniques. Thostenson and Chou [27], in their works, demonstrated that the percolating networks of carbon nanotubes are remarkably sensitive to the matrixdominated failure and can detect damage of the GFRP composite in situ. They assessed the deformation/resistance response under cyclic loading conditions of unidirectional glass fiber and cross-ply glass fiber mat. They also demonstrated that it is possible to identify the nature and progression of damage using this method. Chou et al. [28] examined the resistance response of the cross-ply GFRP composite with varying thickness for cyclic loading and identified its relation with the extent of damages. Nofar et al. [29] investigated the electrical resistance response of woven GFRP composites under static and dynamic loading. They also introduced delamination artificially in some of their samples, as in [30], and tested under cyclic loading for different maximum loads. They concluded that carbon nanotube networks in monitoring damage has better sensitivity than strain measurements. Gao et al. [31] in their works, also studied the damage evolution of cross-ply glass-fiber-reinforced composite under cyclic loading using the electrical resistance method. They made quantitative measurements of accumulated damages and correlated with the change in resistance per unit length for better understanding of the damage evolution. The objective of the work carried out in this project is to compare the effectiveness of using electrical resistance from CNT networks for damage detection, particularly between in-plane resistances and through thickness resistances.

2 Experimental Details Materials and processing Cross-ply laminates were fabricated using unidirectional glass fibers with epoxy resin following hand lay up with autoclave molding process. To produce the conductive networks, multiwall carbon nanotubes (MWCNTs) are dispersed within

Fatigue Damage Behavior of Glass/Epoxy Composites Using CNTs as Sensors

191

the epoxy with the Three-roll Mill calendering machine following the mixing scheme suggested by Rosca and Hoa [32]. DSC (Differential Scanning Calorimeter) tests were carried out on the samples to ensure that they were cured properly. TGA (Thermo Gravimetric Analysis) tests were performed to measure the fiber volume fraction, the typical value of which was found to be around 66%. Arrangement of electrical connections, mechanical testing and measurement The specimens for fatigue test were prepared according to ASTM 3039-76 standard. Screen sandpapers were bonded to both ends of the samples. Silver epoxybased glue was used as the conductive contact for attaching the electrical probes in order to measure electrical resistances. To measure the mechanical deformation, a strain gauge was installed at the middle of the surface of the samples. The dimensions and the arrangement of the electrical connections and strain gauge are shown in figure 1. There are four contact points (1,2,3,4) in-plane of the sample and one contact point (4') in the through thickness plane as shown in the figure.

Fig. 1 Arrangement of electrical connections and dimensions of samples.

The electrical resistances are measured between the following points 1-3 point 2-4 point

2-4' point 4-4' point

Fatigue tests were performed on MTS 100KN machine. The maximum load was varied from 3000N to 12000N while the minimum load was kept at 250N for all tests. The tests were done for different number of cycles. During the tests, electrical resistance measurements were made using a high resistance Agilent meter and, at the same time, strain readings were recorded. The percent change in electrical resistance was calculated as

% ΔR = where, in Eqn (1),

(Ri − R0 ) × 100 R0

(1)

R0 refers to the electrical resistance before loading and Ri re-

fers to the resistance after unloading at a specified number of cycles. Similarly, the change in strains are calculated as

192

H. Hena-Zamal and S.V. Hoa

ΔS = Si − S0 where, in Eqn (2),

(2)

S0 refers to the reference strain value and Si refers to the

strain value recorded after a fixed number of cycles. All reference values are taken while the minimum load 250N was maintained on the sample. Cross-ply [02/902]S GFRP composite laminate samples were tested under different maximum loads up to different number of cycles. The test schedule for various samples are shown in table 1. Each test was carried out on two replicates. Table 1 Test schedule of different samples.

Sample No. I II III IV

Maximum load (N) 6000 8000 10000 12000

Minimum load (N)

250

No. of cycles 500 200 150 100

3 Results and Discussion The samples were observed under SEM in the undamaged state as shown in figure 2. During the fatigue tests, the percentage change of electrical resistance in both the in-plane and through thickness positions with loading cycles are shown in figure 3 (a) for sample I. The corresponding variation of strains with loading cycles are shown in figure 3 (b). In figure 3 (c-1, c-2), the damaged condition as observed under optical microscope is shown for the same sample.

(a) (500x)

(b) (300x)

Fig. 2 SEM images of undamaged sample before testing.

From figures 3(a) and 3(b), it can be observed that the changes in resistances (%) in in-plane positions and changes in strain increase almost linearly up to 500 cycles while in through thickness positions the resistance increases linearly up to 100 cycles and then jumps suddenly. With the stress analysis of glass/epoxy

Fatigue Damage Behavior of Glass/Epoxy Composites Using CNTs as Sensors

193

composite samples, it was found that the first ply failure occurred at around 3500N load. As the tests are done with the maximum loads always above the first ply failure load, the changes in strain from the beginning of loading cycles are found to be high for all samples. The corresponding optical images of the samples in figure 3 (c-1, c-2) show propagation of matrix crack around the fibers which leads to the initiation of delamination. The conductive percolated networks of CNTs are disturbed due to the occurrence of these damages which leads to the variation of resistances. In terms of sensitivity, the changes in in-plane resistances and strains follow the similar trend while the changes in through thickness resistances jump much earlier in the loading cycles. This jump in electrical resistance at certain cycles probably corresponds to the accumulation of micro damages into macro damage events.

(c-1) 4-4' point (200x)

4-4'

1-3 2-4 2-4'

(a)

(c-2) 4-4' point (500x)

(b)

Fig. 3 (a, b). Variation of change (%) in electrical resistance and strain with loading cycle for sample I (6000 N).

Fig. 3 (c-1, c-2). Optical images of sample I after 500 cycles (6000 N).

Almost similar trend is found for sample II tested under 8000N maximum load for 200 cycles as shown in figure 4 (a, b, c). This is because the extent and mode of the damages are similar under this condition to the condition (6000N maximum load for 500 cycles) of the previous test as observed from figures 3 (c) and 4 (c). This implies that, when the damage states (modes, severity, crack propagation etc.) are similar, the response of the electrical resistances resembles in terms of magnitude and trend of the changes as seen in figures 3 (a, b) and 4 (a, b).

194

H. Hena-Zamal and S.V. Hoa

It is worth noting that among the in-plane and through thickness resistances, the latter is found more sensitive to the extent of damage. However, the through thickness resistance response between 2-4' points is similar to the in-plane resistance response. The reason is because along the path of 2-4’, most contacts are still good and only a small portion close to the point 4’ is affected.

(c-1) 4-4' point (200x)

4-4'

1-3 2-4 2-4'

(a)

(c-1) 4-4' point (500x)

(b)

Fig. 4 (a, b).Variation of change (%) in electrical resistance and strain with loading cycle for sample II, 8000 N.

Fig. 4 (c-1, c-2). Optical images of sample II, after 200 cycles (8000N).

In the case of samples III and IV, severe damages (propagation of matrix cracks at multiple locations, delaminations etc.) occurred at the beginning of the loading cycles as observed from figures 5 (c) and 6 (c). The rate of change of strains, as shown in figures 5 (b) and 6 (b), is very high (very steep slope of strain curve) in these very high loads. Although the electrical resistances, at these instances, show similar behavior (linear increase initially with sudden jumps after a certain number of cycles) as in the previous cases, the sensitivity of this seems to be reduced when comparing the magnitude, slopes and jump events in figures 3(a), 4(a), 5(a) and 6(a). At very high loads, severe damages occur at multiple modes at multiple locations at the beginning of loading cycles , as shown in figures 5(c) and 6(c), which leads to very high rate of changes of strain. The severe disturbances in the conductive network (e.g. due to crack opening and closing events during cyclic loading), within a very short time, however, might not be reflected in the acquired resistance data because of slow data acquisition leading to relatively lower sensitivity. However, this should be confirmed with further studies.

Fatigue Damage Behavior of Glass/Epoxy Composites Using CNTs as Sensors

195

(c-1) 4-4' point (500x)

4-4'

(a) (c-2) 4-4' point (500x)

(b)

Fig. 5 (a, b).Variation of change (%) in electrical resistance and strain with loading cycle for sample III, 10000 N.

Fig. 5 (c-1, c-2). Optical images of sample III, 10000 N.

(c-1) 4-4' point (200x) 4-4'

(a) (c-2) 4-4' point (200x)

(b)

Fig. 6 (a, b).Variation of change (%) in electrical resistance and strain with loading cycle for sample IV, 12000 N.

Fig. 6 (c-1, c-2). Optical images of sample IV, 12000 N.

196

H. Hena-Zamal and S.V. Hoa

In fact, many more tests under a variety of combinations of loading and number of cycles were carried out to study the effectiveness of this method in monitoring damage behavior of glass/epoxy composites. Only the important findings are shown here based on the representative results. In summary, it is found that at a given load, both the electrical resistance and strain increase with increasing loading cycles. For maximum loads 6000N and 8000N the through thickness resistance show the jumps earlier than strains and inplane resistances. This implies that the through thickness resistance can correspond well to the occurrence of damages like delamination. On the other hand, the in-plane resistance and strains could not detect this damage as effectively as through thickness resistance under these loads. Thus, it might not be safe to use strain gauges and in-plane resistance measurement for health monitoring of composite structures at high loads. However, for very high loads, e.g. at 10000N and 12000N maximum loads, the sensitivity of through thickness resistance itself seems to be reduced. Future studies will deal with the study of fatigue damage behavior of CNTs infused GFRP composites under loads lower than the first ply failure using the electrical resistance method in order to detect, especially, the initiation of damages.

4 Conclusion In this experimental study, it is verified that CNTs network embedded into GFRP composites are effective sensors for monitoring fatigue damages. Although, both strain and electrical resistance respond to the accumulation of damages in the samples, the change in resistance in through thickness positions is found to be more sensitive as it jumps earlier in the loading cycles. Further, it is observed that the change in through thickness resistance corresponds to the extent of damages better than in-plane resistance. However, this benefit is not clear for very high loads. From the above results, it can be concluded that only the conductive percolating network of CNTs can monitor fatigue damage in glass/epoxy composite structures effectively by means of through thickness resistance measurement.

References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10]

Thostenson, E., Ren, Z., Chou, T.: Compos. Sci. Tech. 61(13), 1899–1912 (2001) Che, J., Cagin, T., Goddard III, W.: Nanotechnology 11(2), 65–69 (2000) Thostenson, E., Li, C., Chou, T.: Compos. Sci. Tech. 65(3-4), 491 (2005) Gamstedta, E., Berglund, L., Peijs, T.: Compos. Sci. Tech. 59(5), 759–768 (1999) Chou, T.: Microstructural Design of Fiber Composites. Cambridge University Press, Cambridge (1992) Reifsnider, K.: Int. J. Fract. 16(6), 563–583 (1980) Obrien, K., Reifsnider, L.: J. Compos. Mater. 15(1), 55–70 (1981) Reifsnider, K., Jamison, R.: Int. J. Fract. 4(4), 187–197 (1982) Reifsnider, K., Jamison, R.: Int. J. Fract. 16(6), 563–583 (1980) Hahn, H., Kim, R.: J. Compos. Mater. 10, 156–180 (1976)

Fatigue Damage Behavior of Glass/Epoxy Composites Using CNTs as Sensors

197

[11] Reifsnider, K.: Fatigue of Composite Materials. Elsevier, Amsterdam (1991) [12] Chou. T., Gao, L., Thostenson, T., Zhang, Z., Byun, J.: Compos. Sci. Tech. 70(1) , 1–19 (2010) [13] Jamison, R.: Compos. Sci. Tech. 24(2), 83–99 (1985) [14] Schulte, K.: J. Phy. 3(7) pt 3, 1629–1636 (1993) [15] Wang, X., Chung, D.: Smart Mater. Struct. 6(4), 504–508 (1997) [16] Irving, E., Thiagarajan, C.: Smart Mater. Struct. 7(4), 456–466 (1998) [17] Gojny, H., Wichmann, G., Fiedler, B., Bauhofer, W., Schulte, K.: Compos. Part A. Appl. Sci. Manuf. 36(11), 1525–1535 (2005) [18] Seo, C., Lee, J.: Compos. Struct. 47(1-4), 525–530 (1999) [19] Zhang, W., Sakalkar, V., Koratkar, N.: Appl. Phys. Lett. 91(13), 1–3 (2007) [20] Lee, S., Yoon, D.: Key Eng. Mater. 321-323( pt.1), 290–293 (2006) [21] Zhang, W., Picu, C., Koratkar, N.: Appl. Phys. Lett. 91(19), 193109-1-3, 5 (2007) [22] Zhang, W., Suhr, J., Koratkar, N.: J. Nanosci. Nanotechnol. 6, 960–964 (2006) [23] Kang, I., Schulz, M., Kim, J., Shanov, V., Shi, D.: Smart Mater. Struct. 15(3), 737–748 (2006) [24] Antonios, V., Panagiota, T., Vassilis, K., Petros, K., Angelos, M., Nikolaos, N.: MWCNT-modified fiber reinforced composites with nano-sensing capabilities: A way towards the development of the new functional materials for space applications. In: Proceedings of the AIAA 57th International Astronautical Congress, IAC, vol. 8, pp. 5523–5530 (2006) [25] Kang, J., Kim, R., Tandon, G.: Nondestructive damage detection of cfrps using 4probe measurement of resistivity. In: Proceedings of the International SAMPE Symposium and Exhibition, vol. 52 (2007) [26] Boger, L., Wichmann, M., Meyer, L., Schulte, K.: Compos. Sci. Tech. 68(7-8), 1886–1894 (1886) [27] Thostenson, E., Chou, T.: Nanotechnology. 19(21), 215713-1-6, 28 (2008) [28] Gao, L., Thostenson, E., Zhang, Z., Chou, T.: Adv. Func. Mater. 19(1), 123–130 (2009) [29] Nofar, M., Hoa, S., Pugh, M.: Compos. Sci. Tech. 69(10), 1599–1606 (2009) [30] Thostenson, E., Chou, T.: Adv. Mater. 18(21), 2837–2841 (2006) [31] Gao, L., Thostenson, E., Zhang, Z., Byun, J., Chou, T.: Philos. Mag. 90(31-32), 4085–4099 (2010) [32] Rosca, I., Hoa, S.: Carbon 47(8), 1958–1968 (2009)

26th ICAF Symposium – Montreal, 1-3 June 2011 DSTO – NLR Collaborative Programme on Fatigue Properties of β-Annealed Ti-6Al-4V: Preliminary Results E. Amsterdam1, A. Shekhter2, S.A. Barter2, M. McDonald2, and R.J.H. Wanhill1 1 2

National Aerospace Laboratory NLR, Amsterdam, the Netherlands Defence Science and Technology Organisation, Melbourne, Australia

Abstract. β-annealed Ti-6Al-4V ELI (Extra Low Interstitial) alloy thick section plate is used in primary fatigue-critical structures of advanced military aircraft. However, little has been generally published about its properties, particularly the fatigue and crack growth behaviour under both constant amplitude loading and variable amplitude loading representative of service load histories. In 2009 the DSTO and NLR set up a collaborative programme of durability and damage tolerance fatigue testing to address this problem and to provide basic data for OEMindependent fatigue design analyses. The programme takes account of the data requirements for the two most relevant analysis methods. The first is strain – life analysis, generally used to estimate durability lives. The second is fatigue crack growth analysis, primarily used in damage tolerance assessments and sometimes also used to estimate durability lives. This paper presents preliminary results of the test programme, comprising (a) constant amplitude strain – life fatigue test results, (b) constant amplitude fatigue crack growth data from ΔK = 11.3 -13 MPa√m (for R = 0.7, 0.4 and 0.1) up to high ΔK values, (c) long crack threshold tests with decreasing ΔK from 11.3 -13 MPa√m down to the threshold, and (d) ΔK-increasing short-to-long crack growth data using quantitative fractography.

1 Introduction β-annealed Ti-6Al-4V ELI has a chemical composition and manufacturing process that are intended to optimise its fatigue and fracture properties, notably in the thick sections required for large primary structures in advanced military aircraft. However, the thickness of the plate and the manufacturing process results in a very coarse grain size, and little is known in detail about the fatigue and fracture properties of this form of the alloy. To predict the fatigue behaviour, a test programme was put together that focuses on two fatigue analysis methods; i) strain-based fatigue life analysis, used to estimate the fatigue durability lifetime, and ii) fatigue crack growth (FCG) analysis, used to assess the damage tolerance behaviour. Strain-based fatigue analysis determines the local stress-strain responses of structural areas under representative load histories and employs a cumulative damage rule (usually a linear Miner’s type) to estimate the fatigue life. The basic

200

E. Amsterdam et al.

inputs are: cyclic stress-strain curves (CSSC), load history analysis by "rainflow" counting, and a strain−life curve of the material developed from constant amplitude (CA) strain-life testing. Standard FCG analysis requires basic material data in the form of CA fatigue crack growth rates versus ΔK, including ΔKth, the threshold for sustained fatigue crack growth of long cracks. Fracture toughness data (plane stress and/or plane strain), or at least an estimate, are also required. The accuracy of the threshold and crack growth values is of the most importance since cracks in aircraft primary structures typically spend most of their lives growing at low growth rates. These data serve as inputs in fatigue crack growth models such as those proposed by Paris [1], Walker [2], Forman [3], and de Koning [4], among many others. The preliminary results of the constant amplitude (CA) strain-life, CA fatigue crack growth and long crack threshold testing are presented in this paper, with some comments on an alternative method for the measurement of slow crack growth rates.

2 Experimental Details The investigation included metallographic examination, strain- life testing and crack growth/threshold testing of β-annealed Ti-6Al-4V ELI thick (100 mm) plate. Metallographic specimens were prepared using standard metallographic techniques for titanium alloys. Strain-life tests were performed using small cylindrical specimens according to guidelines provided by ASTM E 606-04E1. The specimen loading axes were the longitudinal (L) and long-transverse (T) orientations with respect to the parent plate material. The tests were done under completely reversed strain at six microstrain amplitudes: 6000, 5000, 4500, 4000, 3750, and 3500; and in the frequency range 0.5 – 5 Hz. The lower frequency of 0.5 Hz was used for micro-strain amplitudes higher than 4000, to prevent excessive heating (> 5°C) of the specimens. Failure was deemed to have occurred when the force required to maintain a particular strain had either reduced to half its initial value, or increased to 1.25 times its initial value. Tests were stopped when specimens survived 106 cycles. Long-crack FCG tests were carried out using 10 mm thick compact tension (CT) specimens. The tests were performed at constant load and therefore increasing ΔK for increasing crack length. R-ratios of 0.1, 0.4 and 0.7 were used, and testing started with a pre-crack and load that gave initial ΔK values of 13, 12 and 11.3 MPa√m, respectively. All samples had well-polished side surfaces and the crack lengths were measured both optically (on both sides) and automatically using the D.C. potential drop (PD) method. The specimen load and crack growth orientations were L-T and T-L (ASTM Standard E 647 definition), the cycle waveform was sinusoidal, and the frequency was 30 Hz. Long crack threshold tests on CT samples were performed according to ASTM Standard E 647, with a) constant R-ratio and decreasing Kmax, and b) constant Kmax

Fatigue Properties of β-Annealed Ti-6Al-4V: Preliminary Results

201

and increasing R-ratio. Crack lengths were measured using the PD method or optically, and load shedding occurred automatically using INSTRON software installed on the test machine. The tests started at the same ΔK as the long-crack FCG tests in order to cover the entire ΔK range. Thresholds were deemed to have been reached when ≤0.1 mm crack extension occurred during 106 cycles. Crack growth tests starting from the indicated long-crack threshold ΔK level were also carried out with R = 0.7 constant amplitude loading interspersed with marker loads with the same Kmax value and R = 0.1. The R = 0.7 loads were in blocks of 20,000 followed by 10 R = 0.1 loads intended to produce marker bands. Crack growth was measured as above and also post-test by Quantitative Fractography (QF).

3 Results Microstructure

strain amplitude (microstrain)

Metallographic examination showed that the microstructure was fairly uniform with a large transformed β grain size, up to at least 1mm, with no primary α. This is known as a fully lamellar microstructure, since the transformed β grains consist of packets of co-oriented α lamellae surrounded by lamellar grain boundary α. The co-oriented α lamellae were delineated and separated by thin “ribs” of remanent β. The type of microstructure and the sizes of the transformed β grains are important since it is known that the fatigue crack growth properties of titanium alloys depend significantly on their microstructures, e.g. [5] – [7].

7000 6000 5000 DSTO-L 4000

DSTO-T NLR-L

3000

NLR-T Ref. 8

2000 1.0E+02

1.0E+03

1.0E+04

1.0E+05

cycles Fig. 1 Strain-life results.

1.0E+06

1.0E+07

202

E. Amsterdam et al.

Strain-life tests The strain-life results for different strains are shown in Figure 1. The DSTO results are similar to the results obtained at the NLR. There is also no difference between the L- and T-oriented samples. The data coincide with the results reported in ESDU Data Item No. 11003 [8], and are generally consistent with the limited data from the open literature. At low strains there is a large scatter in the results, probably because of the large grain size. Fatigue crack growth measurements The average optical crack lengths for all specimens are correlated in Figure 2 with the PD measurements. The data points at the high end of the curve were measured from the fracture surfaces and are in agreement with the extrapolated optical measurements. The data show limited scatter and are fitted very well by a fourth order polynomial. In turn, this final fit was used to recalculate (adjust) the PDobtained crack lengths from the raw PD data. The recalculated PD crack lengths were then used to obtain the fatigue crack growth rates.

0.9

relative crack length (-)

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.9

1.1

1.3

1.5

1.7

1.9

2.1

potential drop (-)

Fig. 2 Average optical crack length measurements versus the measured D.C. potential drop.

Long-crack growth tests Figure 3 shows the long-crack FCG rate versus ΔK data. There is a large effect of R. For example, at ΔK =20 MPa√m, the crack growth rates for R = 0.1 and R = 0.7 differ by almost an order of magnitude.

Fatigue Properties of β-Annealed Ti-6Al-4V: Preliminary Results

203

1.0E-05

da/dn (m/cycle)

1.0E-06

1.0E-07

1.0E-08 R=0.7 R=0.4 R=0.1 1.0E-09 10

100 dK (MPam1/2)

Fig. 3 Long-crack FCG rates versus ΔK.

Threshold tests The constant R-ratio and decreasing Kmax tests showed a wide range of crack growth rates at low ΔK values. Also, the optical crack length measurements indicated crack lengths 3 mm larger than those indicated by the PD measurements. Consequently, the (adjusted) ΔK values were significantly higher than those obtained from the PD measurements. The “erroneous” PD measurements were most probably due to crack closure originating from the decreasing Kmax value and the observed tortuous crack paths produced by the large grain size. Therefore a constant Kmax test was performed to avoid (or minimise) the crack closure. The crack growth rate decreased, but again the optically measured crack length was larger than the PD-indicated crack length. In view of this, it has been decided to rely on optical measurements in future. Figure 4 shows the threshold results for R = 0.1 and a short-to-long crack growth (StLCG) (ΔK-increasing) test for R = 0.7 and with QF. The R = 0.1 threshold data show a smooth transition to the long-crack growth R = 0.1 data. The threshold was reached at a ΔK value of 8.1-8.4 MPa√m. The short-to-long R = 0.7 crack growth rates lie below those for the long-crack R = 0.7 tests. The reason for this is unknown at present.

204

E. Amsterdam et al.

1.0E-05 LCG R=0.7 LCG R=0.1

1.0E-06

da/dn (m/cycle)

StLCG R=0.7 dK decreasing R=0.1

1.0E-07

1.0E-08

1.0E-09

1.0E-10

1.0E-11 1

10

100

dK (MPam1/2)

Figure 4 da/dn versus ΔK results for an R = 0.1 ΔK-decreasing threshold test (one LT sample) and a short-to-long crack growth (StLCG) R = 0.7 ΔK-increasing test with QF. The long-crack growth data from Figure 3 are included in the graph for comparison and reference.

4 Discussion Strain-life results The test results in Figure 2 show good inter-laboratory reproducibility, despite high scatter at low strains. The results also agree very well with those reported in ESDU Data Item No. 11003 [8]. Long-crack growth results The data in Figure 3 show a strong effect of stress ratio, and fall into reasonably narrow bands for R = 0.4 and R = 0.7. However, for R = 0.1 there is wide scatter in the data below ΔK = 20 MPa√m. This scatter could be due to variations in crack closure that result from the tortuous crack path. Threshold results The test showed that there were significant problems in obtaining accurate crack lengths as the threshold was approached, and these problems were only partly alleviated by conducting a short-to-long crack growth test. It appears most

Fatigue Properties of β-Annealed Ti-6Al-4V: Preliminary Results

205

likely that the tortuous and irregular crack path, caused by the large grain size, is responsible for the measurement difficulties. If so, then these difficulties must be regarded as intrinsic to the material, and that there will be inevitable and considerable scatter in the threshold ΔK values. (It is worth noting here that although crack closure during high-R constant Kmax threshold tests should not be present near the threshold, Newman et al. observed similar behaviour for constant Kmax threshold tests on an aluminium alloy with a very rough crack surface profile [9].) Future work will include threshold tests with high R-ratios and more short-tolong crack growth tests with QF. As may be seen from Figure 4, there is a definite indication that short-to-long crack growth begins at about ΔK = 6 MPa√m. This may be below the long-crack threshold. If so, then this is important for FCG analyses that have to account for short-to-long crack growth. These tests will include FCG under realistic load histories and the dependence of fatigue crack initiation and early crack growth on local grain and α lamellae orientations. The proposed short-to-long crack growth tests will consist of natural cracks with a naturally increasing ΔK as they grow. The crack length, geometry and growth can be determined with high accuracy via QF of marker bands. Figure 5 shows an example of markers that resulted from the short-to-long crack growth test, which had a basic R = 0.7 and sets of 10 R = 0.1 loads, as mentioned earlier. The markers are often visible, despite the rough fracture topography.

Fig. 5 Marker bands (arrowed) used for QF short-to-long crack growth measurements.

206

E. Amsterdam et al.

5 Conclusions Preliminary results of a test programme for assessing the durability and damage tolerance of β-annealed Ti-6Al-4V ELI thick plate have been obtained. The results lead to the following conclusions: (1) The strain-life results showed good inter-laboratory agreement and were generally consistent with (limited) open literature data. (2) Long fatigue crack growth measurements were performed on CT samples and the crack lengths were measured optically and by D.C. potential drop. The PD data corresponded well to the optical crack length data and were used to calculate the crack growth rate. As would be expected, the crack growth rates showed a clear dependence on the stress ratio R. (3) Threshold crack growth measurements were subject to measurement difficulties and uncertainties owing to the intrinsic crack growth behaviour of the material, namely tortuous and irregular crack paths. Thus there will always be inevitable and considerable scatter in the threshold ΔK values. Finally, we note that the test programme is continuing, and the use of marker loads with Quantitative Fractography (QF) should enable the determination of short-tolong fatigue crack growth rates and be a useful adjunct to the spectrum testing which is to follow.

References [1] Paris, P.C., Gomez, M.P., Anderson, W.P.: The Trend in Engineering 13, 9 (1961) [2] Walker, K.: In: Effects of Environments and Complex Load History on Fatigue Life. ASTM STP 462, p. 1 (1970) [3] Forman, R.G.: Journal of Basic Engineering, Transactions ASME D89, 459 (1967) [4] de Koning, A.U.: In: Fracture Mechanics: 13th Conference ASTM STP 743, p. 63 (1981) [5] Yoder, G.R., Cooley, L.A., Crooker, T.W.: Engineering Fracture Mechanics 11, 805–816 (1979) [6] Wanhill, R.J.H., Galatolo, R., Looije, C.E.W.: International Journal of Fatigue 11, 407–416 (1989); Also as NLR MP 88028 U, Amsterdam, the Netherlands (June 1988) [7] Wanhill, R.J.H., Looije, C.E.W.: In: Fractographic and microstructural analysis of fatigue crack growth in Ti-6Al-4V fan disc forgings. AGARD Engine Disc Cooperative Test Programme, AGARD Report 766 (Addendum), pp. 2-1 – 2-40, Advisory Group for Aerospace Research and Development, Neuilly sur Seine (1993) [8] Walker, S.: In: Cyclic Stress-Strain and Strain-Life Properties of Metallic Materials. ESDU Data Item No. 11003 (2010), http://www.esdu.com [9] Newman Jr., J.C., Yamada, Y., Newman, J.A.: Journal of ASTM International 7(4), 16 (2010) [10] McClung, R.C., Chan, K.S., Hudak Jr., S.J., Davidson, D.L.: In: ASM Handbook, Fatigue and Fracture, vol. 19, p. 153 (1996)

26th ICAF Symposium – Montreal, 1-3 June 2011 Damage Tolerance Demonstration of Flange Joint for Aircraft Engine Composite Fan Case Y. Ueda1, H. Kuroki1, T. Murooka1, A. Tanaka1, K. Miyazawa1, I. Okumura2, Y. Shigenari2, K. Oikawa1, and H. Morita1 1

IHI Corporation, Japan 2 IHI AEROSPACE Co., Ltd., Japan

Abstract. Fan Case of aircraft turbo fan engines is located in the front end of an engine, and surrounds a periphery of fan blades. When a fan blade fails for some reasons, which is called fan blade out event, fan case needs to prevent high energy debris from penetrating through it. When a blade out event occurs, fan case is subjected to a very large load, which is called FBO load. FBO load is followed by repeated unbalance load due to wind milling while the aircraft returns to an airport after the event. Therefore fan case needs to endure not only the FBO load but also repeated unbalance load after the event. This is called “Fly home capability”. Also it is necessary to confirm that defects, such as delamination caused during manufacturing process, do not have the detrimental effects on the structure integrity requirements. In this paper, damage tolerance capability of a flange joint of a composite fan case is discussed. Subcomponent tests of flange joints are conducted. Artificial delaminations, which can be detected by non-destructive inspection, are introduced to the test specimens. Test results show that the detectable delamination does not have detrimental effects on durability, ultimate load capability, or fly home capability.

1 Introduction In order to improve fuel consumption rate, recent aircraft turbo fan engines tend to have larger bypass ratio, which results in the increase of fan module weight. Therefore the weight reduction of fan module is a key issue, and many engine manufactures are studying the application of composite material to fan case. Fan case has two important functions. One is to shape a flow passage of fan module, and the other is to prevent the high energy debris from penetrating through it when a fan blade fails for some reasons, which is called fan blade out (FBO) event. Additionally, it is a major techinal issue of composite materials that damage tolerance capability for the defects which occurred in manufacturing process must be demonstrated because the small defects which are not detected by nondestructive inspection (NDI) are inevitable. In other words, it is necessary to show that the inherent defects which is smaller than the maximum detectable size don’t

208

Y. Ueda et al.

have the detrimental effect on the durability of the composite fan case and that the fan case including such defects has sufficient capability under the FBO load and the following windmilling condition. In this research, the damage tolerance capability of a flange of a composite fan case was demonstrated under the normal operating load at first. This test is called ‘Flaw Tolerant Demo.’. And then the capability was demonstrated under the FBO load and the following unbalance load. This is called ‘Fly Home Capability Demo.’. High stress was loaded at the flange of the fan case under these loads. The subcomponent test specimens including artificial defects were prepared and Flaw Tolerant Demo. and Fly Home Capability Demo. were conducted.

2 Experiment Test specimen The test specimens were cut from the ring with flange which was made in the same process with the actual composite fan case. The fan case was made of ±45 degree laminates and 90 degree laminates, and their fiber was carbon, and their resin was thermoset. Figure 1 shows the schematic views of the specimens. Type A specimens have the artificial defects at its corner area, and Type B specimens have the artificial defect at its cylindrical area. The sizes of the artificial defects were determined from the capability of NDI for each area. The width of Type A and Type B specimen is twice and three times of the bolt interval respectively. Type B specimen is wider because it contains a larger defect.

90 deg. +45 deg.

90 deg. +45 deg.

0 deg.

0 deg.

-45 deg.

-45 deg.

artificial defects

artificial defects A: Corner area of a flange B: Smooth area of a cylinder

(a) Type A specimen

(b) Type B specimen

Fig. 1 Schematic views of specimens.

Damage Tolerance Demonstration of Flange Joint for Aircraft Engine

209

Test equipment The test equipment was 8800 Servohydraulic Testing Systems manufactured by Instron. The test specimen was fixed to the metal jig through the metal plate with the bolts and then attached to the test equipment, as shown in Figure 2.

Fig. 2 Test equipment.

Test condition The tests were conducted at room temperature. The speed of the cross head at the static tensile test was 1.0mm/min. The shape of cyclic load at the fatigue test was a sine wave and the frequency was 1.0Hz. Test sequence Test sequence was determined by modifying FIGURE 12.2.2.6 of CMH-17 as shown in Figure 3[1]. The first half , which starts with limit load test followed by one life test, corresponds to Flaw Tolerant Demo. The second half, ulitimated load test and half life test, corresponds to Fly Home Capability Demo. The test sequence is shown in Figure 4. For Flaw Tolerant Demo., the number of the normal fatigue cyclic load was assumed to be 60,000 cycles as the service life of an engine, and the limit load was assumed to be 0.325kN per bolt based on the existing engine design results. Normal fatigue load is so small for fan case that the limit load was applied as normal fatigue load. Factor of load is considered as 1.2. For Fly Home Capability Demo., the ultimate load was defined as the FBO load in this research. Then, the cyclic load after ultimate load was assumed to be the unbalance load due to windmilling following the FBO event. The number of this unbalance load was assumed to be 100,000 cycles from the period for the airplane to return to an airport after the FBO event, and the amplitudes of FBO load and unbalance load were assumed to be 5.7kN and 1.35kN per bolt respectively based on the existing engine design results. Based on these cyclic loads, the test for Flaw Tolerant Demo. and Fly Home Capability Demo. was carried out. These two tests were continuously conducted using the same specimen.

210

Y. Ueda et al.

Flaw Tolerant Demo.

CMH-17,

Fly Home Capability Demo.

Volume 3F, Chapter 12 Damage Resistance, Durability, and Damage Tolerance

Fig. 3 Test sequence reffering to CMH-17-3.

Fly Home Capability FBO (5.7kN per bolt pitch)

Flaw Tolerant Demo

±

Normal Load ( 0.325kN

±

Unbalance Load ( 1.35kN per bolt pitch) : 100,000cycle

per bolt pitch) : 60,000cycle

Limit Load (0.325kN

Limit Load (0.325kN

per bolt pitch)

per bolt pitch)

Fig. 4 Load cycle for the verification of Flaw Tolerant Demo and Fly Home Capability.

Artificial defects sizes and locations The delamination is the typical defects which occurrs in the manufacturing process. The artificial delaminations were made by means of putting kapton films or teflon films between the laminates. The allowable defect sizes were determined by the capability of the NDI which was used in the actual manufacturing process as shown in Table I and Figure 5. They can be detectable with sufficient confidence. The locations of the defects were determined by using FEM analysis. The location where the interlaminer tensile stress was highest was selected as explained in the next section.

Damage Tolerance Demonstration of Flange Joint for Aircraft Engine

211

Table 1 Allowable defect size on drawing A llow able defect size on draw ing C orner 12.7m m area (0.5inch) C ylindrical 25.4m m area (1.0inch) Loaction

Corner area Cylindrical area

Fig. 5 Area division for allowable defect size.

Analysis model and analysis results The FEM analysis was conducted to calculate the interlaminer tensile stress. Based on the analysis results, the loactions of the defects were determined. The analysis software was ANSYS Rev10.0. Figure 6 shows the FEM model. It consisted of solid elements for the specimen and the jig and beam elements for the bolt. The contact area between the specimen and the jig was modeled using gap elements. Figure 7 shows interlaminer tensile stress distribution. The interlaminer tensile stress was highest near the fillet R runout at the bolt position. The artificial defects were inserted between ±45 degree laminate and 90 degree laminate where the interlaminer tesile stress was large. Two defects were put near the fillet R runout at each bolt position for Type A specimen. One defect was put in cylindrical area for Type B specimen, as shown in Figure 1.

Metal plate Fillet R runout

Bolt

Fig. 6 FEM analysis model.

Jig

212

Y. Ueda et al.

High inter-laminar stress

Cross-section at a bolt position

－ 0

+

Fig. 7 Distribution of inter-laminar stress on cross-section at a bolt’s position.

Strain gauge To verify the analysis results, four strain gauges were instrumented to the specimen and the measured strain data was compared. Figure 8 shows the positions of strain gauges on Type A specimen.

1

2, 4 3

1

2, 3 4

Fig. 8 Position of strain gauge on Type A specimen.

Damage Tolerance Demonstration of Flange Joint for Aircraft Engine

213

3 Results Static tensile test To study the effect of the defect on the ultimate load capability, the static tensile tests were conducted. The sizes of the artificial defects are shown in Table II. To confirm the effect of defect size, two different sizes of defects are introduced. Table 2 The sizes of the artificial defects. Specim en Type A Type B

A rtificialdefect size 6.4m m , 12.7m m N o defect, (0.25inch) (0.5inch) 12.7m m , 25.4m m N o defect, (0.5inch) (1.0inch)

Figure 9 shows the behavior of the fracture in the static tensile test of Type A specimen. At first, the delamination occurred from the surface crack in the outer 90 degree laminate(Figure 9 (a)). Next, the delamination occurred at the center 90 degree laminate near the corner(Figure 9 (b)), and then this delamination was propagating(Figure 9 (c),(d)). Following the propagation of the delamination, the deformation became larger and the flange was stretched to the loading direction(Figure 9 (e)). Finally, the load started to decrease. And the highest load is determined to be the ultimate load capability. Figure 10 shows the relation between the displacement and the load of each specimen during the static tensile tests. The displacements at ultimate load of Type A specimen showed large variation, but there was not clear relationship between the displacements and the defect sizes. In the same way, there was no significant difference due to the defect sizes for Type B specimen. As a result, it was verified that there were no decrease of tensile strength due to the defects. And it was also observed that the strength per bolt of Type B specimen was higher than that of Type A. Average ultimate load capability of Type A specimen was 25.6kN per bolt. For Type B spcimen, bolt failure occurred before the failure of composite flange. Average load of Type B at the displacement of 5.0mm was 27.9kN, and the load was still increasing. It can be said that the tesile strength per bolt of Type B specimen was higher than that of Type A.

214

Y. Ueda et al. 30 ) Nk 25 ( lt o B 20 r ep da 15 o L 10 eil s n e 5 T

(d) (c) (b)

(e)

(a) 0

0

1

2

3

4 5 6 7 D isplacem ent　(m m )

Delamination from surface crack at 90 degree laminate

(a) Disp. 0.0mm

(b) Disp. 1,5mm

8

9

10

Delamination at center 90 degree laminate near the corner

(c) Disp. 2.0mm

Load Direction

(d) Disp. 3.0mm

(e) Disp. 6.0mm

Fig. 9 Behavior of Type A specimen in the static tensile test.

Figure11 shows the ultimate load capability of Type A specimens. There was no significant difference in ultimate load capability due to the defects.

Damage Tolerance Demonstration of Flange Joint for Aircraft Engine

(a) Type A specimen

215

(b) Type B specimen

Fig. 10 Relationship between displacement and load in static tensile tests.

)N30 k( tlo 25 Br ep 20 da oL 15 leis 10 ne T. 5 xa M0

1

2

No defect

3

1

2

Defect Size =6.4mm

1

2

Defect Size =12.7mm

Fig. 11 Ultimate load capability of Type A.

Comparison of the measured strain and the analysis results The measured strain during the static tensile test for Type A specimen were compared to the FEM analysis results. Figure 12 shows the relation between the load and the strain for Type A specimen. The relationship among strains at each gage loacation showed relatively good agreement. Therefore, the deformation shape was predicted correctly.

216

Y. Ueda et al.

Fig. 12 Comparison of strain between analysis and test.

Test for Flaw Tolerant Demo. and Fly Home Capability Demo. Type A specimen which had the defects with the size of 12.7mm and Type B specimen which had the defect with the size of 25.4mm were prepared for the test for Flaw Tolerant Demo. and Fly Home Capability Demo. Figure 13 shows the displacement-load curves under normal fatigue cyclic load for Flaw Torelant Demo. No damage occurred and the stiffness of the specimens didn’t change. Then, it was concluded that no detrimental damage occurred under normal fatigue cyclic load. 0.5 0.4

) 0.3 N k( 0.2 tl 0.1 o B 0 re p-0.1 da-0.2 oL -0.3

0.5

First cycle Last cycle

-0.4 -0.5 -0.04

First cycle Last cycle

0.4

) 0.3 N k( 0.2 tl 0.1 o B0 re p-0.1 da-0.2 o L-0.3 -0.4

-0.03

-0.02

-0.01

Displacement (mm)

(a) Type A specimen

0

0.01

-0.5 0

0.01

0.02

0.03

Displacement (mm)

0.04

0.05

(b) Type B specimen

Fig. 13 Displacement-load curve under normal fatigue load cycle for Flaw Tolerant Demo.

Damage Tolerance Demonstration of Flange Joint for Aircraft Engine

217

Figuer14 shows the displacement-load curves under unbalance fatigue cyclic load for Fly Home Capability Demo. Under the FBO load, the stiffness of Type A specimen decreased, but the stiffness of Type B specimen didn’t change. Because the tensile strength per bolt of Type B specimen was lager than that of Type A, the damage was decreased in Type B specimen. The stiffness didn’t change under unbalance fatigue cyclic load in both specimens. It was concluded that there was no growth of a damage. The difference between tensile stiffness and compressive stiffness of Type A specimen resulted from the damage caused by the FBO load. These results verified that the capability of NDI was sufficient for the manufacturing process of composite fan case. 2

2

) 1.5 N (k 1 lto 0.5 B re 0 p da-0.5 o -1 L

1.5

First cycle Last cycle

-1.5 -2 -0.1

) N k( 1 tl 0.5 o B0 re p-0.5 da o -1 L

-0.05

0

0.05

0.1

Displacement (mm)

0.15

(a) Type A specimen

First cycle Last cycle

-1.5

Stiffness before FBO 0.2

-2 -0.1

-0.05

0

0.05

0.1

Displacement (mm)

0.15

(b) Type B specimen

Fig. 14 Displacement-load curve under unbalance fatigue load cycle for Fly Home Capability Demo.

cracks Artificial crack defect

(a) Articial defect on cross-section A

Artificial defect

(b) Artificial defect on cross-section B

Fig. 15 Observation by cutting Type A specimen after test.

0.2

218

Y. Ueda et al.

Cut up of specimen after test Type A specimen whose stiffness was decreased under FBO load was cut and observed. Figure 15 shows the results. No growth from the artificial defect was found. It was confirmed that the defect didn’t have detrimental effects. There were the local cracks, which were generated under the FBO load resulted in the decrease of the stiffness. However, there were no severe damage such as a generation of delamination.

4 Conclusions In this paper, the subcomponent specimens of the composite fan case were prepared, which had the artificial defects at the corner or the cylindrical area. And, the condition of the cyclic load for the Flaw Tolerant Demo. and Fly Home Capability Demo. was determined by referring to the test sequence for strength, durability, and damage tolerance demonstration of CMH-17. The following results were obtained. (1) The behavior of the failure in the flange of the composite fan case was made clear by conducting the static tensile tests. (2) Flaw Tolerant Demo. was conducted with the defect, whose size was detectable with sufficient confidence in consideration of the inspection method for the actual manufacturing process. (3) It was verified that the delaminations which was smaller than the allowable defect size on drawing didn’t have the detrimental effect on the durability, the ultimate load capability and Fly Home Capability. (4) The capability of NDI at the manufacturing process was verified to be sufficient for the composite fan case. In the future work, Fly Home Capability will be verified with the full scale fan case after the containment test for the FBO event.

Acknowledgements This research was conducted as a part of the project, “Advanced Materials & Process Development for Next Generation Aircraft Engines” under the contract with Japanese Aero Engines Corporation (JAEC), funded by Ministry of Economy, Trade and Industry (METI) of Japan.

Reference [1] Composite Material Handbook, Polymer Matrix Composites Materials Usage, Design, And Analysis, Department Of Defense, United States Of America vol. 3

26th ICAF Symposium – Montreal, 1-3 June 2011 *

The Formation/Nucleation of Fatigue Cracks in Aircraft Structural Materials David W. Hoeppner

Professor of Mechanical Engineering, U. of Utah, 50 S. Central Campus Drive, Room 2110, Salt Lake City, UT 84112 [email protected]

Abstract. The issue of crack formation/nucleation and propagation related to model development for accurate fatigue life estimation/prediction of aircraft materials has been and is currently a vexing challenge. In the 1800’s Sorby used the optical microscope to study fatigue deformation of materials. The early investigators of fatigue wanted to learn the mechanism(s) of fatigue crack formation/nucleation and propagation. Many investigators oriented toward developing this understanding aided progress. Many of the studies that were made were “static”. That is, materials were exposed to cyclic loading and then viewed in a light microscope. The paper will illustrate the importance of understanding the formation/nucleation process and its impact on life prediction issues. A few cases will be mentioned where the formation/nucleation of cracks and its relationship to intrinsic and extrinsic factors in aircraft will be presented. The development of the scanning electron microscope aided the understanding fatigue of materials. Starting in 1970 numerous investigators developed fatigue machines that were either placed in the chamber of an SEM or were attached to the SEM and conducted “dynamic” studies. Many contributions have been made and some will be reviewed briefly in this paper. The past 45 years has seen the development of numerous in-situ systems for studying the process of fatigue crack formation/nucleation and the study of microstructural effects on the early stages of crack propagation of aircraft and other structural materials. This has led to increased understanding of fatigue crack formation/nucleation and propagation of cracks that is based on direct observation rather than speculation. Recently a new in situ SEM fatigue system has been developed at the University of Utah and work also has been done which allows accurate determination of local grain orientation effects and the role of relative orientation on crack nucleation and early crack propagation. The background of many of the above studies will be presented in the paper. Subsequently, some of the major contributions these workers have made to understand fatigue deformation in aircraft structural materials will be presented. The paper reviews progress of these studies and insights into fatigue with emphasis on aircraft structural materials. The paper concludes with a discussion of some of the opportunities available to develop a greater understanding of the fatigue crack formation/nucleation process and also to develop a greater understanding of microstructurally short crack propagation. *

Oral presentation.

220

D.W. Hoeppner

1 Introduction The formation of fatigue cracks has been a subject of some interest over nearly 200 years in general but the last 90 with more intensity and focus. However, due to some use of terminology such as “initiation” the specific mechanisms by which fatigue cracks form and grow has evaded many of the fatigue community. Even though Griffith [1] indicated the manner in which cracks may form is a critical issue it appears that in around 1950 many scientists and engineers started to indicate the fatigue process involves three stages; viz. initiation, propagation and final fracture. Many books and papers have indicated this point and many still do (2). On numerous occasions this writer has indicated the lack of precision or physics connected with the use of the word initiation [3-17]. In the 1970- current period it was indicated that actually the fatigue process should be divided into at least four stages; viz. 1)formation or nucleation, 2) short crack propagation, 3) long crack propagation that is characterized by either LEFM, EPFM or FPFM. [Where LEFM is linear elastic fracture mechanics, EPFM is elastic plastic fracture mechanics, and FPFM is fully plastic fracture mechanics], and 4) final failure which may be fracture or some other instability. [3-16]. Even so this imprecision persists and regrettably is a part of the reason for some significant aircraft accidents. At many symposia and meetings the question often has been asked “what do you mean by “initiation” and what is your rational for selecting it?” Often the query is answered with a description of detection of a crack or cracks of some size that is often arbitrary. One investigator who reported his research at a USAF Structural Integrity Conference cited his “initiation crack size” at “0.05” inches and when asked why he chose this at the end of his presentation he said he put all the values in quotes and if he does that he can “define it any way he desires”. As A. Freudenthal noted in a paper previously this is reminiscent of the mad hatter tea party in Alice in Wonderland [17].Steve Swift [18] points out in this very ICAF meeting that exactly how the term is defined and presented is absolutely critical. Although Halford [19] stated “Crack initiation is considered to be a singular event. Details of how the event evolved are considered to be irrelevant” it was done in part to show the concept of “initiation is flawed. An example of how flawed is related to the engine failure from a Hard Alpha Defect on flight UA 232 [20].

The Formation/Nucleation of Fatigue Cracks in Aircraft Structural Materials

221

Origin

Fig. 1 Fracture surface showing origin at Ti-6 Al 4 V disc bore of UA 232 no. 2 engine disc that caused crash of UA 232 in Sioux City, IA 1989.

The initial radial crack depth of the HAD defect was 0.05 inch. The final radial crack depth was 0.56 inch with a surface crack length of 1.24 inches. The low cycle fatigue “initiation” life was 54,000 cycles and the “initiation” crack size was 0.0325 inches. The crack in the disc was missed in seven non directed inspections that were “inspections of opportunity”. (All measurements above and the photo were taken in the QIDEC laboratories at the UU by Dr. David Hoeppner and Dr. Saeed Abidnazari as consultants to the investigation of the crash. The measurements are similar to those reported in [20]. The disc was assumed to have either no defects or any discontinuities should have been captured in the LCF data base used for the LCF design allowables. Thus, the disc possessed an intrinsic defect larger by a significant margin than the END of INITIATION life crack size. The shows, as Swift has indicated [18] how dangerous our words and concepts can be. What is interesting is that this is a form of denial alluded to by Gumbel in [21] where he states “Sint ut sunt aut non sint” - Accept them as they are or deny their existence. The safe life “initiation” based paradigm has resulted in far too many accidents of aircraft as well as other structures. Much of the research and development my students, post docs, colleagues and I have done for nearly 50 years was focused on showing that fatigue cracks are formed at many sites of origin and at various fractions of life that are quite short relatively. We have shown this over and over and over again but we keep finding it and then losing it again and again. (This is in a sense a paraphrase of T. S. Elliott in his great work East Coker of the Four Quartets [22] in which he reminds us “There is only the fight to recover what has been lost and found and lost again and again; ....For us, there is only trying....” The FAA Ti Task force [23] produced a report after the UA 232 crash and there has been much activity at the FAA, the Ti industry, engine manufacturers, other certification authorties, the NDI community, and the research and development community. This includes the fatigue communtity. The ASTM has started using the term crack formation and as one would see if the time were taken to absorb ASTM E 1823 [24] the only way “initiation” is defined is in the initiation of crack growth from a preexisting crack. On many occasions this author has suggested the use of initiation to define the start of any process such as fatigue, corrosion, wear,

222

D.W. Hoeppner

fretting, creep, and synergisms of these [5, 7, 11, 13-16]. It is noteworthy that two recent standards of ASTM now use the phrase crack formation in two recently approved items as well [25, 26].

2

The Framework for Formation/Nucleation Studies and Structural Integrity Lifing

Four distinct phases of life were proposed for the fatigue process and related mechanisms of degradation in previous papers by this writer [5, 11, 13, 15] and by Freudenthal [17]; viz. 1. Nucleation or formation of damage. 2. Micrcostructurally dominated crack link up and short crack propagation. 3. Crack propagation where fracture mechanics similitude is applicable in general. 4. Final instability. Similar divisions were made by Robert Jeal [12] and were put in use by Rolls Royce Aeroengine-Derby around 1976 as part of their life and Methods predictions system. In a recent conversation with Steve Willliams of Rolls Royce [27] it was indicated that RR still uses it as a guide and he emphasized the criticallity of phase 2 and this writer heartily agrees. The challenge is to characterize, understand, model, and develop design methods for fatigue and structural integrity that is reliable to the desired probability of survival with the needed confidence. There are a very large number of cases in the anals of ICAF elsewhere where the understanding of both phase 1 and 2 above were not, and are not at hand. Even though starting in the 1950 period some investigators put forth the idea that persistent slip bands form in fatigue and sometimes are the origin of cracks in structural materials. But it turns out that structural materials have many other sources of cracks including porosity, blowholes, laps, seams, particles that are incohent with the matrix, particles that are coherent with the matrix, aligned phases, true defect particles often inclusions that are undesirable in the alloy of interest, orientation effects, and more. Thus, the total life (LT) of a structure is LT= L1+L2+L3+L4. Figure 1 presents a depiction of the degradation process. The regions shown, e.g. 1, 2, 3, and 4, illustrate the portion of life, on the abscissa, and the corresponding growth in discontinuity size plotted schematically on the ordinate. Note a crack propagates from either an intrinsic initial discontinuity state (IDS) or is formed by a pure fatigue mechanism or other related process that may act concomitantly or sequentially with fatigue.

The Formation/Nucleation of Fatigue Cracks in Aircraft Structural Materials

223

The degradation process

4 1

2

3

Discontinuity Size A

Life A= "FIRST" detectable crack 1. Nucleation phase, "NO CRACK" 2. "SMALL CRACK" phase-steps related to local structure (Anisotropy) 3. Stress dominated crack growth, LEFM, EPFM 4. Crack at length to produce instability

Fig. 2 Depiction of tbe degradation process .

Examples are corrosion fatigue, fretting fatigue, creep fatigue, etc. Exampes of the corrosion process is the occurence of a specific form of corrosion such as pitting and it forms a specific form of evolving discontinuity state (EDS) that is not necessarily a crack like discontinuity. Eventually the EDS may transition to a bona fide crack resulting in either short crack or long crack propagation or both (phase 2 and 3 above). and the development of short cracks and their propagation. The requirement of the community to come up with design methods to deal with corrosion or other degradation, fatigue, creep, and wear, is essential and some of the elements are depicted in Figure 2. This figure illustrates that most of the quantitative methods that have been developed used the concepts of mechanics of materials with an incorporation of fracture mechanics as appropriate. Phase 1 and portions of phases 2 and 3 are usually included in both stress –life and strain life approaches to fatigue design although those they are notusually recognized as such. This is one of the continuing challenges of tbe safe life approach to fatigue design. In attempting to understand the nucleation and phase 2 propagation Wood [28] stated "What the cyclic stress does do in practice depends so much on the experiemntal conditions that it is not easy to distinguish general principles. Nor is it easy to resist the temptation of drawing general conclusions from observations that may hold only for limited experimental conditions. For example, much has

224

D.W. Hoeppner

been made by some authors, including this one, of fatigue cracks that originate in slip bands, and by others of cracks that grow according to “stage I” and “stage II” processes. But not all cracks originate or grow in those ways. Evidently a much needed first step in the study of fatigue, one that would clear away contradictions of the kind just noted, is to find why merely the way in which fatigue cracks form should vary. Then it might be possible to recognize specific types of fatigue, and thus to deal with the subject consistently.” (note he uses the term form too!). The work of Griffith, Freudenthal, Schijve, Wood and many others suggested the idea of attempting to develop systems of fatigue experiments that would allow much higher magnifications and allow it to be done as the dynamic loads were applied. Thus the motivation for moving beyond optical microscopy to the Scanning Electron Miccroscope. Initially it was great challenge to find support for this type work but in 1974 Rolls Royce Aerengine Co. Took it on and started major financial and intellectual support for in situ fatigue systems withni an SEM. As well, the U. Of Toronto and the Connaught Foundation in Canada provided major stimulus for the study of formation and stage 2 of fatigue crack propagation. Much progress has been made and insight into how to improve and enhance the fatigue and lifing methods has been gained over nearly 40 years of effort using these techniques. During this period there has been much progress using these in situ systems [29-48]. All of the systems are described in great detail in the references cited and also many M.S, M.Ap. Sc., and Ph.D. theses as well as reports. In recent years the eighth major evolution of the fatigue system attached to the SEM has emerged [48]. In addition work has recently begun on using OIM (orientation imaging microscopy) in conjunction with our in situ SEM systems. Many of the observations made over this period of time have assisted in the evolution of HOLSIP [49-51] along with a great body of work. In recent work Yamagiwa [52, 53] has shown the utility of serial sectioning in conjunction with studying the evolution of creep damage. It would appear that until a non destructive method of detecting fatigue crack formation subsurface and similar methods of detecting creep and creep fatigue damage subsurface occurs we are left to speculate how subsurface degradation occurs in the crack formation and stage 2 crack propagation processes. The serial sectioning approach provides some insight until such tools emerge. Thus, for the time being, the words of Wood and others can be reiterated to continue to focus our efforts on finding the fatigue crack formation and propagation processes.

3 Conclusions From the period 1955 to the present many studies have been made on formation and propagation of fatigue cracks. These studies have included optical and replication observations of crack formation, structurally dependent crack propagation, and crack propagation where principals of both linear elastic and elastic plastic fracture mechanics are applicable. As well, numerous studies have been done on corrosion fatigue, with concentration on pitting and intergranular

The Formation/Nucleation of Fatigue Cracks in Aircraft Structural Materials

225

attack, and fretting fatigue. Some of those have been reported at ICAF symposia in the past. In situ studies of fatigue crack formation and propagation have been done on both model materials and structural materials since 1970. From these studies the following conclusions may be drawn: 1. Cracks most often form at IDS (initial discontinuity state) in structural materials used in airframes, landing gears, and engine materials. 2. Some of the IDS are dislocations and dislocation networks and arrays but most often the IDS at which cracks form are pores, particles, grain or phase boundaries, aligned phases, and particle boundaries with the matrix. 3. Slip bands form in most materials evaluated and often are connected with the formation of the initial cracks. 4. In some cases intrinsic cracks are present around particles and precipitates and only conditions to propagate them are necessary. These cracks are in the nano and micro scale. 5. When cracks form at an IDS or from slip they can be observed on the surface of model specimens as short as 1 μm. There is no doubt shorter cracks will be observed as techniques and equipment improve. 6. Crack deflection occurs related to grain size, grain orientation, phase orientation, and phase control element packet size and their orientation in some materials. 7. Fretting degradation occurs in hard vacuums. 8. Fretting degradation does not always result in immediate formation of cracks. 9. There is often a formation period to fretting fatigue cracking in contact surfaces. 10. It is not possible at present to observe cracks forming subsurface and this is a great need for the future. 11. At more elevated temperatures evolution of the microstructure occurs that is related to creep effects (Yamagiwa). 12. Numerous cracks are most often observed in the very early stages and link up over cycles and time to form one incipient crack in all cases observed. Some initially formed cracks, both in model materials and actual structural materials, become non propagating cracks. 13. Initial studies in situ and with OIM (Orientation imaging microscopy) show great promise for more quantitative evaluation of effects of microstructure on the formation and propagation of fatigue cracks. 14. The use of OIM in conjunction with the in situ fatigue systems provides the potential for automated studies of orientation/misorientation effects on the phase II of fatigue crack propagation.

226

D.W. Hoeppner

4 Future Paths to Better Understanding of Formation/ Nucleation of Fatigue Cracks and Synergistic Degradation ¾ Use of orientation imaging microscopy (OIM) to ascertain specific quantitative effects of microstructural influence such as grain orientation, phase orientation etc. (work in underway at UU with support from TSL/EDAX Corp. and also at IAR-NRC in conjunction with a German laboratory). ¾ Application of automated software to ascertain Schmid factors, Taylor factors automatically in conjunction with in situ fatigue and related experiments must be expanded to assure complete understanding of the next items. ¾ Determination of transitions from slip to fatigue cracking in more precise detail. ¾ Evolution of microstructures in creep and creep fatigue similar to the work of Dr. Yamagiwa [52, 53] should be accelerated in many laboratories for all the time dependent failure mechanisms of aircraft materials (corrosion, creep, fatigue, and wear). ¾ A great deal of emphasis needs to be placed on detecting the sub-surface deformation that occurs in fatigue and other failures mechanisms. Until this is accomplished a complete understanding of formation and structurally dependent crack propagation will not be possible. ¾ Much greater emphasis needs to be placed on the transition from formation to the first stages of fatigue crack propagation. This is true for all the formation mechanisms such as fatigue from slip, fatigue from particles or pores, fatigue from corrosion of various types, fatigue from wear of various types, fatigue from fretting, fatigue from FOD (foreign object damage) and other extraneous mechanisms of formation of fatigue cracks {given they are not present initially}.

References [1] Griffith, A.A.: The Phenomena and Rupture and Flow in Solids. Philosophical Transactions of the Royal Society of London, Series A 221, 163–198 (1921) [2] Schijve, J.: Analysis of the Fatigue Phenomenon in Aluminum Alloys. NLR-TR M.2122, National Aerospace Laboratory, Amsterdam (1964) [3] Schijve, J.: Significance of Fatigue Cracks in the Micro-Range and Macro-Range, Fatigue Crack Propagation. In: ASTM STP, vol. 415, pp. 415–459. ASTM (1967) [4] Hoeppner, D.W.: The Effect of Grain Size on Fatigue Crack Propagation in Copper. Fatigue Crack Propagation. In: ASTM STP, vol. 415, pp. 486–504. ASTM (1967) [5] Hoeppner, D. W.: Corrosion Fatigue Considerations in Materials Selection and Design. Invited Keynote paper. International Conference on Corrosion Fatigue (June 1971); Published in the Conference Proceedings, NACE, pp. 3–11 (1972) [6] Hoeppner, D. W.: Initiation of Fatigue in Aluminum Alloys. American Institute of Mining, Metallurgical, and Petroleum Engineers, Symposium on Fatigue of Metals, Atlanta, GA (April 1971)

The Formation/Nucleation of Fatigue Cracks in Aircraft Structural Materials

227

[7] Hoeppner, D. W.: Metallurgical Aspects of Fatigue. In: American Society of Metals, WESTEC Conference, Los Angeles, CA (March 1971) [8] Hoeppner, D. W.: The Mechanisms of Fatigue - Part 1 The Effect of Grain Size on Fatigue (Preliminary Results), Lockheed-California Company, LR 24368 (December 1971) [9] Hoeppner, D.W., Krupp, W.: Prediction of Component Life by Application of Fatigue Crack Growth Knowledge. Engineering Fracture Mechanics 6, 47–70 (1974) [10] Hoeppner, D.W.: Comments on Initiation and Propagation of Fretting Fatigue Cracks (letter to the editor). Wear 43, 267–270 (1977) [11] Hoeppner, D.W.: Estimation of Component Life by Application of Fatigue Crack Growth Threshold Knowledge. In: Fatigue, Creep of Pressure Vessels for Elevated Temperature Service, MPC, vol. 17, pp. 1–85. ASME, N.Y (1981) [12] Jeal, R.: Defects and Their Effect on the Behavior of Gas Turbine Engine Discs. In: Maintenance in Service of High Temperature Parts, AGARD Proceedings, vol. 317 (1981); Structures and Materials Panel, Noordwijkerhoot, The Netherlands, NATOAGARD, France (1981) [13] Hoeppner, D., Venter, R., McCammond, D., Ekvall, J.: Aircraft Structural Fatigue, four volumes of notes for FAA two week course held at U of Toronto (1979-85) and U of Utah (1985-92), FAA, Oklahoma City, OK or available from the author [14] Hoeppner, D. W.: Application of Damage Tolerance Concepts to ’Short Cracks’ in Safety Critical Components. In: ICAF Proceedings of International Committee on Aeronautical Fatigue Symposium, Toulouse, France (May 1983) [15] Hoeppner, D. W.: Parameters that Input to Application of Damage Tolerance Concepts to Critical Engine Components. AGARD Conference, San Antonio, Texas, (April 1985); Published in Conference Proceedings AGARD-CP 393, Damage Tolerance Concepts forCritical Engine Components, pp. 4-1 – 4-16, NATO-AGARD, France, (August 1985) invited keynote paper [16] Hoeppner, D. W.: Damage Tolerance in Gas Turbine Engines–Future Technology Requirements. Presented at NATO-AGARD-SMP, meeting held in Mierlo, The Netherlands, October 1988, AGARD/SMP Review Damage Tolerance for Engine Structures, 2. Defects and Quantitative Material Behavior, AGARD report No. 769, NATO-AGARD, Neuilly Sur Seine, France (1989); paper no. 7. Invited keynote paper [17] Freudenthal, A.: Fatigue and Fracture Mechanics. Engineering Fracture Mechanics 5, 403–414 (1973) [18] Swift, S.: Sticks and Stones (Could the words of aeronautical fatigue hurt us?). In: 26th ICAF Symposium, Montreal, Quebec, Canada, June 1-3 (2011) [19] Halford, G.: Low-Cycle Thermal Fatigue. NASA TM 87225, p. 11 (February 1986); Halford G.: Low-Cycle Thermal Fatigue. Thermal Stresses II, ch. 6, pp. 330-428. Elsevier Science, Amsterdam (1987) [20] NTSB/AAR-90/06, PB90-910406, Aircraft accident report, UA 232, McDonnell Dougas DC-10-10, Sioux Gateway Airport, Sioux City, Iowa (July 19, 1989) [21] Gumbel, E.J.: Statistics of Extremes. Columbia University Press, N.Y (1958) [22] Elliott, T. S.: East Coker in the Four Quartets, the Complete Poems and Plays- 19091950, p. 128. Harcourt, Brace and World, Inc., N.Y (1971) [23] Costa, J., et al.: Titanium Rotating Components Review Team Report, USA FAA, Engine and Propellor Directorate, Burlington, MA (December 14, 1990) [24] ASTM, E 1823-10, Standard Terminology Relating to Fatigue and Fracture Testing, ASTM International, West Conshohocken, PA, USA (2010)

228

D.W. Hoeppner

[25] ASTM E2714 – 09, Standard Test Method for Creep-Fatigue Testing, International, West Conshohocken, PA, USA 9. (This test method is the responsibility of Subcommittee E08.05 on Cyclic Deformation and Crack Formation) [26] ASTM Standard Guide for Fretting Fatigue Testing ASTM E2789-10, International, West Conshohocken, PA, USA [27] Williams, Steve.: Associate Fellow - High temperature lifing, Critical Parts Lifing and Integrity, Rolls Royce Aeroengine Co., discussion at HOLSIP 10, Snowbird, UT (February 27, 2011) [28] Wood, W.A.: Four Basic Types of Metal Fatigue, Treatise in Materials Science and Technology, vol. 5, pp. 129–179. Academic Press, New York (1974) [29] Cameron, D.W., Jeal, R.H., Hoeppner, D.W.: SEM Investigations of Fatigue Crack Propagation in RR 58 Aluminum Alloy. Transactions of ASME, Journal of Engineering for Gas Turbines and Power 107, 238–241 (1985) [30] Cameron, D.W., Jeal, R.H., Hoeppner, D.W.: SEM Investigations of Fatigue Crack Propagation in RR 58 Aluminum Alloy. Transactions of ASME, Journal of Engineering for Gas Turbines and Power 107, 238–241 (1985) [31] Smith, F., Hoeppner, D.W.: Observations on Fatigue Crack Growth/Microstructure Interactions Using Advanced Techniques. In: Proceedings of the 16th Annual Meeting of IMS/ASM, Corrosion, Microstructure and Metallography, Microstructural Science, Northwood, White, and Vanderwoort ASM, Metals Park, OH, vol. 1, pp. 435–443 (1985) [32] Hoeppner, D.W., Sherman, I.: Fractographic Observations of Corrosion Fatigue and Fretting Fracture Surfaces. In: Corrosion, Microstructure and Metallography, Northwood, White, and Vanderwoort, pp. 117–125. American Society for Metals (1985) [33] Wu, D., Hoeppner, D.W.: Observations and Characterization Considerations of Fatigue Crack Growth in a Single Crystal Nickel-Base Superalloy. Scripta Metallurgica, vol. 19, pp. 493–498. Pergamon Press Ltd, USA (1985) [34] Smith, F., Hoeppner, D.W.: Observations on Fatigue Crack Growth/Microstructure Interactions Using Advanced Techniques. In: Proceedings of the 16th Annual Meeting of IMS/ASM Corrosion, Microstructure and Metallography, Microstructural Science, Northwood, White, Vanderwoort ASM, Metals Park, OH, pp. 435–443 (1985) [35] Smith, F.M., Hoeppner, D.W.: Quantitative Representation of Microstructural Contributions in Fatigue Crack Nucleation and Growth. In: The 1987 ASME Design Technology Conferences – 7th Biennial Conference of Failure and Prevention and Reliability, Boston, Mass, September 27-30 (1987); Published in Conference Proceedings, DE – Vol.9, pp. 87-90 [36] Wu, D.C., Cameron, D.W., Hoeppner, D.W.: Observations of Microstructural and Geometrical Influences of Fatigue Crack Growth in Single Crystal and Polycrystal Nickel-Base Superalloys. In: Duhl, D.N., et al. (eds.) Superalloys 1988, AIME, pp. 605–614. The Metallurgical Society (1988) [37] Song, Z., Hoeppner, D.W.: Dwell Time Effects on Material Fatigue BehaviorTitanium Alloys. International Journal of Fatigue 10(4), 211–218 (1988) [38] Stephens, R.R., Hoeppner, D.W.: A New Apparatus for Studying Fatigue Deformation at High Magnifications. Review of Scientific Instruments 59(8), 1412–1419 (1988) [39] Stephens, R.R., Grabowski, L., Hoeppner, D.W.: Situ/SEM Studies of Short Crack Growth Behavior at Ambient and Elevated Temperature in a Nickel Base Superalloy. In: Miller, K.J., de los Rios, E.R. (eds.) Short Fatigue Cracks, ESIS, vol. 13, pp. 335–348. Mechanical Engineering Publications, London (1992)

The Formation/Nucleation of Fatigue Cracks in Aircraft Structural Materials

229

[40] Hoeppner, D.W.: History and Prognosis of Material Discontinuity Effects on Engine Components Structural Integrity. Published in Proceedings of 74th AGARD Structures and Materials Panel Meeting, Patras, Greece, May 25-29, 1992, AGARD Report No. 790, NATO-AGARD, Neuilly Sur Seine, France, Paper No. 1-1, pp. 1–8 (1993) (Invited keynote paper) [41] Elliott, C., Hoeppner, D.W.: A Fretting Fatigue System Usable In a Scanning Electron Microscope. In: International Conference on Fretting Fatigue, England, April 19. University of Sheffield, Sheffield (1993) ; Fretting Fatigue. In: ESIS, vol.18, pp. 211–218.Mechanical Engineering Publications, London (1994) [42] Stephens, R.R., Grabowski, L., Hoeppner, D.W.: The Effect of Temperature on the Behaviour of Short Fatigue Cracks in Waspaloy Using an In Situ SEM Fatigue Apparatus. International Journal of Fatigue 15(4), 273–282 (1993) [43] Thomsen, M.L., Hoeppner, D.W.: Microstructurally Based Variations on the Dwell Fatigue Life of Titanium Alloy IMI 834. Presented at the FAA/NASA International Symposium on Advanced Structural Integrity Methods for Airframe Durability and Damage Tolerance, Part 2, Hampton, Virginia, May 4-6, vol. 3274, pp. 871–889. NASA Conference Publication (1994) [44] Elliott, III C. B., Hoeppner, D. W.: Fretting As a Fatigue Crack Nucleation Mechanism-A Close-up View. Presented at the USAF Conference on Structural Integrity (December 1997); published in the Conference Proceedings [45] Taylor, A. M. H., Hoeppner, D. W.: The Effect of Prior Corrosion Damage on the Short Crack Growth Rates of Two Aluminum Alloys. Presented at the USAF conference on Structural Integrity (December 1997); published in the Conference Proceedings [46] Thomsen, M.L., Hoeppner, D.W.: The Effect of Dwell Loading on the Strain Accumulation Behavior of Titanium Alloys. International Journal of Fatigue 20(4), 309–317 (1998) [47] Okada, T., Hoeppner, D.W.: The Behavior of Short Cracks in Corrosive Environments for 7075 Al Alloy. In: Donne, C.D. (ed.) Proceedings of the 23rd Symposium of the International Committee on Aeronautical Fatigue, Presented at ICAF 2005 Structural Integrity of Advanced Aircraft and Life Extension for Current Fleets - Lessons Learned in 50 Years after the Comet Accidents, Hamburg, Germany, vol. 2, pp. 613–622 (2005) [48] Smiltneek, L., Shinde, S., Hoeppner, D.: A Single Cylinder In-Situ SEM Fatigue System. Review of Scientific Instruments, American Institute of Physics 77, 1–4 (2006) [49] Hoeppner, D.W.: From No-life to safe life to HOLISTIC Structural Integrity Based Design. Invited Presentation and Paper for the workshop on Structures, Materials, and Propulsion, Held at National Research Council-Canada, Ottawa, Ontario, Canada (July 2002); published in the workshop proceedings [50] Brooks, C. et al.: AP/ES, http://apesolutions.com/ ;There are many relevant publications [51] NRC-IAR Structures and Materials, http://www.nrc-cnrc.gc.ca/eng/ news/iar/ 2005/08/04/aircraft-assessment.html [52] Yamagiwa, K., Kataoka, S., Izumi, S., Sakai, S.: Measurement of Three Dimensional Geometry of Creep Void and Grain Boundary with Combining 3D-EBSD Method and SEM Images. Trans. of Japan Society of Mech. Eng. (A) 76(772), 1799–1805 [53] Yamagiwa, K.: Evolution of Creep Damage. Presentation at HOLSIP 10 held in Snowbird, UT (February 27-March 3, 2011)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Modelling of Continuing Damage for Damage Tolerance Analysis Yan Bombardier, Min Liao, and Guillaume Renaud Institute for Aerospace Research, National Research Council Canada

Abstract. This paper presents and compares two continuing damage modelling approaches: a simplified approach, which approximately models diametrical cracks as two independent radial cracks, and an improved approach that accurately models the interaction between the cracks. Stress intensity factor closed-form solutions were developed for the improved approach to accurately model continuing damage growth in a structure. Through several examples, it was shown that the simplified approach, commonly used by the aircraft industry, always resulted in smaller continuing damage sizes and shorter fatigue lives compared to the improved approach. For the tested cases, the continuing damage sizes calculated using the simplified approach were underestimated by up to 24 times and the fatigue lives were overestimated by up to 1.4 times.

1 Introduction As specified by the United States Air Force damage tolerance design guidelines provided in the Joint Service Specification Guide JSSG-2006 [1], safety of flight structures shall be capable of maintaining adequate residual strength in the presence of initial defects and damages induced during normal usage and maintenance. For slow crack growth and fail safe primary elements at fastener hole locations, JSSG2006 recommends that an initial primary flaw of 1.27 mm (0.050 inch) at the most critical side of the hole should be assumed. In addition, a continuing damage, or secondary flaw, equivalent to a 0.127 mm (0.005 inch) crack should be assumed on the opposite side of the hole. The initial flaw is normally assumed to be a quartercircular corner crack for structural elements thicker than the initial flaw; otherwise, a through-the-thickness crack is assumed. The damage tolerance analysis of a hole in a thick panel consists of simulating unequal diametrical cracks that would evolve to an edge crack when one of the cracks reaches the edge of the panel. The typical two-stage crack growth scenario is illustrated in Figure 1, where and are the primary and secondary cracks, respectively. Due to the complexity and the lack of closed-form stress intensity solutions for unequal diametrical crack problems, a simplified approach has been commonly used by the aircraft industry. This approach, detailed in the following section, assumes that a diametrical crack can be modelled as two independent radial cracks. Recent advances in the development of numerical and closed-form stress *

Oral presentation.

232

Y. Bombardier, M. Liao, and G. Renaud

Stage A: Unequal diametrical cracks

Stage B: Edge crack through a hole

σ

σ

Primary Secondary crack

Broken Secondary crack

H

A

H

A

σ

B

B

σ Δ

Cross-section A-A

Cross-section B-B

= 1.27 mm (0.050 inch) = 0.127 mm (0.005 inch)

Fig. 1 Typical damage tolerance analysis scenario with a continuing damage.

intensity factor solutions for unequal diametrical cracks at loaded and unloaded holes can be used to improve the continuing damage size calculation. In this paper, new closed-form stress intensity factor solutions are proposed to improve the two-stage crack growth process illustrated in Figure 1. Using the compounding method, these new solutions were developed to be accurate and simple to use while minimizing computational time. Through several examples, this paper compares the simplified approach of continuing damage modelling with the proposed improved approach that simulates unequal diametrical crack growth simultaneously.

2 Simplified Modelling Approach The simplified approach consists of independently propagating the primary and secondary flaws. This approach, illustrated in Figure 2, requires two crack growth curves: one for the primary flaw and one for the secondary flaw. The continuing damage size is obtained by calculating the amount of growth, Δ , that accumulates prior to ligament failure, plus the initial secondary flaw size of 0.127mm (0.005 inch), as specified in JSSG-2006. For the example shown in Figure 2, the initial crack size used for stage B would be 0.477 mm, considering that the secondary crack, , grew by 0.350 mm from an initial flaw of 0.127 mm while the primary crack, , grew until ligament failure ( = 26 mm). Note that the crack growth curve used to calculate the continuing damage was the primary crack growth curve, but starting from a 0.127 mm crack and growing towards the closest edge as opposed to the farthest edge. This approximation is often used because the crack growth simulation can be conducted using only one crack growth curve starting from 0.127 mm (0.005 inch) and the resulting continuing damage is larger. The simplified approach neglects the interaction between the primary and secondary cracks.

Modelling of Continuing Damage for Damage Tolerance Analysis

233

100 Crack growth from

=1.27 mm

Crack size (mm)

∅8 Primary flaw

10

30 250 mm

1 0.005+Δa=0.477 mm

Crack growth from

=0.127

Δa=0.350 mm

Secondary flaw

0.1 0

30

60 Cycles

90

120 x 10000

Fig. 2 Calculation of the continuing damage size using the simplified approach.

3 Improved Modelling Approach The improved approach consists of propagating the primary and secondary cracks simultaneously to accurately calculate the continuing damage size. The improved approach includes the development of two capabilities: the development of the stress intensity factor solutions for stages A and B shown in Figure 1 and the development of a crack growth analysis program capable of modelling multiple cracks growing simultaneously. The stress intensity factor solutions for stages A and B were developed using the superposition and compounding methods presented in [2-4]. Those solutions are detailed in Appendix A, where the stress intensity factors are: for √

for

1,2

(1)

2

(2)

where and are the stress intensity factors for stages A and B, respectively. The subscript refers to the left ( 1) and right ( 2) crack tips as shown and terms are the geometric correction factors, or -factors, Figure 1. The expressed in terms of the crack size measured from the edge of the hole. The quarter-circular corner crack solutions were obtained by multiplying Eqns (1) and (2) by the crack shape correction factor, , provided in Appendix A. The diametrical crack solution was benchmarked against results obtained using AFGROW, ABAQUS/Zencrack, and StressCheck for through-the-thickness and part-through cracks [2, 3]. The stress intensity factor results obtained from the closed-form solution were within 2% of the results obtained with StressCheck and

234

Y. Bombardier, M. Liao, and G. Renaud

AFGROW. The edge crack solution growing through a hole was also in good agreement with AFGROW and StressCheck results [4] for through-the-thickness cracks. Preliminary results for the quarter-circular corner edge crack solution growing through a hole were within 5% of the StressCheck results. For stage A, the developed crack propagation algorithm can propagate each crack tip simultaneously, while considering the interaction effect through the use of the stress intensity factor solutions provided in Appendix A. For stage B, any crack growth analysis program capable of importing user-defined β-factors is sufficient. The crack growth simulations presented in this paper were conducted using an in-house crack growth analysis program developed by NRC, CanGROW. CanGROW has the capability of propagating an unlimited number of cracks simultaneously using a library of β-factors that are automatically compounded according to the geometry of the simulated problem. Therefore, CanGROW automatically switches from Solution A to Solution B once the ligament breaks, resulting in a continuous crack growth simulation.

4 Examples Various geometries, stress levels, and fastener loads were considered in order to quantify the difference between the simplified and improved approaches in terms of the calculated continuing damage size and fatigue life. The continuing damage size was obtained by extracting the size of the secondary damage, , after the primary crack, , reached the left edge of the plate and failed the ligament. The fatigue lives were compared at the end of stage A (ligament failure) and stage B (complete plate failure). The problem considered was a 6 mm (0.236 inch) thick 7075-T73 aluminium plate with a width, , of 300 mm (11.81 inches) and a hole with diameter, , of 8 mm (0.315 inch), located at distances from the left edge ranging from 20 mm (0.787 inch) to 80 mm (3.150 inches). Three levels of applied remote stress, , were evaluated: 100, 150, and 200 MPa. For this study, a fastener load was also applied in some cases to investigate its impact on the calculated continuing damage sizes. The fastener load, , was expressed in terms of bearing-to-bypass stress ratio, , as follows: (3) For each case, the continuing damage sizes and fatigue lives were obtained using the simplified and the improved approaches. The results were compared by calculating the ratio of the improved approach results to the simplified approach results. For example, a fatigue life ratio smaller than 1.0 means that the life calculated using the improved approach was shorter than the one obtained with the simplified approach. The continuing damage sizes and fatigue life ratios are presented in Table 1 for the tested cases. As shown, the continuing damage sizes calculated using the simplified approach were consistently and significantly underestimated based on the results obtained using the improved approach. The continuing damage sizes obtained using the improved approach were 5 to 24 times

Modelling of Continuing Damage for Damage Tolerance Analysis

235

larger than the ones calculated using the simplified approach. Consequently, the improved approach, which models the interaction between the primary and the secondary cracks, provided higher stress intensity factors and crack growth rates, resulting in a shorter fatigue life. The fatigue life calculated with the improved approach was 5% to 36% shorter than the ones obtained using the simplified approach. Larger discrepancies between the simplified and improved approaches were found for the cases where fastener loads were applied (Cases 1b*, 2b*, and 3b*). Table 1 Continuing damage sizes and fatigue life ratios calculated using the simplified and improved approaches. Parameters Case 1a 1b 1c 2b 3b 1b* 2b* 3b*

20 20 20 40 80 20 40 80

100 150 200 150 150 150 150 150

0.0 0.0 0.0 0.0 0.0 1.0 1.0 1.0

Continuing damage size (mm) Simplifi Improv Ratio ed ed 0.31 1.89 6.04 0.51 2.57 5.08 0.59 3.01 5.10 1.03 15.29 14.86 1.81 42.71 23.65 1.08 4.72 4.38 2.99 19.78 6.61 7.39 46.80 6.33

Fatigue life ratio Ligame Plate nt failure failure 1.00 0.96 0.98 0.94 0.98 0.93 0.91 0.89 0.85 0.85 0.93 0.88 0.83 0.81 0.74 0.74

The closed-form stress intensity factor solution for Case 2b was compared with results calculated using AFGROW and StressCheck to ensure that the stress intensity factors obtained using the improved approach were accurate and to better understand the interaction effect between the primary and secondary cracks. As shown in Figure 3, very good agreement was obtained between the closed-form solution, AFGROW, and StressCheck for stages A and B. For Case 2b, the relative differences between the closed-form solution and StressCheck were within 2% for both crack tips whereas most results were within 1 %. Most AFGROW results were within 4% of the closed-form solution. As shown, the β-factors calculated by AFGROW for the secondary crack were marginally higher (3.4%) for small crack sizes compared to the results of the closed-form solution and StressCheck. Interpolation error may also have shifted the AFGROW results = 7 mm and = 20 mm, contributing in increasing the relative between differences for 1.

236

Y. Bombardier, M. Liao, and G. Renaud

Fig. 3 Comparison between the β-factor solutions obtained using the closed-form solution, AFGROW, and StressCheck (through-the-thickness cracks).

The radial crack solution used to calculate the continuing damage using the simplified approach is also shown in Figure 3. Compared to the simplified solution which uses a radial crack solution, the diametrical crack solution significantly increases the stress intensity factors at the secondary damage, resulting in a higher crack growth rate. For diametrical cracks problems, small cracks, such as the secondary damage simulated in this paper, are influenced by the presence of larger diametrically opposed cracks, while the small cracks have insignificant effect on the large cracks. This is due to the fact that the stress intensity factors at diametrical cracks are highly dependent on the tip-to-tip crack size. Consequently, the stress intensity factor at the primary crack is similar to the one calculated using the simplified approach (radial crack model) for a relatively small secondary crack. The effect of the secondary crack on the primary crack increases as the secondary damage grows faster than the primary crack due to higher stress intensity factors. Generally, the stress intensity factors calculated using the improved approach at the primary and secondary cracks should always be higher than the ones calculated using the simplified approach. Consequently, the fatigue life can be overestimated if the simplified approach is used to carry out the damage tolerance analysis. The improved approach was extended to simulate multiple-site fatigue damage scenarios by compounding additional effects to the stress intensity factor solutions, such as the effect of crack interactions and the effect of approaching holes. The MSD examples proposed in this paper combines Cases 1b, 2b, and 3b, with and without fastener loads, where the dimensions are given in Figure 4. This

Modelling of Continuing Damage for Damage Tolerance Analysis

237

scenario is particularly important as the Federal Aviation Administration (FAA) recently issued a final rule amending the regulations pertaining to certification and operation of transport category airplanes to prevent widespread fatigue damage in aircraft structures [5]. σ = 300 mm = 20 mm A

= 8 mm = 40 mm

Thickness ( = 6 mm)

= 80 mm

A

σ Cross-section A-A Hole #1 Hole #2 = 1.27 mm (0.050 inch) = 0.127 mm (0.005 inch)

Hole #3 = 0.127 mm (0.005 inch) = 0.127 mm (0.005 inch)

= 0.127 mm (0.005 inch) = 0.127 mm (0.005 inch)

Fig. 4 Multiple-site fatigue damage scenario.

The continuing damage sizes and fatigue life ratios are presented in Table 2 for the MSD scenario using the simplified and improved approaches with fastener loads ( 1.0) and without fastener loads ( 0). The continuing damage sizes were extracted at holes 1, 2, and 3 when the edge crack reached the hole. As shown, the discrepancy between the continuing damage sizes seems to decrease as the hole is further away from the edge. This could be explained by the facts that the crack growth rates calculated using the improved approach were higher and the edge crack reached the following hole faster due to crack interactions. In addition, the simplified approach used the crack growth curve at the first hole towards the edge of the plate, which could have overestimated the crack growth rates at the second and third holes for the MSD scenario. Overall, this example shows that the fatigue lives calculated using the simplified approach may have been overestimated by 27% and 42% for the case without and with fastener loads, respectively. Table 2 Continuing damage sizes and fatigue life ratios for the Multiple-site fatigue damage scenarios.

0.0 1.0

Ratio

Fatigue life ratio Imp.

Simp.

Ratio

Imp.

Simp.

Ratio

Imp.

Simp.

Continuing damage (mm)

0.51 5.03 9.96 0.58 0.81 1.41 0.62 0.50 0.80 1.08 5.70 5.28 1.35 0.75 0.56 1.55 0.49 0.31

Ligament Hole Hole failure No. 2 No. 3 0.84 0.75

0.79 0.68

0.79 0.70

Plate failure 0.79 0.71

238

Y. Bombardier, M. Liao, and G. Renaud

5 Concluding Remarks This paper presented the continuing damage sizes and fatigue lives resulting from two different modelling approaches. The simplified approach has been commonly used by the aircraft industry, whereas the improved approach provides more accurate results. Through several examples involving a finite plate with various hole positions, fastener loads, and remote tensile stresses, it was shown that the simplified approach consistently and significantly underestimated the continuing damage sizes and overestimated the fatigue life as it does not consider the interaction between the primary and secondary cracks. Consequently, the simplified approach presented in this paper may result in un-conservative fatigue life predictions. The stress intensity factor closed-form solutions developed to support the improved approach were in very good agreement with stress intensity factors calculated using finite element analyses (StressCheck) and the advanced models available in AFGROW. Implemented in a crack growth analysis code capable of propagating multiple cracks simultaneously, these new solutions provide a simple and efficient way to calculate the fatigue life considering secondary cracks and multiple-site fatigue damage in airframe structures.

Acknowledgments This work was performed with financial support from DRDC (Defence Research and Development Canada) and NRC (National Research Council Canada) through project 13ph11: Quantitative Risk Analysis of Aircraft Structures, under 13ph: Economic Life Assessment for CF Air Fleets.

References [1] United States of America Department of Defense. Aircraft Structures, United States Department of Defense, JSSG-2006 (1998) [2] Bombardier, Y., Liao, M.: A New Stress Intensity Factor Solution for Cracks at an Offset Loaded Fastener Hole. In: 51st AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference (2010) [3] Bombardier, Y., Liao, M., Bisson, S., and Beres, W.: A New Stress Intensity Factor Solution for Through-Thickness and Part-Through Cracks at an Offset Loaded Hole, National Research Council Canada, LTR-SMPL-2010-0167, Ottawa (2010) [4] Bombardier, Y., Liao, M.: Stress Intensity Factor Solution for a Through-theThickness Edge Crack Growing Through an Open Hole, National Research Council Canada, LTR-SMPL-2011-0015, Ottawa (2011) [5] Federal Aviation Administration, Aging Airplane Program: Widespread Fatigue Damage. Federal Register 75(219), 69746–69789 (2010) [6] Tweed, J., Rooke, D.P.: The elastic problem for an infinite solid containing a circular hole with a pair of radial edge cracks of different lengths. International Journal of Engineering Science 14(10), 925–933 (1976)

Modelling of Continuing Damage for Damage Tolerance Analysis

239

[7] Bombardier, Y., Renaud, G., Liao, M.: Stress Intensity Factor Solutions for Cracks at a Hole in an Infinite Sheet. National Research Council Canada, LTR-SMPL-20100117, Ottawa (2010) [8] Kathiresan, K., Hsu, T. M., and Brussat, T. R.: Advanced Life Analysis Methods. vol. 2. Crack Growth Analysis Methods for Attachment Lugs. Lockheed-Georgia Company, AFWAL-TR-84-3080, vol. II, Marietta, Georgia (1984) [9] Tada, H., Paris, P.C., Irwin, G.R.: The stress analysis of cracks handbook, 2nd edn. Paris Productions Inc., St. Louis (1985) [10] Newman Jr., J.C., Raju, I.S.: Stress-Intensity Factor Equations for Cracks in ThreeDimensional Finite Bodies Subjected to Tension and Bending Loads. In: Computational Methods in the Mechanics of Fracture Mechanics and mathematical Methods First series Computational methods in Mechanics, pp. 311–334. NorthHolland, Amsterdam (1986)

240

Y. Bombardier, M. Liao, and G. Renaud

Appendix A Stress intensity factor solution for diametrical cracks in a finite plate A new stress intensity solution was developed by the National Research Council Canada (NRC) to simulate unequal diametrical cracks at an offset unloaded or loaded hole [2, 3]. The solution was obtained by superposition and compounding of several stress intensity factor solutions. For simplicity, the solution provided in this appendix only presents the solution for the unloaded hole shown in Figure 5.

σ

Thickness (t) B

D=2R

a

a a

a

b

a

W

σ

a

Through-thickness cracks

Fig. 5 Unequal diametrical cracks at an offset fastener hole.

The solution was developed for radial and diametrical cracks simultaneously. In this appendix, the stress intensity factor and the -factor for crack of Solution and respectively, where 1 for the crack tip growing are defined by towards the left side of the plate and 2 for the crack tip growing towards the right side of the plate, as illustrated in Figure 5. The stress intensity factor solution was obtained by compounding the following solutions: the radial or diametrical cracks at a hole in an infinite plate (Solution A1), the offset crack in a finite plate (Solution A2), and the offset hole in a finite plate (Solution A3). The stress , of Solution A for crack tip i is calculated as follows: intensity factor, 1,2

(4)

Modelling of Continuing Damage for Damage Tolerance Analysis

where the geometric correction factor, or -factor, of Solution A,

241

, is given by: (5)

The stress intensity factor solution for diametrical cracks in an infinite plate, Solution A1, was derived by Tweed and Rooke [6]. This solution requires solving a linear algebraic system to approximate the solution of coupled singular integral equations. The procedure, although relatively simple, is computationally intensive, especially considering that it must be used several times during a crack growth simulation. Several approximated solutions were evaluated at NRC [7] and a new approximated solution was developed to accurately estimate the solution developed by Tweed and Rooke for a large range of / and / values [4, 7]. The regression parameters were calculated for any / 64000. The maximum relative error is 1.0% for any / 500, 1.3% for any / 1000, and 2.7% for any / 64000. For completeness, the approximate solution developed by NRC is provided as follows:

,

,

2

2

(6)

(7)

where, 1,2

1,2

1

0

,

1

(8)

1

(9)

(10)

(11)

242

Y. Bombardier, M. Liao, and G. Renaud

The coefficients

and

are given in Table 3.

Table 3 Regression parameters for the calculation of βr and βu.

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14

3.3645 -7.209304 8.230965 -3.500286 -2.923363 4.306705 -1.562110 -

1.00000000 0.98654553 0.03673931 0.90111374 -0.01711464 2.26550938 0.08639597 3.52048868 1.12869498 0.05132827 0.90257160 -0.01714307 2.34770606 0.08386375 3.52544398

Solutions A2 and A3 were directly taken from [8], where A2 models the presence of the crack of length 2 and A3 models the presence of the hole of diameter 2 . The -factors expressed in terms of remote stress and crack length are calculated using the following equations: ,

for

,

(12)

for

⁄2

(13)

for

⁄2

(14)

,

,

⁄2

for

⁄2

(15)

where, 1 2

(16) 1 2

,

1 4

1

(17)

cos

.

sec

(18)

Modelling of Continuing Damage for Damage Tolerance Analysis

,ω

sec

1 1

0.21 sin 8 arctan

243

1 .

(19)

(20)

2ω

2

2ω

(21)

4 λ 7

3 λ 7

(22)

sin

(23)

Stress intensity factor solution for an edge crack growing through a hole The stress intensity factor solution for the through-the-thickness edge crack growing through an open fastener hole, illustrated in Figure 6, was developed by NRC [4]. The solution was developed by compounding the following solutions: the radial or diametrical cracks at a hole in an infinite plate (Solution B1) and the edge crack solution in a finite plate (Solution B2). The closed-form solution presented in this appendix also provides a way to model various boundary conditions on the plate, such as a plate with free edges (Figure 7a), a plate with laterally constrained displacements on the right edge (Figure 7b), and a plate with laterally constrained displacements on the left edge (Figure 7c). The edge constraints applied to the left and right edges prevent the inplane bending of the plate and consequently, reduce the crack tip stress intensity factor. The option to include constraints in the model was added to simulate the effect of adjacent structures on the boundary conditions. For example, the presence of a stringer, a spar cap, or an adjacent panel connected to the one that is analysed would limit the in-plane bending. In [4], the results calculated using the closed-form solution were in very good agreement with the results obtained with the finite element method (StressCheck) and the advanced models available in AFGROW. Based on extensive comparisons, the new solution is accurate, simple to implement, includes three types of boundary conditions, and does not require interpolation or extrapolation.

244

Y. Bombardier, M. Liao, and G. Renaud

σ

Thickness (t) B

D=2R a 2h a0=B+R+ a2 W

σ

Failed ligament

a

Through-thickness

Fig. 6 Through-the-thickness edge crack growing through a hole.

a) Free

b) Right edge constraints

c) Left edge

Fig. 7 Boundary conditions modelled by the closed-form solution.

The stress intensity factor,

, of Solution B is calculated as follows: (24)

where the geometric correction factor, or -factor, of Solution B,

is given by:

(25) The -factor for solution B1 is taken from Solution A1. The edge crack problem covered in this appendix is however not diametrical cracks in a plate and the crack size corresponding to the failed ligament had to be adjusted to model the

Modelling of Continuing Damage for Damage Tolerance Analysis

245

additional compliance caused by the proximity of the free edge. As shown in Figure 8, the crack size corresponding to the failed ligament, , was artificially increased as follows: 2

2

(26)

where 4 for free edges (Figure 7a), 6 for right edge constraints (Figure 7b), and 1 for left edge constraints (Figure 7c). Except for the left edge constraints, the values used for the parameter has no physical meaning and was adjusted in order to provide accurate stress intensity factors based on finite element analysis results with the corresponding boundary conditions.

a a a*

Edge of the plate Fig. 8 Idealized model to calculate the effect of the hole (Solution B1).

Solution B1,

, is consequently defined as (27)

where

is given by Eq. (7) with

.

The edge crack effect, given by Solution B2, was obtained from the literature for the three available types of boundary conditions. The edge crack solution in a plate without boundary conditions (Figure 7a) was directly taken from the literature [9]: tan

·

0.752

1.286

0.37 1 cos

sin

(28)

246

Y. Bombardier, M. Liao, and G. Renaud

where (29)

2

The accuracy of this solution is better than 0.5% for any / value and for . The effect of the plate height, , is practically negligible for / 1.0 [9]. The solution for the right edge constraints (Figure 7b) was modelled using the double edge notch solution as follows [9]: 1.122

0.561

0.205

0.471

0.190

1

(30)

where (31) The accuracy of this solution is better than 0.5% for any / [9]. The solution for the left edge constraints (Figure 7c) was modelled using the centre crack solution as follows [9]: 1

0.025

0.06

sec

(32)

where and are defined in Eqns (31) and (29) respectively. The accuracy of this solution is better than 0.1% for any / [9]. Quarter-circular corner crack shape correction The effect of crack shape can be included by compounding the shape correction factor, , to the through-the-thickness stress intensity factor solution. The crack shape correction factor was derived from the solution developed by Newman and Raju for equal diametrical corner cracks [10] by isolating the crack shape effect and was shown to provide acceptable results for equal and unequal diametrical cracks [3]. a1 a2 t

Fig. 9 Quarter-circular corner cracks at a hole.

Modelling of Continuing Damage for Damage Tolerance Analysis

247

The solution presented in this appendix assumed equal diametrical quartercircular corner cracks with a crack depth to crack length ratio of 1.0. For unequal cracks the opposite crack was assumed to be of equal size as the one analysed. The crack shape correction factor is given by: 0.6625

cos cos √

0.1285 0.0676 1 0.13

(33)

where 1

1

0.1

0.358

1.425

1.04 1

0.1 1

1

0.35

sin

1.578

2.156

0.85

cos

(34)

0.15

/ 1

.

0.13

(35) (36)

(37)

2 cos 0.85

2

2 2

(38)

(39)

(40)

Accurate crack shape correction factors are currently being developed using three dimensional finite element analyses to improve the solution derived from Newman and Raju’s equations for equal and unequal diametrical cracks with and without fastener loads as well as a specific solution for the edge crack growing through a hole and linking with a corner crack.

26th ICAF Symposium – Montreal, 1-3 June 2011 *

The Critical Importance of Correctly Characterising Fatigue Crack Growth Rates in the Threshold Regime K.F. Walker and S.A. Barter Defence Science Technology Organisation, Fishermans Bend, Australia

Abstract. The majority of the fatigue life for high strength metallic aircraft structures is often spent in the small crack near threshold region. Accurate modelling of this regime is therefore essential if we are to gain the maximum possible life from existing structures while maintaining adequate levels of safety. The threshold and very low growth rate region poses a number of challenges including small crack behaviour, Kmax effects and the apparent breakdown of similitude. Standardised test methods such as the load reduction technique in ASTM E647 introduce a range of problems including load history effects and remote closure. The end result is that typical thresholds are over estimated and rates in the threshold and near threshold regions are under estimated. This paper considers data collected by two very different methods, both of which appear to avoid the problems of the load reduction method and therefore allow for considerably improved near threshold crack growth data, which, as is shown, can improve accuracy in life predictions. Unlike the standard method, the first alternate method generates the cracks under remote compression-compression loading. The second alternate method relies on specially designed sequences to mark the fracture surface on very small/short, natural cracks such that Quantitative Fractography is possible. Excellent correlation between test and analysis are shown for 7050-T7451 aluminium alloy examples including simple coupons and cracking from a full-scale fighter aircraft centre section test.

1 Introduction The ability to accurately understand and characterise the behaviour of fatigue cracks in the near threshold regime is crucial for assuring structural integrity for high strength metallic aircraft structures. This is because the majority of the total fatigue life is often spent in this regime. In the past, attempts to model crack growth initiating from small surface breaking constituent particles present in 7050-T7451 aluminium alloy coupons tested under a fighter aircraft Variable Amplitude (VA) loading spectrum, using fracture mechanics based methods have produced poor results. See for an example Figure 1 re-produced from [1], which compares quantitative fractography (QF) results post test with calibrated (to the test lives) predictions. The shape of the growth curves is incorrect, and in order to get the life predictions close to those of the coupons, a relatively large initial crack *

Oral presentation.

250

K.F. Walker and S.A. Barter

size was required. The worst aspect is that the analytical predictions are nonconservative. This means that they under estimate the crack growth rate while the crack is small and as a result over estimate the life (if a realistic value of initial crack size were to be used). Conversely, in the Non-Destructive Inspection (NDI) region (> about 1 mm) they over estimate the crack growth forcing short inspection intervals (where required), which is expensive and wasteful.

10

Crack Depth, mm

1

0.1

0.01

0.001 0

5000

10000

15000 20000 25000 Time, Flight Hours

30000

35000

40000

Fig. 1 Example (from [1]) of poor analytical results compared with experiment. The experimental data are shown as symbols, and the analyses as solid curves. The four sets are for tests under the same load sequence at different stress scaling levels.

Researchers at DSTO have noted the often poor results obtained from traditional crack growth models and have recently conducted further investigations. Work at DSTO by Wallbrink and Walker and reported in [2] has confirmed that the threshold and near threshold region of the crack growth rate curve is critical for successful crack growth predictions, particularly when such cracks commence from very small initial discontinuities, typical of aircraft structure, for a number of different aircraft spectra and stress level combinations. Recent work, the subject of this paper, has shown that the principal reason for the poor analytical results is that the baseline material crack growth rate characterisation data as collected by ASTM Standard E647 long crack tests when used in predictive programs on a cycle-by-cycle basis are inadequate. The generation of Constant Amplitude (CA) crack growth rate data in the threshold and near threshold regime is usually performed using the constant R load reduction method [3]. The load reduction method however has been demonstrated to produce higher thresholds and lower crack growth rates in the near-threshold regime than steady-state CA data on a number of materials (two aluminium alloys,

The Critical Importance of Correctly Characterising Fatigue Crack Growth Rates

251

two titanium alloys and a superalloy), especially those that develop rough and torturous crack surfaces [4-8]. In addition, the load reduction method produces “fanning” with the load ratio in the threshold regime for some materials (fanning gives more spread in the ΔK-rate data with load ratio in the threshold regime than in the mid-rate regime) [9]. It has also been shown that the test method introduces a load-history effect which may be caused by remote closure [4, 10, 11]. It is suggested that the load reduction method therefore does not produce reliable rate data in the threshold and near threshold regimes, and this unreliability may extend into the mid ΔK growth data for high strength aluminium alloys, as will be shown. In this paper we detail two separate, independent methods for determining the rate data. The methods were applied to 7050-T7451 Alloy. Both methods avoid problems known to exist in methods such as the ASTM Standard E647 load reduction technique [3]. The first alternative method (Method 1) involves establishing the baseline CA crack growth rates by testing under compression-compression pre-cracking, which avoids the problems of remote closure which are known to compromise the results under conventional load reduction testing [9]. Data obtained by the earlier methods usually also need to be corrected to account for non-plasticity induced closure, i.e. closure caused by other mechanisms such as fretting and debris formation, which can be estimated by comparison of lower stress ratio (R=min stress/max stress) results with high R conditions as the threshold is approached. The second method (Method 2) involves growing natural cracks from typical discontinuities in un-notched specimens. The specimens are subjected to specially designed simple load sequences which consist of bands of CA loading with shorter bands of cycles with a changed R inserted periodically. It has been found [12] that the changes in R affect the crack path in such a way that individual CA bands can be identified on the fracture surface when it is examined under a high powered microscope (optical and/or Scanning Electron Microscope; SEM) and accurately measured. This process, known as QF, is carried out on the fractures surface produced by the testing. It produces CA growth rates for naturally occurring small cracks since these bands can be identified at very small crack depths. Other methods have also been used to establish crack growth rates in the near threshold region. The most notable, for natural cracks growing from natural discontinuities is to measure the growth, during the testing, on the surface of the specimens. The main problem with these data, for example [13], is that significant scatter is evident when the cracks are small since the plane stress conditions at the surface influences a larger portion of the total crack front than is the case when the crack is larger. Micro structural effects are also more significant for very small cracks since the growth rate is being influenced by a very much smaller number of grains. Another problem is that the crack needs to be found while it is very small requiring that the surface must be highly finished, and the area where cracking is to be measured usually needs to be confined by a notch so that the cracks can be found with reasonable ease. The rate curve data in the threshold/near threshold region from the two methods explored here has been found to demonstrate very consistent crack growth rate

252

K.F. Walker and S.A. Barter

behaviour. In both cases the results were crack growth rates that are faster and thresholds that are lower in the threshold/near threshold region than those established using earlier techniques. When these apparently more accurate data are used as an input to a fracture mechanics based crack growth analysis to predict growth under VA spectrum loading the results are found to compare very well with the test data. Examples detailed in this paper include coupon specimen results and a case of representative cracking in the main wing carry through structure on a fighter aircraft, both subjected to fighter aircraft spectra.

2 Method 1 In order to produce improved steady-state CA data, compression pre-cracking methods have been proposed [14-16]. A pre-notched specimen is subjected to a large compression load followed by moderate compression-compression cyclic loading. The initial loading is such that local plasticity occurs at the notch and a small, localised zone of residual tension is developed. The remote loading is compression-compression, but a cycle with a tension component sufficient to initiate a fatigue crack is present locally at the notch due to the high residual tension created by the initial compressive load. The crack grows initially, but it slows and then stops when it grows near to the edge of residual tension field. At this point, regular tension-tension cycling can be commenced at a very low level to produce low ΔK crack growth. In this way it is possible to produce fatigue crack growth in the threshold and near threshold regions that is unaffected by the high ΔK grown crack wake problems of the standard load reduction methods. The cracks are actually physically large, of the order of several millimetres, but the behaviour is similar to that exhibited by physically short cracks although many grains in the material are still being tested [17]. This method has been used to characterise the effective growth rate curve for 7050-T7451 [9]. It was found that the FASTRAN analytical crack closure model which accounts for Plasticity Induced Crack Closure (PICC) was able to collapse the data for different R ratios very effectively. A comparison of the collapsed rate data compared with results from standard load reduction testing is shown in Figure 2. It was also found that closure still existed under certain conditions in the compression-compression test, measured using Elber’s method of direct measurement of the strain ahead of the crack tip [9]. As explained in [9], it is considered that the additional closure was due to Roughness Induced Crack Closure (RICC) and/or Debris Induced Crack Closure (DICC). RICC and DICC have not been successfully modelled explicitly, but by the use of local strain gauges the closure level could be measured, and so the effective rate curve could be determined as shown in Figure 3.

The Critical Importance of Correctly Characterising Fatigue Crack Growth Rates

253

Fig. 2 Effective rate curve for 7050-T7451. Provided by the authors of [9]. Data denoted by the red circles were derived by the ASTM Load Reduction method. The open symbol data were derived through compression pre-cracking methods.

254

K.F. Walker and S.A. Barter

Fig. 3 Effective rate curve, demonstrating that accounting for plasticity effects collapses the data, but does not account for other forms of closure. There is significant closure due to the other effects (roughness and debris) which moves the curve further to the left in the near threshold region as shown in this Figure. Reproduced from [9].

3 Method 2 An alternative to the semi-indirect measurement of crack growth during fatigue tests of varying types is the direct measurement of crack growth from a fatigue fracture surface. This method is usually referred to as QF. While the most common method of measuring crack growth rates on fracture surfaces is to look for and measure the widths of individual striations that can be correlated to individual loads, this is both time consuming and is limited by the size of the features that can be observed on a fatigue fracture surface. Typically single striations can not be found at growth rates below about 2x10-8 m/cycle, and even then such features can be extremely difficult to confirm as being striations, as can be seen in Figure 4. An alternative to this is the marker load method developed primarily by Barter at DSTO and Wanhill at NLR in the Netherlands [18], which relies on finding markings produced by bands of loading cycles rather than individual cycles. Such bands can be measured at sizes similar to individual

The Critical Importance of Correctly Characterising Fatigue Crack Growth Rates

255

striations, i.e. about 2x10-8 m, see Figure 4, but since they are made up of many cycles this allows much smaller crack growth rates to be measured albeit for the average growth rates of short regions of crack growth. This is consistent with the standard tests which usually also average the growth over many cycles.

Fig. 4 SEM images of fatigue progression markings on an AA7050-T7451 fatigue fracture surface. The image on the left is of an area produced by R = 0.3 loads at a ΔK of about 4.4 MPa√m. The image on the right shows a band produced by 10 cycles of R = 0 loads at a ΔK of about 1.5 MPa√m. In both cases the smallest width of a progression marking is about 2×10-8m.

To produce measurable marks on a fracture surface, it is necessary to vary the loading. This arguably could result in “spectrum effects” altering the growth rates for the bands of loads of interest. It has been found that the strongest spectrum effect is produced by overloads [12]. Consequently, if a constant maximum load is applied to the test specimen, this effect is negated. The most obvious other influence, crack closure, appears to be less important when cracks are very small (<1 mm in each dimension), so this method is most applicable to measuring data from small cracks. In 7050-T7451, the apparent absence of crack closure affecting crack growth can be demonstrated fractographically on a crack produced by a simple spectrum with bands of CA with decreasing R. This is shown in an example (Figure 5). In this figure the bands marked; A-E separated by marker bands; M1-M5 are growing steadily larger, whereas if closure was having a significant influence it may be expected that the low R bands would start to have about the same amount of growth as they approach R=0. This region of the crack was at a depth of about 1mm and the peak stress applied was 200MPa. Other tests with the spectrum shown in Figure 5, modified with progressively larger M1 marker bands did not reveal any notable influence on the bands of the remainder of the spectrum suggesting that for these types of sequences spectrum effects that may be due to factors other than closure are minimal.

256

K.F. Walker and S.A. Barter

M1 1 Normalised load level

M5 M4 E

M3 M2

D

M1 0 -0.5

C

M1

-1 0

B E

A M2 B M3 C M4 D M5 E

0.5

200

400

600 800 Turning point

1000

1200

Crack growth

A

Fig. 5 SEM image of fatigue progression markings on an AA7050-T7451 fatigue fracture surface created with the spectrum shown in the adjacent schematic.

Markers can be produced by varying the R ratio of blocks of loads. It has been found that this can result in slight crack growth plane changes that are visible and measurable. Apart from the example shown in Figure 5, Figure 6 shows another sequence that was used to produce crack growth data for 7050-T7451. While both these examples yielded good data, the data used in the remainder of this paper were generated by a simpler type of loading block that uses only two R ratios. A typical case was 5000 cycles of R=0.1 and 1500 cycles of R=0.7. Alternatively, the more complex loading block shown in Figure 6 was very effective at very short crack lengths. In all cases the maximum stress applied was constant. 1.2 1 0.8 Normalised 0.6 load level 0.4 0.2 0 0

200

400

600 Turning points

800

1000

Crack growth direction

Fig. 6 SEM image of R-induced changes in crack growth planes on the fatigue fracture surface of an AA7050-T7451 specimen tested with four CA bands. These bands had R values of 0.7, 0.3, 0.5 and 0 respectively. The large change from R = 0 to R = 0.7, which occurs twice in the image, was particularly effective at producing a marker.

The Critical Importance of Correctly Characterising Fatigue Crack Growth Rates

257

4 Method 1 and 2 Rate Curve Results for 7050-T7451 Compared with an Earlier Estimate Based on the Load Reduction Method Recall that the principal information required for a cycle-by-cycle crack growth prediction is the baseline, steady-state crack growth rate under CA versus Stress Intensity Factor (SIF) Range relationship. The Method 1 CA results were taken directly from [9]. Testing for Method 2 was conducted on simple, flat hour glass style coupons with shallow side radii to give a stress concentration factor, Kt close to 1, and a test section cross section of 73.3 x 11 mm. To promote cracking from a small initial size at a good representative rate, the coupons were etched producing a high density of small (about 0.01mm) pits in the surface. The largest, and in some cases the second largest, cracks produced were measured and Finite Element Analysis (FEA) was used to generate the beta factors for calculating the K values [19] The cracks were found to generally be semi-circular in shape and so the geometry (beta) factors were about 0.713, i.e. very close to that for a semi-circular surface flaw in a semi-infinite sheet [20]. The curve estimated by Newman [9] based on the Method 1 approach is compared with individual data points determined using Method 2 in Figure 7. The curve obtained and reported in [9] for the load reduction technique is also shown. Some scatter is evident in the data as expected, but the trend is clear. The consistency between the Method 1 and Method 2 approaches is clearly evident. Based on Methods 1 and 2, the threshold for this material is estimated at around 0.6 MPa√m, which is significantly lower than earlier methods would suggest. A threshold of around 1.8 MPa√m is evident from ASTM E647 load reduction test results also shown in Figure 7. There is also a pronounced “knee” in the rate curve such that it flattens out below about 3 MPa√m. These features are clearly apparent in both the Method 1 and 2 results, and are absent in the ASTM E647 load reduction results. Although the Method 2 data are consistent with the Method 1 curve, consideration of the total data set did indicate that the curve could be refined further and this curve is also shown in Figure 7 as a combined curve. Method 1 and combined rate curves were applied to coupon results and the bulkhead example to examine any improvement that the combined curve has over the Method 1 curve.

258

K.F. Walker and S.A. Barter

7050-T7451 Crack Growth Rates 1.E-05

Curve Based on Method 1 Data Symbols are Method 2 Data 1.E-06

da/dN, dc/dN, m/cycle

Curve Based on ASTM Load Reduction Method Curve Based on Combined Method 1 and 2 Data

1.E-07

1.E-08

1.E-09

1.E-10 0.1

1

ΔK or ΔKeff, MPa√m

10

100

Fig. 7 Crack growth rate curve from Method 1 [9] compared with data from Method 2 and from the load reduction technique A crack growth rate curve that is a combination of method 1 and 2 is also included.

5 Variable Amplitude Spectrum Loading Analytical Predictions Using Rate Data Obtained from Methods 1/2 The baseline, steady-state CA rate data obtained using the Method 1 and the combined Method 1/2 results were then used as the principal input in predictions for example cases involving simple coupons and a more complex full-scale aircraft structure, both of which were subjected to VA spectrum loading. The analyses were conducted using FASTRAN [21]. FASTRAN is based on an analytical crack closure model which accounts for load interaction and closure effects. The model was set up as suggested in [9], except that the near threshold region was included in the rate curve (as per Figure 7). The constraint factor, α, was set at 1.3 as per [9]. This is a low value, and as discussed in [9] this is considered to be needed because the FASTRAN model accounts for PICC only, but DICC and RICC are expected in the threshold region, even though as seen in the example shown in Figure 5 their effect is not strong when the crack is small. The low value of α essentially causes the FASTRAN model to introduce more plasticity induced closure than is truly present to account for the lack of explicit modelling of any other forms of closure such as roughness or debris induced closure. Analyses were performed for both simple coupons and the full scale aircraft bulkhead example. Details follow.

The Critical Importance of Correctly Characterising Fatigue Crack Growth Rates

259

Coupon Example Coupon test data from [1] as detailed earlier in Figure 1 were used for the comparison. The coupon geometry was similar to the baseline Method 2 testing described earlier, although the coupons were physically smaller. The coupons were tested with a fighter wing root bending moment spectrum (a block) that was derived from in-service flying and the crack growth was measured by QF on a block-by-block basis after the coupons had failed. Comparisons between the experimental results as per Figure 1 and the FASTRAN predictions using the Method 1 and the combined Method 1/2 rate curves are shown in Figure 8. The updated analyses produced a significantly better correlation with the coupon data than that which was reported in [1], both in terms of shape of the curve and life. The combined rate curve analyses improved the shape of the crack growth curve relative to those using the Method 1 rate curve, particularly in the later stages as the crack depth reached about 1 mm

10

Crack Depth, mm

1

0.1

0.01

0.001 0

5000

10000 15000 Time, Flight Hours

20000

25000

Fig. 8 Representative example detailing improved analysis compared with test data. The experimental data (from [1] and as per Figure 1) are shown as symbols. The analyses using the Method 1 rate curve are shown as solid lines, and the analyses using the combined Method 1/2 rate curve are shown as dashed lines. As for Figure 1 the four sets are for testing under the same sequence but at different stress scaling levels.

260

K.F. Walker and S.A. Barter

Fighter Aircraft Wing Carry-Through Bulkhead Example Another example involving 7050-T7451 material where an earlier FASTRAN based analysis produced poor comparison between experiment and prediction is for a feature in the main carry-though bulkhead of a fighter aircraft as detailed in [22]. The same case has been reported in several other papers [23-25]. The case was originally reported in [26]. The general location is shown in Figure 9. The test used a fighter wing root bending moment spectrum, similar to that applied to the coupons. The loads were applied to the wing attachment lugs of each of the three bulkheads. As was the case for the coupons, the crack growth was measured by QF on a block-by-block basis after the bulkhead failed. The result from the earlier FASTRAN analysis of the location of interest here is reported in [22]. A recreation of that analysis compared with the experimental result is shown in Figure 10. Note, that the original analysis used a rate curve similar to the load reduction method based curve shown in Figure 7. As was the case for the coupon example, the current re-analyses were performed using both the Method 1 and the combined Method 1/2 rate curves (see Figure 7), and the constraint from [9], which resulted in a great improvement as shown in Figure 11. Also as was the case for the coupons, the combined rate curve analyses improved the shape of the crack growth curve, particularly in the later stages as the crack depth reached about 1 mm.

Fig. 9 Fighter aircraft wing carry through structure.

The Critical Importance of Correctly Characterising Fatigue Crack Growth Rates

261

10 Test data Re-creation of Earlier FASTRAN Analysis as per [22]

Crack Depth, mm

1

0.1

0.01

0.001 0

5000

10000

15000

20000

25000

30000

35000

40000

45000

Time, Flight Hours

Fig. 10 Comparison between test result and earlier analysis showing a poor correlation. Reproduced from [22].

10

Crack Depth, mm

1

0.1 Test data Updated Analysis Using Method 1 Rate Data

0.01

Updated Analysis Using Combined Method 1/2 Data

0.001 0

2000

4000

6000

8000

10000

12000

14000

16000

Time, Flight Hours

Fig. 11 Comparison between test data and updated analysis showing greatly improved result.

262

K.F. Walker and S.A. Barter

6 Discussion and Conclusion In investigating examples of poor predictive performance of traditional crack growth models such as FASTRAN, DSTO researchers confirmed the critical importance of the threshold and near threshold region for typical aircraft spectra and stress levels. Two very different and independent methods have been demonstrated to produce consistent crack growth rate data for 7050-T7451 material in this crucially important threshold region. The data thus obtained are considered to avoid some of the shortcomings which other researchers have identified with traditional load reduction methods of data collection. Using these data in a fracture mechanics based crack growth analysis under loading by a complex VA spectrum, representative of a fighter aircraft, has been shown to produce significantly better results than those obtained previously. Further work is needed to identify if the methods also work for other materials. Titanium alloys are of particular interest since they are extensively used in aircraft structures and they also exhibit rough and torturous fracture surfaces. It is also important to continue work developing analytical methods to account for nonplasticity induced closure mechanisms. The results from this work offer an improved potential to maximise the life of aircraft structures while maintaining adequate levels of safety and therefore contribute to a reduction in the capability and environmental impact of premature fatigue related retirement and/or replacement.

Acknowledgements The authors would like to thank Professor J. Newman from Mississippi State University for his assistance and advice and the provision of detailed data, and also Mr M. McDonald from DSTO for data and advice and Dr C. Wallbrink from DSTO for technical review.

References [1] McDonald, M., Molent, L., Green, A.J.: Assessment of Fatigue Crack Growth Prediction Models for F/A-18 Representative Spectra and Material. DSTO-RR-0312 (2006) [2] Jackson, P., Wallbrink, C., Walker, K., Mongru, D., Hu, W.: Exploration of questions regarding modelling of crack growth behaviour under practical combinations of aircraft spectra, stress levels and materials. DSTO-RR, DSTO (2011) [3] Anon: Standard Test Method for Measurement of Fatigue Crack Growth Rates. ASTM E 647-00, USA, ASTM (2000) [4] Newman, J.C.J.: A nonlinear fracture mechanics approach to the growth of small cracks. AGARD CP-328 pp. 6.1–6.27 (1983) [5] Forth, S.C., Newman Jr., J.C., Forman, R.G.: On generating fatigue crack growth thresholds. International Journal of Fatigue 25, 9–15 (2003) [6] Newman, J.C.J., Schneider, J., Daniel, A., McKnight, D.: Compression pre-cracking to generate near threshold fatigue crack growth rates in two aluminium alloys. International Journal of Fatigue 27, 1432–1440 (2005)

The Critical Importance of Correctly Characterising Fatigue Crack Growth Rates

263

[7] Ruschau, J., Newman Jr., J.C.: Compression pre-cracking to generate near threhold fatigue crack growth rates in an aluminum and titanium alloy. Journal ASTM International 5(7) (2008) [8] Ruschau, J., Newman Jr., J.C.: Improved test methods for very low fatigue crack growth rate data. In: American Helicopter Society International 64th Annual Forum and Technology Display, Montreal, Canada (2008) [9] Newman, J.C., Yamada, Y., Newman, J.A.: Crack-closure behaviour of 7050 aluminum alloy near threhold conditions for wide range in load ratios and constant Kmax tests. In: ASTM/ESIS Fatigue and Fracture Mechanics Symposium, Vancouver, Canada (May 2009) [10] Newman, J.C.J.: Analysis of fatigue crack growth and closure near threshold conditions. In: ASTM STP, vol. 1372, pp. 227–251 (2000) [11] McClung, R.C.: Analysis of fatigue crack closure during simulated threshold testing. In: ASTM-STP, vol. 1372, pp. 209–226 (2000) [12] White, P.A., Barter, S.A., Molent, L.: Observations of crack path changes caused by periodic underloads in AA 7050-T7451. International Journal of Fatigue (2007) [13] Newman, J.C., Wu, X.R., Venneri, S.L., Li, C.G.: Small-Crack Effects in HighStrength Aluminium Alloys. NASA Reference Publication 1309, NASA (1994) [14] Pippan, R.: The growth of short cracks under cyclic compression. Fatigue and Fracture of Engineering materials and Structures 9, 319–328 (1987) [15] Pippan, R., Plochl, L., Klanner, F., Stuwe, H.P.: The use of fatige specimens precracked in compression for measuring threshold values and crack growth. ASTM Journal of Testing and Evaluation 22, 98 (1994) [16] Topper, T.H., Au, P.: Fatigue test methodology. AGARD Lecture Series, vol. 118. The Technical University of Denmark, Denmark (1981) [17] Suresh, S., Ritchie, R.O.: Propagation of short fatigue cracks. International Metals Reviews 29(6), 445–476 (1984) [18] Barter, S.A., Wanhill, R.J.H.: Marker Loads for Quantitative Fractography of Fatigue Cracks in Aerospace Alloys. In: 25th ICAF Symposium, Rotterdam, The Netherlands, May 27-29 (2009) [19] Jones, R., Barter, S., Chen, F.: Experimental studies into short crack growth. Engineering Failure Analysis (2011) [20] Tada, H., Paris, P., Irwin, G.: The Stress Analysis of Cracks Handbook. St Louis, Missouri (1985) [21] Newman Jr., J. C.: FASTRAN II - A fatigue crack growth structural analysis program. NASA TM-104159, NASA (1992) [22] Jones, R., Molent, L., Pitt, S.: Crack growth of physically small cracks. International Journal of Fatigue (2007) [23] Jones, R., Molent, L., Pitt, S., Soires, E.: Recent developments in fatigue crack growth modelling. In: European Conference on Fracture (2006) [24] Jones, R., Pitt, S., Peng, D.: The generalised Frost-Dugdale approach to modelling fatigue crack growth. Engineering Failure Analysis (2008) [25] Jones, R., Molent, L., Pitt, S.: Proceedings of International Conference on Fatigue Damage of Structural Materials VI, Hyannis, USA (2006) [26] Molent, L., Dixon, B., Barter, S.: The FINAL program of enhanced teardown for agile aircraft structure. In: 8th NASA/FAA/DOD Conference on Aging Aircraft, Palm Springs CA USA, January 31 - February 3 (2005)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

The Relationships between Crack Closure, Specimen Compliance and ‘Effective’ Fatigue Crack Growth Rate D.L. Ball1, J.K. Donald2, M.A. James3, and R.J. Bucci3 1

Lockheed Martin Aeronautics Co., Fort Worth TX, USA Fracture Technology Associates, Bethlehem PA, USA 3 Alcoa Technical Center, Alcoa Center PA, USA

2

Abstract. The recent utilization of new, large aluminum forgings for critical airframe structural components has prompted the need for improved fatigue crack growth rate (FCGR) characterization methods. This need is most evident for cases in which the FCGR data are confounded by the presence of manufacturing process induced bulk residual stresses. The test method that has been the most successful at removing these confounding effects is the adjusted compliance ratio (ACR) method. In this paper we review two of the relationships that play a pivotal role in the success of the method. The first is the relationship between specimen compliance and fatigue crack closure, and the second is the relationship between closure and the so called ‘effective’ fatigue crack growth rate. We conclude with a brief discussion of the manner in which ACR data may be used in design.

1 Background Recent advances in the design and manufacture of advanced metallic structures have been enabled, in part, by the development of advanced techniques for the characterization of fatigue crack growth rate. In one design approach intrinsic, residual stress free, fatigue crack growth rate (FCGR) data are developed using the adjusted compliance ratio (ACR), Kmax normalization method together with a reconstruction process based on plasticity induced closure. The ACR method was introduced by Donald in the late 1990s and has been extensively studied in the years since. At the coupon and design feature level the ACR methodology has demonstrated ability to account for the effects of crack closure (both plasticity and roughness induced), as well as the effects of residual stress on FCGR. Since the generation of this data is a critical component of the design method, validation of the ACR, Kmax normalization method is of vital importance.

2 Specimen Compliance and Crack Closure The relationship between the elastic compliance of a solid and the size of a crack within that solid was one of the first and most important discoveries in fracture *

Oral presentation.

266

D.L. Ball et al.

mechanics. For materials that exhibit purely elastic behavior, the compliance at a given crack size has a single value, given by the slope of the load vs. displacement diagram. For most metallics, however, the occurrence of plasticity, the formation of asperities, the oxidation of exposed surfaces, etc., tend to complicate load vs. displacement relationships and thus preclude the existence of this single (constant) value of compliance at a given crack size. Elber [1] was the first to demonstrate that the formation of a plastic wake behind the tip of a growing fatigue crack would cause the crack faces to come into contact at applied loads greater than the cycle minimum load. Furthermore, this ‘crack closure’ was clearly evident in a standard load vs. displacement diagram: at the high end of the cycle, the crack would be fully open and the measured compliance would be equal to the theoretical, elastic value for the current crack length. However, at the low end of the cycle, the crack would be partially or fully closed (creating a load path across the crack faces) and the compliance would decrease, ultimately approaching the value for the uncracked specimen. Elber defined the crack opening stress as the applied stress at which the crack became fully open and described the manner in which this stress could be identified on a displacement diagram. From the beginning, however, it was clear that crack opening (and closure) do not occur at a single applied load, but rather that they occur over a load range, as shown in Fig. 1. In the years since Elber’s groundbreaking work, a vast amount of research has been conducted on crack closure; we make no attempt here to give a comprehensive survey. Please refer to [2]. In the context of specimen compliance, however, we can highlight several important findings. First, as stated above, it is well accepted that there are several different mechanisms of fatigue crack closure, ranging from roughness induced to viscous fluid induced, to plasticity induced [3]; the last being by far the most thoroughly studied. (It is important to note that while the many different types of closure manifest themselves as changes in compliance, changes in compliance alone are not sufficient to distinguish between mechanisms.) Second, in the case of plasticity induced crack closure (PICC), it is generally accepted that during loading, the crack opening process begins away from the crack tip and the crack faces ‘peel’ apart as the applied stress approaches the crack opening stress. Closure can occur over any and all parts of a growing fatigue crack, with that occurring in the vicinity of the crack tip being referred to as local closure, and that occurring away from the crack tip as remote closure. This distinction can be important because some of the methods used for measuring closure are insensitive to small scale behavior at the crack tip, i.e. to local closure.

The Relationships between Crack Closure, Specimen Compliance

267

P D

Pmax

crack fully open C

Popen

crack partially closed B crack fully closed Pmin

A δ

Fig. 1 Evidence of Crack Opening and Closure on a Typical Load vs. Displacement Diagram.

Crack closure can be measured in a variety of ways ranging from direct observation to compliance to indirect or inferred observation techniques. Of the compliance techniques, it is the mechanical compliance methods that are of interest to us here. In general load or stress is plotted against displacement or strain, and the inverse slope of the relationship is referred to as the compliance. Clip gages mounted at the open end (or mouth) of the crack measure crack mouth opening displacement (CMOD), while those mounted at or near the crack tip measure crack tip opening displacement (CTOD). Similarly, strain gages can be mounted on or near the crack in the vicinity of the tip, or away from the tip, on the back face. The extent to which any crack closure phenomenon may be observable as a compliance change on a load vs. displacement (or stress vs. strain) diagram depends on the physical scale of the closure region (relative to the body dimensions) as well as on the sensitivity of the measurement devices. For example, numerous researchers have reported, and it is now generally accepted, that bulk compliance techniques (i.e. those using load line or CMO displacements) are not sensitive to local closure phenomena [4]. Crack closure has been the subject of a very large number of model development and numerical simulation studies. Newman was one of the first to conduct detailed finite element analyses (FEA) of the closure phenomenon [5] and has developed a closure-based fatigue crack growth model based on the Dugdale strip yield model [6]. Newman’s closure model is capable of simulating many of the important features of fatigue crack closure and growth, including crack opening / closing levels as well as crack growth rates. Likewise McClung [7, 8] has performed numerous detailed FEA simulations of closure under various conditions. These studies allow direct calculation of the relationships between

268

D.L. Ball et al.

both local (to the crack tip) and remote closure parameters and the bulk response parameters such as load line or CMO displacements. The relationship between closure and compliance has been the subject of increased scrutiny of late, especially since the introduction of the adjusted compliance ratio (ACR) method for the measurement of fatigue crack growth rate (FCGR) [9, 10]. The ACR is defined as

ACR =

Δδ eff − Δδ i Δδ app − Δδ i

(1)

Cs − Ci C o − Ci

(2)

or

ACR =

where Δδeff, Δδapp and Δδi are the actual measured, the closure free and the measured uncracked displacement ranges respectively, and Cs, Co and Ci are secant, closure free and uncracked compliances respectively. The displacements and compliances used are shown schematically in Fig. 2. Donald [9] has argued that the ACR method is capable of removing the effects of closure from FCGR data, and Donald [11] and Lados et al. [12] have shown that the method is also capable of removing the effects of residual stress. In each of these statements, the claim that the ACR method ‘removes the effects’ of closure or residual stress is based on the observation that when FCGR is plotted against an effective stress intensity factor (SIF) range, defined as

ΔK eff = ACR ⋅ ΔK app ,

(3)

the data collapse to a single curve. This implies that a more fundamental property has been revealed. We leave for the following section, the discussion of just what the ‘effective’ SIF range is. For now, we note that the method relies on clearly defined relationships between crack size and compliance and that it includes the transitional region (between Smin and Sopen) where the crack is opening (or closing) and is in a state of partial closure.

The Relationships between Crack Closure, Specimen Compliance

Load, P Pmax

Notch(ai)

Δδi

269

an Δδnc

ΔP

ΔP

Pop

ΔP Δδcl

Pmin P0

Δδi

Δδcl Δδnc

Disp. at the load line, δ

Fig. 2 Load vs. Displacement Records Showing the Critical Parameters in the ACR Method.

3 Crack Closure and Fatigue Crack Growth Rate The study of crack closure is useful insofar as it leads to an understanding of the stress intensity factor range that is effective in causing crack advance during fatigue. In much of the early work on closure, the assumption was made that only that portion of the stress cycle during which the crack was fully open, i.e. S>Sopen, was effective in causing crack growth. This in turn led to a great deal of effort being expended toward the prediction and measurement of the crack opening (or closing) stress. However, many researchers have argued based on mechanics principles, as well as from experimental grounds, that stress intensity factors that occur during the transition from a fully closed to a fully open crack can contribute to the effective SIF range. That is to say, the proper definition of ΔKeff is ΔK eff = K max, eff − K min, eff

(4)

where it is generally accepted that Kmax,eff=Kmax. Kmin,eff is to be defined. As early as 1992, Taylor [13] argued, based on strain energy release rate principles, that the definition of ΔKeff based on the opening (or closing) value of stress or SIF was too simplistic. He pointed out that it failed to account for the energy associated with the curved portion of the load or SIF vs. displacement line and therefore underestimated the true ΔKeff. Chen et al. [14] formalized this argument by developing a model for the closure behavior of an edge crack in a finite width strip into which an elastic wedge could be inserted. By calculating

270

D.L. Ball et al.

stress intensity factors for a range of wedge elastic moduli, they showed that the crack tip SIF can be positive at stresses below Sop and that the correct calculation of ΔKeff requires the calculation of the SIF at the cycle minimum stress, Kmin,eff. Chen went on to show that only when the elastic wedge has infinite modulus (rigid) or infinite width, does the SIF go to zero below Kopen. Thus, for virtually all practical applications, and certainly for all applications in which the wedge is a plastic wake, setting Kmin,eff= Kopen results in an under-estimate of ΔKeff. In [12] Lados et al. found Kmin,eff by calculating strain energy release rate directly from the load vs. displacement diagram, and showed that this parameter is positive valued. Again, this implies that SIF excursions below Kopen are effective. In the current study, the authors used the model of Budiansky and Hutchinson [15] to estimate the SIF at the cycle minimum stress, Kmin,eff, for a range of assumed crack closure lengths. The Budiansky-Hutchinson model is a simplified model for plasticity induced closure based on an ideally-plastic DugdaleBarenblatt strip yield model. Their model can be used to study various aspects of closure behavior for a simple (infinite) planar geometry. In particular, it permits the direct calculation of SIF for partially closed cracks. By assuming various crack closure scenarios, it is possible to shed light on the contribution that transitional closure behaviour makes to the effective SIF range. We first consider the case in which the crack closes completely at K= Kopen; in this case Kmin,eff=0 for K
The Relationships between Crack Closure, Specimen Compliance

271

1

0.8

K/ Kmax

closure free zone

0.6 closure affected zone

0.4

0.2

Kopen/Kmax Kmin,eff/Kmax

0 0

0.2

0.4

R

0.6

0.8

1

Fig. 3 Dependence of Kmin,eff on Crack Closure Length for Budiansky-Hutchinson Closure Model.

4 Example Application The ACR method has been evaluated for a variety of materials / conditions and found to consistently remove crack closure effects from FCGR data [16, 17]. (Note that the ACR method is often used in conjunction with a Kmax normalization procedure that can further collapse FCGR data under certain conditions; see Bray and Donald [16] and Donald and Lados [11]. Since the focus of this study was the relationship between crack closure and specimen compliance, the Kmax normalization step was not addressed.) In the current study, the ACR method was evaluated by taking a series of fatigue crack growth rate measurements at constant R in 7000-series aluminum forging. This material presented significant technical difficulty for conventional FCGR testing; as shown in Figs. 4a and 5a, when the data are plotted in terms of the applied SIF range, significant scatter and R-crossing behavior are apparent. This behavior was attributed to the presence of residual stresses that were introduced by the forging process and remained in the test specimens after machining (both C(T) and M(T) specimens were tested). As shown in Figs. 4b and 5b, the data collapse considerably when plotted in terms of the effective SIF range as determined by the ACR method. This collapse is interpreted as an indication that the effects of both closure and residual stress have been removed.

272

D.L. Ball et al.

Another indicator of the utility, if not the validity of the ACR method, is the extent to which the correlation between calculated and measured fatigue crack growth lives can be improved with the use of ACR generated FCGR data. As indicated above, this data often used to support the design and manufacture of advanced metallic structures. For standard damage tolerant design, fatigue crack growth life at any given detail is calculated for a family of design service usage spectra, and design allowable stresses are determined accordingly. This requires the ability to analyze variable amplitude loading and load interaction effects, both of which require modeling stress ratio dependence. Thus, In order to use the ACR generated da/dN vs ΔKeff data in a fatigue crack growth analysis, the analysis must either be modified to calculate ΔKeff for standard geometries, materials and loading profiles, or a method must be developed to re-construct R-dependent FCGR tables for use in an unmodified analysis. To date, only the latter approach has been investigated. The FCGR data re-construction process used here and described in [18] involves the re-introduction of stress ratio effects due to crack closure. While it is recognized that a variety of closure mechanisms are typically at play during fatigue crack growth, only plasticity induced closure has been formally addressed in the reconstructions performed to date. This is easily done using the expressions given by Newman for the relationship between closure factor and applied stress ratio [19]. Efforts are now underway to develop the methods / tools necessary to address both roughness induced crack closure (RICC) and oxide film induced closure. In the mean time, after the re-introduction of plasticity induced closure, the combined (net) effect of the other closure mechanisms is determined empirically by comparing FCG analysis results for constant amplitude loading with data taken using residual stress free coupons. The FCGR mean curves are adjusted (calibrated) as required in order to achieve agreement between analysis and test. The amount of calibration required is a measure of the significance of the other closure mechanisms. The end result is a family of da/dN curves, each at constant R, suitable for design analysis / FCG life prediction for the material in question. Examples of improved correlation with spectrum test data are given in [18].

The Relationships between Crack Closure, Specimen Compliance

273

1.E-01 R=0.5, L-T, C(T) R=0.05, L-T, C(T) R=0.05, L-T, M(T)

1.E-02

R= -0.5, L-T, M(T)

da/dN (mm/cycle)

1.E-03

1.E-04

1.E-05

1.E-06 7000-series Aluminum 5-10 cm Forging, L-T, HHA 1.E-07 1

10 ΔKapp (MPa√m)

100

(a) da/dN vs. ΔKapplied 1.E-01 35-02-35, R=.5, L-T, C(T) 35-02-05, R=.05, L-T, C(T) 35-02-07, R=.05, L-T, M(T)

1.E-02

35-02-13, R= -.5, L-T, M(T)

da/dN (mm/cycle)

1.E-03

1.E-04

1.E-05

1.E-06 7000-series Aluminum 5-10 cm Forging, L-T, HHA 1.E-07 1

10 ΔKeff, ACR (MPa√m)

100

(b) da/dN vs. ΔKeff,ACR Fig. 4 Comparison of da/dN vs. ΔKapplied and da/dN vs. ΔKeff,ACR data for Aluminum Die Forging, 5-10 cm, L-T, HHA.

274

D.L. Ball et al. 35-02-03, R=.5, T-L, C(T) 35-02-04, R=.5, T-L, C(T) 35-02-09, R=.05, T-L, C(T) 35-02-10, R=.05, T-L, C(T) 35-02-2TC, R=.05, T-L, C(T) 35-02-11, R=.05, T-L, M(T) 35-02-12,, R=.05, T-L, M(T) 35-02-15, R= -.5, T-L, M(T) 35-02-16, R= -.5, T-L, M(T)

1.E-01

1.E-02

da/dN (mm/cycle)

1.E-03

1.E-04

1.E-05

1.E-06 7000-series Aluminum 5-10 cm Forging, T-L, HHA 1.E-07 1

10 ΔKapp (MPa√m)

100

(a) da/dN vs. ΔKapplied 1.E-01

35-02-09, R=.05, T-L, C(T) 35-02-10, R=.05, T-L, C(T) 35-02-11, R=.05, T-L, M(T)

1.E-02

35-02-12,, R=.05, T-L, M(T) 35-02-15, R= -.5, T-L, M(T) 35-02-16, R= -.5, T-L, M(T)

da/dN (mm/cycle)

1.E-03

1.E-04

1.E-05

1.E-06 7000-series Aluminum 5-10 cm Forging, T-L, HHA 1.E-07 1

10 ΔKacr (MPa√m)

100

(b) da/dN vs. ΔKeff,ACR Fig. 5 Comparison of da/dN vs. ΔKapplied and da/dN vs. ΔKeff,ACR data for Aluminum Die Forging, 5-10 cm, T-L, HHA.

The Relationships between Crack Closure, Specimen Compliance

275

5 Conclusions In this study we have examined the relationship between crack closure of the type generated by the build-up of a plastic wake behind a growing fatigue crack, and the compliance of a specimen containing such a crack, as measured on a standard load vs. displacement diagram. It is apparent that closure can be detected using compliance methods, though the fidelity of this measurement depends on the crack size and the closure configuration. Compliance methods are most reliable for long cracks and remote closure. Next, an assessment of the relationship between crack closure and ‘effective’ fatigue crack growth rate was made. Here we find evidence, both from the literature and from analytical simulations of the effects of plasticity induced closure using the Budiansky-Hutchinson closure model, that a crack driving force (positive strain energy release rate or SIF) exists below the crack opening stress. Thus leading to the conclusion that estimates of the effective SIF range based on the opening stress alone may under-estimate the true effective SIF range. Finally, the adjusted compliance ratio method for measuring fatigue crack growth rate was evaluated. This method relies on both items 1 and 2 above: specifically it relies on the compliance – crack closure relationship, and it includes the SIF contribution below the opening stress in the measured effective SIF range. The ACR method demonstrated the ability to collapse experimental data with significant scatter into a (virtually) single, intrinsic, da/dN vs ΔKeff curve free of the scatter and bias linked to residual stress and related sampling effects.

References [1] Elber, W.: Damage Tolerance in Aircraft Strucutres. In: ASTM STP, vol. 486, pp. 230–242. American Society for Testing and Materials (1971) [2] Schijve, J.: In: Mechanics of Fatigue Crack Closure, ASTM STP 982, American Society for Testing and Materials, pp. 5–34 (1988) [3] Suresh, S., Ritchie, R.O.: Fatigue Crack Growth Threshold Concepts, pp. 227–261. The Metallurgical Society (1984) [4] Newman Jr., J.C.: Advances in Fatigue Crack Closure Measurement and Analysis. In: 2nd ASTM STP, vol. 1343, pp. 128–144. American Society for Testing and Materials (1999) [5] Newman Jr., J.C.: In: Mechanics of Crack Growth, ASTM STP 590, American Society for Testing and Materials, pp. 281–301 (1976) [6] Newman Jr., J.C.: In: Methods and Models for Predicting Fatigue Crack Growth Under Random Loading, ASTM STP 748, American Society for Testing and Materials, pp. 53–84 (1981) [7] McClung, R.C., Sehitoglu, H.: Eng. Fract. Mech. 33, 237–252 (1989) [8] McClung, R.C., Davidson, D.L.: Advances in Fatigue Crack Closure Measurement and Analysis. In: 2nd ASTM STP, vol. 1343, pp. 106–127. American Society for Testing and Materials (1999) [9] Donald, J.K.: Int. J. Fatigue 19 (supp.1), S191–S195 (1997)

276

D.L. Ball et al.

[10] Donald, J.K., Bray, G.H., Bush, R.W.: In: 29th National Symposium on Fatigue and Fracture Mechanics, ASTM STP 1332, American Society for Testing and Materials, pp. 674–695 (1999) [11] Donald, J.K., Lados, D.A.: Fatigue Fract. Engr. Mater. Struct. 30, 223–230 (2006) [12] Lados, D., Apelian, D., Donald, J.K.: Int. J. Fatigue 29, 687–694 (2007) [13] Taylor, D.: Eng. Fract. Mech. 43(1), 109–115 (1992) [14] Chen, D., Weiss, B., Stickler, R.: Eng. Fract. Mech. 53(4), 493–509 (1996) [15] Budiansky, B., Hutchinson, J.W.: J. App. Mech. 45, 267–276 (1978) [16] Bray, G.H., Donald, J.K.: In: Advances in Fatigue Crack Closure Measurement and Analysis, 2nd Vol., ASTM STP 1343, American Society for Testing and Materials, pp. 57–78 (1999) [17] Donald, J.K., Phillips, E.P. In: Advances in Fatigue Crack Closure Measurement and Analysis, 2nd Vol., ASTM STP 1343, American Society for Testing and Materials, pp. 79–93(1999) [18] Ball, D.L.: In: Fatigue and Fracture Mechanics, 36th Vol., ASTM STP 1508, American Society for Testing and Materials, pp. 216–239 (2008) [19] Newman Jr., J.C.: Int. J. Fracture 24, R131–R135 (1984)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Experimental and Numerical Study of Stress and Strain Field around the Rivet W. Wronicz, J. Kaniowski, B. Korzeniewski, and E. Gadalinska Institute of Aviation, Warsaw, Poland

Abstract. The paper deals with determination of stress and strain field induced during riveting and FEM modelling of this process. Stress and strain measurements during riveting allow for better understanding the phenomena that occur during this process and to determining stress and strain field which exists in the joint after riveting. The results will be also used for validation of FEM models. Stress and strain measurements were carried out with the X-ray diffractometer and strain gauges on the sheet surface near the driven head. The investigation concerned sheets and rivets used in the Polish aerospace industry. This paper presents measurements results for two types of rivets; the brazier rivet (BN70/1121-06) and the rivet with a compensator (OST 1 34040-79 1), as well as the FEM analyses results. Bare sheets made from 2024 T3 aluminium alloy with the nominal thickness of 1,27 mm and rivets with the diameter of 3 mm and 3,5 mm made from Polish aluminium alloy PA25 were used. The influence of squeezing force as well as the rivet type on stress and the strain system was investigated. Two types of strain gauges were used. The strip miniature gauges with the gauge length of 0,51 mm were located outside driven head and worked during the whole riveting process. The micro strain gauges with gauge length of 0,38 mm were applied very close to the rivet hole, in the area which is under the driven head after the the riveting process. The gauges recorded strain to the point when they were destroyed by the driven head. The measurements results were compared with the FEM results.

1 Introduction In the aluminium aircraft structures, riveting process induces high stress in sheets. Typically, it exceeds the yield point near rivet hole [1]. The total stress value in sheets is crucial for fatigue properties since fatigue cracks initiate in this area. An investigation of the stress system after riveting as well as the influence of the process parameters, the rivet type and joint geometry on the stress system is vital for improving fatigue performance of riveted joints as well as accuracy of their fatigue life estimation. The effect of squeezing force on joint fatigue life has been studied by many researchers, e.g. Müller [1], B.Langrand [2] and Lie [3]. Residual stresses resulting from rivet upsetting are compression radial stresses and, for some range *

Oral presentation.

278

W. Wronicz et al.

of squeezing force, compression tangential stresses in the hole vicinity [4]. The presence of compression stresses prevents crack nucleation. After riveting, radial expansion near the manufactured head is lower than near the driven head, which is unfavourable from the fatigue point of view due to lower residual stresses in this area. Rivets with compensators were developed to improve this situation. Compared to normal rivets, rivets with compensators provide better hole filling and radial expansion in the sheet at the manufactured head. This results in higher residual stresses and longer fatigue life. An investigation of fatigue life of specimens with rivets with compensators was carried out by Simenz [5] and Klima [6] among others. The presented works were carried out under the IMPERJA project (Eureka initiative, the project E3496!). The goal of the IMPERJA project is to increase the fatigue life of riveted joints as well as to improve fatigue life estimation methods for such joints. The project included experimental and numerical analyses of stress and strain system around the rivet as well as the impact analysis of joint geometry and squeezing force. The investigation concerned sheets and rivets used in the Polish aerospace industry. This paper presents measurements results for two types of rivets; the brazier rivet (BN-70/1121-06) and the rivet with a compensator (OST 1 34040-79) as well as the FEM analyses. MSC MARC software was used.

2 X-RAY Difractometry Measurements WP6.1 series specimens (Fig. 1) were designed for investigation of the squeezing force influence. There are four fields on the specimen. Rivets inside fields were installed with force control. The force values were attuned to obtain D/Do parameter (D-driven head diameter, Do-rivet shank diameter) equal to 1,2; 1,4; 1,5 and 1,55 respectively in the fields 1 to 4.

Fig. 1 WP6.1.8.1 specimen with numbers of rivets and measurements paths.

The WP6.1. specimens consist of two bare sheets made of 2024-T3 alloy, with nominal thickness of 1,27 mm. and the rivets according to the Polish standard BN-70/1121-06 (WP6.1.8.1) and the Russian standard OS 1 34040-79 (WP6.1.9.1). The rivet diameter is 3 mm, length 6 mm. Rivets are made of polish aluminium alloy PA25. The distance between the rivet axes equals to 5 rivet diameters (15 mm).

Experimental and Numerical Study of Stress and Strain Field around the Rivet

279

2.1 X-ray diffraction measurement In each field one rivet was selected for radial and tangential stress measurements on the sheet surface, near the driven head. Measurements were carried out on paths parallel and perpendicular to the rolling direction (the longest edge of the specimen). The specimen with marked measurements paths was presented in Fig. 1. Stress measurements were performed using the X-ray diffractometer XSTRESS3000, produced by Stresstech Oy. Following measurements parameters were used: pick shift: cross correlation, 2Θ/hkl: 156.7°/311, method: psi, Ф oscillation: 10° / 5, Ψ values: -39°/+39°, Ψ oscillation, angle and number: ±6° ; 5/5, collimator diameter: 0,8 mm. 2.2 Results Results of the measurements on the paths crosswise to the rolling direction were collected in Table 1. The graphs for the specimen WP6.1.8.1 are in the left column, for the specimen WP6.1,9.1 in the right column. The name of series indicates the rivet number. The graphs of radial stress correlate with literature. In the tangential stress graphs, disturbances are present. The influence of squeezing force can be seen in the stress plots for both types. The curves in the graphs are not smooth in many cases. Probably this is a result of residual stresses, which were present in the sheets before the riveting process (after rolling, drilling etc.). Moreover, the measurements points nearest to the driven head are in the area where stress has exceeded the yield point during riveting. The method of stress determination used in the presented measurements concerns the elastic range only. Besides, in this area, the driven head overshadowing occurs, which has limited measurement capability. For these reasons the measurements results in the immediate vicinity of the driven head are less reliable.

280

W. Wronicz et al. Table 1 XRD measurements results.

brazier rivet

rivet with compensator SpecimenWP6 .1.9 .1,radialstress

SpecimenWP6 .1.8 .1,radialstress 1,0

1,5

2,0

2,5

3,0

3,5

4,0

4,5

R/r

5,0

1,0

5,5

0

2,0

2,5

3,0

3,5

4,0

R/r 4,5

5,0

5,5

-100

-100

-200

Sr[MPa]

-200

Sr[MPa]

1,5

0

03_w

-300

06_w -400

13_w

-300

04_w 08_w

-400

14_w

18_w

-500

17_w -500

-600

-600

-700

-700

Radial stress, perpendicular to rolling direction SpecimenWP6 .1.8 .1,tangentialstress 1,0

1,5

2,0

2,5

3,0

3,5

4,0

4,5

R/r

5,0

250

5,5

1,0

1,5

SpecimenWP6 .1.9 .1,tangentialstress 2,0

2,5

3,0

3,5

4,5

R/r 5,0

5,5

200

200

04_w

03_w 06_w

150

150

13_w 100

18_w

50

St[MPa]

St[MPa]

4,0

250

100

08_w 14_w 17_w

50

0

0 -50

-50 -100

-100

Tangential stress, perpendicular to rolling direction 2.3 FEM calculation The axisymmetric FEM models (Fig. 2) were prepared for analyses of the stress system around the rivets. The models refer to the WP6.1 series specimens. The dimension of the models was restricted to 5 rivet radius. Each of the model consists of five contact bodies; three deformable (two sheets and a rivet) and two rigid, (hold on and press punch) and about 2000 linear elements and nodes. In the model of the rivet with a compensator, local adaptivity function (local splitting of elements) was activated near the compensator due to high strain in this area. The materials models, developed based on the monotonic test of specimens cut from sheet (2024-T3) and rivets (PA25) were used (Fig. 3). The tests were carried out by professor Malgorzata Skorupa’s team at the AGH University of Science and Technology in Cracow. The stick-slip coulomb model of friction was selected. For sheets and sheet-rivet contact pair, dynamic friction coefficient equal to 0,34 was assumed [7]. For the contact between the rivet and the rigid bodies, the friction coefficient value was 0,15. Static friction coefficients were 24% higher.

Experimental and Numerical Study of Stress and Strain Field around the Rivet

281

Fig. 2 FEM models. M ate rial mode ls 700 600

Sigma [MPa]

500

2024-T3-iter PA25-eve

400 300 200 100 0 0

0,2

0,4

0,6

0,8

1

1,2

1,4

Epsilon_pl [\]

Fig. 3 Material models.

Boundary conditions. Boundary conditions are presented in Fig. 2. During the whole analysis the hold on did not move. The press punch was able to move only in the axis direction (x) and was controlled by force equal to 10,4 kN (brazier rivet) and 12 kN (rivet with compensator). Radial displacements were restricted for the nodes of the sheets on the outer edge of the model (reaction of not modelled part of the sheets). Riveting on the press is done with the riveting set Fig. 6b). It consists of the punch and the clamping sleeve. During the riveting process, before the punch touches the rivet shank, the plates are pushed together by the sleeve, which is coupled with the punch by a spring. To take this into account during the FEM analysis, the forces equal to the one acting on the sleeve were applied to the nodes belonging to the surface on the upper (inner) sheet (Fig. 2). Results. Results of the FEM calculations for both types of the rivet were collected in Table 2. The figures present deformations of the models and stresses; equivalent (Huber-Mises-Hencky, HMH), radial and tangential. Deformations of the joints during the riveting process are correct. Non-uniform stress distribution along the sheets thickness is visible. For the brazier rivet, the highest stress values are near the driven head while the smallest values on the manufactured head side. For the rivet with a compensator, stress distribution is definitely more uniform and

282

W. Wronicz et al.

stress values are higher. The range of plastic zone is also significantly bigger. Near the rivet hole, compression stresses (radial and tangential) occur, which is beneficial from the fatigue point of view as it limits crack nucleation. Table 2 FEM results.

brazier rivet

rivet with compensator

Equivalent stress, gray colour-elastic zone

Radial stress

Tangential stress

Comparison with the measurements. The numerical calculations were compared with the XRD measurements. In the graphs, the results were compared directly. This way of comparing the FEM results with the neutron diffraction measurements was employed in [8]. It should be noted that the X-ray (neutron) diffractometer measures total stress. Total stress comprises stress induced during

Experimental and Numerical Study of Stress and Strain Field around the Rivet

283

the riveting process but also residual stress after rolling, drilling etc., which makes comparison more difficult. The authors are going to measure the residual stress that existed in the sheets before riveting and develop a method of using the measurements results in the FEM models validation and verification. Quite good correlation of the numerical results with the measurements was obtained regarding both the course and the values. In the case of radial stress around the brazier rivet, the values (absolute) from the FEM calculation are higher than measured. The biggest difference exists near the rivet hole, in the plastic zone where measurement values are less reliable. In the graphs, the equivalent stresses were presented and the plastic zone radius was determined. The graphs were shown in Fig. 4.

SpecimenWP6 .1.8 .1 Brazierrivet 600 500 400 300

St[MPa]

200

1,5

2,0

2,5

3,0

3,5

4,0

SpecimenWP6 .1.9.1 Rivetwithcompensator

R/r 4,5

5,0

1,0

5,5

FEM_S_eq Re

500

r_Re

400

FEM_St

2,0

2,5

3,0

3,5

4,0

4,5

R/r 5,0

5,5

FEM_S_eq Re r_Re 14_w_r FEM_Sr

13_w_t

300

13_w_r

200

FEM_Sr

1,5

600

St[MPa]

1,0

14_w_t FEM_St

100

100 0

0

-100

-100

-200

-200

-300

-300

-400

-400

-500

-500

Fig. 4 Comparison of FEM calculation and XRD measurements.

3 Strain Gauge Measurements Literature and the FEM calculations indicate the existence of the high stress and strain gradient near the rivet hole. The XRD measurements covered the region on the boundary or outside the high gradient stress area. To measure strain in the area of the high stress and strain gradient, the WP6.2 series specimens were designed and manufactured. Strain progress was recorded during the riveting process. Geometry of specimens was presented in fig. 5. Strains were measured on the sheet surface, near the driven head. The specimens consist of two bare sheets made from 2024 T3 aluminium alloy with the nominal thickness of 1,27 mm and three rivets with the diameter of 3,5 mm made from Polish aluminium alloy PA25. The rivets types used in the WP6.1 specimens for the XRD measurements were selected; BN-70/1121-06 (specimen WP6.2.8.1) and OST 1 34040-79 (WP6.2.9.1).

284

W. Wronicz et al.

3.1 Experiment The outer rivets were installed, afterwards the strain gauges were applied and the central rivet was installed. Strain progress was recorded. The strain gauges should have very small gauge length and should be placed as close to the rivet hole as possible due to the high strain gradient occurrence. Two types of gauges produced by Vishay were used; the strip and micro gauge. The strip miniature gauges contain ten gauges, each with gauge length of 0,51 mm, and were located outside the driven head. They worked during the whole riveting process. The 020MT gauges measured radial strains while the 020PF gauges measured tangential strains. Besides those, two micro strain gauges with gauge length of 0,38 mm were applied very close to the rivet hole in the area which is under the driven head after the riveting process. The gauges recorded strain until they were destroyed by the driven head. The 015CK gauge measured radial strain while the 015SE gauge measured tangential strain. The gauges placement and their numbering were presented in fig. 5. Radial positions of the gauge centre on the specimen WP6.2.8.1 were following: 015CK - 1,90 mm, 0,15 SE - 1,97 mm, 020MT – 3,67. 020PF – 3,74 mm. On the specimen WP6.2.9.1 these positions was following: 015CK - 2,15 mm, 0,15SE - 1,95 mm, 020MT – 3,58, 020PF – 3,91 mm. In case of the strip gauges only the positions of first sections were given. The distance between sections is 0,89 mm.

Fig. 5 WP6.2 specimens geometry.

Experimental and Numerical Study of Stress and Strain Field around the Rivet

a)

285

b) Fig. 6 Riveting set, a) standard, b) designed for WP6.2 specimens.

Riveting of the central rivets was performed in the Institute of Aviation with the testing machine. When the standard riveting set (fig. 6a) is used the plates are pushed together by the sleeve. In order to avoid damage of the gauges by the riveting set, a special device was designed and manufactured (fig. 6b). The sleeve was replaced with a part with four legs. The diameter of this part is much larger than that of the standard sleeve, so the specimen would be bending. To avoid this, a support in the form of a ring with the inner radius of 80 mm and width of 10 mm was introduced. 3.2 FEM calculations The riveting processes of the WP6.2 specimens were analyzed with Finite Element Method. The axisymmetric (Axi) models were developed. The models described in point 2.3 were scaled and corrected to obtain the geometry of joints with 3,5 mm rivets. Afterwards the models were extended so that they covered the region of the radius 50 mm. Only the presence of the central rivet was taken into account (outer rivets were neglected). Special boundary conditions resulting from the use of the special riveting set were modelled. The models consisted of about 2600 linear elements and 2900 nodes. Fig. 7 presents the model of the specimen with the brazier rivet. Forces were applied to the nodes on sheet surface to model clamping of the sheets by four-legged part (fig. 6b). The support was modelled as a motionless rigid contact body (curve). At the begging of the analysis there was a 0,5 mm slit between the support and sheet surface. The squeezing force value was 15,35 kN.

286

W. Wronicz et al.

Fig. 7 WP6.2.8.1 specimen FEM model.

3.3 Results The results of the numerical calculations were compared with the strain measurements. The graphs in Table 3 show plots of strain progress during the riveting process obtained for the gauges and nodes of the FEM models which are placed closest to the the gauge centre. The Series name indicates the type of results (Gauge, Axi-FEM axisymmetric model) and distance between the node and gauge centre. Only results for gauges no 1 and 11 were presented The „S” shaped curve can be seen in the radial strain graph for the rivet with compensator. Similar shapes of the strain plots were presented in the paper [3]. This phenomenon has not been explained so far. Table 3 Experimental and numerical strain Progress during riveting process.

brazier rivet

rivet with compensator Radial strain, Gauge 11 2

radial straine [um/m]

Radial strain, Gauge 11

0

2

radial strain [um /m ]

-8000

-7000

-6000

-5000

-4000

-3000

-2000

-1000

0

1000

-2

0 -7000

-6000

-5000

-4000

-3000

-2000

-1000

0

1000

-4

-2

-6

-4 -6

-10

-8

Force [kN]

-8

-10 -12

-12 -14

Axi, -0,04

Axi, -0,04

-14

Gauge_11

-16

Axi, 0,31

-18

-16 -18

Tangential strain, gauge 1 2

Force [kN]

-8000

Gauge_11 Axi, 0,31

Tangential strain, gauge 1

tangential strain [um/m]

2 tangential strain [um/m]

0 -1000

0

1000

2000

3000

4000

5000

6000

0

7000

0

-2

1000

-2 -4

-4

Force [kN]

Force [kN]

-6 -8 -10

-6 -8 -10

-12 -14 -16

Axi, -0,14 Gauge_1 Axi, 0,21

-12 -14

Axi, -0,14 Tens_O1

-16

Axi, 0,21

-18

-18

2000

3000

4000

5000

6000

7000

8000

Experimental and Numerical Study of Stress and Strain Field around the Rivet

287

Good correlation with the experiment was obtained for the model of the specimen with the brazier rivet. In the case of the rivet with a compensator, good correlation with the experiment was obtained for radial stress. For tangential stress, the calculation results are considerably lower than the measurements results. This model needs to be refined. The graphs of strain as a function of radial position were present in Fig. 8. The results after the riveting process were presented. The name of each series indicates the type of the results (Gauge-measurement, FEM-calculation) and the type of stress (Sr - radial, St - tangential). Because of malfunction of one testing card, the strains of the gauges no 3 were not recorded. Very good correlation with the experiment was obtained for both models.

WP6.2.9.1 Rivet with compensator

WP6.2.8.1 Brazier rivet 35 000 40 000

25 000 20 000

5 000

0

-5 000 1

2

3

4

5

-15 000

6

7

8

FEM_BN_St

-25 000

Gauge_BN_St

9

e_t [um/mm]

e_o [um/mm]

15 000

1

2

3

4

5

6

7

8

9

-20 000

FEM_OST_St -40 000

Gauge_OST_St Gauge_OST_Sr

-35 000

Gauge_BN_Sr

-45 000

FEM_BN_Sr

-55 000

r [mm]

-60 000

FEM_OST_Sr

-80 000 r [mm]

Fig. 8 Experimental and numerical strains as a function of radial position.

Deformations of the models were correct. The pictures of stresses obtained in these calculations were similar to the results of the calculation described in point 2.3.

4 Summary The results of the strain and stress measurements around two types of the rivets near the driven head were compared with the FEM calculations. Higher stress and strains after the riveting process were found for higher squeezing force and for the rivets with compensators. Residual stresses existing in the sheets before the riveting process influence the XRD measurements. The measurement of these stresses and the method of using these results in the FEM model validation and verification will be the subject of further works. Also, the methodology of the XRD measurements around the rivets will be developed. Literature and the FEM calculations indicate the presence of the high stress and strain gradient near the rivet hole. The XRD measurements covered the region on the boundary or outside the high gradient stress area. This was the reason for the strain gauge measurements. The high strain value near the rivet hole was proved. The presence of compressive tangential strains in this region was not experimentally confirmed. The „S” shaped plot of radial strain for the rivet with a

288

W. Wronicz et al.

compensator was recorded. Similar shape of strain plots were presented in the paper [3]. This phenomenon has not been explained so far. Good correlation of FEM calculation with the XRD and strain measurements was obtained. Numerical results indicate positive influence of rivet with compensator on stress system after riveting in sheets. Compressive stresses are present also in sheet on manufactured head (more uniform stress distribution through thickness). Area of compressive stress presence is larger in comparison with the brazier rivet. Significantly larger is also the range of plastic zone. The benefits of using the rivets with compensators were confirmed by the fatigue test results [5], [6]. The authors are going to use FEM models to determine the joint geometry and the riveting process parameters which are optimal from the fatigue point of view.

References [1] Muller, R.: An Experimental and Analytical Investigation on the Fatigue Behaviour of Fuselage Riveted Lap Joint, TU Delft, p. 58 (1995) [2] Langrand, B., Patronelli, L., Deletombe, E., Markiewicz, E., Drazétic, P.: An alternative numerical approach for full scale characterization for riveted joint design. Aerospace Science and Technology 6, 343–354 (2002) [3] Li, G., Shi, G., Bellinger, N.C.: Studies of Residual Stress in Single-Row Countersunk Riveted Lap Joints. Journal of Aircraft 43(3) (2006) [4] Muller, R.: An Experimental and Analytical Investigation on the Fatigue Behaviour of Fuselage Riveted Lap Joint, TU Delft, p. 100 (1995) [5] Simenz, R.F., Steinberg, M.: Alloy Needs and Design: the Airframe, In Fundamental Aspects of Structural Alloy Design. In: Jaffee, R.I., Wilcox, B. (eds.) Proceeding of the 10th Battelle Colloquium in the Materials Science. Plenum Press, New York (1977) [6] Klima, Z.: Urceni velikosti nekterych faktoru majicich vliv na unavu nytoveho spojeni, VZLU report no V-1399/80 (1980) [7] Blau, P.: Friction, Lubrication, and Wear Technology ASM Handbook vol. 18 (1995) [8] Rans, C.D.: The Role of Rivet Installation on the Fatigue Performance of Riveted Lap Joints. PhD thesis, Department of Mechanical and Aerospace Engineering Carleton University Ottawa, Ontario, Canada, pp. 51–52 (2007)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Fatigue Analyses of Riveted Lap-Splice Joints in a Narrow-Body Aircraft J.C. Newman Jr.1 and R. Ramakrishnan2 1

Mississippi State University, Mississippi State, MS, USA 2 Delta Air Lines Inc., Atlanta, GA, USA

Abstract. The U.S. Federal Aviation Administration and Delta Air Lines had teamed to conduct a destructive evaluation of a retired narrow-body passenger aircraft that had nearly 60,000 flights. Some of the program objectives were to characterize the state of damage at riveted lap-joint fastener holes in the fuselage of an aircraft at the design service goal; and to develop or verify analysis methods that can correlate and predict the state of cracking at any point in time. The crack-growth model, FASTRAN, was used to evaluate Elber's effective stress-intensity-factor range in terms of crack-tip cyclic hysteresis energies and was found to correctly partition energies associated with crack-tip damage. The model was used to correlate constant-amplitude fatigue-crack-growth-rate data over a wide range in rates and stress ratios (minimum to maximum applied stress) from threshold to near fracture conditions. The model was then used to calculate fatigue lives using small-crack theory and/or crack growth in open-hole laboratory specimens, fastener-loaded holes in curved test panels cut from the aircraft fuselage, and fastener-loaded holes in acutal fuselage lap joints from the retired aircraft made of thin-sheet 2024-T3 aluminum alloy. Equivalent-initial-flaw sizes (EIFS) were established for the laboratory specimens and for the fuselage lap joints that had been subjected to actual service loads and environments. Calculated fatigue lives for the laboratory specimens agree well with test data; and the calculated crack length against flight pressure cycles for the narrow-body aircraft fuselage were quite similar to the results found from fractographic examinations. In this paper, the terms fatigue life and crack-growth life are used synonymously since all fatigue-life calculations were performed using the FASTRAN crack-growth methodology with the EIFS concept and formulations based on small-crack theory.

1 Introduction The Federal Aviation Administration (FAA) and Delta Air Lines [1-6] had teamed to conduct a destructive evaluation of a retired narrow-body passenger aircraft that had nearly 60,000 flights. Some of the objectives of the program were to characterize the state of damage at riveted fastener holes in the fuselage of an aircraft that had achieved its design service goal life; and to develop or verify *

Oral presentation.

290

J.C. Newman Jr. and R. Ramakrishnan

analysis methods that can correlate and predict the state of cracking at any point in time. For the narrow-body aircraft, observations from the destructive examination of the fuselage joints indicated that the left side of the aircraft along stringer 4L appeared to have tight (within specifications) rivets with no detectable cracks present, whereas on the other side of the aircraft, along stringer 4R, the rivets appear to be under driven with a large number of cracks present [5, 6]. For these cracks, faying surface origins were predominant, despite a majority of rivets being under driven. A large number of cracks were examined with a scanning-electronmicroscope (SEM) to count striations and to back-track the cracking history to reconstruct the crack-length-against-cycle behavior [4]. The measured cracklength-against-flight-cycle results tended to fall within a fairly narrow band considering the complexity of the structural joints. Several curved fuselage panels were cut from the other side of the fuselage and two panels underwent fatigue testing in a pressure-box facility that simulated fuselage service loading [3, 7]. The objective of this paper is to use FASTRAN [8] and small-crack theory [9] to calculate fatigue and crack-growth lives for laboratory specimens and structural joints made of 2024-T3 clad aluminum alloy and to compare the calculated results with the test and service findings. Life analyses were made using the crack-closure concept and Elber's effective stress-intenstiy-factor range [10]. An evaluation was also made between the effective stress-intensity-factor range and the cyclic hysteresis energies in the crack-tip region. Three types of configurations were analyzed: (1) fatigue of open-hole laboratory specimens, (2) fatigue and crack growth in the lap joints present in curved fuselage test panels, and (3) fatigue of actual fuselage lap joints in the retired aircraft. For these configurations, equivalent initial flaw sizes (EIFS) were established to fit the fatigue and/or crackgrowth results. The curved fuselage test panels cut from the aircraft were tested in the FASTER Test Facility at the FAA William J. Hughes Technical Center [7]. Calculated fatigue lives and crack-growth behavior were compared with test results. Comparisons were also made between FASTRAN [8] and AFGROW [11] for crack growth in the narrow-body aircraft and the reconstructed crack-lengthagainst-flight-pressure-cycle history.

2 Stress Analyses of Cracks at Open Holes or Aircraft Lap-Joint Rivet-Loaded Fastener Holes Laboratory specimens In open-hole laboratory specimens, the stress-concentration factor is slightly higher in the center of the hole than at the edge. Thus, for polished and deburred specimens, a surface crack located at the center of the hole was assumed as the inititation site. This location has been confirmed by test data in the literature. The stress-intensity factors for a surface crack at a hole in a plate under a remote tension (St) stress are given in Ref. 12. Here the initial semi-circular surface crack was allowed to grow independently in the thickness (a) and width (c) directions.

Fatigue Analyses of Riveted Lap-Splice Joints in a Narrow-Body Aircraft

291

Aircraft lap joints with loose rivets As mentioned, one side of the retired aircraft appeared to have under-driven rivets along stringer 4R (within specification) with a large number of cracks present. For these cracks, faying surface (corner crack) origins were predominant. In the fastener-loaded-hole configuration, the stress-intensity factors for a corner crack or a through crack emanating from the hole under remote applied stress (S = Sp + Sb), remote outer-fiber bending stress (SB) due to the bending moment (M), by-pass stress (Sb), fastener load (Sp = P/(wrB)), where wr is rivet spacing and B is sheet thickness, and interference (Δ), as shown in Fig. 1, are given in Ref. 13. Recently, the influence of biaxial loading (λSb) on stress-intensity factors for cracks emanating from the fastener hole were developed in Ref. 14. Herein, one of the restrictions for the corner-crack equations is that the crack aspect ratio, a/c, is fixed, and the influence of rivet interference is based on a simple approximation [13].

Fig. 1 Crack configuration and loading for fastener-loaded hole.

To calculate the growth of a corner crack initiating at a critically-loaded rivet hole in a lap-splice joint (see Fig. 1), the stress-intensity factors for rivet loading (Sp), by-pass loading (Sb), local bending (M or SB), and interference (Δ) for a through crack must first be obtained and added as K = Kp + Kb + KM + KΔ

(1)

Herein, to calculate the maximum stress-intensity factor, Kmax, it was assumed that the maximum applied loading is such that the rivet will not be in contact due to interference, that is, the applied stress will be greater than the rivet liftoff stress, SLO, see Ref. 13. Thus, at maximum load it was assumed, for simplicity, that KΔ = 0. In addition, for the under-driven rivet side of the aircraft, KΔ was also assumed to be zero. The influence of the rivet being in the hole, however, is

292

J.C. Newman Jr. and R. Ramakrishnan

reflected in the calculation of the other values of stress-intensity factor due to the radial pressure distribution. Therefore, the stress-intensity factor in terms of applied stress is K = Sp √(πc) Fp + Sb √(πc) Gb Fλn + σB √(πc) Hb

(2)

Expressing Eqn. 2 in terms of the total remote stress S with the bending stress expressed as σB = kB S gives K = S √(πc) [ Lp Lf Fp + Lb (1 – Lf) Gb Fλn + kB Hb ] = S √(πc) F

(3)

where Lp is the load factor (Sp/S) for the critical rivet (crack location), Lb is the load factor (Sb/S) for the by-pass loading, Lf is the rivet-load factor, and kB is the Hartman-Schijve bending factor [15]. The term Fλn is the correction factor for biaxial loading and the equation is given in the Ref. 14. Ref. 13 gives the equations for the boundary-correction factors Fp, Gb and Hb, for rivet load, by-pass load and bending, respectively. Thus, F is the boundary-correction factor for a through crack at a rivet-loaded hole in the lap joint. To convert Eqn. 3 to a corner crack at the edge of the rivet-loaded hole, F is multiplied by the corner-crack-to-throughcrack (Kcc/Ktc) ratio [13] as: Kcc/Ktc = 0.8 + 0.2 a/B – 0.2 (1 – a/B) (a/c) – 0.05 (1 – a/B)15

(4)

Eqn. 4 was very accurate for small a/B ratios and the Kcc/Ktc ratio approaches unity as the corner crack becomes a through crack (a/B = 1). Aircraft lap joints with tight rivets From the same retired aircraft, the lap joint along stringer 4L, appeared to have tightly-driven rivets with no detectable cracks present after the 60,000 operational cycles. From this side of the aircraft, some curved fuselage test (FT) panels were removed (with the 4L lap joint included) and tested in a pressure-box facility at the FAA William J. Hughes Technical Center [3, 7]. For this configuration, a different crack-growth model was used because tight-riveted joints tend to produce fretting surface cracks called eye-brow cracks above the rivet hole. Thus, a surface crack in a thin sheet under remote tension and bending loads (without a rivet hole) was used to make fatigue life and crack-growth predictions [14]. (Note that the surface- or through-crack model (without a rivet hole) would not be influenced by biaxial loading.) The stress-intensity factors for a surface crack in a plate under a remote tension (St) and bending (SB) stresses are given in Ref. 12. After extended fatigue testing with no apparent damage, one of these panels (FT-1) had artificial damage induced by saw-cuts at critical rivet-hole locations and tested to failure. Thus, a through crack at a rivet hole was used to conduct a fatigue-crack-growth analyses on the artifically-damaged panel. The stress-intensity factors for surface cracks in thin sheets, corner cracks at open holes or rivet-loaded holes and through cracks at rivet-loaded holes, such as Eqns. 3 and 4, were further modified using the effective stress-intensity factor range based on Elber's crack-closure concept, as discussed herein.

Fatigue Analyses of Riveted Lap-Splice Joints in a Narrow-Body Aircraft

293

3 Elber's Effective Stress-Intensity-Factor Range In 1968, Elber [10] observed that fatigue-crack surfaces contact with each other even during tension-tension cyclic loading and recognized the significance of the crack-closure concept. Using the Paris et al [16] concept, Elber proposed to use an effective stress-intensity-factor range as the crack-tip damage parameter. Prior to Elber’s discovery, Tomkins [17] was using the Bilby et al dislocation model [18] to develop a cyclic crack-tip-opening displacement (CTOD) parameter for fatiguecrack growth. Recently, an application of the FASTRAN crack-closure model based on the Dugdale strip-yield model [19] was used to evaluate the effective stress-intensity-factor range (ΔKeff) for characterizing crack-tip damage. Fig. 2 shows the calculated crack-tip-opening displacements for a crack in an infinite plate under remote uniform applied stress at R = 0 loading. (The CTOD values are proportional to the cyclic plastic strains in the crack-tip region.)

Fig. 2 Cyclic crack-tip-opening displacements with and without crack closure.

In the model simulation, the initial crack length was 2 mm and the crack was grown until the length was 30 mm; and plasticity-induced crack-closure behavior had stabilized (constant crack-opening stress, So). The solid curves show the loading and unloading records during one cycle of loading. The calculated crackopening stress is shown by the lower solid symbol along the loading trace. The smaller solid (upper) symbol shows the applied stress level when the crack-tip element went from compression to tensile stress, and the beginning of element yielding. The enclosed area is the cyclic hysteresis energy for crack growth,

294

J.C. Newman Jr. and R. Ramakrishnan

(Wp)cg. A second simulation was then made with an initial crack length of 30 mm and applying the same cyclic stress. The dashed curves show the loading and unloading traces with no crack growth, but with plastic yielding at the crack tip. Here the cyclic hysteresis energy with no crack growth, (Wp)ncg, is the area traced by the dashed curves. For small-scale yielding, the strain-energy-release rate, G, and the J-integral, are equal to K2/E. Assuming that the ratio of the cyclic hysteresis energies is proportional to (ΔKeff/ΔK)2, the middle circular symbol shows the value of crack-opening stress to satisfy the energy ratio. The ΔKeff values calculated from the model using the contact stresses and the hysteresis energies were within 2%. Thus, the current method of calculating crack-opening stress levels from contact stresses, correctly partitions the hysteresis energies for crack-tip damage. The effective stress-intensity factor against crack-growth rate relation, that was used in the fatigue and crack-growth analyses, was generated on the thin sheet 2024-T3 clad aluminum alloy as described in the following section.

4 Material Crack-Growth Properties The material used in the laboratory specimens and the fuselage structure was 2024-T3 thin-sheet aluminum alloy. Fatigue-crack-growth-rate data on a 2-mm thick clad aluminum alloy were obtained from Schijve et al [20]. The yield stress (σys) was 360 MPa, ultimate tensile strength (σu) was 490 MPa and the flow stress (σo) was 425 MPa. These data covered a wide range in stress ratios (R = -0.1 to 0.73). Newman [21] had previously developed steady-state crack-opening stress equations from the FASTRAN crack-closure model for middle-crack tension, M(T), specimens subjected to constant-amplitude loading at various stress levels (Smax/σo), stress ratios (R), and constraint factors (α). These equations were then used to develop the effective-stress-intensity-factor-range-against-rate relation for the clad alloy, as shown in Fig. 3. The symbols show the test data for the various stress ratio tests. The data correlated very well with the same constraint factors that had been used for the bare material [21, 22] with B = 2.3 mm.

Fatigue Analyses of Riveted Lap-Splice Joints in a Narrow-Body Aircraft

295

Fig. 3 Fatigue-crack-growth-rate properties for 2024-T3 clad material.

The crack transitions from flat-to-slant crack growth (assumed constraint-loss regime) at a value of (ΔKeff)T given by 0.5 σo √B [22], as shown by the vertical dashed lines at B = 1 and 2 mm, respectively. Small-crack data on the thin-sheet 2024-T3 bare alloy [23] were used to estimate the effective stress-intensity factor range relation at extremely low crack-growth rates near threshold. The solid curve shows the ΔKeff baseline relation used in all subsequent fatigue and crack-growth calculations. However, because the fuselage material was thinner (B = 1 mm), the constraint-loss regime was estimated to occur at lower rates than the 2-mm thick material data shown in Fig. 3. The dashed curve shows the relation obtained for the bare material. The clad and bare results agreed quite well. Fatigue-crack-growth rates were calculated from the multi-linear equation as ni (5) dc/dN = Ci (ΔKeff) where Ci and ni are determined from the solid curve in Fig. 3 and ΔKeff = U ΔK = Kmax – Ko. The crack-opening stress-intensity factor, Ko, was calculated from the crack-closure model [8]. Herein, dc/dN and da/dN relations are assumed the same.

5 Laboratory Open-Hole Specimens and Analyses Landers and Hardrath [24] determined fatigue lives on an aluminum alloy with a central hole (Fig. 4) for specimens with two hole diameter-to-width (D/W) ratios.

296

J.C. Newman Jr. and R. Ramakrishnan

An average value of D/W was used in the current life analyses. The calculated results, as shown by the solid curve, was made using an initial semi-circular surface crack (6-μm radius) that had an equal area to the average inclusionparticle sizes that initiated cracks [23]. As shown by Laz and Hillberry [25], the fatigue scatter on 2024-T3 notched specimens can be predicted very well using the extreme inclusion-particle sizes. Herein, a range from 3- to 15-μm flaw fit the scatter band from low- to high-cycle fatigue very well, except in the mid-region.

Fig. 4 Measured and calculated fatigue behavior of open-hole specimens.

6 Curved Panel Configuration and Analyses Two of the curved fuselage test (FT) panels removed from the left side of the retired aircraft (including stringer 4L) after about 60,000 operational (pressure) cycles was tested in a pressure-box facility at the FAA William J. Hughes Technical Center [3, 7, 26]. The first panel, FT-2, had about 43,000 additional pressure cycles applied without any noticeable cracking in the lap joint. Testing was terminated due to loading fixture failures. For the FT-2 panel, a surface-crack model was used to make fatigue life and crack-growth calculations, as described in Ref. 14, and will not be presened here. But calculations were made for two levels of secondary bending. The two levels of bending (γ = 0.85 and 0.38) were estimated from lap joints with and without a bonded doubler (see Refs. 3, 27, 28). The flight-load history was assumed to be 94.5 MPa at R = 0 loading for 59,495 cycles; and the stress applied to the FT-2 and FT-1 panels (due to internal pressure) was 98 MPa at R = 0.1 for about 43,000 and 120,000 additional simulated in-service cycles,

Fatigue Analyses of Riveted Lap-Splice Joints in a Narrow-Body Aircraft

297

respectively. But close examination of the lap-joint region indicated no cracking. The predicted crack-growth behavior on the curved fuselage test panels depended greatly upon the magnitude of secondary bending used in the analyses. The higher bending factor (0.85) indicated that the panels should have cracked in less than 100,000 total (flight plus simulated) cycles, but there was no evidence of cracking. However, using the lower bending factor of 0.38 indicated that the panels should not have cracked in less than 200,000 cycles, as observed in the curved-panel tests. Geometric non-linear analyses may be required to assess the impact of secondary bending. Panel FT-1 with improved loading fixtures withstood about 180,000 total cycles (flight loading plus 120,000 cycles of in-service simulated loading for load condition A [26]), but a detailed examination revealed no cracking in the lap joints. Thus, damage in the form of saw-cuts were inserted at some particular rivet holes [26] and the panel was subjected to residual strength loads (113 MPa at R = 0.1 for load condition B) for 10,000 cycles, but no crack growth was observed at the saw-cuts. Loads were then increased to the design limit loads (130 MPa at R = 0.1 for load condition C) for 5,000 cycles and some crack growth was detected. Finally, loads were increased 147 MPa at R = 0.1 (load condition D) and the panel was cycled for an additional 6,770 cycles. Some measured crack growth against cycles at a critical-rivet location is shown in Fig.5. In the FT-1 analysis, an initial through crack (equal area of the saw-cut, ci = 1.65 mm) was placed at the forward and aft edges of the damaged rivet hole. Because the number of cycles to initiate a crack from the saw-cut was not known, the number of cycles in load condition B was changed until the crack length at 135,000 cycles was about 4.4 mm. Thus, the computed crack growth under load condition D would be compared to the measured crack growth during the same loading sequence. Three levels of secondary bending were assumed. The high bending level (0.85) was obtained from Refs. 3 and 27. However, Fawaz [28] had tested 4 joints and measured secondary bending factors from 0.2 to 0.34. None of these joints, however, were exactly the same as the narrow-body fuselage joints, but Joint I was the closest. Joint I had 1.6-mm skin with 0.64-mm thick doubler, while the narrow-body joint had 1-mm skin and 0.5-mm thick doubler. The calculated results with the highest bending factor (0.85) produced the shortest crack-growth behavior, while the results with the nominal bending factor (0.38) was fairly close. A bending factor of 0.25 was required to match the test results. Further study is needed to determine the appropriate bending factor (or other parameters in the crack-growth analyses) for the panels tested in the pressure-box facility.

298

J.C. Newman Jr. and R. Ramakrishnan

Fig. 5 Influence of secondary bending on predicted and measured crack growth from a loaded rivet hole in FT-1 panel tested at FAA Technical Center.

7 Aircraft Lap-Joint Configuration and Analyses A large number of rivet locations had been examined from the crown stringer 4R lap joint of the retired passenger aircraft (see Fig. 6) and a large number of cracks were found emanating from the rivet holes [2, 4]. A number of these cracks have been examined in a scanning-electron-microscope (SEM) to count striations and to back-track the cracking history to reconstruct the crack length against flight cycle behavior [4]. Some of these results are shown in Fig. 7. The open symbols show results on a surface crack emanating along the faying surface but near the fastener hole and the solid symbols show a corner or surface crack emanating from the edge of the fastener hole. FASTRAN [8] and AFGROW [11] have both been used to calculate the cracking in the retired aircraft during its 60,000-pressure cycle history. Of concern was the restriction in FASTRAN that the a/c ratio had to be held constant, such as a/c = 1. Fig. 8 shows a comparison of the a/c ratio for a test case, which had 37% fastener load, 63% by-pass load and 85% bending [27]. The solid curve is the results from AFGROW for a corner crack growing at a fastener-loaded hole in which crack growth is independent in the a- and c-directions (a/c variable). The predicted a/c-ratio was nearly unity until the crack began to break through the sheet thickness (a/t = 1). For a/t ratios greater than unity, AFGROW analyses modeled an oblique crack front until the crack transitions into a straight-through crack. However, FASTRAN assumes that the a/c ratio is held constant at unity until breakthrough. Thereafter, a straight-through crack is assumed until failure. Because of compensating effects of the remote loads causing higher stress-intensity factors along the depth, a-direction, and bending causing lower stress-intensity factors along the depth direction, the crack in AFGROW was predicted to grow as a nearly quarter-circular (a/c ~ 1) crack.

Fatigue Analyses of Riveted Lap-Splice Joints in a Narrow-Body Aircraft

299

Fig. 6 Crown lap joint destructively examined along stringer 4R between fuselage/frame stations covering eight lap-joint bays.

Fig. 7 Measured and calculated crack growth from riveted joints in narrow-body aircraft.

300

J.C. Newman Jr. and R. Ramakrishnan

Fig. 8 Comparison of AFGROW and FASTRAN crack-shape analyses for a typical riveted lap joint.

In Fig. 7, the dashed curve is a calculated result from AFGROW using an EIFS of 5-μm radius corner crack. The EIFS value was chosen to roughly fit the mean of the measured data. The solid curves show calculations made with FASTRAN for values of EIFS ranging from a 9 to 40-μm radius corner crack. Both codes produced essentially the same results, in that, the shape of the crack-lengthagainst-flights curves were similar for cracks larger than about 300-μm. The only major difference was in the small-crack regime (5 to 30-μm). (Note that the AFGROW calculations did not include the effects of biaxial loading; whereas, the FASTRAN calculations did included the influence of biaxial loading, λ = 0.5, on stress-intensity factors.) FASTRAN calculations were also carried out to failure, which indicated that if not repaired or replaced, the fuselage is predicted to go to failure from 74,000 to 97,000 flights. (Incidentally, the Aloha Airlines Boeing 737 with a similar fuselage lap-joint design, but without a bonded doubler, had a fuselage failure at about 89,680 flight cycles [29].)

8 Concluding Remarks The crack-growth model, FASTRAN, was used to calculate fatigue lives and crack growth in laboratory open-hole specimens subjected to constant-amplitude loading, a curved fuselage test panel cut from the retired aircraft subjected to additional pressure cycles, and actual aircraft fuselage lap joints under operational (pressure) loading and environments. All of these specimens and components were made of a thin-sheet aluminum alloy. For each configuration, equivalent initial flaw sizes (EIFS) were established to fit the fatigue and/or crack-growth behavior using the effective stress-intensity-factor-range-against-rate relation for the aluminum alloy.

Fatigue Analyses of Riveted Lap-Splice Joints in a Narrow-Body Aircraft

301

For the open-hole laboratory specimens, an initial semi-circular surface crack (6-μm radius) located at the center of the hole was used as the crack initiation site and the calculated fatigue lives agree well with the test data from low- to highcycle fatigue behavior. The predicted cycles on the curved fuselage test panels depended greatly upon the magnitude of secondary bending factor used in the analyses. For simulated inservice loading, the higher bending factor (0.85) indicated that both FT-1 and FT2 panels should have cracked in 100,000 to 180,000 cycles, but they had no evidence of cracking. However, using the bending factor of 0.38 indicated that both panels should not have cracked in less than 200,000 cycles. In addition, predicted crack growth from saw-cut induced damage at a critical rivet hole in the FT-1 panel at higher applied loading with the lower bending factors agreed better with test measurements. Further study is needed on the appropriate bending factor or other parameters in the crack-growth analyses for the panels tested in the pressure-box facility. Using the crack-growth relation for the clad alloy, a corner crack at a fastenerloaded hole was used to calculate the fatigue lives and crack growth in the narrowbody aircraft fuselage joints (under-driven rivet side). From FASTRAN analyses, the EIFS values ranged from 9 to 40-μm radius corner cracks. The calculated crack-length-against-flight pressure cycles in the narrow-body aircraft fuselage was quite similar to results found from extensive fractographic examinations. Overall, the paper demonstrated that fuselage lap-joint fatigue-life-prediction methods based on crack growth are very adequate for the aluminum alloy.

References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13]

Steadman, D., Bakuckas Jr., J. G.: DOT/FAA/AR-07/22, V1 (2007) Ramakrishnan, R., Jury, D.: DOT/FAA/AR-07/22, V2 (2007) Mosinyi, B., Bakuckas Jr., J. G., Steadman, D.: DOT/FAA/AR-07/22, V4 (2007) Ramakrishnan, R., Steadman, D., Carter, A.: In: Proc. Int. Fatigue Congress, Atlanta, GA (2006) Jury, D., Ramakrishnan, R., Carter, A.: In: Proc. Int. Fatigue Congress, Atlanta, GA (2006) Ramakrishnan, R., Steadman, D.: In: Proc. Int. Fatigue Congress, Atlanta, GA (2006) Bakuckas Jr., J. G.: DOT/FAA/AR-01/46 (2002) Newman Jr., J. C.: NASA TM 104159 (1992) Newman Jr., J. C.: In: AGARD CP 328, pp. 6.1–6.26 (1983) Elber, W.: ASTM STP 486, pp. 230–242 (1971) Harter, J.: AFGROW Users Guide, Version 4.0005.12.10. Wright-Patterson Air Force Base, OH (2002) Newman Jr., J.C., Raju, I.S.: In: Atluri, S.N. (ed.) Computational Methods in the Mechanics of Fracture, vol. 2, pp. 311–334 (1986) Newman Jr., J.C., Harris, C.E., James, M.A., Shivakumar, K.N.: Fatigue in New and Aging Aircraft. In: Cook, R., Poole, P. (eds.) Proceedings of the 19st ICAF Symposium, vol. I, pp. 523–539. EMAS Publishing, UK (1997)

302

J.C. Newman Jr. and R. Ramakrishnan

[14] Newman Jr., J. C., Steadman, D., Ramakrishnan, R.: submitted to Int. J. Fatigue (2010) [15] Hartman, A., Schijve, J.: NLR TR 69116 U, National Aerospace Laboratory (1969) [16] Paris, P.C., Gomez, M.P., Anderson, W.E.: Trend Engng. 13(1), 9–14 (1961) [17] Tomkins, B.: Phil. Magazine 18, 1041–1066 (1968) [18] Bilby, B.A., Cottrell, A.H., Swinden, K.H.: Proc. Royal Society, 272(A), 304 (1963) [19] Dugdale, D.S.: J. Mech. Phys. Solids 8, 100–104 (1960) [20] Schijve, J., Jacobs, F.A., Tromp, P.J.: NLR-TR 68117 U, National Aerospace Laboratory (1968) [21] Newman Jr., J.C.: Int. J. Fract. 24, 131–135 (1984) [22] Newman Jr., J. C.: In: Fatigue of Aircraft Materials, Delft University Press, pp. 83109 (1992) [23] Newman Jr., J., Edwards, P.: (eds.), Short-Crack Growth Behaviour in an Aluminum Alloy. AGARD R-732 (1988) [24] Landers, C.B., Hardrath, H.F.: NACA TN 3631 (1956) [25] Laz, P.J., Hillberry, B.M.: In: Lutjering, G., Nowack, H. (eds.) Fatigue Berlin, Germany, vol. 96, pp. 1293–1298 (1996) [26] Mosinyi, B., Bakuckas, J., Steadman, D., Awerbuch, J., Lau, A., Tan, T.: In: Ninth Joint FAA/DoD/NASA Aging Aircraft Conf., Atlanta, GA (2006) [27] de Rijck, J.J.M., Fawaz, S.: In: Fourth Joint DoD/FAA/NASA Aging Aircraft Conf., St. Louis, MO (2000) [28] Fawaz, S.A.: AFRL-VA-WP-TR-2000-3024 (2000) [29] Aircraft Accident Report—Aloha Airlines, Flight 243, Boeing 737-200, National Transportation Safety Board, NTSB/AAR-89/03 (1989)

26th ICAF Symposium – Montreal, 1-3 June 2011 Crack Growth Rate Curves: Which Part Dominates Life Prediction and When? C. Wallbrink, P. Jackson, and W. Hu Air Vehicles Division, DSTO, 506 Lorimer St, Fishermans Bend, Australia 3207

Abstract. This paper details investigations into the importance of the crack growth rate data used in conjunction with FASTRAN to make fatigue crack growth predictions. Selected fighter and transport aircraft load spectra have been used in the present investigation to determine the regions of the crack growth rate curve that dominate the crack growth predictions. The results of the analysis show that universally crack growth predictions under typical aircraft load spectra utilize a significant portion of the crack growth rate curve close to the threshold region. Of note is the observation that both fighter and transport aircraft display this same behaviour throughout the entire crack growth prediction. This finding demonstrates the requirement for accurate crack growth rate data for small cracks to obtain appropriate total fatigue life predictions.

1 Introduction Modelling of fatigue crack behaviour in metallic aircraft structures has traditionally been split into an initiation phase and an observable growth phase. Although the analysis of large cracks using fracture mechanics is conducted routinely, there have been recent suggestions that the total fatigue life of an aircraft structure can be modelled from a small crack (i.e. crack growth from an initial flaw assuming little or no initiation phase), using small crack growth theory e.g., [1-3]. This is supported by the observation that fatigue cracks subjected to agile fighter flight load spectra are observed to grow with little or no discernable crack initiation phase [4-6]. In contrast, more benign transport aircraft load spectra appear to be characterised by a period of initiation or slow crack growth followed by the typical crack growth observed in aerospace aluminium alloys [7]. This vast difference in observed growth behaviour has proved to be a major challenge in the prediction of total fatigue lives for both fighter and transport aircraft using a consistent crack growth methodology. Consequently, various empirical relationships have been proposed in the literature [8, 9]. They are generally useful for the specific cases for which they were developed, but attempts in generalising their applicability are usually met with difficulties. The present investigation revisits some of the fatigue crack growth analysis using the well known crack growth code FASTRAN [10] conducted at the Defence Science and Technology Organisation (DSTO) for selected fighter and transport aircraft load spectra. The aim is to ascertain the role

304

C. Wallbrink, P. Jackson, and W. Hu

played by different sections of the crack growth rate curve in an attempt to answer the question “which part of the crack growth curve dominates when?” It is hoped that answering this question will allow engineers to concentrate on developing accurate material data in regions that will have the most benefit to total fatigue life predictions.

2 Methodology In fatigue crack growth analysis based on linear elastic fracture mechanics, the crack growth is predominantly governed by the crack growth rate curve, da / dN versus ΔKeff. Here da / dN is the crack growth rate and ΔKeff is the effective stress intensity range. The effective stress intensity range ΔKeff is used by FASTRAN and is based on the concept of crack closure [11]. It is evaluated by considering only the portion of the load cycle where the crack is open. More formally it can be written as: ΔK eff = ( S max − S x ) π a F

(1)

where Smax is the maximum load in the load cycle and Sx is either equal to the crack opening stress So if So is greater than the minimum load Smin in the load cycle or otherwise Sx is equal to Smin. The crack opening stress So is the minimum load required to open the crack. F is a geometric correction factor and a is the length of the crack. For a general nonlinear crack growth rate curve (see Figure 1 as an example), the resulting crack growth behaviour (crack length a vs time) is dependent on which section of the curve is used and how frequently it is used. For each cycle FASTRAN evaluates ΔKeff and uses the crack growth rate curve to calculate the growth. For one cycle the crack growth increment is equal to da / dN . The summation of these growth increments provides the crack growth behaviour (crack length a vs time). The complexity in computation results from the evaluation of the crack opening stress So based on a plastically induced crack closure model in FASTRAN but is not discussed here in detail. It is also worth noting that all ΔKeff values less than the threshold stress intensity do not contribute to crack growth. It is then of interest to know which segment of the crack growth rate curve is used for a given load cycle. To answer this, a series of typical fighter and transport aircraft load spectra have been considered which include; an F-111 wing splice spectrum [12], an F/A-18 wing root bending spectrum [13, 14], and a P-3C wing bending spectrum from the P-3 service life assessment program [15]. The version of FASTRAN used in this analysis was specially modified at the DSTO to allow output of data at every cycle in the evaluation of crack growth. The data output by the DSTO FASTRAN code will be used to assess the importance of portions of the ΔKeff against crack growth rate curve used in the analysis.

Crack Growth Rate Curves: Which Part Dominates Life Prediction and When?

305

0.001

Crack Growth Rate (m/cycle)

0.0001 1E-05 1E-06 1E-07 1E-08 1E-09 1E-10 1E-11 0.1

1

10

100

Flight Hours

Fig. 1 Crack growth rate data for 7075-76 aluminium alloy.

3 Numerical Data and Their Analysis The analysis involved the total crack growth predictions using the flight spectra mentioned earlier. The ΔKeff value and its associated crack growth increment were recorded for every cycle applied during the crack growth analysis. These data were then analysed to determine the contribution of each cycle to crack growth, what part of the crack growth rate curve was used by which cycle, and which portions of the crack growth rate curve had the most significant contribution to the total crack growth. The cycle-by-cycle data for each of the cases considered were collected and a binning process was applied to the ΔKeff values evaluated for each cycle. Each bin is defined by a small range of ΔKeff values. In each bin the total amount of crack growth and the total number of cycles with a ΔKeff value in the range of each bin were evaluated. For each analysis two histograms are generated, one showing crack growth related to the ΔKeff bin ranges and the other the number of cycles related to the ΔKeff bin ranges. A number above the bars in the cycle count histogram indicates the number of cycles contributing to the growth in the associated bin. In the case of the F/A-18 and P-3C it is the total cycles contributing to crack growth, but in the F-111 case only certain blocks were considered in the analysis due to data size limitations. Only cycles that contributed to crack growth were considered in the analysis, i.e. cycles with ΔKeff ranges below the crack growth threshold were ignored as they produced no analytical crack growth. In this way the analysis provides an indication of the relative significance of the various portions of the input material crack growth rate curve. For all cases considered in this paper, 7075-T6 aluminium was used. The crack growth rate data for 7075-T6 aluminium was sourced from the P-3C service life

306

C. Wallbrink, P. Jackson, and W. Hu

assessment program conducted by Lockheed Martin [15]. The crack growth rate data is presented in Figure 1. The associated constraint factors (α values) used by FASTRAN to model the changing stress state (plane strain to plane stress) were also sourced from [15]. Before the results are presented, it is useful to recall that the values of ΔKeff and crack growth increment are those that are predicted by FASTRAN. The quality and interpretation of the prediction is through the comparison to the experimentally observed crack growth. For the F-111 and F/A-18 examples, the total life correlation is reasonable but the shape differs, for the P-3 example the predictions are very close above 0.5mm (0.020”). It should also be noted that FASTRAN was specifically calibrated for the P-3 analysis by modifying the values of α to obtain a match between experiment and prediction. Experimental crack growth curves were obtained using 7075-T6 aluminium with the test specimen geometry detailed in [15]. All predictions for the following examples used a threshold stress intensity factor of 0.68 MPa m . For the F-111 and F/A-18 an initiating discontinuity observed in the coupons was used as the initial crack size and was measured to be ~25 μm. For the P-3C a value of ~127 μm was used taken from [7]. F-111 Wing Splice spectrum Due to the size of the output files generated by this particular spectrum only certain blocks of the crack growth curve were selected for analysis. Four blocks were considered, block 1, block 17, block 57 and block 87, each block represents crack growth resulting from the application of the complete spectra. The location of the blocks is indicated on the crack growth curve in Figure 2. 10

Block 1

Crack Growth (mm)

Block 17 1

Block 87

0.1

Peak Stress 254 MPa 221610 cycles in a block 2000 hours in a block

Block 52

FASTRAN 3.8 Fractography

0.01 0

50000

100000

150000

200000

250000

Flight Hours

Fig. 2 Experimental and FASTRAN predicted crack growth for F-111 spectrum.

Crack Growth Rate Curves: Which Part Dominates Life Prediction and When?

307

Block 17, 57 and 87 are indicative of crack growth in the first third, second third and final third of the predicted crack growth life. Block 1 is of interest as the spectrum contained a significant ‘overload’ called a cold proof load test (CPLT) load at the end of the spectrum. Hence crack growth in bock 1 is free of the effects of retardation due to the overload. The CPLT load (1.2 times the peak spectrum load) significantly retards the crack growth predicted by FASTRAN. The crack growth data along with the FASTRAN crack growth prediction is presented in Figure 2. The correlation of the crack growth prediction to the experiment is discussed later. The analysis showing the combined contribution to crack growth of individual cycles (grouped into bins of ΔKeff) is shown in Figure 3 whilst the number of cycles that FASTRAN determined to be contributing to crack growth (called ‘active’ cycles) is shown in Table 1. Figure 3 present data collected for crack growth in each spectrum block indicated in Figure 2. The data in Figure 3a shows the growth attributed to the ΔKeff. As the CPLT load occurs at the end of the first block the crack opening stress is still small. Therefore we see significantly more cycles contributing to crack growth than in subsequent blocks. Also identifiable in this figure is the significant contribution to crack growth at very low ΔKeff (in the region between the threshold of 0.68 MPa m and 3 MPa m ). In Figure 3b block 1 contributes to a large number of cycles near the threshold, highlighting the importance of this region in the early stages of the predicted life. Block 17 is indicative of crack growth in the first third of the total crack growth. At this stage the effect of the CPLT load is evident as shown by the reduced number of cycles contributing to crack growth. The CPLT load increases the crack opening stress evaluated by FASTRAN and therefore (referring to Eqn. 1) reduces the ΔKeff evaluated. As such the ΔKeff value for a significant number of cycles drops below the threshold for which crack growth will occur and we see a reduction in the total number of active cycles contributing to crack growth. It is interesting to note that the two CPLT load cycles contributed to a significant proportion of the crack growth while the remaining 3350 cycles that still contribute to crack growth all have a ΔKeff below 3 MPa m . The analysis of Block 52 is indicative of the behaviour of FASTRAN at about half the total predicted life. Again a significant portion of crack growth is attributed to cycles with small ΔKeff ranges. Indeed 96.6% of the cycles contributing to crack growth are below 3 MPa m . Examining block 87 in the final third of the predicted crack growth life we see that the CPLT loads become very dominant in the calculated crack growth life. According to the FASTRAN analysis at this point the importance of the rest of the spectrum is greatly diminished. Only 0.7% of the cycles in the entire spectrum are contributing to the prediction of crack growth, with the vast majority of crack growth due to the CPLT loads.

308

C. Wallbrink, P. Jackson, and W. Hu

b)

a)

Fig. 3 a) Growth and b) cycle counts attributed to different bands of ΔKeff for crack growth under the F-111 spectrum. Table 1 Statistics on FASTRAN analysis of sections of the crack growth curve under the F111 spectrum.

Proportion analysed Block 1 Block 17 Block 52 Block 87

Elapsed Flight Hours 0 35,000 105,000 175,000

No. of cycles in analysis 221,610 221,610 221,610 221,610

No of ‘active’ cycles 17,164 2,801 2,787 1,642

% of cycles ‘active’ 7.75 1.26 1.26 0.74

Total crack growth predicted 0.039 mm 0.093 mm 0.43 mm 2.3 mm

A key point to note in Figure 3 is the contribution of growth attributed to cycles evaluated with small ΔKeff. Indeed a significant portion of cycles active in producing crack growth have a ΔKeff below 3 MPa m . In the final third of the crack growth life the large crack size a is expected to result in an increased ΔKeff, yet 86% of the active cycles were still below 3 MPa m . F/A-18 spectrum FASTRAN cycle-by-cycle output for the F/A-18 spectrum was of a smaller size which also produced a smaller output allowing analysis of all the data. The crack growth life (Figure 4) was divided into thirds. Similar analyses (as used earlier) were performed to establish amounts of crack growth and number of cycles attributable to various ΔKeff. The combined contribution to crack growth of individual cycles is shown in Figure 5 whilst the information showing the number of active

Crack Growth Rate Curves: Which Part Dominates Life Prediction and When?

309

cycles is shown in Table 2. Interestingly the FASTRAN prediction showed that over the first two-thirds of the total predicted crack growth life over 99% of all load cycles had a ΔKeff below 3 MPa m . As shown in Figure 5b, in the first third of crack growth life 100% of cycles contributing to crack growth had a ΔKeff below 3 MPa m and contributed to 100% of the crack growth in the first third of the total crack growth. In the second third of crack growth life, 98.7% of cycles contributing to crack growth had a ΔKeff below 3 MPa m which contributed to 80.7% of crack growth. The final third of the total predicted life still indicated that a considerable proportion of cycles 63.7% still had a ΔKeff below 3 MPa m , however these cycles now only contributed to 1.2% of crack growth. 100

Peak Stress 168 MPa 6728 cycles in a block 324.9 hours in a block

FASTRAN 3.8 Fractography

Crack Growth (mm)

10

1

0.1

First Third

Second Third

Final Third

0.01 0

2000

4000

6000

8000

10000

12000

Flight Hours

Fig. 4 Experimental and FASTRAN predicted crack growth for the F/A-18 spectrum. Table 2 Statistics on FASTRAN analysis of sections of the crack growth curve under the F/A-18 spectrum.

Proportion analysed

Analysis Flight Hours

No. of cycles in analysis

No of ‘active’ cycles

% of cycles ‘active’

Total crack growth predicted

First third Second third Final third

0-3,865 3,865-7,730 7,730-11,596

80,279 80,279 80,279

29,555 37,046 51,265

36.8 46.1 63.9

0.048 mm 0.127 mm 9.37 mm

310

C. Wallbrink, P. Jackson, and W. Hu

a)

b)

Fig. 5 a) Growth and b) cycle counts attributed to different bands of ΔKeff for crack growth under the F/A-18 spectrum.

P-3C spectrum The P-3C spectrum was chosen to assess the output produced by FASTRAN for a typical transport/maritime aircraft wing spectrum and stress level combination. This spectrum consisted of 421,545 cycles in a block with each block equivalent to 15,000 flight hours. The total predicted crack growth results from FASTRAN, see Figure 6, were divided into three equal sections. Analysis was again performed to determine the proportion of crack growth attributed to various ΔKeff. 100 Peak Stress 173 MPa 421,545 cycles in a block 15,000 hours in a block Crack Growth (mm)

10

FASTRAN 3.8 Fractography

1

0.1

First Third

Second Third

Final Third

0.01 0

5000

10000

15000

20000

Flight Hours

Fig. 6 Experimental and FASTRAN predicted crack growth for the P-3C spectrum.

Crack Growth Rate Curves: Which Part Dominates Life Prediction and When?

311

The results are presented in Figure 7 and Table 3. In the first third of crack growth, 99.7% of cycles active in producing crack growth had a ΔKeff below 3 MPa m which contributed to 87.8% of crack growth. In the second third of crack growth a considerable proportion of the cycles (87.6%) still had a ΔKeff below 3 MPa m , but these cycles only contributed to 24.6% of crack growth. In the final third of crack growth 68.9% of cycles had a ΔKeff range below 3 MPa m resulting in 3.6% of crack growth. Table 3 Statistics on FASTRAN analysis of sections of the crack growth curve under the P-3C spectrum.

Proportion analysed

Analysis Flight Hours

First third Second third Final third

0-6,756 6,756 -13,513 13,513-20,269

a)

No. of cycles in analysis 189,879 189,879 189,879

No of ‘active’ cycles 135,513 143,941 109,223

% of cycles ‘active’ 71.37 75.81 57.52

Total crack growth predicted 0.17 mm 2.17 mm 10.4 mm

b)

Fig. 7 a) Growth and b) cycle counts attributed to different bands of ΔKeff for crack growth under the P-3C spectrum.

The probability of a ΔKeff value occurring Using the present analysis it is possible to approximate a probability density function that shows the probability of a certain ΔKeff value being evaluated in the analysis for the three spectra considered. Figure 8 presents a comparison of the probability density functions for the F-111, F/A-18 and P-3C crack growth analyses.

312

C. Wallbrink, P. Jackson, and W. Hu

0.18

P-3C F/A-18 F-111

0.16 0.14

Probability

0.12 0.1 0.08 0.06 0.04 0.02 0 0

1

2

3

4

5

6

7

8

9

10

DKeff

Fig. 8 Probability density of ΔKeff for crack growth under the F-111, F/A-18 and P-3C spectrums.

4 Discussion In this paper, observations of crack growth in 7075-T6 Aluminium are presented for different military aircraft under various load spectra. These results are then compared to standard crack growth predictions from a well known crack closure model (FASTRAN). For both the F-111 and F/A-18 examples, the shape of the crack growth curve predictions did not compare well with the experimentally observed crack growth. The analysis of the fatigue crack growth prediction showed that for high stress fighter-type spectra and the more benign transport aircraft spectra used in the present analysis, almost the entire crack growth life for the first two-thirds of the crack growth was spent in the region of ΔKeff below 4 MPa m . In more recent years it has been suggested that plasticity, oxide and surface roughness induced closure has a significant effect on threshold crack growth rate data [16]. More recent experimental methods that reduce the effects of closure have been suggested to obtain more realistic threshold region crack growth rate data [17, 18]. Applying these methods have shown that data in the threshold region has not been well represented in the past. Examining predictions of crack growth from very small naturally occurring initiation sites under various load spectra for fighter aircraft and the more benign transport aircraft show that for all cases considered the early portion of the crack growth rate curve featured strongly in final crack growth predictions. Indeed the probability density functions predicted through FASTRAN in Figure 8 show considerable correlation. Not only are they closely correlated they show that most of the cycles will probably be evaluated with a ΔKeff below 3 MPa m . As the threshold region of the crack growth rate curve is dominantly used in the analysis it is apparent that poor quality data within this region will result in large cumulative errors. The results of this investigation can be used to guide the process of

Crack Growth Rate Curves: Which Part Dominates Life Prediction and When?

313

subsequent calibrations by refining the material crack growth rate curve in the most critical regions. The importance of the threshold region of the crack growth rate material data is demonstrated and thus efforts to improve data in this region should improve crack growth predictions. In so doing it may be possible to improve predictions of crack growth from the first cycle, while also improving the accuracy of crack growth predictions for cracks with long incubation periods.

5 Conclusion The results and the discussion presented in this paper demonstrate the critical importance of the quality of the threshold region crack growth rate data in crack growth predictions using both transport and fighter aircraft load spectra. For each of the spectra considered the critical and dominant regions of the crack growth rate curve were determined allowing future targeted refinement of input material data. Predominantly for all the spectra considered, predictions of crack growth from an initial nucleated crack utilised the threshold region of the crack growth curve. An interesting observation is that both transport and fighter aircraft spectra produce very similar probability density functions, in that the probability of a particular ΔKeff value occurring is similar for all the spectra considered. The work contained in this paper will serve as a guideline for the judicious use of fatigue crack growth analysis tools by aerospace engineers.

Acknowledgements The authors wish to acknowledge Dr Manfred Heller and Mr Kevin Walker for their advice and input into this paper.

References [1] Newman Jr., J.C., Phillips, E.P., Swain, M.H.: International Journal of Fatigue 21(2), 109–119 (1999) [2] Ranganathan, N., et al.: International Journal of Fatigue 33(3), 492–499 (2008) [3] Wu, X.R., et al.: Fatigue and Fracture of Engineering Materials and Structures 21(11), 1289–1306 (1998) [4] Molent, L., Singh, R., Woolsey, J.: Engineering Failure Analysis 12(1), 13–24 (2005) [5] Goldsmith, N.T., Clark, G.: Analysis and Interpretation of Aircraft Component Defects Using Quantitative Fractography. In: Bernard, S.M., Susil, P.K. (eds.) ASTM International on Quantitative Methods in Fractography (1990) [6] Barter, S., et al.: Journal of Engineering Failure Analysis 12(1), 99–128 (2005) [7] Wallbrink, C., Hu, W.: Advanced Materials Research 41-42, 189–197 (2008) [8] Jones, R., Molent, L., Pitt, S.: In. Proceedings of International Conference on Fatigue Damage of Structural Materials VI, Hyannis, USA (2006) [9] Tiong, U.H., Jones, R.: International Journal of Fatigue 31(6), 1046–1053 (2009) [10] Newman Jr., J. C.: Fastran II - a Fatigue Crack Growth Structural Analysis Program. NASA TM-104159, NASA (1992)

314

C. Wallbrink, P. Jackson, and W. Hu

[11] Elber, W.: The Damage Tolerance in Aircraft Structures. In: ASTM STP, vol. 486. p. 230-242 (1971) [12] Diab, H., Goldsmith, R.: Fractography Results of F-111 Loads Interpretation and Truncation Validation (LITV) Coupon Test Program. DSTO-TR-2000, DSTO (2007) [13] Barter, S.A.: Fatigue Crack Growth in 7050-T7451 Aluminium Alloy Thick Section Plate with Surface Condition Simulating Some Regions of the F/A-18 Structure. DSTO-TR-1458 (2003) [14] Mongru, D.: Direct Current Potential Drop System Calibration of 1015-T6 Notched and Centre Crack Growth Coupons. DSTO-DP-1110, Melbourne, DSTO (2009) [15] Iyyer, N., et al.: International Journal of Fatigue 29(9-11), 1584–1607 (2007) [16] Newman, J.A., Piascik, R.S.: International Journal of Fatigue 26(9), 923–927 (2004) [17] Newman Jr., J.C., Yamada, Y.: International Journal of Fatigue 32(6), 879–885 (2010) [18] Ruschau, J.J., Newman Jr., J.C.: Journal of ASTM International 5(7) (2008)

26th ICAF Symposium – Montreal, 1-3 June 2011 Critical Distance for Fatigue Life Prediction in Aerospace Materials Yoichi Yamashita, Yusuke Ueda, and Hiroshi Kuroki IHI Corporation, Japan

Abstract. This study has investigated a method for estimating the fatigue life of small-notched specimens using the theory of critical distance for aerospace materials of Ti-6Al-4V and In718. Critical distance stress is defined as the average stress within the critical distance from notch root using simple linear-FE results. A good correlation exists between the critical distance stress and fatigue life of small-notched specimens if the critical distance is calibrated by the two notched fatigue failure curves of specimens with different notch root radii. Using these critical distances, the fatigue lives of various small notched specimens can be well predicted for a wide range of fatigue life. Other verification results are shown in a contact edge fatigue problem of the dovetail in aero-engine component. An analogy exists between a flat/rounded contact stress and a small notch stress fields. There have been the possibilities that the predictions using critical distance give the reasonable predicted results for fretting fatigue crack initiation lives.

1 Introduction A phenomenon known as foreign object damage (FOD), results from the impact when the rotating blades in gas turbine engines are struck by debris during takeoff and landing of the aircraft. This FOD often causes small notches on their leading and trailing edges of the airfoils. For this reason, the task of reducing aero-engine component fatigue arising from FOD continues to be one of the most significant challenges facing aero engineers. Figure 1 shows the schematics and photographs of FOD notches on the leading edges of airfoil specimens generated by quasistatic impactor. There is a great deal of concern over whether these defects may cause fatigue failure or not. From these view points, it is important to investigate the effects of FOD on fatigue strength from the perspective of the small notched fatigue problem. The previous paper [1] investigated the method for estimating the fatigue strength of small-notched Ti-6Al-4V specimens using the theory of critical distance (TCD) [2, 3] that employs the stress distribution in the vicinity of the notch root. The TCD assumes that fatigue damage can be correctly estimated only if the entire stress field damaging the fatigue fracture process zone is taken into account. However, the issues still remain that a general procedure is necessary to determine critical distances of aerospace materials and the further verifications are needed. An objective of this study is to consider, for small defects in specimens of aerospace materials, a general method of determining an appropriate critical distance in a wide fatigue life range using simple linear-FE results. In the experimental

316

Y. Yamashita, Y. Ueda, and H. Kuroki

study, small notched round-bar fatigue tests have been conducted with forged Ti6Al-4V alloy and Ni base superalloy In718 to quantify the effect of small notch radius and small notch depth on the fatigue strength of the material. And the analytical study shows that the appropriate method of evaluating fatigue life of the small notched specimens can be constructed from the relationships between the critical distance stress and fatigue crack initiation life for a wide range of fatigue failure cycles if the critical distance is calibrated by the two notched fatigue failure curves of specimens with different notch root radii. Here, Critical distance stress is defined as the average stress within the critical distance from notch root. Other verification results are shown by the application of the critical distance to the contact edge fatigue problem of the dovetail in aero-engine component.

Fig. 1 Schematics and photographs of FOD damage on the leading edge of airfoil; (a) schematic fan blade and (b) example of small FOD damages on the leading edges of airfoil fatigue specimen by quasi-static impactor.

2 Experiment Material and specimen Axial-loading fatigue tests were conducted to investigate the effects of small notches on the fatigue strength of aerospace materials. Table 1 shows the mechanical properties of the materials used. Figure 2(a) shows an alpha-beta titanium alloy microstructure. Figure 2(b) shows the microstructure of In718 used. Figure 3 shows the size and shape of round-bar fatigue specimens used. Specimens were 110mm in length with a grip section diameter of 15mm and a V-notch angle of 45°. Before introducing notches, all cylindrical fatigue specimens were machined using a low-stress grinding technique. Finally, small notches are introduced by machining. Fatigue test method All the tests were carried out under constant stress amplitude using an MTS servohydraulic test system at a nominal frequency of 20Hz. Specimens were cycled under 0-tension fatigue loading of stress ratio R = σ min σ max = 0 where σ min is minimum stress and σ max is maximum stress.

Critical Distance for Fatigue Life Prediction in Aerospace Materials

317

Fatigue test results Fatigue test results are included in Fig. 8 exhibited later. It has been found that the fatigue tests show that the larger notch radius increases fatigue strength and the larger notch depth decreases fatigue strength. Table 1 Mechanical properties of materials used. (a) Ti-6Al-4V (AMS4928)

σ σ ε ψ (M Pa) (M Pa) (%) (%) 935 1006 18.4 44.5 Y0

Note)

uts

f

(b) In718

σ σ ε ψ (M Pa) (M Pa) (%) (%) 1387 1576 17.7 29.4 Y0

uts

f

σ Y0 : Yield stress, σuts :Ultimate tensile strength, ε f : Elongation, ψ

(a) Ti-6Al-4V

: Reduction of area.

(b) In718

Fig. 2 Microstructures of Ti-6Al-4V alloy and In718 alloy.

Fig. 3 Shape and dimensions of the small notched round-bar fatigue specimens; (a) specimen size and geometry, (b) schematics of circumferential notch, and (c) definition of notch size and geometry.

318

Y. Yamashita, Y. Ueda, and H. Kuroki

3 Fatigue Life Prediction Using Critical Distance Critical distance determined by the conventional procedure The TCD is based on the assumption that the only way of accurately estimating the fatigue damage is to take the entire stress field affecting the fatigue process zone into account because there is a direct correlation between fatigue damage and the stress field distribution in the area around the stress concentrator. Instead of calculating point values for the actual quantities concerned, average values are calculated in the conventional approach, which is known as the "line method" (LM) [2,3]. The condition for determining the critical distance, over which the average stress range in the notched specimen is equal to the stress range in the smooth specimen Δσ 0 , is expressed as the following; 1 LLM ∫ Δσ yy ( x)dx = Δσ 0

LLM 0

(1)

where Δσ yy ( x) is the axial stress range distribution along the x axis by FEA and

LLM is the critical distance in the line method. According to the line method, a notched component is in its 107 cycles high-cycle fatigue limit condition when the average axial stress range within a distance from the notch tip of the empirical critical distance, LLM = 2 L , equals the plain fatigue limit where L is defined by the following;

L=

1 ΔK th ( )2

π Δσ 0

(2)

where Δσ 0 is 107 cycles plain fatigue limit stress and ΔK th is threshold value of stress intensity factor range. However, the critical distance based on the conventional line method is targeted at high-cycle fatigue, so the method may not be applied to a range from the medium-cycle low-cycle fatigue regime to the high-cycle fatigue regime without alteration. In Ref. [4], it has been found that the critical distance, over which the average stress range in the notched specimen is equal to the stress range in the smooth specimen, varies from 0.0186mm to 0.715mm for Ti-6Al-4V alloy. The critical distances determined by the conventional line method are different in different life ranges and depend on the shape and size of a small notch. Thus, the general procedure to determine critical distance for a wide range of fatigue of fatigue failure life is not established.

Critical Distance for Fatigue Life Prediction in Aerospace Materials

319

Critical distance calibrated using two notch-fatigue failure curves A method of determining the critical distance in a wide life range from the lowcycle fatigue life range to the high-cycle fatigue range is considered. Since the dominant factor for fatigue strength of a specimen with a small notch is the stress gradient at the root of the notch, this study proposes a method for determining the critical distance using the fatigue curves of two type specimens each of which has a notch with a different notch radius, as shown in Fig. 4. In this study, at a fatigue failure cycle N f , the distance L1 at which S0 becomes equal to S1 is assumed to be the critical distance defined as the following;

σ CD =

1 L1 1 L1 ∫ σ yy ( x, ρ = ρ0 )dx = ∫ σ yy ( x, ρ = ρ1)dx L1 0 L1 0

(3)

Figure 5(a) and(b) show the critical distances determined by using two notchfatigue failure curves for Ti-6Al-4V and In718 alloy respectively. In Ti-6Al-4V alloy, two notch-fatigue failure curves of specimens with d=0.3, ρ =0.2mm and specimens with d=0.3, ρ =0.05mm were used. In In718 alloy, two notch-fatigue failure curves of specimens with d=0.3, ρ =0.2mm and specimens with d=0.3, ρ =0.6mm were used. In Ti-6Al-4V alloy, the calibrated critical distances did not vary clearly over a wide range of fatigue failure cycles from medium-cycle low-cycle fatigue regime to high-cycle fatigue regime and have an almost constant value as shown in Fig. 5(a). Its average is 0.032 mm. The average critical distance corresponds to the depth of the crystallographic facets at the crack initiation sites observed by SEM as shown in Fig. 6(a). It can be seen that the cracking of alpha-grain is the main cause of fatigue failure in Ti-6Al-4V alloy. On the other hand, the critical distances in In718 alloy calibrated by the two notched fatigue failure curves are shown in Fig. 5(b). The longer fatigue life gives the smaller critical distance. In In718, it can be seen that the critical distance is dependent on the yielding scale near notch tip such as plastic zone size. Figure 6(b) shows the SEM photographs of crack initiation site in the small notch specimen of In718 alloy. The fatigue damage region can be seen in the ahead of the notch root region.

Y. Yamashita, Y. Ueda, and H. Kuroki

Small notch

σyy(x)

Φ

15

R60

7.5

320

Φ

10 110

Axial stress distributions of specimens containing two different notch root radii to calibrate the critical distance at the same fatigue cycle, Nf.

S0 S1

y d

ρ

o

L1

L0

L1 is determined as S0=S1 = 0 = 1

ρρ ρρ

Distance from notch tip, x (mm)

Fig. 4 Calibration procedure of critical distance for fatigue strength of small notched Ti6Al-4V specimens using two notched fatigue failure curves.

) 1 m m ( 1L ,e cn at 0.1 isd la cit ir C

0.01 1.E+03

Ti-6Al-4V, R.T.

L1(average) = 0.032 mm

1.E+04 1.E+05 1.E+06 Crack initiation life, Ni (cycles)

(a) Ti-6Al-4V alloy

1.E+07

) m (m L1, ec ant isd la icit rC

1

IN718, R.T. L1 = 0.9535 Nf-0.2087

0.1

0.01 1.E+03 1.E+04 1.E+05 1.E+06 1.E+07 1.E+08 Number of cycles to failure, Nf (cycles)

(b) In718 alloy

Fig. 5 Critical distances determined for aerospace materials.

Critical Distance for Fatigue Life Prediction in Aerospace Materials

321

0.03mm

(a) Ti-6Al-4V alloy

(b) In718 alloy

Fig. 6 SEM photographs of crack initiation sites.

The critical distances were used to compute the critical distance stress for each fatigue specimen with a small notch and to obtain a relationship with fatigue life. Fig. 7 shows the relationship. It has been found that a good correlation exists between the critical distance stress and crack initiation life of small-notched specimens if the critical distance is calibrated by the two notched fatigue failure curves of specimens with different notch root radii.

DC σ ,s esr ts ) ec aP na M tis ( dl ac tiri C

3000 2500 2000

CD=1962

σ

Ni

-0.0344

d=0.1mm, ρ=0.05mm d=0.3mm, ρ=0.05mm d=0.5mm, ρ=0.05mm d=0.3mm, ρ=0.2mm d=0.3mm, ρ=0.6mm

1500 1000 500

In718, R.T., R=0 Round bar tensile fatigue specimen

0 1.E+03 1.E+04 1.E+05 1.E+06 1.E+07 1.E+08 Crack initiation life, Ni (cycles)

(a) Ti-6Al-4V

(b) In718

Fig. 7 Relationship between critical distance stress and fatigue life.

Fig. 8 compares the predicted results of fatigue strength Δσ which is determined from the relationship between the crack initiation life and critical distance stress shown in Fig. 7 in comparison with the experimental results. Using these critical distances, the fatigue lives of various small notched specimens can be predicted for a wide range of fatigue life from medium-low-cycle regime to highcycle regime for Ti-6Al-4V and In718 alloy.

322

Y. Yamashita, Y. Ueda, and H. Kuroki

1 200 ) a 1 000 P M ( 800 σ Δ , 600 e g n ar 400 s s ret 200 S

E xp. d=0 .1 , ρ =0 .0 5

E xp. d=0 .3 , ρ =0 .0 5

E xp. d=0 .5 , ρ =0 .0 5

E xp. d=0 .3 , ρ =0 .2

P r e dic t , d=0 .1 , ρ =0 .0 5

P r e dic t , d=0 .3 , ρ =0 .0 5

P r e dic t , d=0 .5 , ρ =0 .0 5

P r e dic t , d=0 .3 , ρ =0 .2

Ti- 6Al- 4V, R.T.

0

10

2

10

3

10

4

10

5

10

6

10

7

10

8

Numbe r of c yc le s t o c ra c k init ia t ion, Ni (c yc le s )

(a) Ti-6Al-4V

1200

)a PM ( σ Δ

Exp. d=0.1mm, ρ=0.05mm Exp. d=0.6mm, ρ=0.05mm Exp. d=0.3mm, ρ=0.6mm Predict, d=0.1mm, ρ=0.05mm Predict, d=0.3mm, ρ=0.2mm

Exp. d=0.3mm, ρ=0.05mm Exp. d=0.3mm, ρ=0.2mm Predict, d=0.3mm, ρ=0.05mm Predict, d=0.6mm, ρ=0.05mm Predict, d=0.3mm, ρ=0.6mm

1000 800 600 400 200 0 1.E+03

In718, R.T.

1.E+04

1.E+05

1.E+06

1.E+07

Number of cycles to failure, Nf (cycles)

(b) In718

Fig. 8 Predictions for fatigue strength in comparison with experimental results.

4 Application to Dovetail Fretting Fatigue Tests Notch analogy in dovetail fretting fatigue In this section, other verification results for fatigue life prediction using critical distance are shown in a contact edge fatigue problem of the dovetail in aeroengine component. As shown schematically in Fig. 9, it has become common practice in recent years to approximate the dovetail geometry by a half-plane equivalent of a flat punch with rounded corners on a half plane [5]. And the analogy exists between the contact stress fields and near-notch tip stress fields. Nowell et al. [6,7] have explored the notch analogy for generating the equivalent stress gradients to those in a typical dovetail contact without the associated surface damage. And they concluded that notch fatigue tests and fretting fatigue tests

Critical Distance for Fatigue Life Prediction in Aerospace Materials

323

conducted with equivalent stress gradients are possible and a surface damage effects on fatigue crack initiation may be distinguished from the stress gradient effects. Fretting fatigue may be predicted by using critical distance calibrated by the two notched fatigue failure curves of different notch root radii. T V Notch

σ

Q P Q

σ

0

0

σxx

P

σxxy Dovetail

y

Notch analogy

Flat pad approximation

Fig. 9 Approximation of dovetail contact by flat/rounded pad on a half-plane and analogy between stresses at a contact and a notch.

Results Figure 10 shows the predictions using critical distance compared to fretting fatigue test results [8] for flat and rounded contact pads on Ti-6Al-4V alloy specimens. Although the limited predicted results are shown in Fig. 10, there have been the possibilities that the prediction using the critical distance gives the reasonable fretting fatigue crack initiation lives.

)s lec yc ( efi l eu igt af la to T

Experiment, Arau'jo & Nowell [3] Prediction

1.00E+09

Ti-6Al-4V

1.00E+08

Q

P

σ

2a

0

σ

0

Tensile specimen

1.00E+07 Fretting pad

1.00E+06

Run-out

1.00E+05

Stress gradient

1.00E+04 0

0.5

1

1.5

Contact half width, a (mm)

Figure 10 Predictions using critical distance compared to fretting fatigue test results for contact pads on Ti-6Al-4V alloy specimens.

324

Y. Yamashita, Y. Ueda, and H. Kuroki

5 Conclusions This study investigated the critical distance for fatigue life predictions in aerospace materials. The following conclusions can be made. (1) It has been found that a good correlation exists between the critical distance stress and crack initiation life of small-notched specimens if the critical distance is calibrated by the two notched fatigue failure curves of different notch root radii. (2) In Ti-6Al-4V, the calibrated critical distances did not vary clearly over a wide range of fatigue failure cycles and have an almost constant value. The critical distance corresponds to the depth of the crystallographic facets at the crack initiation sites. It can be seen that the cracking of alpha-grain is the main cause of fatigue failure in Ti-6Al-4V alloy for a wide range of fatigue life. (3) In In718, the longer fatigue life gives the smaller critical distance. It can be seen that the critical distance is dependent on the yielding scale ahead of the notch tip, such as plastic zone size. (4) Using these critical distances, the fatigue lives of various small notched specimens can be predicted for a wide range of fatigue life from medium-cycle regime to high-cycle regime. (5) Other verification results are shown by the application of the critical distance to the contact edge fatigue problem of the dovetail in aero-engine component. There have been the possibilities that the prediction using the critical distance gives the reasonable fretting fatigue crack initiation lives.

Acknowledgements This study was conducted under contract with the New Energy and Industrial Technology Development Organization (NEDO) as a part of "aircraft and space industry innovation program" and "energy innovation program" of the Ministry of Economy, Trade and Industry (METI).

References [1] [2] [3] [4] [5] [6] [7] [8]

Yamashita, Y., Ueda, Y., Kuroki, H., Shinozaki, M.: Proceedings of ICAF (2009) Taylor, D.: Engng. Fract. Mech. 75, 1696–1705 (2008) Susmel, L.: Engng. Fract. Mech. 75, 1706–1724 (2008) Yamashita, Y., Ueda, Y., Kuroki, H., Shinozaki, M.: Engng. Fract. Mech. 77, 1439–1453 (2010) Dini, D., Nowell, D.: Int. J. of Mch. Sci. 46, 1635–1657 (2004) Nowell, D., Dini, D.: Stress gradient effects in fretting fatigue. Tribology Int. 36, 71–78 (2003) Nowell, D., Dini, D., Hills, D.A.: Engng. Fract. Mech. 73, 207–222 (2006) Araújo, J.A., Nowell, D.: Int. J. of Fatigue 24, 763–775 (2002)

26th ICAF Symposium – Montreal, 1-3 June 2011 A Unified Variable-Amplitude Model for Crack Initiation and Crack Propagation A.B. Chattopadhyay and G. Glinka University of Waterloo

Abstract. Fatigue crack growth under variable amplitude loading is of interest to the aerospace industry as the number of ageing aircraft in use rises. The model proposed in this paper aims to provide accurate fatigue crack growth estimates from the crack initiation phase through to final failure. Most variable amplitude fatigue crack growth models require tuning with variable amplitude fatigue crack growth data in order to provide reliable estimates for fatigue life [1, 2]. The model proposed in this paper requires only constant amplitude fatigue crack growth data in order to operate, and therefore requires much less material testing before it can be used to provide a good fatigue crack growth estimate. The model describes the crack as having a blunted tip of radius r*, and uses the Smith-Watson-Topper model to calculate the fatigue damage in r* sized elements ahead of the crack tip. Whenever the damage in one of these elements reaches the value of one, the element breaks, and the crack is extended. The r* value is a constant for a given material in a given environment; for example: steel in air, or aluminum in salt water. The residual stress field affecting the crack tip is what allows for the model to account for variable amplitude loading. The paper outlines a set of five rules which determine which loading cycles affect the residual stress field. By taking the residual stress field generated by each successive loading cycle into account, the model gains a structural memory. This structural memory combined with the material memory provided by the fatigue damage in the r* sized material blocks allows the model to handle a wide array of variable amplitude loading spectra. Since the proposed fatigue crack growth model uses the Smith-Watson-Topper fatigue damage parameter to propagate the crack, it is capable of growing a crack from a r* sized notch on the order of a few microns through to the final failure. A short crack correction factor is also used, which provides a smooth transition from the short crack to long crack fatigue crack regimes.

1 Introduction The research detailed in this article describes a variable-amplitude model that is capable of handling the fatigue analysis of a component from crack initiation through to final failure. The model aims to achieve superior accuracy in its fatigue crack growth predictions by incorporating two types of memories: a structural memory that records the residual stress field behind the crack front, and a material memory that records the fatigue damage in elementary material blocks ahead of the crack tip. In order to be useful to industry, the proposed model is based on

326

A.B. Chattopadhyay and G. Glinka

commonly available fatigue properties derived from smooth materials subject to constant amplitude loading. It should be noted here that the exponents in the fatigue life equation must be determined using constant-amplitude crack growth data, as the fatigue properties of a smooth fatigue testing coupon will be different from the fatigue properties of an elementary block. Fatigue crack growth occurs when a cracked body is subject to repeated loadings. Most often, the presence of the initial crack is not sufficient to cause a catastrophic failure. Multiple loading reversals must be applied for the crack to reach a critical size. Even structures that do not initially contain cracks will eventually form them if they are subjected to cyclic loadings. The number of cycles required for such an initiation is given by the strain-life approach. Most fatigue crack growth analyses study the effects of constant-amplitude fatigue: situations when the load level varies between two known values for the course of a component’s life. Constant amplitude fatigue crack growth is reasonably well understood, and provides good estimates for the life of a component in situations where the constant amplitude method is applicable. In contrast to the constant-amplitude scenario, prediction of fatigue lives under variable amplitude loading can be considerably more problematic. Variable amplitude loading generates memory effects in the material following overloads and underloads in the loading spectrum. Minor changes in the arrangement of the loads in otherwise similar loading spectra can result in large differences in the fatigue life of a component.

2 Basic Assumptions The model assumes that the material in question may have its stress-strain response modelled by the Ramberg-Osgood equation [3], given below:

σ

⎛σ ⎞ ε = +⎜ ⎟ E ⎝ K'⎠

1/ n '

(1)

The values of the cyclic strength coefficient K’ and the cyclic hardening coefficient n’ are assumed to be stable during the course of the analysis. The proposed model functions by determining the fatigue damage in elementary material blocks ahead of a blunted crack tip of radius r*. In addition to being equal to the radius of the notch tip, the r* parameter gives the size of the elementary material blocks. The damage in the elementary material blocks is determined using the SmithWatson-Topper fatigue damage accumulation model. If an initial crack length is provided, a crack with an initial length ai and a notch tip radius r* is used. In the case of an uncracked specimen, the material is assumed to have a r* sized flaw at the surface, which acts like a notch tip. The number of cycles to failure for any material element in the model is given by the Smith-Watson-Topper expression, shown below [4]:

A Unified Variable-Amplitude Model for Crack Initiation and Crack Propagation

σ 'f 2 2b Δε σ max = 2N f ) + ε ' f σ ' f ( 2N f ( E 2

)

327

b +c

(2)

Given that the above conditions are satisfied, the model should give reasonable projections for the fatigue life of a component.

3 General Analysis Technique The stress and strain at the crack tip is determined by first finding the maximum stress intensity factor and the stress intensity amplitude associated with the given loading and geometry. It should be noted that in the case of variable-amplitude loadings, a stress intensity amplitude of DKTotal is used in the place of the applied stress intensity amplitude, DKApplied. Similarly, rather than using the applied maximum stress intensity factor, KMax,Applied, the total maximum stress intensity factor is used, KMax,Total. The expressions for KMax,Total and DKTotal are given in Eqn. 3 and Eqn. 4:

K Max,Total = K Max, Applied + KRe s

(3)

ΔKTotal = ΔK Applied + KRe s

(4)

The weight function technique is used to find the value of the residual stress intensity factor, KRes. The residual stress fields generated ahead of the crack tip contribute to the build up of this stress intensity term. The details of the method by which KRes is found is discussed in the Structural Memory section. The elastic stresses at the notch tip are found by entering the values of KMax,Total and DKTotal into the Creager-Paris equation [5]. Stresses are determined at the centroid of four r* sized elements ahead of the crack tip. The fatigue damage buildup in elements further than four times r* from the crack tip is negligible [6]. Once the elastic stresses from the Creager-Paris solution are generated, the multi-axial Neuber rule is employed to return the actual elasto-plastic stress-strain response at the crack tip. The stress state in each of the four material blocks ahead of the crack tip is monitored, along with the amount of damage accumulation in the said blocks. Once a material block breaks, the crack propagates by the number of blocks broken multiplied by the distance r*, providing the basis for this fatigue crack growth model. Further details on the stress analysis and fatigue damage accumulation techniques used in this model are included in the Material Memory section. A program implementing this model is being written, and is already functional for constant-amplitude loading cases. Validation is currently underway for variable-amplitude loading spectra using experimental data provided by the Office of Naval Research.

328

A.B. Chattopadhyay and G. Glinka

4 The r* Parameter There are several options when it comes to representing the crack tip in continuum mechanics. A sharp crack tip with a radius of r* = 0 leads to a singularity in the stress field at the crack tip. In order to avoid such unrealistically high stresses in the process zone ahead of the crack tip, a crack with a finite radius of r* was used. Several attempts to model the crack tip with respect to microstructure have been attempted [7-13]. Neuber [12] proposed that the material ahead of the crack tip could be represented by elementary material blocks. Forsyth, however, suggested that an elementary material block size could be related to microstructure [14]. Glinka and Noroozi developed the expression for r* given below [15]:

1.6332 ⎛ ΔKTh ⎞ ρ∗ = ⎜ ⎟ 2π ⎝ Δσ Tha ⎠

2

(5)

Recently, Mikheevskiy showed that the r* parameter may be found by taking constant-amplitude fatigue crack growth data for a material at various R-ratios and finding the r* value at which the fatigue crack growth rate vs. driving force collapses along a single curve [16]. It is this final method that will be used to determine the value of this parameter. The r* value tends to be of the same order of magnitude of grain size in both steels and aluminium alloys, once again indicating that there may be a relation to microstructure.

5 Structural Memory The structural memory behind the crack front is found by recording the residual stress fields generated by successive loading and unloading cycles. As described in the General Analysis section, the crack only propagates when it breaks one or more material elements ahead of the crack tip. Therefore, in the case of a virgin crack or smooth material (one containing a surface flaw of size r*) there would be no residual stress field acting on the crack length. Accordingly, in the case of virgin material, the values of KMax,Total and DKTotal would respectively be equal to KMax,Applied and DKApplied. It should be noted that as loads are applied to the material, residual stress fields will be generated ahead of the crack tip. The proposed model works by recording the largest residual stress field generated ahead of the crack tip for any given crack increment. When the material block ahead of the crack tip final breaks and the crack advances, it grows into the residual stress field that was created ahead of the crack tip. At this point, the stress field and appropriate weight function are integrated over the crack length to yield a KRes value. The following general expression is used to find the value of KRes:

A Unified Variable-Amplitude Model for Crack Initiation and Crack Propagation

329

a

K Re s = ∫ σ r ( x ) ⋅ m ( x, a ) dx

(6)

0

Mikheevskiy showed that the one-dimensional universal weight function given in Eqn. 7 could be used in cases in which the region of integration is less than half of the crack length [16].

m ( x, a ) =

2 2π ( a − x )

(7)

The residual stress field typically takes the form of a saw-tooth pattern generated as the crack propagates through a material. The stress field is generated by linking the maxima and intersection points of successive residual stress fields. Which residual stress fields are included in the KRes calculation is given by the following memory rules. The first rule was developed specifically for this model, whereas the remaining four rules were proposed by Mikheevskiy [17]. Rule 1: The residual stress field recorded while at any given crack increment is the largest one that has been generated at that crack increment. All other fields must be inside the largest stress field, therefore the material does not ‘feel’ their occurrence. Rule 2: Only the compressive portion of the residual stress field is incorporated into the calculation of the KRes value. Rule 3: If the compressive part of the residual stress field induced at a crack increment is completely inside the residual stress field generated at the previous crack increment, the material does not ‘feel’ it and the current minimum stress distribution should be neglected. Rule 4: If the compressive part of the residual stress distribution of the current crack increment is fully or partly outside the residual stress fields generated at previous crack increments, they should be combined as shown in Fig. 1.

Fig. 1 Linking of adjacent residual stress fields. [17].

330

A.B. Chattopadhyay and G. Glinka

Rule 5: Each residual stress distribution should only be included in the KRes calculation so long as the crack tip is inside its compressive stress zone. Therefore, if the crack tip has propagated all the way through a given residual stress field’s compressive zone, the residual stress field may be neglected in all following analyses.

6 Material Memory In addition to recording the residual stress fields behind the crack front, the proposed model also tracks the state ahead of the crack tip. The condition of the elementary material blocks ahead of the crack tip is recorded by the material memory model.

Fig. 2 Material blocks ahead of the crack tip.

Since stress intensity factor expressions exist for most commonly encountered crack geometries, the remote load and stress intensity factor solution is used to determine the stress intensity factor created by the corresponding geometry and loading condition. Given the stress intensity factor for a given load level, the Creager-Paris equation is used to obtain the elastic stress field ahead of the crack tip of radius r*. The elastic stress is found for the centre of each of the four damage accumulation elements.

A Unified Variable-Amplitude Model for Crack Initiation and Crack Propagation

331

The above equations are effective for situations when the crack tip is small in comparison to the crack size. The proposed model handles short crack growth using a short crack correction factor. As the length of the crack becomes comparable to the crack tip radius r*, the Creager-Paris solution returns a stress concentration value of two. In reality, the stress concentration caused by a circular notch in an infinite plate is three. The short crack correction magnifies the result of the Creager-Paris solution when r* is comparable to the crack length, thereby reflecting the accelerated growth rates demonstrated by short cracks. The expression is given below:

SCC = 1 +

1 ρ∗ 2 a

(8)

After determining the elastic stress ahead of the crack tip, the elasto-plastic stress may be found using the multiaxial Neuber equation. The damage generated by a given loading cycle is obtained by first rainflow counting to the first damaging cycle. The stress state for the damaging cycle is then obtained by tracking the stresses from the beginning of the history to the counted cycle. It should be noted that even when a component is being subject to constant amplitude loading, the elements ahead of the crack tip experience a variable-amplitude loading. This is because as the crack tip advances, the loads acting on the material element intensify. A typical hysteresis loop of one of the material elements ahead of the crack tip is shown in Fig. 3.

Fig. 3 Constant-amplitude loading results in a variable-amplitude hysteresis response in the elements ahead of the crack tip.

332

A.B. Chattopadhyay and G. Glinka

Fig. 4a An element ahead of the crack tip reaches its critical damage value.

Fig. 4b The failed element is removed, the crack advances, and an undamaged element is added to the end of the stack.

Once the actual stresses corresponding to the damaging cycle are obtained the strains associated with them are found using the Ramberg-Osgood approach. The damage value associated with the damaging cycle is then saved. Please note that the stress analysis and damage determination are carried out for each of the elements ahead of the crack tip. The damaging cycle is then removed from the spectrum and the next damaging cycle is found via rainflow counting of the modified

A Unified Variable-Amplitude Model for Crack Initiation and Crack Propagation

333

history. This is the method by which damage accumulates in the material elements ahead of the crack tip. The process is repeated until the element closest to the crack tip reaches a damage value of one, and the crack propagates. The point of fracture in the model is shown in Fig. 4a. Once the crack propagates, the damage values in the unbroken elements are retained, and the damaged elements are shifted closer to the crack tip. An element with a damage value of zero is added to the end of the stack of elements. This is shown in Fig. 4b. The process described above works well for cases in which linear-elastic fracture mechanics are applicable. However, towards the end of the fatigue life of a component the crack can begin to propagate by distances greater than four r* in a single loading cycle. When this occurs, the program switches to determining the point at which the strain ahead of the crack tip equals the true strain at fracture. The crack is then propagated by the corresponding distance.

7 Conclusions The proposed method shows promise for handling variable amplitude loading scenarios. Both crack initiation and crack propagation phases of crack growth may be predicted using the same model. The method requires only smooth-specimen constant-amplitude fatigue crack growth data in order to operate. It should be noted that the exponents in the Smith-Watson-Topper expression must be found by fitting constant-amplitude fatigue crack growth data. The model produces its fatigue crack growth estimates by maintaining two forms of memory: a structural memory of the residual stress field behind the crack front, and a material memory in the elements ahead of the crack tip. This combination of techniques allows for good accuracy in its fatigue crack growth estimates.

References [1] Newman, J.C.: Prediction of fatigue crack growth under variable amplitude and spectrum loading using a closure model. In: ASTM STP, vol. 761, pp. 255–277. American Society for Testing and Materials (1982) [2] Suresh, S.: Fatigue of Materials. Cambridge University Press, Cambridge (1991) [3] Landgraf, R.W., Morrow, J.D., Endo, T.: Journal of Materials 4(1), 176 (1969) [4] Smith, K.N., Watson, P., Topper, T.H.: Journal of Materials 5(4), 767–778 (1970) [5] Creager, M., Paris, P.C.: International Journal of Fatigue 3, 247–251 (1967) [6] Glinka, G.: Engineering Fracture Mechanics 21(2), 245–261 (1985) [7] Lal, D.N., Weiss, V.: Metallurgical Transactions 9A, 413–426 (1978) [8] Majumder, S., Morrow, J.D.: Fracture Toughness and Slow-stable Cracking. In: STP, vol. 559, pp. 159–182. American Society for Testing and Materials (1974) [9] Chakrabortty, S.B.: Fatigue of Engineering Materials and Structures 2, 331–344 (1979) [10] Glinka, G.: International Journal of Fatigue 4, 59–67 (1982)

334 [11] [12] [13] [14] [15]

A.B. Chattopadhyay and G. Glinka

Irwin, G.R.: Journal of Applied Mechanics 24, 109–114 (1957) Neuber, H.: Kerbspannungslehre. Springer, Berlin (1958) Harris, W.J.: Metallic Fatigue. Pergamon Press, London (1961) Forsyth, P.J.: International Journal of Fatigue 5, 3–14 (1983) Glinka, G., Noroozi, A.H., Lambert, S.: International Journal of Fatigue 29, 1616–1634 (2007) [16] Mikheevskiy, S.: Elastic-Plastic Fatigue Crack Growth Analysis Under Variable Amplitude Loading Spectra. University of Waterloo, Waterloo (2009) [17] Mikheevskiy, S., Glinka, G.: International Journal of Fatigue 31, 1828–1836 (2009)

26th ICAF Symposium – Montreal, 1-3 June 2011 Development of an Efficient Methodology and Tool to Determine Stress Intensity Correction Factors for Complex Aircraft Structures Guillaume Renaud, Min Liao, and Yan Bombardier 1 blank between names and affiliations Institute for Aerospace Research (IAR), National Research Council Canada (NRC), Ottawa, Canada

Blank lines Abstract. Damage tolerance analysis (DTA) of complex aircraft structures requires stress intensity correction factor (β-factor) solutions that cannot be accurately or practically determined with handbook solutions. Typically, these β-factors can be related to the load transfer to adjacent structures and to irregular geometries. A methodology based on ratios between stress intensity factors obtained from StressCheck finite element models is proposed to efficiently determine such factors, which can then be compounded with handbook solutions. Three CC-130 fatigue critical location examples are presented, showing that the results from the proposed methodology are in good agreement with those of other approaches and that the studied β-factors have an important impact on the total compounded stress intensity factor. Various approaches for calculating the β-factors for the complex structures are investigated. Finally, an automated parametric tool, which is able to efficiently generate special factors for typical aircraft critical locations, is presented.

1 Introduction Advanced damage tolerance analysis (DTA) and quantitative risk analysis (QRA) methods and tools are being developed at the National Research Council of Canada (NRC) to assist the Canadian Forces (CF) for aircraft structures life-cycle management [1-3]. In order to be able to provide support on a regular basis and with quick turn-around times, efficient methods and generic tools are needed, especially for problems involving multi-site fatigue damage (MSD) and multielement damage (MED). The conventional DTA approach relies on the compounding of handbook closed-form or empirical solutions to obtain stress intensity correction factors (βfactors). Using this approach, each individual correction factor is usually assumed independent and related to a specific aspect of the problem. However, some complex aircraft structural locations require β-factor solutions that may not be accurately or practically estimated by handbook solutions. These β-factors typically represent configurations with load transfer to adjacent structures, non-planar crack paths, complex boundary conditions, and irregular three-dimensional (3D) geometries. For example, the analysis of the three typical CC-130 centre wing locations shown in Figure 1 requires the development of such β-factor solutions.

336

G. Renaud, M. Liao, and Y. Bombardier

Location 1 T-shape junction

Location 2 Spar cap

Stringer

Location 3 Spar web Wing skin

3 add’l stringers

crack in riser

crack in panel

Cap corner angle crack in cap b) E: adjacent structures c) E: adj. struct. + geo

a) E: geometry

Fig. 1 Typical CC-130 centre wing critical locations.

One approach to solve this type of problems is to ignore some of the complex features, or to simplify them using existing handbook solutions. However, these features often play a major role in the total stress intensity factor (SIF) formulation, especially for MSD/MED scenarios, and should be calculated as accurately as possible. Another approach is to use detailed 3D finite element (FE) models of the cracked component, including the adjacent structures that affect crack growth. Because this approach is completely FE based, all the characteristics of the problem must be embedded in the model, including the part-through thickness cracks, pin loading, etc. Although potentially very accurate, the development and solution times associated with this type of models make this approach inefficient.

σtotal

Plate Crack

c σtotal

Multi-phase single crack growth analysis:

Ligament failure Load path

σbypass BBR=σbearing/σbypass

Stiffener

Phases I & II Phases III & IV Phases V & VI

3 add’l stringers

Stringer/Cap ∅ 0.339” (BBR=1.587) effect

σbearing

∅ 0.267”

c c B

Phases VII & VIII B

7075-T7351 0.22” thick W

VIII

VII VI V

IV III

I

D

W

II

σtotal

σtotal B

Edge crack through hole

a φ

2c

Thickness (T) D=2R

c

Corner crack Crack approaching a hole

Fig. 2 Compounding of β-factors for location 2.

Crack at hole with bearing load

Development of an Efficient Methodology and Tool

337

The approach being developed at NRC relies on the compounding method using available handbook β-factor solutions and β-factors developed by relatively simple FE models, which only consider the “special” features that cannot be accurately covered by the handbook solutions. As an example, β-factors associated with Location 2, shown in Figure 1b, are presented in Figure 2. In this case, the stringer/cap effect is calculated with simple FE models and handbook solutions are used for the other effects. This paper presents recent work performed at NRC on the development of an efficient methodology and tool to determine the stress intensity correction factors associated with “special” features. The development of the β-factor solutions for the locations shown in Figure 1 are described through three examples and an automated parametric tool for these locations is presented.

2 Methodology A finite element based methodology was developed to efficiently estimate the βfactor solutions corresponding to the isolated effects of the “special” features to be considered. For each configuration, two FE models are built, the first one including the considered features, and the other representing a baseline model, without the features. The ratio of the stress intensity factors (SIF) obtained for different crack length from both models equals to the sought β-factor solution, β f, as described by

Kf Kb

=

βTf σ πa βTbσ πa

=

β Tb β f σ πa βTbσ πa

= βf

(1)

where Kf is the crack tip SIF of the structure including the features, Kb is the crack tip SIF of the baseline model, β Tf and βTb are the total correction factors of the structure with and without the features, respectively, σ is the applied stress, and a is the crack length. To compound the handbook solutions with βf, the loading, geometry, and boundary conditions used in the baseline FE model must be compatible with those used for the closed-form or tabular solutions. Once βf is known, the total stress intensity correction factor, β T, is calculated as

β T = β f × β HDBK

(2)

where β HDBK is the handbook solutions, typically including part-through crack effects, pin loading effects, and finite width effects. The p-version FE package StressCheck [4] was used to develop simplified twodimensional (2D) planar and 3D models for several typical fatigue-critical locations of the CC-130 aircraft wing. The models were designed to isolate the effect of the “special” features using simplified geometries. Additional effects, such as the effect of the crack shape, holes, and fastener loads, were considered by

338

G. Renaud, M. Liao, and Y. Bombardier

compounding available closed-form and tabular solutions with the beta-factor corresponding to the “special” features. Depending on the problem, the baseline solution could also be obtained using a library of handbook solutions or from a fracture mechanics software such as AFGROW [5] or CanGROW [6]. The results and tool presented in this paper were however all developed completely with StressCheck.

3 Examples A quick survey indicated that most aircraft locations, such as the CC-130 centre wing fatigue critical locations, can be represented by a limited number of typical configurations. To take advantage of this grouping, the FE models developed for the locations shown in Figure 1 were parameterized and automated. Location 1: Geometry effect As a first example, the models built to isolate the effects of a complex geometry for a crack propagating from a lower surface riser into the wing panel (Figure 1a) is presented. In this case, Eqn. 1 was applied directly using a simplified 3D FE model for determining Kf and 2D planar baseline FE model for determining Kb. The ratio between the SIF from these two models was taken to determine βf for the T-shape junction. The part-through crack and the hole were not modelled in the model as these features were covered by handbook β-factors in the compounding formulation. The geometric effect βf and the compounded βT solutions are shown in Figure 3 and Figure 4, respectively. 1.5

1 Ef 0.5

0 0

50

100

150

Cracklength(mm)

Fig. 3 Stress intensity correction factors βf for Location 1.

E7

Development of an Efficient Methodology and Tool

339

2.5 NRC(bendingrestrained)

Hole

2

NRC(5risersͲ bendingnotrestrained) NRC(1riserͲ bendingnotrestrained) DSTOHandbook

1.5

DSTO3DFE(bendingrestrained) DSTO3DFE(bendingnotrestrained)

E7 1

0.5

Corner

0 0

50

100

150

Cracklength(mm)

Fig. 4 Total stress intensity correction factor (Location 1).

The β f solution presented in Figure 3 shows that the T-shape geometry would significantly affect the β T calculations for all considered crack lengths. The total stress intensity correction factor solution, presented in Figure 4 shows that the NRC methodology is in good agreement with results obtained by the Defence Science and Technology Organisation of Australia (DSTO) [7] when the same out-ofplane bending assumptions are used. In this case, the handbook solutions did not consider bending, which effect is shown in Figure 4 to affect the β T solution when the crack tip is between the hole (after ligament failure) and the junction (i.e. a = 15 to 35mm). It is also seen that the effect of out-of-plane bending on the solution was dependent on the number of risers considered in the analysis. In this case, a parametric study showed that the difference between the use of three and five risers was minimal. For this location, the use of five risers was therefore considered sufficient. Location 2: Adjacent Spar Cap and Stringers Effect The second example presents the models developed to isolate the effect of adjacent structures on a crack growing in a skin panel under a spar cap and a series of four stringers (Figure 1b and Figure 2). The adjacent structures effect was calculated using Eqn. 1, using the Kf SIF from a model including the panel, the spar cap, and the stringers, and the Kb SIF from a baseline model of the panel only. To be consistent with the compounding methodology provided in Eqn. 2, the geometry and the boundary conditions applied on the baseline model were made identical to the ones used for the handbook solutions. Both the Kf and Kb FE models used simplified 2D planar representations of the components. Only the hole from which the primary crack nucleated was modelled. In the Kf model, the spar cap and stringers were connected to the panel using nonlinear contact fastener and link elements. The effect of stringer failure

340

G. Renaud, M. Liao, and Y. Bombardier

(multi-element damage) on the β-factor solution was also considered by removing broken stringers from the Kf model. The cracked panel was assumed 652 mm wide, which corresponds to a single lower skin panel. In reality, the lower wing skin is composed of three panels and in-plane bending deformations should be restricted to simulate attachment with the adjacent panels, as opposed to the handbook solutions that allow in-plane bending deformations. Boundary conditions can be applied to the Kf FE model by either directly restricting lateral displacements, or by increasing the width of the panel to simulate the presence of adjacent panels. As an example, the effects of using the width of a single panel (W1) and of the three panels (W2) are shown in Figure 5. 450 BaselinemodelandfiniteͲwidth HDBKsolutionͲ W1

400

BaselinemodelandfiniteͲwidth HDBKsolutionͲ W2

K/V(mm)

350

Cap/stringermodelͲ allstringersintactͲ W1

300

Cap/stringermodelͲ 1ststringerbrokenͲ W1

250

Cap/stringermodelͲ 1stand2ndstringersbrokenͲ W1 Cap/stringermodelͲ allstringersintactͲ W2

200

W1 = 652 mm (1 panel) W2 = 2032 mm (3 panels)

150 100 50 0 0

100

200

300

400

500

Cracklength(mm)

Fig. 5 Stress intensity factor solutions associated with Location 2.

It is seen that the effect of the width on the cap/stringer stress intensity factor was found to be negligible as the spar cap and the stringers significantly reduced the in-plane bending caused by the presence of a long crack. The baseline model stress intensity factor, however, was highly sensitive to the assumed width, which shows why the baseline model must use the same geometry and boundary conditions as those assumed in the handbook solutions. Assuming a baseline FE model width larger than that used in the handbook solutions may lead to βf overestimations as the crack becomes long relatively to the width of the plate (a ≥ 100 mm in this example). Similar observations can be made by comparing results with and without restraining boundary conditions in the baseline model. The Location 2 adjacent structures effect, β f, is shown in Figure 6, along with the effect of stringer failures (MED). The width of a single panel was assumed in the baseline model and appropriate boundary conditions were applied to the perimeter of the stiffened model, allowing the compounded solution to include the effect of the surrounding structural components [2].

Development of an Efficient Methodology and Tool

341

1.2 1

Allstringersintact 1ststringerbroken

0.8 1stand2ndstringersbroken E f 0.6 0.4 0.2 0 0

100

200

300

400

500

Cracklength(mm)

Fig. 6 Cap and stringers stress intensity correction factor βf for Location 2.

It is seen that the adjacent cap and stringers induce a major SIF reduction as the crack length increases towards the edge of the panel. This reduction is directly related to the panel width assumed in baseline and handbook models and counteracts the increasingly high SIF due to the finite-width correction factor, which is greatly lowered by the cap and stringers. Tests showed that when the width of the entire lower wing skin was assumed, the handbook SIF did not increase as much, the βf did not decrease as much, and the compounded β f remained similar. In this example, β f was not limited to a cap and stiffeners effect, but also included boundary condition effects that were not considered by the handbook solutions. In general, effects including loading conditions, as well as special geometry or material configurations, can be combined in the Kf model. Location 3: Geometry and Adjacent Structures Effect The third example combines the effects of adjacent structures and a complex geometry. It consists of an L-shaped cracked spar cap, fastened to a wing panel and a spar web. This problem can be addressed by several approaches using different levels of decomposition. One approach would be to combine all effects in a single 3D FE model. Another approach could use 2D planar and 3D models to estimate the isolated adjacent structures and geometry effects, which would then be combined to form the complete β f solution. If no coupling exists between the various effects, the problem could be further decomposed to consider basic effects from single features, using a library of simple models. Various levels of decomposition are shown in Figure 7.

342

G. Renaud, M. Liao, and Y. Bombardier

Level 1: Single Model

Beta FE total

+ HDBK Betas

Level 2: Effect types

Beta Adj. Struct.

X

Beta Geometry

Level 3: Effects from Beta Struct. 1 X Beta Struct. 2 X Beta Geo 1 X Beta Geo 2 single features

Complexity, knowledge required (coupling effects, etc.)

Beta total

Time (development, solving)

Specific

Level 0: Single Model

Generic

Fig. 7 Decomposition for combined adjacent structures and geometry effects.

The level 1 decomposition combines all the effects excluded from the handbook solutions but the development and solving times of 3D models including complex geometries, adjacent structures, and fastener contact were found to be very high. The level 2 decomposition involves simple 3D and planar models similar to those used for the locations 1 and 2. As such, a library of generic models can be used to calculate combined effects. Once set up in an automated tool, this approach is more flexible and faster than the level 1 decomposition. However, care must be taken to consider the coupling between the different types of effects by making the various models compatible. For instance, the cracking of a standalone complex 3D component can induce large out-of-plane bending that is not compatible with the presence of stiffening adjacent structures. The Location 3 β f solution developed at NRC contained the separate effects of: a) the spar web and lower panel adjacent structures; herein referred to as β f1, and b) the non-planar, L- shaped cap geometry, referred to as βf2. Displacements compatibility between the models was ensured using boundary conditions that simulated the presence of the surrounding structures [3]. The combined β f was calculated as

β f = βf1 ×β f 2 =

K f 1 _ BC K f 2 _ BC × K b _ BC K b _ free

(3)

where the subscripts _BC and _free refer to the configuration modified by the special boundary conditions and a free configuration, respectively. Three of the four K solutions were calculated using the displacement-compatibility boundary conditions. However, the baseline model of βf2 used a free configuration to ensure compatibility of the combined β f solution with the handbook solutions. Since both βf1 and βf2 assumed the same unstiffened flat baseline model, the boundary conditions and free configuration could have been switched between the βf2 and βf1 models, i.e., ( K f 1 _ BC / K b _ BC ) × ( K f 2 _ BC / K b _ free ) = ( K f 1 _ BC / K b _ free ) × ( K f 2 _ BC / K b _ BC ) .

Development of an Efficient Methodology and Tool

343

The combined geometric and adjacent structures effect βf is shown in Figure 9. Similar to Location 2 (Figure 6), the Location 3 βf solution gets very low when the crack length approaches the width of the component, assumed to be 118 mm for this example. As shown in Figure 10, the compounded β T solution calculated by NRC is in good agreement with the solution calculated by DSTO, which used a 3D FE model [7] that considered all the effects but was much slower. 1

0.75

Ef E

0.5

0.25

0 0

25

50

75

100

Cracklength(mm)

Fig. 8 Stress intensity correction factors βf for Location 3.

2.5

Hole

NRC

2

E7 E7

DSTO3DFE

1.5

1

0.5

Corner

0 0

25

50

75

100

Cracklength(mm)

Fig. 9 Total stress intensity correction factor for Location 3.

4 Automated Parametric Tool To maximize the calculation efficiency of the various FE-based β-factors, a userfriendly VBA-based analysis tool, implemented in MS Excel, was developed.

344

G. Renaud, M. Liao, and Y. Bombardier

This tool, linked with StressCheck through its component object model application programming interface (COM API), contains basic parametric model definitions of typical critical locations, including the geometry, materials, and crack growth analysis range. It allows solving of user-defined problems and automatic generation of correction β-factors as a function of crack length. With this tool, engineers can quickly assemble models for structures of similar configurations and efficiently calculate β-factors related to “special” features. Modelling assumptions can easily be tested and sensitivity analysis can easily be performed. A screenshot of the tool is shown in Figure 10 for Location 2, defined by a set of 32 geometric and basic material parameters. The outputs are the Kf and Kb curves, as well as the βf solution, as a function of crack length.

Input parameters Model sketch

Ecurve(s) K curves

Fig. 10 Screenshot of the generic FE-based beta tool for Location 2.

5 Conclusions The DTA of complex aircraft structures requires β-factor solutions that cannot be accurately or practically determined with available handbook solutions. An accurate yet efficient determination of the β-factors related to complex aircraft structural configurations is therefore highly desirable. The NRC methodology and tool are based on simplified FE models, and are demonstrated to provide good estimations of the β-factors considering adjacent structures and complex geometry effects. The use of fully parameterized representations of the actual problems makes the developed methodology computationally

Development of an Efficient Methodology and Tool

345

efficient and easily adaptable to various problems of similar configurations. As a practical application, the developed method/tool has been successfully used by NRC to conduct QRA of built-up structures containing MSD and MED in support of the CF aircraft structural life cycle managements.

References [1] Liao, M., Bombardier, Y., Renaud, G.: Quantitative Risk Assessment for the CC-130 Centre Wing Structure (Phase II) - Final Report. NRC Lab. Tech. Report, LTR-SMPL2009-0098 (2009) [2] Liao, M., Bombardier, Y., Renaud, G.: Updated Quantitative Risk Analysis for the CC-130 Center Wing Structures: Part 1 CW-1 Location. NRC Lab. Tech. Report, LTR-SMPL-2010-0222 (2010) [3] Liao, M., Bombardier, Y., Renaud, G.: Updated Quantitative Risk Analysis for the CC-130 Center Wing Structures: Part 2 CW-14B Location. NRC Lab. Tech. Report, LTR-SMPL-2010-0232 (2010) [4] Engineering Software Research and Development Inc. StressCheck Master Guide, Release 9.0 (2009) [5] Harter, J.A.: AFGROW Users Guide and Technical Manual, Version 4.0012.15 AFRL-VA/WP-TR-2008-XXXX (2008) [6] Bombardier, Y., Liao, M., Renaud, G.: A New Crack Growth Analysis Tool for the Assessment of Multiple Site Fatigue Damage. In: Proceedings of the, Aircraft Airworthiness & Sustainment Conference, Austin, TX (2010) [7] Evans, R., et al.: Computational Approaches for the Development of Improved Beta Factor Solutions for C-130J-30 DTA Locations. In: Proceedings of the, Aircraft Airworthiness & Sustainment Conference, Austin, TX (2010)

26th ICAF Symposium – Montreal, 1-3 June 2011 Improved SIF Calculation in Riveted Panel Type Structures Using Numerical Simulation S.C. Mellings1, J.M.W. Baynham1, and T.J. Curtin2 1

C.M. BEASY, Southampton, England 2 C.M. BEASY, Boston, USA

Abstract. As damage tolerance methods continue to evolve it has become possible to evaluate the interaction of riveted connections on the predicted SIF values in airframe structures. This current work is focused on the prediction of SIF values and crack growth paths in riveted structural members subject to complex loading, taking into account the influence of contact loading in the rivet hole, by-pass loading and changes to the load path which occur when a growing crack causes a rivet to completely cease transferring load. The SIF calculation and simulated crack growth trajectory are performed using a boundary element based fatigue and crack growth toolset. Newly developed modelling tools and analysis capabilities are applied to demonstrate how load transfer between different structural members may influence the calculated SIF values and crack growth direction. The results from analysis of different airframe type models are presented. The first study examines the effects of changing the rivet geometry, using perfect fit rivets, clearance fit rivets and push fit rivets, and determining the effect on SIF values calculated for a half-penny shaped crack. A second study looks at load transfer between more complex riveted components, including effects of the significant load redistribution which occurs when a rivet is lost as a result of ligament failure.

1 Introduction Many aerospace structures are fabricated using mechanically fastened joints. In order to perform an accurate damage tolerance analysis of a multi-fastener connection it is important to understand how the combined effect of bearing loads and loads that bypass the hole can impact the behaviour of a crack. In most applications there is some degree of clearance between the hole and fastener. This clearance affects the contact area at the fastener-hole interface in a nonlinear manner and influences the stress state around the hole. In order to accurately analyze this behaviour an iterative contact analysis is required. If friction within the fastener hole is considered then the way in which the load is applied also becomes quite important due to the load path dependency associated with frictional contact analysis.

348

S.C. Mellings, J.M.W. Baynham, and T.J. Curtin

The following discussion describes the use of the BEASY Fatigue and Crack Growth software to investigate the damage tolerant behaviour of a mechanically fastened airframe structure. In the first example a simple lap joint example is shown where different contact connection types are represented using different BEASY contact conditions. In the second example, a more detailed connection joint is represented.

2 Methodology In this paper, the analysis uses the Boundary Element Method rather than the usual finite element method to analyse the behaviour of the structures involved. The Boundary Element Method (BEM) has significant advantages for the modelling of fracture and is also ideally suited for the analysis of multi-body contact. In fracture analysis, the crack is a new surface that is part of the structure and as the crack grows this surface changes and evolves. In the BEM analysis of cracks [2-7], only the surface of the crack itself needs to be modelled. As the crack grows the surface mesh on the crack and the surrounding area is modified, representing the growth of the crack. The simplicity of the boundary element mesh required to represent behaviour of cracks is a major advantage when using BEM for fracture analysis. The method allows very refined meshing near the crack front without any difficulty at all. By contrast FE meshes often suffer from the unwanted side-effect that the refinement tends to propagate through the volume of the structure. The simplicity provided by the BEM can especially be appreciated during the growth of a crack, since the only parts of the model that are affected are the surface mesh on the crack and the mesh on the immediately adjacent surfaces. In the examples presented in this paper, multi-body contact analysis is also included. This type of analysis is also ideal suited to the BEM methodology as contact simulations require information about the surface behaviour of the model. In boundary element analysis, the displacements and stresses are computed directly on the surface with no need to extrapolate from internal results. Some other methods derive stresses by differentiation of displacement, and often calculate results at “Gauss points” inside the volume, thereafter extrapolating to the surface. It is well known that differentiation tends to dilute accuracy, reducing the accuracy for the contact simulation. In the examples presented here the BEASY Fatigue and Crack Growth software is used to model and analyse the structures. All computed stress and SIF results are provided using this software.

Improved SIF Calculation in Riveted Panel Type Structures

349

3 Lap Joint Model Model description In the first example a simple lap joint model has been defined as show in Figure 1. This structure has two flat plates connected by 4 connecting pins. The lower plate is restrained, whilst the upper plate has a traction load applied.

Fig. 1 Simple lap joint geometry.

The two plates are connected by “slider” elements on the shared interface. Pins are used to join the plates together and contact conditions are applied at each of the pin-panel connections. In this model the contact connections are defined using “conforming contact” – this means that the pins are modelled to fit exactly into the holes in each panel. This type of contact allows users to select the exact contact behaviour at each interface. In the analysis 3 different initial contact conditions have been considered as follows: • • •

Perfect fit pins: Here the pins are represented with a perfect fit to the panels. This is represented with an initial gap of zero in the model Interference fit pins: Here the pins are assumed to be a “push fit” – so that there is an initial pre-stress in the holes. This is represented with a negative initial gap. Clearance fit pins in one zone: Here the connection between the pins and the restrained plate are modelled with a clearance fit, while those in the loaded plate are represented with a perfect fit.

Initial crack added to model The model has a single half-penny shaped crack positioned on the side of one of the interface holes at the point with peak maximum principal stress value. The crack has been added into the lower, restrained plate as shown in Figure 2, which also shows contours of stress.

350

S.C. Mellings, J.M.W. Baynham, and T.J. Curtin

Fig. 2 Stresses around half-penny-shaped crack in bore of rivet hole in the lap joint assembly.

Stress intensity factors caused by the three different contact conditions The SIFs are computed directly from the stresses and displacements in the BEM analysis by using a decomposition of the J-Integral, to give the three required SIF values[8-9]. Variation along the crack front of the mode 1, 2 and 3 stress intensity factors for the three different contact assumptions are shown in Figures 3 to 5.

Fig. 3 Variation along the crack front of Mode 1 Stress Intensity Factor (K1) for different assumed rivet sizes.

Improved SIF Calculation in Riveted Panel Type Structures

351

Fig. 4 Variation along the crack front of Mode 2 Stress Intensity Factor (K2) for different assumed rivet sizes.

Fig. 5 Variation along the crack front of Mode 3 Stress Intensity Factor (K3) for different assumed rivet sizes.

352

S.C. Mellings, J.M.W. Baynham, and T.J. Curtin

4 Beam Fastener Model Model description The component geometry shown in Figure 6 consists of a beam cap fastened by rivets to a lower panel and vertical web.

Fig. 6 Mechanically fastened structural panel geometry.

A simulation model was created using a short section (3.2 inch) of this geometry to represent the load interaction between the various mechanically fastened components when load is transferred from the beam cap to the lower panel and vertical web. Cracks of different size were inserted into the model and stress intensity factor (SIF) solutions were obtained. This approach offers significant advantage in terms of accuracy compared to current handbook based solutions. Loading and Boundary Conditions The model shown in Figure 7 is a boundary element surface mesh. Only 7500 elements were required to accurately model the three structural plates and 12 rivets. A perfect fit (i.e. no initial gap) and frictionless contact boundary conditions were applied at all rivet locations. Rivet contact boundary condition were applied under the rivet heads and along the shank of the rivet. Applying contact boundary conditions to all the rivet surfaces was numerically more intensive than for example the methods used in the study by Evans[1] but provides a different insight into the actual load transfer mechanisms active in this particular structural assembly. A rivet pre-stress was not applied in this modelling exercise although it could easily be implemented through a simple change in the contact gap parameter applied under the rivet head.

Improved SIF Calculation in Riveted Panel Type Structures

353

A slider boundary condition (allowing in-plane displacement without any transfer of shear stress) was applied on the contacting regions between the beam-lower panel and beam-vertical web surfaces. This is a numerically more efficient approach compared to an iterative contact solution and may be appropriate when inplane loading is the dominant mode of stress transfer.

Fig. 7 BEASY Model Geometry Showing Loading and Restraint.

A traction of 10,900 psi is applied to one end of the beam cap as shown in Figure 7. The panel/web sections are restrained at the other end to prevent axial displacement. Additional restraints, provided to prevent rigid body motion, are not shown in the figure. Rivet Loading Load transfer through the rivets from the beam to the panel (and similarly to the web) is automatically taken care of in the model, including redistribution of load as the crack grows. However, it is assumed that when a ligament fails the local rivet will no longer transfer load, and the rivet is therefore removed from the model. The by-pass loading ratio occurring in the model is the real value, which corresponds to the geometry, loading, and crack location and size. It would of course be possible to model a single panel and to load it with both remote traction and rivet pin loads, to determine effects of specific by-pass ratios, but that process is not covered in this paper. Crack location The various cracks assumed in the beam cap are in the plane of the rivets shown in red in Figures 8 and 9. The first of these figures shows the rivets (in the plane of the crack) which are transferring load in crack cases 1 and 2, while the second shows those which are transferring load in crack cases 3 and 4.

354

S.C. Mellings, J.M.W. Baynham, and T.J. Curtin

Fig. 8 Rivets in the plane of the crack which are assumed to be transferring load for Crack Cases 1 & 2 (4 rivets active in cross section).

Fig. 9 Rivets in the plane of the crack which are assumed to be transferring load for Crack Cases 3 & 4 (3 rivets active in cross section).

Improved SIF Calculation in Riveted Panel Type Structures

355

Crack Case Scenarios and SIF Solutions Four different crack sizes were studied ranging from a 0.1 inch radius corner crack to a 0.6 inch through crack with an adjoining failed ligament (Figure 10). A rivet was removed from the model once a crack grew to a size large enough to completely fracture a ligament in the beam. The coupled contact and SIF solution method used improved the understanding of the impact of load redistribution on SIF values as crack lengths were increased in size.

Fig. 10 Different Crack Cases Investigated.

A plot of the stress intensity factor solutions (KI) along the crack front is shown in Figure 11. The plot in Figure 11 indicates the magnitude of KI increases as the initial corner crack (Crack Case 1) grows to the point of becoming a through crack (Crack Case 2). However once the remaining ligament in the beam fails and a secondary corner crack initiates (Crack Case 3) there is a reduction in the magnitude of KI . The corresponding break in the beam ligament at this location also results in a redistribution of load. It can be seen from Figure 12 that the rivet holes closest to the beam connection with the lower panel experience an increase in stress whereas the upper row of rivets perpendicular to the crack surface show a corresponding decrease in stress. As this secondary corner crack then continues to grow eventually becoming a large through crack (Crack Case 4) there is another phase of increasing KI as this crack front approaches a second highly stressed rivet hole in the beam.

356

S.C. Mellings, J.M.W. Baynham, and T.J. Curtin

Crack Case 1

Crack Case 2 Crack Case 3

Crack Case 4

Fig. 11 SIF Solutions for Crack Cases 1 - 4 (K Solutions for combined bearing and by-pass load).

PSI Crack Case 1

Crack Case 3

Crack Case 2

Crack Case 4

Fig. 12 Change in Maximum Principal Stress on Beam for Different Crack Cases.

Improved SIF Calculation in Riveted Panel Type Structures

357

Benefits of BEASY SIF Solution Method The coupled contact and SIF solution methodology used in this work automatically accounts for the interrelationship between the contact stress state and increasing component compliance caused by an opening crack. This capability provides an added degree of accuracy in terms of analyzing mechanically fastened plates since the presence of a crack at the edge of contact may alter the contact solution. It is likely that the compliance of the component will change once a crack begins to open and propagate and this will impact the contact length and contact stress state (i.e. pressure and shear traction). This effect is not accounted for using simplified fracture mechanics approaches such as generalized weight functions or the method of distributed displacement discontinuities. Previous work by the authors also suggests that friction can dramatically influence the crack growth life estimates. Although this analysis assumed frictionless contact the modelling method used could be easily adapted to investigate the impact of frictional loading in the rivet holes.

References [1] Evans, R., Gravina, R., Heller, M., Clarke, A., Rock, C., Burchill, M.: In: Rouchon, J. (ed.) Proceedings of the Aircraft Airworthiness & Sustainment Conference, vol. I, pp. 124–138. Cépaduès, Toulouse (2010) [2] Mellings, S., Baynham, J., Adey, R.A.: Advances in crack growth modelling of 3D Aircraft Structures. In: International Committee on Aeronautical Fatigue, Rotterdam, Netherlands (May 2009) [3] Mellings, S., Baynham, J., Adey, R.A., Curtin, T.: Durability Prediction Using Automatic Crack Growth Simulation. In: International Committee on Aeronautical Fatigue, Toulouse, France (June 2001) [4] BEASY User Guide, Computational Mechanics BEASY Ltd, Ashurst, Southampton, UK (2011) [5] Portela, A., Aliabadi, M.H., Rooke, D.P.: The Dual Boundary Element Method: Efficient Implementation for Cracked Problems. International Journal for Numerical Methods in Engineering 32, 1269–1287 (1992) [6] Mi, Y., Aliabadi, M.H.: Three-dimensional crack growth simulation using 6EM. Computers & Structures 52(5), 871–878 (1994) [7] Neves, A., Niku, S.M., Baynham, J.M.W., Adey, R.A.: Automatic 3D crack growth using BEASY. In: Proceedings of 19th Boundary Element Method Conference, Computational Mechanics Publications, Southampton, pp. 819–827 (1997) [8] Rigby, R., Aliabadi, M.H.: Mixed-mode J-integral method for analysis of 3D fracture problems using BEM. Engineering Analysis with Boundary Elements 11(3), 239–256 (1993) [9] Rigby, R., Aliabadi, M.H.: Decomposition of the mixed-mode J-integral—revisited. International Journal of Solids and Structures 35(17), 2073–2099 (1998)

26th ICAF Symposium – Montreal, 1-3 June 2011 An 1D-Beam Approach for Both Stress Analysis and Fatigue Life Prediction of Bonded Joints E. Paroissien, A. Da Veiga, and A. Laborde SO GE TI HIGH TECH, TRPE, PE6, Blagnac, France

Abstract. An approach for both stress analysis and fatigue life prediction of bonded joints, based on a 1D-beam model, is presented. Only the adhesive is supposed to fail. The Goland and Reissner framework [1] is extended to unbalanced laminar or monolithic adherends under thermal loads. The J-integral is derived and employed in a modified Paris law, leading to fatigue lives, which are assessed w.r.t. published experimental results [2, 3].

1 Introduction In the frame of the structural component design, bonding can be considered as a suitable assembly method or an attractive complement to conventional ones as mechanical fastening. Bonding offers the possibility of joining without damaging various materials, such as plastics or metals, as well as various combinations of materials. This first advantage is reinforced by a large choice of adhesive families and by the possibility to formulate adhesives to meet at best the joint specifications. Compared to bolting, bonding shall allow for mass benefits, since the continuous distribution of load transfer all over the overlap implies that additional concentrated materials are not required to sustain loads. Nevertheless, the main restriction to a more widespread application of bonding could be the lack of assessment ability of its reliability. To our knowledge, non destructive test methods allow for detecting possible adhesive absences but not the adhesion absences. As a result, to control the design of bonded joints, it is necessary to predict its strength, including both stress and fracture analyses. In this paper, a 1D-beam approach, allowing both for stress analysis and fatigue life prediction of bonded joint, is presented. Only the adhesive is supposed to fail. The single-lap bonded joint described in [2] (see Figure 1) allows for exemplifying the approach. Firstly, a general 1D-beam model for bonded joint stress analysis is presented. The model can be related to the Goland and Reissner framework [1], which is extended by considering unbalanced overlaps made of laminated monolithic beams under thermal loading (thermal mismatch effect). The computation method [4], inspired by the finite element method (FEM), enables solving the full set of equations. It is based on the analytical formulation of macro-element with four nodes, called bonded-beams (BB) element, able to simulate an entire bonded overlap. The model provides the distribution in the adherends of normal displacements, deflections and bending angles and of normal forces, shear forces and bending moments, as well as the distribution of adhesive shear and peeling stresses along the overlap.

360

E. Paroissien, A. Da Veiga, and A. Laborde

Elements of validation are then presented, in order to show that same hypotheses lead to same results. Secondly, the presented approach is employed to predict fatigue life of bonded joints, through elementary manipulations consisting in the introduction of adhesive cracks at both overlap ends. A modified form of Paris law [2, 3] allows for linking the fatigue cycle crack growth rate and the maximum energy release rate per cycle. The maximum energy release rate is related to the computation of the J-integral, the analytical simplified expression of which is derived, based on [5-7], in the presented framework. The approach is assessed with regard to experimental fatigue test results on isotropic balanced single-lap bonded joints, provided in [2, 3]. A way to simply approximate the thermal mismatch effect is suggested and remains to be assessed. adherend 1

b

25.4 mm

e

0.4 mm

ei

2 mm

L

12.7 mm

li

76.2 mm

E

3 GP a

Ei

210 GPa

!

0.4

!i

0.35

b=width

adherend 2

adhesive

y1 e1

x y2

e2

e l1

beam element

L

BB element

x

l2

beam element

Fig. 1 Idealization of a single-lap bonded joint with of beam and BB elements. Geometrical and mechanical parameters [2].

2 1d-Beam Model for Bonded Joint Stress Analysis Overview of the approach [4] The presented approach allows for the resolution of the set of differential equations. The bonded joint is meshed in elements (see Figure 1). While the parts outside the bonded overlap are simulated by beam elements, the bonded overlap is simulated by a four nodes macro-element, called bonded-beams (BB) element; this macro-element is the model core and is specially formulated. After finding the stiffness matrices of each element type, the stiffness matrix of the full structure – termed K – is assembled. The boundary conditions are then introduced. The vector of displacements – termed U – and the vector of forces – termed F – including the thermal equivalent nodal forces – termed FT – are determined; the stiffness matrix is updated. The resolution consists then in inverting the linear system F=KU.

An 1D-Beam Approach for Both Stress Analysis and Fatigue Life Prediction

361

Hypotheses The model is based on the following hypotheses: (i) the thickness of the adhesive layer is constant along the overlap, (ii) the adherends are considered as linear elastic Euler-Bernoulli laminated or monolithic beams, (iii) the adhesive layer is simulated by a linear two-parameter uncoupled elastic foundation and consists thus in a continuously distributed layer of shear and transverse normal springs, (iv) the temperature is uniformly distributed on the adherends. In particular, the hypothesis (iii) implies that the adhesive stress field is reduced to the shear and peeling stress only, constant in the adhesive thickness. A quasi-static analysis is considered. Formulation of BB element Governing equations. The subscript i refers to the ith adherend; i=1,2. Each adherend is associated to a local referential x, yi, zi (see Figure 1); the origin of which is located at its neutral line; the neutral line is oriented according to an x-axis, while the y-axis is defined according to its thickness. In the frame of the classical Euler-Bernoulli model of beams, the assumed displacement field is under the shape:

ui ' ( x , y j ) = ui ( x ,0 ) − yi

(

)

dwi = ui − yiθ i ; wi ' x , y j = wi (x ) dx

(1)

where ui’ and wi’ are the displacement of any points of the ith adherend crosssection according to the x- and yi-axis, respectively; ui and wi are the displacement of points located at the ith adherend neural line according to the x- and yi-axis, respectively; θi is the bending angle. By taking into account the thermal strain due to a variation of temperature ΔT, the tensile stress can be expressed as:

⎡ du i

σ i = Ei ( yi )⎢

⎣⎢ dx

− yi

d 2 wi dx

2

⎤ − α i ( yi )ΔT ⎥ ⎦⎥

(2)

where Ei is the Young’s modulus and αi the thermal expansion coefficient. The integration on the cross-section of tensile stresses and elementary bending moments induced by these tensile stresses allows for the computation of the normal force Ni and bending moment Mi: ⎧ du i ⎧ dui d 2 wi ⎪ N i = Ai = Di N i + Bi M i − (Bi M Ti − Di N Ti )ΔT − Bi − N Ti ΔT ⎪ Δi dx dx ⎪ ⎪ dx 2 ⇔ ⎨ ⎨ 2 dui d 2 wi ⎪ ⎪Δi d wi = Ai M i + Bi N i + (Bi N Ti − Ai M Ti )ΔT + M Ti ΔT ⎪M i = − Bi dx + Di ⎪⎩ 2 dx 2 dx ⎩

(3)

where Ai is the extensional stiffness, Di is the bending stiffness, Bi the extension bending coupling stiffness, NTi is the thermal force per °K, and MTi is the bending moment per °K, Δi=AiDi-Bi²≠ 0.

362

E. Paroissien, A. Da Veiga, and A. Laborde

The local equilibrium of adherends is performed according to [1] (see Figure 2):

dN i dVi = (− 1)i T ; = (− 1)i +1 S ; bdx bdx

dM i e + Vi + i bT = 0 2 dx

(4)

where b is the overlap width and Vi is the shear force. V 1 (x+dx)

M 1 (x+dx)

bdxS

N 1 (x)

V 2(x+dx)

bdxT

M 2 (x+dx) M1 (x)

N1 (x+dx)

N2 (x)

bdxT V 1 (x)

N 2(x+dx) M 2 (x)

bdxS

V 2 (x)

Fig. 2 Free body diagrams of infinitesimal adherend elements of the overlap.

The adhesive shear stress T and the adhesive peeling stress S are then given by:

T = Gγ =

e e G⎛ E ⎞ ⎜ u 2 − 2 θ 2 − u1 − 1 θ 1 ⎟; S = Eε = (w1 − w2 ) e⎝ 2 2 ⎠ e

(5)

where G and E are the adhesive Coulomb’s and Young’s moduli. In the case of an enclosed adhesive layer, the effective Young’s modulus could be used instead of the Young’s modulus. System of differential equations in terms of adhesive stresses. By combining Eqn. 3, Eqn. 4 and Eqn. 5, the following differential equation system is obtained in terms of adhesive stresses:

d 3T dx

3

= k1

dT + k2 S ; dx

d 4S dx

4

= −k 4 S − k 3

dT dx

(6)

where the constants are: k1 =

⎡ A e 2⎞ D ⎛ A e 2 ⎞ ⎛e B e B Gb ⎢ D1 ⎛⎜ 1 + 1 1 ⎟ + 2 ⎜ 1 + 2 2 ⎟ + ⎜⎜ 1 1 − 2 2 Δ2 e ⎢ Δ1 ⎜ 4 D1 ⎟ Δ 2 ⎜ 4 D2 ⎟ ⎝ Δ 1 ⎝ ⎠ ⎝ ⎠ ⎣

Gb ⎡ e1 A1 e 2 A2 ⎛ B1 B 2 − +⎜ + k2 = ⎢ e ⎢⎣ 2 Δ1 2 Δ2 ⎜⎝ Δ 1 Δ 2

⎞⎤ Eb ⎡ A1 A2 ⎤ ⎟⎥; k 4 = + ⎢ ⎥ ⎟⎥ e ⎣ Δ1 Δ2 ⎦ ⎠⎦

⎞⎤ Eb ⎡ e1 A1 e 2 A2 ⎛ B1 B 2 ⎟⎥ ; k 3 = − +⎜ + ⎢ ⎟ e ⎢⎣ 2 Δ1 2 Δ2 ⎜⎝ Δ 1 Δ 2 ⎠⎦⎥

⎞⎤ ⎟⎥ ⎟ ⎠⎦⎥

(7)

This system of differential equations in terms of adhesive stresses can be uncoupled by consecutive differentiations and combinations as:

An 1D-Beam Approach for Both Stress Analysis and Fatigue Life Prediction

⎧d 6S d 4S d 2S + S ( k 2 k 3 − k1 k 4 ) = 0 ⎪ 6 − k1 4 + k 4 dx dx 2 ⎪ dx ⎨ 6 4 2 ⎪ d ⎛⎜ d T − k d T + k d T + T ( k k − k k ) ⎞⎟ = 0 1 4 2 3 1 4 ⎟ ⎪ dx ⎜ dx 6 dx 4 dx 2 ⎠ ⎩ ⎝

363

(8)

The Cardan’s method is employed to solve the characteristic equation of the differential equation system in Eqn. 8 (see Appendix A) and find its root r² and (s±it)² – r, s and t are positive real numbers – so that the adhesive shear and peeling stress are given by:

⎧ ⎡ K e sx sin(tx) + K 2 e sx cos(tx) + K 3 e − sx sin(tx)⎤ ⎪S ( x) = ⎢ 1 ⎥ rx − sx − rx ⎪⎪ ⎣⎢+ K 4 e cos(tx) + K 5 e + K 6 e ⎦⎥ ⎨ − ⎡ K1 e sx sin(tx) + K 2 e sx cos(tx ) + K 3 e sx sin(tx)⎤ ⎪ ⎥ ⎪T ( x) = ⎢ ⎢⎣+ K 4 e − sx cos(tx) + K 5 e rx + K 6 e − rx + K 7 ⎥⎦ ⎪⎩

(9)

Nodal displacements and forces. The computation of the BB element stiffness matrix takes place through the determination of nodal displacements and forces (see Figure 3). The second term of equivalency in Eqn. 3, together with Eqn. 4, allows uncoupling the expressions of derivatives of u1, u2, w1 and w2 (and then θ1 and θ2) as a function of linear combinations of adhesive stress derivatives and polynomial expressions; following the resolution scheme in [8], the total number of independent integration constants can be reduced to 12: ⎧ 2 2 ⎪u ( x ) = β~ T + β dS − bL K 7 − 6 B1 J 0 L ⎛ x ⎞ + J x + J ⎟ ⎜ 1 1 1 5 6 ⎪ L dx 2 A1 ⎝L⎠ ⎪ 2 ⎪ J bL2 K 7 + 6 B2 J 0 L ⎛ x ⎞ ~ ⎛ ⎞x ⎪u 2 ( x ) = β 2 T + β 2 dS + ⎜ ⎟ + ⎜⎜ J 5 + 1 ( e1 + e 2 ) ⎟⎟ L dx 2 A2 ⎪ ⎝L⎠ ⎝ ⎠L ⎪ e e J ~ ~ ⎛ ⎞ ⎪ + J 6 + 2 ( e1 + e2 ) − K7 ⎜⎜ 1 β 5 + 2 β 6 ⎟⎟ ⎪ 2 2L ⎝ 2 ⎠ ⎪ 3 2 ⎪ 2 ⎞ ⎛ ~ d S dT x ⎪ ⎟ + β S + J ⎛⎜ ⎞⎟ + J ⎛⎜ x ⎞⎟ + J x + J + k2 ⎨w1 ( x ) = β 3 ⎜⎜ k 4 5 0 1 2 3 dx L ⎝L⎠ ⎝L⎠ dx 2 ⎟⎠ ⎪ ⎝ ⎪ 3 2 ⎪ ~ ⎛ d 2 S ⎞⎟ dT x ⎛x⎞ ⎛x⎞ + k2 + β6 S + J 0 ⎜ ⎟ + J 1 ⎜ ⎟ + J 2 + J 3 ⎪w2 ( x ) = β 4 ⎜⎜ k 4 2 ⎟ dx L L⎠ L⎠ ⎝ ⎝ dx ⎪ ⎝ ⎠ ⎪ ~ ~ J2 x2 dS x ⎪ ⎪θ 1 ( x ) = β 5 T + β 5 dx + 3 J 0 3 + 2 J 1 2 + L − K 7 β 5 L L ⎪ ⎪ 2 J ~ ~ x dS x ⎪θ 2 ( x ) = β 6 T + β 6 + 3J 0 + 2J 1 + 2 − K7 β 6 dx L L3 L2 ⎩⎪

(10)

364

E. Paroissien, A. Da Veiga, and A. Laborde

with: ⎧~ ⎪β 1 ⎪ ⎪~ ⎪β 2 ⎪ ⎪~ ⎪β 5 ⎪ ⎪ ⎪β~ ⎨ 6 ⎪ ⎪~ ⎪β 3 ⎪ ⎪ ⎪ ⎪ ⎪J 0 ⎪ ⎪ ⎩

=

C k − D10 k 1 C 10 k 4 − D10 k 3 ; β 1 = 10 2 k 1k 4 − k 2 k 3 k 1k 4 − k 2 k 3

=

C k − D 20 k 1 C 20 k 4 − D 20 k 3 ; β 2 = 20 2 k 1k 4 − k 2 k 3 k 1k 4 − k 2 k 3

=

A k − B10 k 1 A10 k 4 − B10 k 3 ; β 5 = 10 2 k 1k 4 − k 2 k 3 k 1k 4 − k 2 k 3

A k − B 20 k 1 A20 k 4 − B 20 k 3 ; β 6 = 20 2 k1k 4 − k 2 k 3 k 1k 4 − k 2 k 3 ~ ~ β5 β6 ~ ; β4 = = k 1k 4 − k 2 k 3 k1k 4 − k 2 k 3 ⎛ 1 1 ⎞ ⎟ bL3 ⎜⎜ + ⎟ ⎝ A1 A2 ⎠ K7 = ⎛ B1 B 2 ⎞ ⎜ ⎟ 3( e1 + e 2 ) + 6 ⎜ − ⎟ ⎝ A1 A2 ⎠ =

⎧ ⎪ A10 ⎪ ⎪ ⎪ A20 ⎪ ⎨ ⎪C ⎪ 10 ⎪ ⎪C ⎪ 20 ⎩

b [2 B1 + e1 A1 ] ; B10 = − A1b Δ1 2 Δ1 A2 b b = [2 B2 − e2 A2 ] ; B20 = Δ2 2 Δ2

=−

=− =

b 2 Δ1

[e1 B1 + 2 D1 ] ; D10 = − B1b

(11)

Δ1

b [− e2 B2 + 2 D2 ] ; D20 = B2 b Δ2 2Δ2

The 12 nodal displacements are then the values at x=0 and x=L of the previous expressions of displacements, as a function of 12 integration constants. The relationship U=MC can be written in the form of a matrix, where C is the integration constant vector. By introducing Eqn. 10 in Eqn 3 and with Eqn. 4, the normal and shear forces and the bending moment in both adherends can be computed as a function of the 12 integration constants (Eqn. 12), leading to the expressions of nodal forces (see Figure 3), which can be written F=NC at ΔT=0°K. ⎧ J J dT d 2S ⎪ N 1 ( x ) = a~1 + a1 − bK 7 x − 2 B 1 21 + A1 5 − N T 1 ΔT L dx L dx 2 ⎪ ⎪ 2 J e +e ⎞ 2 B dT d S ⎛ 2 ~ ⎪N 2 ( x ) = a 2 + a2 + bK 7 x + J 1 ⎜⎜ − 2 + A2 1 2 2 ⎟⎟ + A2 5 − N T 2 ΔT ⎪ L dx L dx 2 ⎠ ⎝ L ⎪ 2 ⎞ ⎛ 6 J J B J Δ dT d S x ⎪ ~ + ⎜⎜ 20 1 + bLK 7 1 ⎟⎟ + 2 D1 21 − B 1 5 + M T 1 ΔT ⎪ M 1 ( x ) = a 3 dx + a 3 2 A1 ⎠ L L ⎝ L A1 L dx ⎪ 2 ⎪⎪ ⎞ ⎛ 6 J ~ dT + a d S + x ⎜ 0 Δ2 − bLK B 2 ⎟ ⎨M 2 ( x ) = a 4 4 7 dx A2 ⎟⎠ dx 2 L ⎜⎝ L2 A2 ⎪ ⎪ e +e ⎞ J ⎛ 2D ⎪ + J 1 ⎜⎜ 2 2 − B 2 1 2 2 ⎟⎟ − B 2 5 + M T 2 ΔT ⎪ L L ⎠ ⎝ L ⎪ 2 3 ⎛ 6 J B1 ⎞ e 1 b Δ d S 1 d T ⎪ ~ ⎟ ⎜ 0 1 ⎪V1 ( x ) = − a 3 dx 2 − a 3 dx 3 − L ⎜ L2 A + bLK 7 A ⎟ − 2 T 1 1 ⎠ ⎝ ⎪ ⎪ B ⎞ e b d 3 S 1 ⎛ 6 J 0 Δ2 d 2T − a4 − ⎜⎜ 2 − bLK 7 2 ⎟⎟ − 2 T ⎪V 2 ( x ) = − a~4 2 3 L A A2 ⎠ 2 ⎪⎩ dx dx 2 ⎝ L

(12)

An 1D-Beam Approach for Both Stress Analysis and Fatigue Life Prediction

365

with: ⎧a~1 = A1 β~1 − B1 β~5 ; a1 = A1 β 1 − B1 β 5 ⎪ ~ ~ ⎪⎪a~2 = A2 β 2 − B2 β 6 ; a 2 = A2 β 2 − B2 β 6 ⎨~ ~ ~ ⎪a3 = − B1 β 1 + D1 β 5 ; a3 = − B1 β 1 + D1 β 5 ⎪~ ~ ~ ⎪⎩a4 = − B2 β 2 + D2 β6 ; a4 = − B2 β 2 + D2 β 6 wi

w 1 (x)

&i

& 1 (x)

ui

wk

&k

u 1 (x)

uk

node i

(13)

M 1 (x)

V1 (x) node i

node k

Si

Ri

N 1 (x)

Sk

Rk

node k Qi e

node j

Qk

e

Qj

Ql

node l

wj

w 2 (x)

&j

& 2 (x)

wl

&l

Sl Sj

Rj

M 2 (x)

V2 (x)

Rl

node j u 2 (x)

uj dh x

y 0

ul

node l N 2 (x)

y x

x

L

0

x

L

Fig. 3 Bonded-beams element: a four-nodes macro-element with three degrees of freedom per node (uα, wα, θα)α=i,j,k,l.

Stiffness matrix of BB element. The coefficients of the stiffness matrix are obtained by differentiating each nodal force by each nodal displacement. Of course, the stiffness matrix is not modified by the consideration of a thermal load:

K BB

⎛ ⎡ ∂Qσ ⎜⎢ ⎜ ⎣ ∂uτ ⎜⎡ ∂R =⎜⎢ σ ⎜ ⎣ ∂uτ ⎜ ⎜ ⎡ ∂Sσ ⎜ ⎢ ∂u ⎝⎣ τ

⎤ ⎡ ∂Qσ ⎤ ⎡ ∂Qσ ⎤ ⎞ ⎥ ⎢ ⎥ ⎢ ⎥⎟ ⎦ ⎣ ∂wτ ⎦ ⎣ ∂θτ ⎦ ⎟ ⎤ ⎡ ∂Rσ ⎤ ⎡ ∂Rσ ⎤ ⎟⎟ ∂K BB =0 ⎥ ⎢ ⎥ ⎢ ⎥ ⎟ ,σ ,τ = i , j , k ,l ⇒ ∂ΔT ⎦ ⎣ ∂wτ ⎦ ⎣ ∂θτ ⎦ ⎟ ⎤ ⎡ ∂Sσ ⎤ ⎡ ∂Sσ ⎤ ⎟ ⎥ ⎢ ⎥ ⎢ ⎥⎟ ⎦ ⎣ ∂wτ ⎦ ⎣ ∂θτ ⎦ ⎠

(14)

The twelve nodal displacements (uγ, γ = 1:12) and the twelve nodal forces (Qα, α = 1:12) are expressed as functions of the twelve independent integration constants (Cβ, β = 1:12) at ΔT=0°K. The nodal forces depend linearly on integration constants as well as the nodal displacements. Thus, the integration constants depend linearly on the nodal displacements (Eqn. 15), enabling the determination of 144 coefficients of KBB (Eqn. 16):

366

E. Paroissien, A. Da Veiga, and A. Laborde 12

Qα =

∑ β

12

nαβ C β

and

Cβ =

=1

∂Qα = ∂uδ

m' βγ uγ ∑ γ

(15)

=1

∂uγ

∑ nαβ ∑ m' βγ ∂uδ

(16)

But: 12 ⎧1 if γ = δ ∂Q ∂uγ ⇒ α = nαβ m' βδ = δ γδ = ⎨ ∂uδ ∂uδ ⎩0 if γ ≠ δ β =1

∑

(17)

m' βδ = C β ( uδ = 1 , uγ ≠δ = 0 ) = C β ( 0 ,…0 ,uδ = 1,0 ,… ,0 )

The coefficients of KBB are thus obtained through:

[K BB ]α ,δ

∂Qα = nαβ C β ( 0 …0 ,uδ = 1,0 …0 ) ∂uδ β =1 12

=

∑

(18)

Practically, Cβ(0…0,uδ=1,0…0) is automatically generated by looping on the twelve canonical vectors of displacement, through the following inversion Cβ[(0…0,uδ=1,0…0)]=M-1(0…0,uδ=1,0…0). In other words, the stiffness matrix of the BB element KBB is such that F=KBBU. With U=MC, this becomes F=KBBMC; thus KBB=NM-1, since F=NC at ΔT=0°K. Resolution

Stiffness matrix of the single-lap bonded joint. The single-lap bonded joint (for example) stiffness matrix is then assembled, using the FEM conventional assembly rules. The beam stiffness matrix is provided in Appendix B. The total number of nodes is 6, resulting in a total number of 6*3=18 degrees of freedom (DoF). Equivalent thermal nodal forces for the BB element. The thermal load is classically transformed in terms of equivalent thermal nodal forces, resulting in the same displacements caused by the actual thermal loads. This does not change the element stiffness matrices. In the case of the BB element, the equivalent thermal nodal force vector can be computed as (without any transverse temperature gradient): ⎞ ⎛L FT = ⎜ B t dx ⎟(N T 1 ΔT ⎟ ⎜ ⎠ ⎝0

∫

N T 2 ΔT

0 0 0 0 )t

(19)

An 1D-Beam Approach for Both Stress Analysis and Fatigue Life Prediction

367

where: ⎧ E t = ( BU )t ⎪ ⎨ t ⎛ du1 du 2 ⎪ E = ⎜⎝ dx dx ⎩

0 0

dθ 1 dx

(

)

−1 t ⇒ B = HM (20) dθ 2 ⎞ t ⎟ = (HC ) = HM −1U dx ⎠

Boundary conditions. The stiffness matrix is then classically reduced by removing rows and lines, which correspond to fixed – thus known – DoF. Various boundary conditions can be easily applied, such as those for the simply supported (u=w=0 at one joint end and w=0 at the other joint end, leading to a total number of 15 DoF; see Figure 4) or clamped (u=w=θ=0 at one joint end and w=θ=0 at the other joint end, leading to a total number of 13 DoF). The vector of nodal force is then constructed taking into account the applied mechanical forces and replacing the thermal load by the equivalent nodal thermal forces.

w=0

Δ T

u=w=

Fig. 4 Simply supported boundary conditions and applied loads.

Computation. A computer programme, implemented in SCILAB [9], was produced to solve the analysis. The resolution consists simply in the computation of the nodal displacement vector U=K-1F, allowing for the determination of the integration constant vector. The adherend displacements, rotations, forces and moments, and adhesive shear and peeling stresses can be then deduced at any abscissa. Elements of validation

Goland and Reissner. The adhesive stress distribution predicted by the Goland and Reissner theory [1] are compared to the model predictions for the single-lap bonded joint defined in Figure 1 and Table I. In order to perform a comparison on exactly the same hypotheses, the length outside the overlap is chosen equal to 59.66 mm, resulting in a same bending factor of 0.9038 (for a beam approach) at an applied force of 1 kN and simply supported boundary conditions. The superimposition of curves shown in Figure 5 allows for the conclusion that the same hypotheses lead to the same results. Thermal loading. In order to evaluate the adhesive stress distributions predicted by the present model under a pure thermal loading, a FE model of a single-lap bonded joint is developed using the SAMCEF FE code [10] to be as close as

368

E. Paroissien, A. Da Veiga, and A. Laborde

possible to the present model. Indeed, the adherends are simulated by beam elements; in the overlap region, they are connected through springs working in shear and transverse tensile mode in order to simulate the adhesive layer; both stiffnesses of these springs are assessed according to [11]. The computation is linear (geometry and materials). The simply supported boundary conditions are chosen. The geometrical and mechanical parameters are given in Table I, some of which are replaced by E1=72 GPa and ν1=0.33; moreover: α1=24.10-6 °K-1 and α2=12.10-6 °K-1. A very good agreement is shown. 7

shear - Goland & Reissner shear - present model peeling - Goland & Reissner peeling - present model

adhesive stress in MPa

6 5 4 3 2 1 0 -1 -2 0

0.2

0.4

0.6

0.8

1

normalized overlap abscissa

Fig. 5 Comparison of adhesive stresses predicted by Goland and Reissner theory by the present model under a pure mechanical loading (f=1 kN; ΔT=0°K).

shear - by FEM shear - present model peeling - by FEM peeling - present model

adhesive stress in MPa

12 8 4 0 -4 -8 -12 0

0.2

0.4 0.6 normalized overlap abscissa

0.8

1

Fig. 6 Comparison of adhesive stresses predicted by a FE model and by the present model under a pure thermal loading (ΔT=100°K; f=0 N).

An 1D-Beam Approach for Both Stress Analysis and Fatigue Life Prediction

369

4 Fatigue Life Prediction Method for crack growth prediction under fatigue load cycle

The presentation of the method is performed on a single-lap bonded joint configuration, for which a crack in the adhesive of length is present at both ends of the adhesive. The idealization of this balanced cracked single-lap bonded joint is illustrated in Figure 7: the bonded overlap length is reduced of 2a and each length outside the overlap is increased of a. Elementary modifications of the structure stiffness matrix are thus involved.

l1

a

L-2a

a

l2

Fig. 7 Idealization of a single-lap bonded joint, cracked at both overlap ends.

Modified Paris law

The fatigue cycle crack growth rate is related to the maximum energy release rate, through the modified Paris law employed in [2, 3] (Eqn. 21). D, n1, n2, n are material parameters, Gth is the threshold strain energy release rate, Gc is the critical strain energy release rate, a0 is the Griffith flaw size, af is the crack length at the final failure, Nf is the number of cycles at failure and Gmax is the maximum strain energy release rate applied in a fatigue cycle. If Gmax is known, the fatigue life can be computed by numerical integration (e.g. rectangle method).

⎡ ⎛ G ⎢ 1 − ⎜ th da ⎢ ⎜⎝ Gmax =⎢ dN ⎢ ⎛ G max ⎢ 1 − ⎜⎜ G ⎣ ⎝ c

n ⎡ ⎛ G ⎞1⎤ th ⎟ ⎥ af ⎢1 − ⎜ ⎟ ⎥ ⎜ ⎠ DG n ⇒ N = ⎢ ⎝ Gmax ⎢ max f n ⎥ ⎞ 2⎥ ⎛ Gmax a0 ⎢ ⎟ ⎥ ⎢ 1 − ⎜⎜ G ⎟ ⎠ ⎦ ⎣ ⎝ c

∫

n ⎞1⎤ ⎟ ⎥ ⎟ ⎥ da ⎠ n ⎥ n ⎞ 2 ⎥ DGmax ⎟ ⎥ ⎟ ⎠ ⎦

(21)

Computation of J-integral

According to [5], in the Goland and Reissner framework, if the adherends are considered as beams subjected to low levels of rotation, the adhesive stress field is

370

E. Paroissien, A. Da Veiga, and A. Laborde

assumed constant in the thickness and the adhesive constitutive law are explicit without any dependence on loading history, then the J-integral is nearly pathindependent. Moreover, the J-integral is equal to the product of the joint thickness by the energy density at bond termination, so that the mode I and mode II components, where J=JI+JII, can be approximated by: ε

∫

γ

∫

J I = e Sdε and J II = e Tdγ 0

(22)

0

The J-integral parameters are then computed, based on [6, 7], in the frame of the previous set of governing equations (Eqn. 3 to Eqn. 5) without any thermal strain contributions. The slope of ε with respect to x and the shear force contributions are then neglected. JI and JII are then approximated, as a function of loading conditions, through the computer programme output data:

E JI ≈ 2ek 4

J II

⎛ A1 M 1 + B1 N 1 A2 M 2 + B2 N 2 ⎜⎜ − Δ1 Δ2 ⎝

dε dx

⎞ 2ek 3 ⎛ dM 1 dM 2 ⎞ ⎛ dε ⎞ ⎟⎟ + + ⎜ ⎟d ⎜ ⎟ ( ) k b e e dx ⎠ ⎝ dx ⎠ + 4 1 2 0 ⎝ dx ⎠

∫

⎛ ⎛ D2 N 2 + B2 M 2 D1 N 1 + B1 M 1 ⎞ ⎜⎜ ⎟⎟ − ⎜ Δ2 Δ1 G ⎜⎝ ⎠ ≈ ⎜ 2ek1 ⎜ 1 ⎛ A1 M 1 + B1 N 1 ⎞ 1 ⎛ A2 M 2 + B2 N 2 ⎟⎟ − e2 ⎜⎜ ⎜ − 2 e1 ⎜⎜ Δ1 Δ2 ⎝ ⎠ 2 ⎝ ⎝

⎞ ⎟ ⎟ ⎟ ⎞⎟ ⎟⎟ ⎟ ⎠⎠

(23)

2

(24)

The last term of the right hand side of Eqn. 23 represents the contribution when the joint is unbalanced; it appears difficult to express without any simplifying hypotheses. For balanced cases and B=0, the previous approximations are not required to obtain simple expressions of JI and JII: JI =

J II =

G 2ek1

E 2ek 4

⎡⎛ (M − M ) ⎞ 2 (V − V2 ) dε ⎤⎥ 2 ⎢⎜ 1 ⎟ + 2e 1 D D dx ⎥ ⎢⎣⎝ ⎠ ⎦

⎡⎛ (N − N ) e (M + M ) ⎞ 2 (V + V2 ) γ ⎤⎥ 1 1 2 − 1 ⎢⎜ 2 ⎟ − ee1 1 A 2 D D ⎠ ⎢⎣⎝ ⎥⎦

(25)

(26)

Gmax is then computed as J at the crack tip at the maximal load in a fatigue cycle. The thermal mismatch effect could be related to the thermal loading application, as mechanical loading conditions, through the equivalent thermal nodal

An 1D-Beam Approach for Both Stress Analysis and Fatigue Life Prediction

371

forces; in this way, the simulated thermal mismatch effect is seen as external mechanical work. Results

The model predictions are compared (see Figure 8) to experimental fatigue test result on single-lap-bonded joints (see Figure 1), provided in [2, 3], as well as the Paris law parameters and the Griffith flaw size required (see Table 1). J is computed with Eqn. 21 and Eqn. 22. An encouraging correlation is then shown. Table 1 Paris law parameters and Griffith flaw size employed.

Gc 450 J.m-2

Gth 85 J.m-2

D 3.64.10-20 m²/N cycle

n

n1

n2

5.61

3.20

9.34

a0 85 µm

maximum load per unit width, f/b in N.mm-1

250 model prediction 200

experimental test

150

100

50

0 1.E+03 1.E+04

1.E+05 1.E+06

1.E+07 1.E+08 1.E+09

1.E+10 1.E+11

number of cycles to failure, Nf

Fig. 8 Comparison of fatigue life predicted by the model (Eqn. 25 and Eqn. 26) with experimental test data extracted from [2, 3].

5 Conclusion A 1D-beam approach for both stress analysis and fatigue life prediction of bonded joints is presented. Only the adhesive is supposed to fail. The 1D-beam model is developed in an extended Goland and Reissner framework [1] by considering

372

E. Paroissien, A. Da Veiga, and A. Laborde

unbalanced laminated or monolithic beams under thermal loading. The method employed [4] takes benefit of the flexibility of FE method, since it allows, thanks to a computer programme, both for the resolution of the entire set of equations and for the simple simulation of crack propagation in the adhesive layer through simple manipulations. It is underlined that one macro-element is enough to simulate a full bonded overlap. Simplified expressions of the J-integral parameters are expressed as a function of the load conditions and employed as a fracture criterion. This is then introduced through a modified Paris law for the crack propagation simulation. An encouraging correlation with published [2, 3] experimental results is shown. The thermal mismatch effect could be simply approximated by applying the equivalent thermal nodal forces; it remains to be assessed.

Acknowledgments The authors gratefully acknowledge the SOGETI HIGH TECH engineers and managers – especially the “bolted joint method and research team” – involved in the development of JoSAT (Joint Stress Analysis Tool) internal research program.

References [1] Goland, M., Reissner, E.: J. Appl. Mech. 11, A17–A27 (1944) [2] Curley, A.J., Hadavinia, H., Kinloch, A.J., Taylor, A.C.: Int. J. Fract. 103, 41–69 (2000) [3] Hadavinia, H., Kinloch, A.J., Little, M.S.G., Taylor, A.C.: Int. J. Adhesion Adhesives 23, 463–471 (2003) [4] Paroissien, E., Sartor, M., Huet, J., Lachaud, F.: AIAA J. Aircraft 44(2), 573–582 (2007) [5] Fraisse, P., Schmidt, F.: Int. J. Fract. 63, 59–73 (1993) [6] Tong, L.: Acta Mech. 117, 101–113 (1996) [7] Tong, L.: Int. J. Solids Structures 35(20), 2601–2616 (1998) [8] Högberg, J.L.: Thesis for the degree of licentiate of engineering, Chalmers University of Technology, Göteborg, Sweden (2004) [9] SCILAB, v4.1.2, 23/10/2007, INRIA/ENPC [10] SAMCEF, v13.1-01, 25/06/2009, Samtech Group [11] Dechwayukul, C., Rubin, C.A., Hahn, G.T.: AIAA J. 41(11), 2216–2228 (2003)

An 1D-Beam Approach for Both Stress Analysis and Fatigue Life Prediction

373

Appendix A This appendix details the resolution of the differential in Eqn. 8, which is identical for both adhesive stresses. The characteristic polynomial expression is: ⎧ P( R ) = aˆR 3 + bˆR 2 + cˆR + dˆ = 0 ⎪ ⎪R = r 2 ⎪ˆ ⎪a = 1 ⎨ˆ ⎪b = −k1 ⎪cˆ = k 4 ⎪ ⎪dˆ = k k − k k 2 3 1 4 ⎩

(27)

To determine these roots, the Cardan’s method is employed. Then, Eqn 25 is modified as: ⎧ 3 ⎪ R′ + ˆpR′ + qˆ = 0 ⎪ k1 2 ⎪ˆ p = − + k4 ⎨ 3 ⎪ k1 ⎪ 2 ⎪⎩qˆ = − 27 ( 2k1 − 9 k 4 ) + k 2 k 3 − k1k 4

(28)

where: R′ = R −

k1 3

(29)

Δˆ = qˆ 2 +

4 3 ˆp 27

(30)

and the determinant is:

By defining: ⎧ ˆ ⎪uˆ = 3 − qˆ + Δ ⎪ 2 ⎨ ⎪ − qˆ − Δˆ ⎪vˆ = 3 2 ⎩

(31)

374

E. Paroissien, A. Da Veiga, and A. Laborde

The roots of the reduced equation are written as: ⎧ R′ = uˆ + vˆ ⎪⎪ 1 ⎨ R2′ = juˆ + jvˆ ⎪ ⎪⎩ R3′ = j 2 uˆ + j 2 vˆ

(32)

Consequently, the roots of the characteristic equation (Eqn 25) are given by: k1 ⎧ 2 ⎪ R1 = uˆ + vˆ + 3 = r ⎪ ⎪ k1 3 1 +i ( uˆ − vˆ ) = ( s + it )2 ⎨ R2 = − ( uˆ + vˆ ) + 2 3 2 ⎪ ⎪ k 3 1 ( uˆ − vˆ ) = ( s − it )2 ⎪ R3 = − ( uˆ + vˆ ) + 1 − i 2 3 2 ⎩

(33)

Finally, the adhesive stresses have to be determined through Eqn. 29 where: r = uˆ + vˆ +

k1 ; s= 3

1 (Re( R2 ) + R2 ) ; t = 2

1 ( R2 − Re( R2 )) 2

(34)

Appendix B The stiffness matrix of a beam element KB can be expressed in the base u, w, θ: ⎛ Ai ⎜ ⎜ li ⎜ Ai ⎜− ⎜ li ⎜ ⎜ 0 ⎜ KB = ⎜ ⎜ 0 ⎜ ⎜ B ⎜− i ⎜ li ⎜ ⎜ B ⎜⎜ l ⎝ i

A − i li Ai li 0

0

0

0

0

12 Δi l 3 Ai i

−

12 Δi l 3 Ai i

0

12 Δi − l 3 Ai

12 Δi l 3 Ai

Bi li

6 Δi l 2 Ai

6 Δi − l 2 Ai

B − i li

6 Δi l 2 Ai

−

i

i

B − i li Bi li 6 Δi l 2 Ai

i

i

6 Δi l i 2 Ai

i

−

6 Δi l 2 Ai i

⎞ ⎛ Δi ⎟ ⎜3 ⎜ A + Di ⎟ i ⎠ ⎝ ⎞ 1 ⎛ Δi ⎜3 − Di ⎟⎟ l i ⎜⎝ Ai ⎠

1 li

⎞ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ i ⎟ 6 Δi − ⎟ 2 Ai li ⎟ ⎞ ⎟⎟ 1 ⎛ Δi ⎜3 − Di ⎟⎟ l i ⎜⎝ A i ⎠⎟ ⎟ ⎞⎟ 1 ⎛ Δi ⎟ ⎜3 D + i⎟⎟ l i ⎜⎝ Ai ⎠ ⎟⎠ Bi li B − li 6 Δi l 2 Ai

(35)

26th ICAF Symposium – Montreal, 1-3 June 2011 Cyclic Stress-Strain and Strain-Life Properties of Aerospace Metallic Materials S.K. Walker1 and A.C. Quilter2 1

Jesmond Engineering Ltd, Brough, East Yorkshire, UK 2 IHS ESDU, London, UK

Abstract. IHS ESDU decided to approach a number of organisations in order to compile a reference source of cyclic stress-strain and strain-life properties for the benefit of the wider aerospace industry. Work began in 2006 on the collection and collation of the raw data points for commonly-used aerospace metallic materials. Beside the data available in the literature and other readily-accessible sources, efforts were made to encourage organisations with their own data to contribute to a database from which they would subsequently benefit. Considerable support and enthusiasm were expressed for the project and generous donations of data were received from a number of organisations. These data were combined with those gathered from public domain sources and those considered to be the most robust were retained. Cyclic stress-strain curves fitted using the Ramberg-Osgood equation were found to correlate well with the test data. The Coffin-Manson strain-life model was generally less successful but curves are provided due to their widespread use. The resulting set of cyclic stress-strain and strain-life properties of commonly-used aerospace metallic materials permits a wider appreciation of material fatigue behaviour not previously available. The results are to be published in forthcoming IHS ESDU Data Item Number 11003. Analysis of specimen geometries led to some simple design recommendations that may be useful to those considering future strain-controlled testing.

376

S.K. Walker and A.C. Quilter

Nomenclature Symbol b c C, K D, d, r, g, L, T E K′ KTN n′ Nf Rε Δε Δεp Δσ ε′f σ′f γ

Fatigue Strength Exponent Fatigue Ductility Exponent Chaboche Characteristic Coefficients Test specimen dimensions Young’s Modulus Cyclic Strength Coefficient Net-Section Stress Concentration Factor Cyclic Strain-Hardening Exponent Number Of Cycles To Failure Strain Ratio = εmin / εmax Total True Strain Range Plastic True Strain Range Total Stress Range Fatigue Ductility Coefficient True Fatigue Strength Coefficient Chaboche Characteristic Coefficient

N/m2 mm N/m2 N/m2

cycles

N/m2 N/m2

1 Introduction Local strain-based fatigue analysis has been in use since the 1970s, its applications being initially rooted in the military aircraft and automotive industries of the US where much of the developmental work was performed. This approach requires cyclic stress-strain and strain-life relationships to be known for each material considered. The technique is now widely accepted throughout the international aerospace industry, yet the quantity of material data in the public domain remains limited. Much of the large amount of data produced over the years has been commissioned and held privately by industry. IHS ESDU decided to approach a number of organisations in order to compile a reference source of cyclic stress-strain and strainlife properties for the benefit of the wider aerospace industry.

2 Methods Work began in 2006 on the collection and collation of the raw cyclic stress-strain and strain-life data points for commonly-used aerospace metallic materials. Beside the data available in the literature and other readily-accessible sources, efforts were made to encourage organisations with their own data to contribute to a database from which they would subsequently benefit. The resulting data were assessed and those considered to be the most robust were retained.

Cyclic Stress-Strain and Strain-Life Properties of Aerospace Metallic Materials

377

Ramberg-Osgood and Coffin-Manson lines were drawn through the data based on a combination of traditional numerical techniques and engineering judgement. Monotonic tensile test values have been retained where available for reference purposes, but they were not used in the cyclic stress-strain or strain-life curve fits because they neglect the effects of cyclic stabilisation.

3 The Cyclic Stress-Strain Relationship A cyclic stress-strain curve is often defined by the locus of the tips of stable (or sometimes near-half-life) hysteresis loops from several tests performed at different completely reversed constant strain amplitudes. This is often known as the Companion Test method and is used in many of the data of the present study. Some data, however, use the Block Incremental Step method. These methods (and others) are described by Landgraf, Endo & Morrow [1] who concluded via experimental data that the different methods produce approximately coincident curves. Comparisons made in the present study appear to support this conclusion, although they are made between sets of test data from different sources. Cyclic Stress-Strain Equations A commonly used equation for describing the cyclic stress-strain curve follows the form proposed by Ramberg and Osgood, [2], with monotonic constants replaced by their cyclic equivalents. The equation, Eqn. 1, splits total strain into a summation of elastic and plastic parts. 1

Δ ε Δ σ ⎛ Δσ ⎞ n ' = +⎜ ⎟ 2 2E ⎝ 2K ' ⎠

(1)

An alternative to the Ramberg-Osgood equation that is sometimes used is that proposed by Chaboche, [3], Eqn 2.

⎛ Δε p ⎞ C Δσ ⎟⎟ − k = tanh⎜⎜ γ γ 2 ⎝ 2 ⎠

(2)

It was decided to use the Ramberg-Osgood equation for the purposes of at least the initial issue of the Data Item because this is currently considered to be the form in most widespread use. Suitability of the Cyclic Stress-Strain Equation for Modelling Test Data No particular problems were found when fitting empirical curves to the test data using the Ramberg-Osgood equation.

378

S.K. Walker and A.C. Quilter

4 The Strain-Life Relationship Strain-Life Equation Strain-Life testing is normally carried out on specimens loaded under controlled constant amplitude total strain conditions. Total strain amplitude can be divided into elastic and plastic parts, where the plastic part represents the width of the hysteresis loop. Plotted separately, the elastic strain versus life and plastic strain versus life relationships are often approximated as being exponential: linear on loglog scales. The approximate log-log linearity of the stress versus life relationship was noted by Basquin, [4], and his observation can be applied to elastic strain versus life, since elastic stress is directly proportional to elastic strain. Later Coffin [5] and Manson [6] noticed a similar relationship between plastic strain and life. The combination of the equations for elastic and plastic strain provides a relationship between total strain and life and is often simply referred to as the CoffinManson Relationship, Eqn (3).

Δε σ ' f = ( 2 N f )b + ε ' f ( 2 N f ) c E 2

(3)

Suitability of the Strain-Life Equation for Modelling Test Data Whilst the form of the Coffin-Manson relationship is conveniently simple and in common use, it is well-known that in practice both the elastic- and plastic-strain versus life relationships are often non-linear on log-log scales. This is demonstrated in Figure 1 for two example materials. Some authors have suggested nonlinear or bi-linear models to better represent the data (see for example [7], [8], [9] & [10]) whilst others advocate simple interpolation of the raw test data, [11]. The curved nature of the elastic- and plastic-strain lines made the fitting of straight lines a somewhat judgemental process. In general, attempts were made to provide reasonable fit to mean fatigue strength around the 105 cycle region, and this often resulted in a conservative fit elsewhere. On the positive side (as far as the use of Coffin-Manson model is concerned) the amount of inherent scatter in the test data often tended to mask deviations of the experimental curve away from the modelled shape and in most cases the model fitted the test data reasonably well. It should be noted, however, that the Coffin-Manson curve shape was initially developed for use in Low Cycle Fatigue and its suitability for extrapolation into the ultra-high-cycle regime remains questionable. Further investigation may be required here.

Cyclic Stress-Strain and Strain-Life Properties of Aerospace Metallic Materials

379

Strain Amplitude

1E-1

1E-2

Total Strain Elastic Strain Plastic Strain

1E-3 1E+0

1E+1

1E+2

1E+3 1E+4 Cycles, Nf

1E+5

1E+6

1E+7

Fig. 1 Non- Linearity of Elastic Strain and Plastic Strain Lines (after [11], 7075-T6).

Strain Amplitude

1E-1

Total Strain Elastic Strain

1E-2

Plastic Strain

1E-3 1E+0

1E+1

1E+2

1E+3

1E+4

1E+5

1E+6

1E+7

Cycles, Nf

Fig. 2 Non-Linearity of Elastic Strain and Plastic Strain Lines (present study, collated 2014-T6 & 2014-T651 data from 7 sources).

5 Effects on Fatigue Performance It is important to note that the data are developed from testing of largely unnotched specimens, whose behaviour is generally equated to that at the root of a

380

S.K. Walker and A.C. Quilter

notch. Use of the data to simulate the fatigue behaviour of structures invariably requires the incorporation of all effects on fatigue performance. These may include, but may not be limited to those due to: geometry, surface condition, environment, frequency and fretting. In addition the proper accounting of mean stress, residual stress and variable amplitude loading must be made. The vast majority of testing was carried out at room temperature on axiallyloaded specimens under R=-1 constant amplitude total strain controlled conditions. Where data covering some of the above effects were available they were presented separately for comparison.

6 Specimen Design Specimen geometries (where available) were generally different for each set of test data examined. The vast majority of the tested specimens were generally of either machined solid-cylindrical or flat sheet (or plate) form, Figure 3 and Figure 4 respectively, although one set of tests used a hollow tubular specimen. r

d

D

L

g

Fig. 3 Typical Machined Solid Cylindrical Specimen.

d

r D

L

g

T

Fig. 4 Typical Flat Plate / Sheet Specimen.

The varying geometries of the specimens meant that correspondingly varied stress concentration factors may be present. Fine-mesh mathematical models [12] were constructed for those specimens with dimensions available. Symmetric constraints were used such that the models could be limited to half-length, quartercircular for the cylindrical specimens and half-length, half-width for the flat-sheet specimens. The models were given elastic properties with Poisson’s ratio 0.3. Maximum axial net stress concentration factors, KTN, were calculated with the external end of the specimen subjected to uniformly distributed loading. An example results contour plot is shown in Figure 5.

Cyclic Stress-Strain and Strain-Life Properties of Aerospace Metallic Materials

381

KTNET=1.05

Fig. 5 Example Stress Contour Results (2014_7 low strain specimen).

With a few exceptions the tested solid-cylindrical specimens were found to have KTN in the region 1.0-1.05, whilst KTN for the flat plate specimens was generally higher: around 1.05-1.15. In addition to the tested geometries, KTN was calculated for some nominal geometries of the ASTM [13] and British Standard [14] recommendations. The results are shown in Table 1. Table 1 KTN for Nominal Standard Geometries.

Shape BS7270 ASTM

Flat Round Flat Round

L 50 50 50 50

Dimensions (mm) d r g D 5 10 11.25 20 10 20 22.5 20 10 10 15 20 10 40 30 20

T 5 N/A 5 N/A

D/d

r/d

KTN

4 2 2 2

2 2 1 4

1.13 1.089 1.29 1.045

It was noted that the ratio D/d was either close to or greater than 2 for all of the geometries investigated. By inspection of published data for stress concentrations in notched bars and plates, [15], it may be concluded that:

382

• • •

S.K. Walker and A.C. Quilter

variation in D/d above 2 does not significantly affect KTN; the driving parameter for KTN is r/d; increasing g reduces KTN (g=0 being the worst case).

The relationship between r/d and calculated KTN is shown in Figure 6 where excellent agreement with [15] was found for g=0, cylindrical specimens. 1.4 Tested Flat-Sheet Specimens (g>0) Tested Cylindrical Specimens (g=0 & g>0) ASTM E606/BS 7270 Nominal Flat-Sheet Geometries (g>0) 1.3

ASTM E606/BS 7270 Nominal Cylindrical Geometries (g>0)

KTN

Peterson [15] Curve for Flat-Sheet Specimens (g=0) Peterson [15] Curve for Cylindrical Specimens (g=0) 1.2

1.1

Tested Specimens with g=0

1.05 1 0

1

2

3

4

5

6

7

8

9

r/d

Fig. 6 Relationship between r/d and KTN.

Effect of KTN on Fatigue Performance One dataset had a high KTN (1.35) and its fatigue performance was noticeably inferior to others. This dataset aside, the differences in fatigue performance of the two specimen types appeared not to be great when compared with inherent scatter and this tended to be supported by their broadly similar stress concentrations. Although only one third of the geometries were known, their observed fatigue performance was similar to that of the remaining geometries suggesting that stress concentrations did not differ greatly overall. Conformity with ASTM & British Standard Recommendations The Standard Recommendations allow relatively low r/d and correspondingly high KTN. From the authors’ point of view this was surprising given that stress concentration is a major consideration in the design of plain specimens. Many of the tested geometries did not conform to Standard Recommendations.

Cyclic Stress-Strain and Strain-Life Properties of Aerospace Metallic Materials

383

Suggested Guidelines for Strain-Controlled Test Specimen Design Some specimen design guidelines can be derived based on Figure 6, where it can be seen that KTN may be kept beneath 1.05 (a nominally low value) by ensuring that r/d is greater than either 4.5 (round specimens) or 6.5 (flat-sheet specimens). The ratio D/d is considered relatively unimportant in terms of KTN for the reasons described earlier. Low D/d values tend to reduce KT, but they increase grosssection stress which may promote grip failures. It is therefore suggested to limit D/d to a minimum of 2. Table 2 summarises the suggested guidelines. Table 2 Suggested Guidelines for Strain-Controlled Test Specimen Design.

Specimen Shape Flat sheet or plate Cylindrical

r/d ≥6.5 ≥4.5

D/d ≥2 ≥2

The design guidelines associated with effects of thickness in flat specimens and overall specimen length require further work.

7 Results The materials covered in ESDU Data Item 11003 are given in Table 3. Table 3 Materials Covered in ESDU Data Item 11003.

Aluminium Alloys

Titanium Alloys

Steels

2014-T6, 2014-T651 2024-T3, 2024-T351, 2024-T4, 2024-T42 2124-T851 7010-T7451, 7050-T7451, 7050-T7651 7075-T6, 7075-T65, 7075-T651, 7075-T73, 7075-T7351 7475-T7351, 7475-T76, 7475-T761 Ti-6Al-4V Beta Annealed, Mill / Recrystallisation Annealed prEN 3354*(TI-P64001) IMI685 (Ti-6Al-5Zr-0.5Mo-0.2Si) S99 PH13-8Mo (AMS 5629) 300M (BS S155) 17-4PH H1150

The following two sections present example cyclic stress-strain and strain-life properties of one of the materials contained in the Data Item, namely 2014 -T6/-T651.

384

S.K. Walker and A.C. Quilter

8 Example Cyclic Stress-Strain Data Figures 7 and 8 present measured and fitted cyclic stress-strain data for 2014-T6/T651.

Stress Amplitude (MN/m2)

1000

100 0.001

ID 2014_1 ID 2014_4 ID 2014_6 ID 2014_7 ID 2014_8 ID 2014_9 ID 2014_10 ID 2014_11 ID 2014_12 ID 2014_13 ID 2014_15 Fitted Curve 0.01 Total Strain Amplitude 0.1

1

Fig. 7 Cyclic Stress-Strain Data for 2014-T6/-T651. 1000

Stress Amplitude (MN/m2)

E = 70353 MN/m^2, K' = 654.7MN/m^2, n' = 0.0705

100 0.001

Total strain test data Points used in elastic strain line fit Fitted elastic strain line Plastic strain Points used in plastic strain line fit Fitted plastic strain line Fitted total strain curve Monotonic test total strain (not used in curve fit) Monotonic test plastic strain (not used in line fit) 0.01

Strain Amplitude

0.1

Fig. 8 Cyclic Stress-Strain Curve Development for 2014-T6/-T651.

1

Cyclic Stress-Strain and Strain-Life Properties of Aerospace Metallic Materials

385

9 Example Strain-Life Data Figures 9 and 10 present measured and fitted strain-life data for 2014-T6/-T651.

1 ID 2014_1

Total Strain Amplitude

ID 2014_4 ID 2014_5 ID 2014_7

0.1

ID 2014_8 ID 2014_14 ID 2014_15 Fitted Curve

0.01

0.001 1E-1

1E+0

1E+1

1E+2

1E+3

1E+4

1E+5

1E+6

1E+7

1E+8

Cycles, Nf

Fig. 9 Strain-Life Data for 2014-T6/-T651, Rε=-1.

1

Total strain test data Elastic Strain Points used in elastic strain line fit Fitted elastic strain line Plastic Strain Points used in plastic strain line fit Fitted plastic strain line Fitted total strain curve Monotonic test total strain (not used in curve fit) Monotonic test elastic strain (not used in line fit) Monotonic test plastic strain (not used in line fit)

Ee = 0.0105 (2Nf) ^ -0.1082

Strain Amplitude

Ep = 0.3041 (2Nf) ^ -0.6478

0.1

0.01

0.001 1E-1

1E+0

1E+1

1E+2

1E+3

1E+4

1E+5

1E+6

1E+7

Cycles, Nf

Fig. 10 Strain-Life Curve Development for 2014-T6/-T651, Rε=-1.

1E+8

386

S.K. Walker and A.C. Quilter

10 Conclusions Considerable support and enthusiasm were expressed for this project and generous donations of data were received from a number of organisations. These data were combined with those gathered from public domain sources. Curves fitted through the data using the Ramberg-Osgood equation were found to correlate well with the test data. Although, as presented, the curves obtained using the Coffin-Manson equation do not generally correlate well, they are provided due to their widespread use. The combined set of data and the associated curves form the basis of forthcoming IHS ESDU Data Item Number 11003. The Data Item compares material data obtained from numerous sources for the first time. An appreciation of the amount of scatter is now clear. The resulting set of cyclic stress-strain and strain-life properties of commonlyused aerospace metallic materials permits a wider appreciation of material fatigue behaviour not previously available. Analysis of specimen geometries led to some simple design recommendations that may be useful to those considering future strain-controlled testing.

Acknowledgement The author wishes to thank the following: without whose help the work would have not been possible:•

The contributors of the data - many of whom had to put significant work into being given permission to release the data into the public domain. This was particularly the case in large organisations.

•

The IHS ESDU Fatigue Committee for their continued support and helpful suggestions.

•

Paul Griffiths for his Finite Element Analyses and William Lennox for proof reading.

References [1] Landgraf, R.W., Morrow, J., Endo, T.: Journal of Materials. JMSLA 4(1), 176–188 (1969) [2] Ramberg, W., Osgood, W.R.: NACA Tech. Note 902 (1943) [3] Lemaitre, J., Chaboche, J.-L.: Mechanics of solid materials. Cambridge University Press, Cambridge (1990) [4] Basquin, O.H.: In: Proceedings of the American Society for Testing and Materials, vol. 10, pp. 625–630 (1910) [5] Coffin Jr., L.F.: Trans. ASME 76(6), 931–949 (1954); Discussion, pp. 949-950 [6] Manson, S.S.: NACA Report 1170 (1954)

Cyclic Stress-Strain and Strain-Life Properties of Aerospace Metallic Materials

387

[7] Fatemi, A., Plaseied, A., Khosrovaneh, A.K., Tanner, D.: Int. J. Fatigue 27, 1040–1050 (2005) [8] Sanders, T.H.: In: Jaffee, R.I., Wilcox, B.A. (eds.) Fundamental Aspects of Structural Alloy Design. Plenum Publishing, NY (1977) [9] Stephens, R.I., Koh, S.K.: In: Stephens RI, editor. Bi-linear log–log elastic strain-life model for A356-T6 cast Aluminium alloy round-robin low cycle fatigue data. Fatigue and fracture toughness of A356-T6 cast Aluminium alloy. SAE SP-760 (1988) [10] Wong, W.A.: SAE 840120, Presented at International Congress & Exposition Detroit, Michigan (1984) [11] Endo, T., Morrow, J.: Journal of Materials, JMLSA 4(1), 159–175 (1969) [12] Griffiths, C.P.: JRP-10-014-1, Jesmond Engineering Ltd. internal report (2011) [13] ASTM International, Standard Practice for Strain-Controlled Fatigue Testing, Designation E606-04e1 (2004) [14] British Standard, B.S.: 7270:2006, Metallic materials – Constant amplitude strain controlled axial fatigue – Method of test (2006) [15] Peterson, R.E.: Stress Concentration Factors. John Wiley & Sons, Chichester (1974)

26th ICAF Symposium – Montreal, 1-3 June 2011 Crack Propagation Calculation for Aluminium Aircraft Structures Considering the Influence of Load Sequences R. Buchholz IMA Materialforschung und Anwendungstechnik GmbH Dresden, Germany

Abstract. The load sequence has an important influence on the crack propagation in aircraft structures. The estimation of this influence is a particular challenge in the life time prediction of aircraft components. Current calculation methods consider this inadequately. Therefore a new calculation method was developed to determine the crack propagation step-by-step for each load point considering the influence of former loads on the current crack propagation rate. This method bases on the assumption that the current load distribution in front of the crack tip determines the effective stress intensity factor directly for the current load step. Therefore, it is mandatory to possess a very detailed knowledge of the stress-strain-behaviour of the used aluminium alloy under cycled loadings. The mathematical description of the load distribution in the structure is carried out for the ligament in front of the crack tip by a function of the distance to the crack tip by special assumptions. These functions result from FE analyses or a specific simplified simulation method. Supported by particular weight functions the effective stress intensity factor for the current load step is calculated directly from the current load distribution function. Based on the effective stress intensity factor the crack propagation rate is calculated by the Paris-equation. The applicability of the new crack propagation calculation method was demonstrated by a number of plausibility checks and an extensive test program. The analysis of the test and calculation results shows very good correlations.

1 Introduction Civil aircrafts are designed according to the damage tolerance concept for a design service goal of 25 years, up to 90 000 flights depending on aircraft mission. Verification for the whole aircraft life time is needed to demonstrate that there will be no failure of structural part. The verification is done by tests and calculation methods. The life time of a component designed according to the damage tolerance concept consists of time frames for crack initiation and crack growth up to a critical crack length. Based on this knowledge inspection programs are prepared.

390

R. Buchholz

Considering a safety factor, they define the location, the inspection methods, time for initial inspection and inspection interval. In the case of verification by calculation or simulation methods an increased safety factor is used as for verification by tests. The load sequence and the load history have an important influence on the crack propagation in aircraft structures. In the case of verification by calculation methods the estimation of this influence is a particular challenge in the life time prediction of aircraft components. Current calculation methods consider this inadequately. In this project a new calculation method was developed to consider the influence of load sequences while the crack propagation calculation. It is based on the assumption that the load distribution directly in front of the crack tip determines the effective stress intensity factor. Based on theoretical analysis a calculation approach was developed and verified by conservative calculation methods and tests on coupons and simple structural details. The results were analysed in detail qualitative and quantitative considering known phenomena. The implementation of a calculation method which is able to show the influence of load sequences leads to an improved prediction of the crack propagation in aircraft structures. Hence, a replacing of verification tests by verification calculations is feasible.

2 Calculation Method The new calculation method developed within this project determines the crack propagation step-by-step for each load point considering the influence of former loads on the current crack propagation rate. This method bases on the assumption that the current load distribution near the crack tip determines the stress intensity factor directly for the current load step. The mathematical description of the load distribution in the structure is carried out for the ligament in front of the crack tip by a function of the distance to the crack tip by special assumptions. These functions result from FE analyses or a specific simplified simulation method. Supported by particular weight functions the effective stress intensity factor for the current load step is calculated directly from the current load distribution function. Based on the effective stress intensity factor the crack propagation rate is calculated by the PARIS-equation, an established crack propagation equation without any adjustment factor of influences already included in the effective stress intensity factor. The calculation process is shown in Figure 1. It is mandatory to possess a very detailed knowledge of the stress-strainbehaviour of the used aluminium alloy under cycled loadings. This behaviour was determined by tests (constant amplitude tests and incremental step tests) for different materials. The description of the stress-strain-behaviour is done by the

Crack Propagation Calculation for Aluminium Aircraft Structures

391

Ramberg-Osgood- equation considering the Masing-criteria and memory behaviour of the material. The splitting of the strain into elastic and plastic strain is a precondition to calculate the stress distribution. Therefore the parameters elasticity modulus E, cyclic strain hardening coefficient K’ and cyclic strain hardening exponent n’ were determined.

Fig. 1 Calculation process.

3 Load Distribution The mathematical description of the load distribution in the structure is carried out for the ligament in front of the crack tip by a function of the distance to the crack tip by special assumptions. These functions result from FE analyses or a specific simplified simulation method. In the loading case there is a stress peak in front of the crack tip. Resulting, there is a plastication of the material in a limited area (Figure 2). This effect determines the stress intensity near the crack tip. In the case of a cyclic loading an alternating plastic zone is formed inside the plastic zone. In case of increasing a material depending limit value for the stress intensity the crack is growing.

392

R. Buchholz

Fig. 2 Stress distribution near the crack tip.

Figure 3 shows some examples for the stress distribution for some load case sequences determined by FE-analysis 600 500 400 300 200

σy (MPa)

100 0 -100 -200 -300 0 MPa - 75 MPa - 0 MPa 0 MPa - 75 MPa - 0 MPa - 75 MPa 0 MPa - 100 MPa - 0 MPa 0 MPa - 100 MPa - 0 MPa - 100 MPa 0 MPa - 100 MPa - 0 MPa - 75 MPa

-400 -500 -600 0,00

0,05

0,10

0,15

0,20

x (mm)

Fig. 3 Stress distribution examples.

0,25

0,30

Crack Propagation Calculation for Aluminium Aircraft Structures

393

4 Stress Intensity Factor Supported by particular weight functions the effective cyclic stress intensity factor for the current load step is calculated from the current stress distribution function. Therefore, it is assumed that the loading path is damage relevant while the unloading path does not generate any damage. Furthermore, there are two shares of interest to determine the effective stress intensity factor: the crack opening shear and the crack closure share. The stress distribution of previous load step determines the crack closure share. This share includes the complete information of the load history. The variation of the stress distribution function of the current load case determines the crack opening share. The difference of both shares represents the effective cyclic stress intensity factor. The shares of the effective stress intensity factor are shown in Figure 4 exemplary for different load levels corresponding to a constant amplitude base level with a load ratio of R = 0. It is assumed that up to the previous load step a constant amplitude load acted (load level of 100%).

Fig. 4 Shares of the effective cyclic stress intensity factor.

5 Crack Propagation Based on the effective stress intensity factor the crack propagation rate is calculated by the Paris-equation, an established crack propagation equation without any adjustment factor of influences already included in the effective cyclic stress intensity factor. In parallel crack propagation calculation were performed by a conventional procedure using Paris- or Forman-equation. The principle of the calculation procedures is shown in Figure 5. The crack propagation is changing the crack geometry. This effect needs to be considered for the following load points because it leads to a reduction of the sectional area and affected the stress distribution in front of the crack tip.

394

R. Buchholz

The applicability of the new crack propagation calculation method was demonstrated by a number of plausibility checks. For instance the effect of different load ratios are investigated as well as the known effects of tension and compression overloads on the crack propagation behaviour. Both are shown in Figure 6.

Fig. 5 Calculation procedures.

Fig. 6 Plausibility check.

Crack Propagation Calculation for Aluminium Aircraft Structures

395

The sequence of load events influences the crack propagation. For the same load path on different positions in the load sequence different load distributions are calculated. Following, there are different values for the effective cyclic stress intensity factor und different crack propagation rates. This is shown for an example in Figure 7. If the stress strain behaviour is reviewed for a location very short in front of the crack tip the different locations for the stress strain paths are visible (Figure 8).

Fig. 7 Stress distributions for the same load path on different positions of the load sequence.

Fig. 8 Stress strain paths for different load paths.

6 Results and Conclusion To validate the developed calculation procedure a test series was conducted. The crack propagation of different crack geometries in specimens made of different materials was determined. Therefore the tests were performed in an uniaxial servo-hydraulic test rig controlled by a special control system. The nominal load values (load programs of typical aviation load sequences, flight-by-flight) were generated and controlled by a separate external computer.

396

R. Buchholz

The crack length was measured by visual testing methods (VT) using a travelling microscope and by eddy current testing (ET). Furthermore a marker load procedure was used for further investigations on the fractured surface by a scanning electron microscope after test conclusion. The results of a direct comparison of the crack propagation rate and the component life time up to a critical crack length are shown in Figure 9 and Figure 10.

Fig. 9 Results, example 1.

Fig. 10 Results, example 2.

Crack Propagation Calculation for Aluminium Aircraft Structures

397

The conventional calculation methods could predict an increased or a reduced crack propagation rate. The new calculation procedure gives a considerable improved crack propagation prediction. Within the developed and numerically implemented analysis, a tool was prepared which is able to determine the crack propagation in certain structural areas better than it is possible with traditional methods. It offers the possibility to represent the crack propagation realistically for different loads up to real load spectra. Based on the individual change of stress distribution an effective cyclic stress intensity factor is computed. By using the Paris-equation as usual crack propagation equation, however without correction approaches the crack propagation is calculated load step wise. The developed procedure was investigated regarding plausibility of its results. Thus it was proven that the calculated crack propagation for cyclic loading under a stress ratio R = 0 corresponds to the results of conventional methods. By testing cyclic loadings with greater stress ratios the expected higher crack propagation rates were observed and vice versa. Also the phenomena of retarded crack propagation after a tension overload and the accelerated crack propagation after a compression overload could be demonstrate with the developed procedure. In the further process of the study it was shown, that the crack propagations determined in the various test series could be simulated by using the new calculation model. Both for cyclic loads under different stress ratios and especially for uniaxial flight program sequences very good agreements with the experimental crack propagation curves were detected. While the component live according to conventional equations differs in average about 50% from the experimental data the differences by using the developed method is in average only 10%. Compared with conventional crack propagation calculation methods the new procedure needs additionally information about the material behaviour. This applies especially for the stress-strain-behaviour resulting from the respective load. Often these values are not available and have to be determined in additional experimental test series. In addition all material characteristics are subject to variation. The calculation method is based generally on the average values of the individual characteristics. For this reason small deviations between the computed and the practically determined results are to be expected. In conclusion, it is to be noted that – by the crack growth calculation method developed during this study – a basis was established to predict the crack propagation very exactly for different loads up to load spectra considering the effects of the load histories.

26th ICAF Symposium – Montreal, 1-3 June 2011 Analysis of Fatigue Crack Growth under Random Load Sequences Derived from Military In-flight Load Data C. Mattrand1, J.-M. Bourinet1, and D. Théret2 1

Clermont Université, IFMA & UBP, EA 3867, Laboratoire de Mécanique et Ingénieries, BP 10448, F-63000 Clermont-Ferrand, France 2 DGA - Techniques Aéronautiques 47 rue Saint-Jean, BP 53123, 31131 Balma Cedex, France

Abstract. The crack growth process of a crack initiating at a hole of a skinstringer panel in a fighter aircraft subjected to random variable amplitude loading is investigated in this paper. The generation of synthetic random load sequences hinges on Markovian models for which their parameters are identified from real in-service recorded flights, which gives these models a real sound basis. The ability of these stochastic models to capture and recreate the load scatter measured on a French fighter aircraft fleet is first appraised. The most efficient model serves then to probabilistic crack growth analyses under random loading. It is found that the extreme load values as well as the length of load sequence being repeated through the load spectra must be considered as driving parameters for crack growth analyses under repeated load sequences.

1 Introduction The fatigue management of an aircraft starts in the design process with a design philosophy. It hinges on the choice of the most appropriate damage theory to estimate the fatigue life, the definition of suitable load spectra and the choice of appropriate values for material properties and initial defect sizes. Inputs to calculation generally come from a statistical approach based on test results (e.g. initial crack sizes based on the applicable non destructive evaluation method NDE, average material fatigue crack growth rate …). In some cases experience-based safety factors are applied to these values in order to cover inherent uncertainties observed in the microstructure, in material properties, in crack growth rates and in applied loads. The estimated life needs then to be validated through a global demonstration including both tests and calculations concluded by a full-scale fatigue test. This design is expected to comply with some military Defence Standards for fighters or military transport aircraft or with Civil Airworthiness Requirements such as CS for civil aircraft in Europe. The structure of interest is supposed to be designed with an acceptable safety level for instance, i.e. the probability of occurrence of individual lives shorter than the target life, albeit undefined, needs to be kept small enough.

400

C. Mattrand, J.-M. Bourinet, and D. Théret

Load Uncertainties Mission types Operational Environments (e.g. gust, maneuver, …)

Structural Uncertainties

Material Uncertainties

EIFS, EPS, … Inspection PoD Geometric dimensions

Crack growth parameters Fracture toughness

Damage Tolerance Analysis Crack growth models FE models

Structural Response Crack sizes Residual Strength

Inspection PoD Design Requirements PROBABILISTIC ANALYSIS

Failure Function Critical Failure Probability

Fig. 1 Global probabilistic fracture mechanics approach (grey boxes for areas investigated in this paper).

The current design philosophies are mostly deterministic and, as a consequence they can not support the quantitative computation of the structural reliability under operational conditions which are themselves known with some degree of uncertainty. Such analyses appear also inappropriate to assess the respective contributions of the many parameters involved in such problem, from a stochastic view point. They are in addition insufficient to assist decision making in the maintenance of airframe structures. In an effort to address the above mentioned shortcomings the so-called probabilistic fracture mechanics has received an increasing attention in the past few years in the field of damage tolerance, see e.g. references [1-5] for rather comprehensive guidelines on this topic. Figure 1 summarizes/depicts how to perform such probabilistic analyses. A stochastic analysis of a crack initiating at a hole of a skin-stringer panel of a fighter aircraft subjected to random variable amplitude loads is performed in this paper. A suitable crack propagation model needs first to be selected for computing the structural response, which is the aim of the first section (see Figure 1). The material properties (crack growth rates and fracture toughness), which are kept constants to their mean values in this study, are introduced in the second section as well as the test specimen geometry. This paper especially investigates the effect of scatter in applied loads on the crack growth process. A methodology for modelling random fatigue loads based on real in-flight load data is proposed in the third section for this purpose, which is rarely addressed in the literature. This methodology contributes to the specific field of probabilistic fracture mechanics offering new tools to generate random load sequences from real observed load data. It is based on the choice of either First order Markov chains or Hidden Markov chains. The

Analysis of Fatigue Crack Growth under Random Load Sequences

401

Hidden Markov Model appears to be the most accurate and efficient model. It is then used in the last section to explore the influence of load sequence parameters such as the length of the load sequences (i.e. number of applied flights) and the extreme values on the crack growth process in a skin-stringer panel of a fighter aircraft.

2 Crack Growth Model Crack growth concepts A basic step in a probabilistic fracture mechanics analysis consists in the selection of an appropriate damage theory among available fatigue crack growth models. Crack growth theory for variable amplitude loading ranges from global analysis, which tries to predict the fatigue crack growth considering the whole loading cycles at the same time, to cycle-by-cycle analysis [6]. In the latter case crack growth is assumed to be a summation of crack extensions in each cycle. A cycleby-cycle analysis can be performed with or without taking interaction effects into account. PREFFAS crack closure model The PREFFAS crack closure model [7] proposed by Aliaga, Davy and Schaff in 1985 is selected here to compute the crack growth. This model is selected due to its ability to capture interaction and retardation effects resulting from a crack growth under variable amplitude loads. This is generally a key point in the damage tolerance analysis of a structural component of an aircraft. Moreover this model is characterized by a straightforward and fast calculation procedure, which alleviates the computational burden specific to probabilistic analyses. The PREFFAS model hinges on the Elber crack closure concept [8]. In cycle i, the crack extension reads:

δ ai = Ceff ( K max, i − K op, i ) = Ceff ( ΔK eff, i ) with Ceff = m

m

CR ⎡⎣U ( R ) ⎤⎦

m

(1)

where Kmax, i is the maximum value of the Stress Intensity Factor K (SIF) during the loading cycle i, Kop, i is the opening stress intensity factor, m is the exponent of the Paris law, CR is the C-parameter of the Paris law determined from a constant amplitude test at stress ratio R and U(R) = aR+b is the Elber effective SIF range ratio. In PREFFAS, it is often assumed that a+b = 1, where the parameter b is found to be material and thickness dependent. The opening level Kop, i at a given cycle i depends on the previous load history and calculating its variation cycle-by-cycle is a key feature of the PREFFAS model. The PREFFAS model also incorporates the so-called Rainflow effect, which is another essential aspect of this method. The reader is invited to refer to [7, 9] for details on these specific features. Finally, the calculation effort is

402

C. Mattrand, J.-M. Bourinet, and D. Théret

relatively small in the PREFFAS model because it is assumed that the K-levels are not affected by the crack growth during the whole sequence of applied loads. This allows to fully separate the effects of the crack length and geometry from those of the stress history. This in fact implies a stationary assumption during the application of this N-cycle length sequence.

3 Material and Test Specimen Material The material considered in the present study is a 2024-T351 aluminum alloy. Its mechanical properties are summarized in Table 1. Elber parameters a and b are selected as recommended in [7]. Table 1 Material properties of the structural component.

σy = 350 MPa K1c = 1185.8 MPa

Yield stress Fracture toughness Paris constants

-13

C0.1 = 2.417·10 a = 0.35, b = 0.65

Elber parameters

mm MPa mm , m = 3.42

It should be noted that the scatter of the crack growth propagation is not addressed in this paper. Definition of the structure

The given example considers a crack initiating at a hole in a skin-stringer panel of a fighter aircraft. This problem is modelled by a corner crack from an offset hole in a plate (CC02 specimen in reference [10]). The width of the specimen is W = 59 mm and its thickness is t = 21 mm, see Figure 2. The other geometry parameters are D = 10 mm and B = 25 mm. All structural components are assumed to have an initial crack size a = 0.5 mm, with shape ratio a/c = 1. The role of the scatter in the initial crack length is not investigated in this paper.

Fig. 2 Geometry of the structure.

Analysis of Fatigue Crack Growth under Random Load Sequences

403

4 Scatter in Loads Randomness of the load spectra

Most aircraft experience in-service random variable amplitude loads which need to be properly accounted for in a safe design against fatigue. A common practice of the aircraft industry consists in deriving standardized and/or simplified load sequences from in-service load data representative of real complex loads. Examples of such standardized load sequences are TWIST (stresses in the lower wing skin at the wing root of a transport aircraft) and FALSTAFF (stresses in the lower wing skin near the wing root in a fighter aircraft) [11, 12]. These load sequences are deterministic despite the statistical nature of the experimental observations they are derived from. As a consequence, they cannot help for assessing the scatter in the crack growth life and the respective contributions of load parameters from a probabilistic view point. This paper addresses this shortcoming. The objective is to propose a methodology for modelling random fatigue loads appropriate and accurate enough to capture the scatter in loads in a specific aircraft fleet. Variable amplitude loads sustained by a specific aircraft fleet are in fact of random nature, due to substantial variability in their missions (commercial vs. military, military transport aircraft vs. fighter aircraft, various missions in a same fleet for a specific type of aircraft) and to inherent uncertainties of their operational environments (e.g. gust, maneuvers and ground-air-ground loads). In-flight load data

The present work is based on the operational usage of a fleet of a fighter aircraft in the French Air Force. A set of 27458 flights was recorded in the form of acceleration time series of the aircraft. These acceleration time series were then postprocessed such as to obtain stress time series at the hole of a skin-stringer panel of a fighter aircraft under study. Each stress time serie is characterized by a sequence of loads, a whole time length, i.e. the duration of the flight, and recording timesteps which are automatically varied along the flight. Recorded flights have been split into 27 classes which altogether are representative of the activity of the aircraft fleet, see Figure 3. A class gathers flights recorded on aircraft with same mission profiles and similar external load considerations. According to the unequal repartition of the flights through the classes (several of them with a very few flights) we decided to split the overall 27458 flights into two major groups with obvious distinct flight domains in order to increase the sample size. This selection was made based on experts’ judgment and the results were confirmed by a clustering analysis. The first group, called subsequently “group A” gathers 24320 flights recorded on aircraft loaded with two additional on-board fuel tanks. The second group, called “group B” gathers 3138 flights recorded on lighter aircraft loaded with at most one additional on-board fuel tank.

404

C. Mattrand, J.-M. Bourinet, and D. Théret

0.5 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0

Fig. 3 Relative class frequencies.

5 Stochastic Models for Random Loads in Fatigue Since the important information lies in the sequence of max-min stresses for fatigue and crack growth applications, the choice made here is to model the observed fatigue loads by means of discrete-time Markov chains. Discrete-time Markov chain processes have indeed been found interesting tools to model random sequences of cycles, i.e. succession of local peak and trough stresses also called sequences of turning points [13, 14]. For each group of in-flight data (“group A” and “group B”) we propose to model a N-cycle load sequence (X1, X2, …, Xn, …, XN) with Xn = {Mn; mn}, n=1, …,N, where Xn represents the nth cycle and Mn (resp. mn) is the peak stress value (resp. the trough stress value) of this nth cycle, either using discrete-time first order Markov chains or hidden Markov chains, which are described in the next subsection. It is worth noting that the definition of this N-cycle load sequence differs from previous work of the literature, to the authors’ knowledge. Markov chain models

Discrete-time first order Markov chain model. A First-order Markov Chain (FMC) [15, 16] with finite space E is a sequence of E-valued random variable ( X n )n∈`* such that the conditional distribution of Xn+1 knowing the discrete-time

process ( X m ) m≤ n is the same as the conditional distribution of Xn+1 given only Xn:

P ( X n +1 = en +1 X n = en , X n −1 = en −1 ,..., X1 = e1 ) = P ( X n +1 = en +1 X n = en ) (2)

The finite state space E for Xn = {Mn; mn} is here defined such as a valley mn systematically follows a peak Mn:

{ (

)

}

E = en = si ,s j , i > j and i, j ∈ {1, 2,..., K c }

where si and sj are selected stress levels.

(3)

Analysis of Fatigue Crack Growth under Random Load Sequences

405

If Kc is the number of load classes, i.e. if K = Kc (Kc –1)/2 is the cardinal number of the load cycles states E (|E| = K), the transition probabilities (moving from a given cycle to the following one) define a K×K-real matrix P such that: ⎛ p ⎜ 1,1 ⎜p P = ⎜ 2,1 ⎜ .. ⎜ ⎜p ⎝ K ,1

p p

1,2

2,2 ..

p

K ,2

...

p

...

p

..

⎞

1,K ⎟

⎟

2,K ⎟ and .. ⎟

⎟ ⎟ ... p K ,K ⎠

(

K

∑ pi, j = 1 , 0 ≤ pi, j ≤ 1 for i = 1,..., K (4) j =1

)

where pi , j = P X n +1 = e j X n = ei . In the following we restrict our analysis to time-homogeneous Markov chains, i.e. P is constant over time. The transition matrix P is estimated from observations of trajectories of X, i.e. from sequences of turning points obtained from filtered load time series. The quality of these estimations depends obviously of the available amount of data. A homogeneous first order Markov chain is completely defined by its transition matrix P, its initial distribution X1 describing the starting probabilities of the various states and its length N. Hidden Markov chain Model. Hidden Markov chains Models (HMM) [17, 18] are an extension of the concept of Markov chains for which the observation of X is not directly the state pertaining to E but a probabilistic function of this state. It is a bivariate discrete-time process {Sn, Xn}n>0, where Sn = {M’n; m’n} for our specific use. Sn corresponds here to the first order Markov chain defined above with the

{ (

)

}

finite state space E h = en = ci , c j , i > j , i, j ∈ {1, 2,..., K c } (ci and cj are not stress levels but only load classes), transition matrix P, initial distribution X1 and length N. Conditional on Sn, Xn = {Mn; mn} is a sequence of independent random variables such that the conditional distribution of Xn depends on Sn only. Observed values of Mn and mn are then obtained from a set of Kc-probability distributions ( f k , k ∈ {1, 2,..., K c }) switching along the sequence according to M’n and m’n. Inference of model parameters

The parameters of the previously introduced two Markovian models, i.e. E, P, X1, and N for the FMC model and Eh, P, X1, N and the set of Kc-probability distributions ( f k , k ∈ {1, 2,..., K c } ) for the HMM model are identified for each group of in-flight load data (“group A” and “group B”) presented in the previous section.

406

C. Mattrand, J.-M. Bourinet, and D. Théret

0.01

c

c

3

1

0.009 0.008 0.007

σ

f (σ)

0.006 0.005 0.004

c

2

0.003

c

4

0.002

c

5

0.001

c

c7

6

0 0

0.1

0.2

0.3

0.4

σ/σ

0.5

0.6

c

8

0.7

0.8

0.9

m

Fig. 4 Histogram of normalized peak/trough stresses of “group A” split into 8 ck-classes, k ∈ {1, 2,..., K c = 8} (all stresses divided by the maximum stress σ m of “group A” and “group B” recorded data sets).

Based on the shape of the histogram of stress levels, see Figure 4 for “group A” data set, M- and m-stress values are altogether split into Kc load classes. The structure mainly experiences tension loads and hence a very few compressions. All negative stresses are then truncated to zero which corresponds to class c1 in Figure 4. For the FMC model, all stresses of load sequences pertaining to a given class ci are replaced by a unique stress value si. Table 2 summarizes selected stress levels

{ (

}

)

si of the state space E = en = si ,s j , i > j and i, j ∈ {1, 2,..., K c }

for the two

groups of data. Table 2 Number of load classes and selected stress levels of the FMC model for “group A” and “group B” data sets.

“group A”: “group B”:

K c = 8 , si = {0, 0.051, 0.110, 0.186, 0.321, 0.482, 0.617, 0.761} K c = 5 , si = {0.039, 0.113, 0.248, 0.507, 0.840}

In the second model, the underlying unobserved process ( Sn )n∈`* of the HMM has the same state space as the FMC model, here termed Eh, except that no specific values are affected to Mn and mn but only their classes.

{ (

)

}

E h = en = ci , c j , i > j , i, j ∈ {1, 2,..., K c } (“groups A and B” data sets) (5)

Analysis of Fatigue Crack Growth under Random Load Sequences

407

Observed values of M n and mn are here obtained from a set of Kc-probability distributions

( fk , k ∈ {1, 2,..., Kc }) summarized in Table 3.

Table 3 Set of probability distributions f k , k ∈ {1, 2,..., K c } for “group A” and “group B” data sets.

c1 : 0 “group A”: c − c : Truncated Gaussian Distributions 2 7 c8 : Generalized Pareto Distribution ε = −0.08, β = 3.9, u = 0.676 “group B”: c1 − c4 : Truncated Gaussian Distributions c5 : Generalized Pareto Distribution ε = −0.20, β = 14.7, u = 0.676 For each group of data set, the transition matrix P of both FMC and HMM models is obtained by transforming the load sequences of recorded data set into load cycle states and estimating probabilities pi,j by the maximum likelihood method. The distribution X1 is obtained in a straightforward manner for both models from the first state of the sequences. The distribution of N is here the empirical distribution of the number of cycles per flight transformed into cycle load states. Accuracy of synthetic load sequences

The ability of these two inferred models to capture the statistical properties of the observed load flights is now appraised. We here consider the crack depth extension Δa, of the cracked specimen introduced in the second section, produced by load sequences with Nc sequential flights, Nc ={1, 200, 500, 1000} obtained either by simulations based on the two models, referred as case S, or by random sorting out of the load sequences of data sets, referred as case R. The number of simulated crack growths is taken to Nsim=10000 in each analysis (i.e. Nsim=10000 for each Nc value and for each group of data set “A” and “B”) in order to assess the two first statistical moments (mean and variance) and the following quantiles of Δa distribution: 0.1, 1, 10, 50, 90, 99 and 99.9 percentiles. Relative errors ( •case S − •case R ) •case R obtained between the statistical properties in case R considered as reference and those obtained from synthetic load sequences in case S are here represented in Figure 5 for the “group A” data set. For one single flight neither the FMC model nor the HMM model is able to adequately fit the reference Δa-distribution. For several cumulative flights Nc = {200, 500, 1000} a good agreement is obtained between the statistical properties of Δa based on the observed data and those based on the HMM model. The statistical properties of Δa obtained from simulations with the FMC model becomes less accurate with an increasing number of flights, compared to reference values. This is mainly due to the fact that the crack growth retardation model is highly sensitive to extreme values in the load history, which are not correctly modelled with the FMC model with finite state space: extreme values collapse in the single state value of cKc class.

408

C. Mattrand, J.-M. Bourinet, and D. Théret

80

80 Nc=1

N =1 c

Nc=200

60

Nc=200

60

Nc=500

20

0

20

0

−20

−20

−40

−40

−60

Nc=1000

40 Relative error (%)

Relative error (%)

Nc=500

Nc=1000

40

−60 Mean Std.

0.1

1

10

(a) FMC

50 90 Percentiles

99

99.9

Mean Std.

0.1

1

10

50 90 Percentiles

99

99.9

(b) HMM

Fig. 5 Relative error (in %) between statistical properties of Δa -distribution (Case S vs. Case R) for “group A” data set

Comprehensive methodology for modelling fatigue loads

A methodology for modelling random fatigue loads of an aircraft fleet under specific operational requirements given in-flight data is proposed based either on the FMC model or the HMM model. In this paper this methodology is applied on two groups of flights, named “group A” and “group B”, with clearly distinct flight domains. We denote FMC(A) and HMM(A) (resp. FMC(B) and HMM(B)) the defined Markovian models identified from in-flight load data of “group A” (resp. “group B”) in the following. An overall fatigue load model representative of the global activity of the aircraft fleet that encompasses altogether the random fatigue load models derived for each operational requirement can also be defined with a HMM model. The manner the whole random load models, here HMM(A) and HMM(B) or FMC(A) and FMC(B), switch along the flight sequence according to this global HMM model is shown in Figure 6. The reader may refer to [19] for details on the inference of these models. It is worth noting here that the proposed methodology for modelling random loads of a specific military aircraft fleet can be used not only for crack growth problems but also for fatigue and crack initiation applications under variable amplitude loading. The underlined Markovian models can also be appropriate in fields where data is either available or easily measured, e.g. structures subjected to sea or wind loads, road or driver-induced loads on cars…

Analysis of Fatigue Crack Growth under Random Load Sequences

0.886

0.114

A

409

0.114

B 0.886

One flight simulated with FMC(A) or HMM(A)

One flight simulated with FMC(B) or HMM(B)

Fig. 6 Comprehensive fatigue loads model for a French aircraft fleet.

6 Effect of Load Scatter on Crack Growth in a Skin-Stringer Panel of a Fighter Aircraft This section illustrates the stochastic crack growth in a structural component of an aircraft, introduced in the second section, considering the sole scatter in the loading. The previous HMM models, which have been found to properly render the scatter in loads, observed at the location of the aircraft structure under study, are used for the generation of random load sequences which gives this probabilistic analysis a real sound basis. Numerical simulations

Two analyses are performed here. In a first analysis only the scatter in loads measured on the French aircraft fleet subjected to the operational requirement “A” (i.e. the load scatter of flights in “group A”) is accounted for. In that case the stochastic model used for the generation of the random load sequences is the HMM(A) model, which has been found to be the most accurate of the two FMC(A) and HMM(A) models. The scatter in loads representative of the global activity of the aircraft under considerations (i.e. loads from both “group A” and “group B”) is simulated in the second analysis. The comprehensive fatigue loads model described in Figure 6 is used for this purpose as well as the underlying two HMM(A) and HMM(B) models. Crack growth calculations are performed using the PREFFAS crack closure model briefly recalled in the first section. In order to meet the assumption of stationary spectra of the PREFFAS model, see [7], it is required to limit the length of the load sequences which are repeatedly applied. We consider various lengths, here Nc = 100, 200, 500 or 1000 flights in order to assess the effect of this parameter. These Nc - flight long load sequences are then repeated sequentially, in order

410

C. Mattrand, J.-M. Bourinet, and D. Théret

to obtain a given total number of flights, NT = 6000 flights here. Four samples of Nsim = 5000 load sequences are simulated either from the HMM(A) model in the first analysis or from the comprehensive HMM model with underlying HMM(A) and HMM(B) models in the second analysis, each sample being Nc-flight long, where Nc = 100, 200, 500 or 1000. Hence, an overall 40000 crack growth calculations with NT = 6000 cumulated flights have been carried out. Results

Figure 7 gathers the results obtained in the two analyses. The coefficient of variation (c.o.v.) of the crack sizes (respectively the crack depth a and the crack length c) is reported every 1000 flights until the total number of applied flights NT = 6000 is reached. 0.4

0.4 Nc=100

0.35

Nc=100 0.35

Nc=200 Nc=500 a

0.25

a

0.2

a a

0.15

c

c

a

0.05

0

a

c

0.2

a

0.15

a 0.05

2000 3000 4000 5000 Number of applied flights (a) group A

c

c

0.1

c

1000

c

0.25

a a

0.1

c

c

Nc=1000

c

c

a

Nc=500

0.3

Nc=1000

cov (%) of crack sizes

cov (%) of crack sizes

0.3

a

Nc=200

6000

0

c

1000

2000 3000 4000 5000 Number of applied flights (b) groups A and B

6000

Fig. 7 c.o.v (%) for crack depth a and crack length c.

The c.o.v. of the crack sizes increases with the number of flights, in the range 1000-6000 flights. It is also noticed from these results that the dispersion on the crack sizes abruptly decreases with the length of the load sequence Nc. Hence, the length of the sequence which is repeated through the whole spectrum can be viewed as a critical parameter for crack growth analyses performed with the PREFFAS crack closure model. This remark is even more marked for the second analysis for which load sequences have been generated from the comprehensive model (i.e. the load scatter accounted for is representative of all the activities of the aircraft). This phenomenon might be explained by the c.o.v. of the value of the highest peak Nc-load sequence, noted Pmax, and the correlation between the crack size (a or c) and Pmax. The correlation between the crack depth a and Pmax is plotted on Figure 8. A strong negative correlation is appraised in both analysis which decreases to –1

Analysis of Fatigue Crack Growth under Random Load Sequences

411

with an increasing number of flights Nc per sequence. The crack growth process is hence highly sensitive to extreme load values in the Nc-flight load sequences. These global extreme values in the whole load sequences are responsible for the paramount retardation effect, which is accounted for with the PREFFAS model. The c.o.v. of Pmax decreases with the number of flights Nc per sequence, see Figure 9, explaining why the c.o.v. of the crack sizes decreases with the number of flights Nc per sequence, Nc being then a critical parameter.

−0.55

−0.55 1000 flights 2000 flights 3000 flights 4000 flights 5000 flights 6000 flights

−0.6

−0.65

−0.65

max

−0.7

−0.75

−0.75

r

r

a; P

max

−0.7

a; P

1000 flights 2000 flights 3000 flights 4000 flights 5000 flights 6000 flights

−0.6

−0.8

−0.8

−0.85

−0.85

−0.9

−0.9

−0.95 100

200 500 Number of flights N per seqence

1000

−0.95 100

200 500 Number of flights N per sequence

1000

c

c

(a) group A

(a) groups A and B

Fig. 8 Bravais-Pearson correlation ra; Pmax between the crack depth a and Pmax.

8 group A 7

groups A and B

cov (%) Pmax

6

5

4

3

2 100

200 500 Number of flights per block Nc

Fig. 9 c.o.v (%) of Pmax .

1000

412

C. Mattrand, J.-M. Bourinet, and D. Théret

Finally, a higher dispersion on Pmax together with a higher correlation between the crack depth a and Pmax are estimated in the second analysis (i.e. for “groups A and B”) than the ones obtained in the first analysis (i.e. for “group A”). It partially explains why the crack growth sizes results are more scattered in the second analysis. Consequently the global extreme values must really be considered as driving parameters for crack growth analyses under repeated Nc-flight load sequences.

6 Conclusion The framework of probabilistic fracture mechanics approaches is introduced to a certain extent in this paper. For this purpose the growth of a corner crack in a skinstringer panel of a fighter aircraft subjected to random variable amplitude loads is studied. The sole role of the scatter in loads on the crack growth process is addressed, which therefore neglects the potential impact of structural and material uncertainties. The novelty of this paper specifically lies in the description of the statistical randomness in loads from measured in-flight data. Two Markovian models are proposed for the modelling of random fatigue loads sustained by the aircraft fleet under two specific operational requirements. Their parameters are identified from an overall 27458 in-flight load sequences measured on a French military aircraft fleet. The ability of the two models to capture and reproduce the observed scatter in loads is then assessed by means of crack growth simulations using the PREFFAS crack closure model. Hidden Markov Models (HMM) is found to be the most appropriate tools for conveying the statistical properties of the real load sequences. The proposed methodology, which is done for each operational requirement of the aircraft fleet, is then applied for varying operational requirements, still in the framework of HMMs. Finally, the influence of load sequence parameters on the crack growth in a skin-stringer panel of a fighter aircraft is investigated. It can be pointed out that: • •

•

•

The crack growth process is highly sensitive to extreme load values of load sequences which are repeated sequentially in the whole load spectra. The sensitivity degree of the crack growth process to these extreme loads seems however to be dependent on the length of the load sequences, sequentially repeated in the whole load spectra, which has been found to have a strong effect on the crack growth dispersion. The repetition of these simulated load sequences introduces in fact an artificial sequence effect, i.e. the maximum overload is repeated at the same interval, which is of course unrealistic for aircraft in real operational conditions. Addressing fully random load sequences such as those generated by HMM models requires modifications of the PREFFAS model in order to circumvent this restrictive assumption on load sequences, which are currently undertaken.

Analysis of Fatigue Crack Growth under Random Load Sequences

413

To conclude it can be said that this work represents a step in a more global stochastic damage tolerance approach, which aims at defining optimal inspection schedules based on safety requirements.

Acknowledgments The work presented in this paper is financially supported by DGA which is gratefully acknowledged. The authors want also to thank DGA/Techniques Aéronautiques in Toulouse for providing them with the load data.

References [1] Provan, J.: Fracture, fatigue and mechanical reliability, MECH 471 & 571 Lecture Notes (2008), http://www.engr.uvic.ca/~mech471/ [2] Tong, Y.: Tech. Rep. DSTO-TR-1110, DSTO Aeronautical and Maritime Research Laboratory, Airframes and Engines Division (2001) [3] White, P.: Tech. Rep. DSTO-TR-1916, DSTO Platforms Sciences Laboratory, Air vehicule Division (2006) [4] McClung, R., Riha, D.: Tech. Rep. Probabilistic fracture mechanics guidelines and templates, Southwest Research Institute (2006) [5] Riha, D., McFarland, J., McClung, R.: Tech. Rep., Initial assessment for ”level 1” probabilistic fracture mechanics analysis approach, Southwest Research Institute (2008) [6] Sander, M., Richard, H.A.: Fatigue Fract. Engng. Mater Struct. 39, 303–319 (2006) [7] Aliaga, D., Davy, A., Schaff, H.: In: ASTM STP, vol. 982, pp. 491–504 (1988) [8] Elber, W.: In: ASTM STP, vol. 486, pp. 230–242 (1971) [9] Schivje, J.: Tech. Rep. LR-537, Faculty of Aerospace Engineering (1987) [10] NASGRO Manual (2006), http://www.nasgro.swri.org/ Version 5.0 [11] de Jonge, J., Schütz, D., Lowak, H.: Tech. Rep. NLR-TR-73029-U, National Aerospace Laboratory (NLR), Amsterdam (1973) [12] de Jonge, J.: Tech. Rep. NLR-TR-79056-U, National Aerospace Laboratory (NLR), Amsterdam (1979) [13] Rychlik, I.: Int. J. of Fatigue 18, 429–438 (1996) [14] Johannesson, P.: Probabilistic Engineering Mechanics 17, 123–130 (2002) [15] Bickenbach, F., Bode, E.: In: Proceedings of Congress of the European Regional Science Association (2002) [16] Craig, B., Sendi, P.: Health economics 11, 33–42 (2002) [17] Cappé, O., Moulines, E., Rydén, T.: Inference in Hidden Markov Models. Springer Series in Statistics (2005) [18] Rabiner, L.: In: Readings in Speech Recognition, A tutorial on hidden Markov Models and Selected Applications in Speech Recognition (1990) [19] Mattrand, C., Bourinet, J.-M.: In: Proceedings of AIAA Non Deterministic Approaches Conference, Denver, Colorado (2011)

26th ICAF Symposium – Montreal, 1-3 June 2011 Statistical Analysis of Fatigue Crack Growth Based on the Unigrow Model S. Mikheevskiy, S. Bogdanov, and G. Glinka University of Waterloo, Waterloo, Canada

Abstract. A variety of fatigue crack growth models have been developed during the last three decades aiming at capturing the effect of variable amplitude loading. However, all of them require using as the base a set of experimental fatigue crack growth data obtained under constant amplitude loading which can be highly inaccurate especially in the ‘near threshold’ region. The main purpose of the paper is to illustrate how the scatter of the input data (material constants) can affect the prediction of fatigue lives. Constant amplitude fatigue crack growth parameters were considered to be random variables with statistical distributions determined from the experimental data. The UniGrow fatigue crack growth model based on the local crack tip stress/strain material behaviour was combined with the MonteCarlo method in order to obtain the distribution of the final fatigue life based on the probability distributions of the input material parameters. A large set of experimental fatigue crack growth data for an aluminum alloy (7075-T6) was used for the verification of this methodology. The fatigue crack growth analysis was carried out for central through crack specimens. The simulated fatigue life distributions enable to determine the fatigue life corresponding to a given probability with a given confidence level. Comparison between theoretical and experimental results confirms the ability of the UniGrow fatigue crack growth model to simulate the load-interaction effect and shows the advantages of probabilistic approach for the fatigue crack growth analysis.

1 Introduction One of the difficulties arising while modeling the fatigue crack growth (FCG) process is sufficiently accurate estimation of residual stresses and strains ahead of the crack tip resulting from the cyclic plastic deformation of the material volume in the crack tip region. As a consequence of cyclic plastic deformations at the crack tip compressive residual stresses are induced around the crack tip by entirely tensile applied remote stresses or loads. Therefore, it is necessary to include the effect of the actual crack tip stresses and strains on the subsequent fatigue crack growth. Most of the existing fatigue crack growth models [1] are empirical in nature and they emphasize on the effect of the applied stress range without direct relation to the crack tip stress-strain affairs. On the other hand, most of the fatigue crack growth models are deterministic and based on the assumption that cracks propagate in an ideal continuum domain.

416

S. Mikheevskiy, S. Bogdanov, and G. Glinka

However real materials consist of randomly distributed microstructural elements which can greatly affect the fatigue crack growth. In addition, even when a series of tests on identical specimens are performed, wide scatter of the final data can still be observed. Therefore, deterministic approaches can be considered only as an approximation of the actual random fatigue crack growth process. Thus, the use of probabilistic models becomes necessary to make an accurate and reliable prediction of fatigue lives. This leads to the idea that certain deterministic parameters should be replaced by random variables with appropriate statistical distributions. However, the estimation of these distributions may be a very challenging task by itself due to the fact that the amount of fatigue crack growth data collected through the sets of experiments is usually not sufficient to distinguish between different types of statistical distributions. The importance of the scatter of fatigue crack growth parameters under both constant and variable amplitude loading spectra is discussed in the paper. The statistical analysis is based on the well-known Monte-Carlo method associated with the deterministic UniGrow model. The set of experimental data obtained by Porter [2] was simulated in order to verify the accuracy of the UniGrow model for both constant and variable amplitude loading spectra, and to show the difference between probabilistic and deterministic approaches.

2 The Unigrow Fatigue Crack Growth Model The UniGrow fatigue crack growth model, proposed by Noroozi and Glinka [3], is based on the idea that the fatigue process near cracks and notches is governed by highly concentrated strains and stresses. Therefore, the fatigue crack growth can be subsequently considered as a process of successive crack increments resulting from the material damage in the crack tip region. The two parameter driving force postulated by Vasudevan et.al [4] was also incorporated. It was postulated that the real material can be modeled as a set of elementary particles or material blocks of a finite dimension, ρ*. The assumption of the elementary material block implies that the actual stress-strain and fatigue response of the material near the crack tip is such as the crack had a blunt tip with the radius of ρ*. Therefore, the usual notch stress-strain analysis techniques can be applied in order to determine stresses and strains in the crack tip region. The following assumptions and computational rules form the base for the UniGrow fatigue crack growth model. • The material consists of elementary blocks of a finite dimension, ρ*. • The fatigue crack is regarded as a deep notch with a finite tip radius, ρ*. • The stress-strain analysis is based on the cyclic Ramberg-Osgood material stress-strain curve. • The number of cycles necessary to fracture the material over the distance, ρ*, ahead of the crack tip can be obtained using the cumulative fatigue damage concept and the Smith-Watson-Topper [5] fatigue damage parameter.

Statistical Analysis of Fatigue Crack Growth Based on the Unigrow Model

417

Based on the assumptions above Noroozi and Glinka [3] have analytically derived the fatigue crack growth expression in the form of

(

1− p 1− p da = C ( ΔK appl + K r ) ( K max,appl + K r ) dN

)

γ

= C ( Δκ )

γ

(1)

where, Kmax,appl and ΔΚappl, is the applied maximum stress intensity factor (SIF) and the stress intensity range respectively, and Kr is the residual stress intensity factor accounting for the effect of the crack tip residual stresses resulting from reversed plastic deformations. Very similar fatigue crack growth equation has been proposed by Walker [6] based on empirical fitting of observed constant amplitude fatigue crack growth data obtained at various stress ratios, R. However, Walker’s expression do not take into account the fact that the correlation between the stress intensity factor and the crack tip stress/strain field is often altered by the residual stress resulting from reversed plastic deformations. It was also found [7] that the instantaneous fatigue crack growth rate depends not only on the residual stresses produced by the recent loading cycle, but on a number of stress fields generated by preceding cycles. Therefore, set of “memory rules” has been formulated [7] based on the experimental observations of fatigue crack growth under variable amplitude loading spectra. Detailed description of the UniGrow model and additional verification data can be found in references [3-7].

3 Material, Load, and Specimen Geometry The experimental fatigue crack growth data for constant and variable amplitude were taken from reference [2]. Porter collected fatigue crack growth data for central through crack specimens made of AL 7075-T6 alloy. Specimens were made of 305mm wide, 915mm long, and 4.1mm thick panels for which the material properties were: modulus of elasticity, E = 69000MPa, yield limit, Sys = 520MPa, and ultimate strength, Sult = 575MPa. The Ramberg-Osgood cyclic material properties were obtained from the strain controlled constant amplitude experimental data provided by Jiang [8]. The constant amplitude fatigue growth data obtained at seven different stress ratios were provided by Jiang [9] and Newman [9]. All constant amplitude fatigue crack growth data sets are presented in Figure 1 in terms of applied stress intensity range. All experimental constant amplitude fatigue crack growth data sets were finally presented in terms of the total driving force, Δκtot, (Eqn. 1) resulting in one ‘master curve’ shown in Figure 2. The ‘master curve’ was divided into two segments and each segment was subsequently approximated by a line fitted using the linear regression method. These lines represent mean values of the constant amplitude

418

S. Mikheevskiy, S. Bogdanov, and G. Glinka

fatigue crack growth data for the ‘near threshold’ and the Paris regions. The combined set of data presented in Figure 2 consists of about 2000 data points scattered around mean lines. The ‘master curves’ or mean curves are shown in Figure 2 together with the scatter band corresponding to 99% probability. Step-wise loading spectra shown in Figure 3 were chosen for investigating the effect of multiple overloads on the fatigue crack growth. Constants ‘m’ and ‘n’ denote the amount of large and small cycles respectively. Deterministic (using mean values of C1 and C2) and probabilistic (10000 simulations with random C1 and C2) fatigue crack growth simulations were performed and compared with experimental data for 2 constant amplitude loading spectra ([n = 0, m = 50] and [n = 50, m= 0]) and 6 variable amplitude spectra (n = 50, m = [1, 3, 6, 10, 25, 50]). The results will be discussed later.

Fig. 1 Constant amplitude FCG data in terms of applied SIF range.

4 Estimation of Statistical Distribution of Material Parameters As mentioned earlier, the paper addresses the stochastic nature of the constant amplitude fatigue crack growth data and its effect on final life predictions under variable loading. In other words, the constant C in Eqn. 1 should be replaced by random variable estimated from the scatter of the experimental data (Figure 2). It has been found earlier that the random variable Y=log(C) is distributed normally, and therefore the random variable C should have lognormal distribution.

Statistical Analysis of Fatigue Crack Growth Based on the Unigrow Model

419

Fig. 2 Constant amplitude FCG data in terms of total driving force.

L oa

m - cycles n - cycles

Tim

One block Fig. 3 Schematic loading history.

First, the set of experimental fatigue crack growth data was divided into two parts corresponding to the ‘near threshold’ and the Paris regions and described by two pairs of constants (C1,γ1) and (C2,γ2) . The numerical values of these parameters were obtained using the least square method: C1=1.158e-13,

420

S. Mikheevskiy, S. Bogdanov, and G. Glinka

γ1=26.9, C2=2.89e-10, γ2=3.48. The following expression allows estimating the standard deviation of error.

σε =

∑ε

⎛ ⎛ da ⎞ ⎞ and ε = ⎜ log ⎜ ⎟ − log(C ) − m log ( Δκ i ) ⎟ n−2 ⎝ ⎝ dN i ⎠ ⎠ 2 i

2 i

2

(2)

Where

σ ε denotes

points,

da dNi is fatigue crack growth rate for experimental point ‘i’, and

the standard deviation of the error, n is the number of data

Δκ i is the total driving force corresponding to point ‘i’. The value of σ log C ,1 was found to be equal 0.74 and

σ log C ,2 is

equal 0.21. The standard deviation of the

data in the ‘near threshold’ region is much higher than in the Paris region due to the larger scatter of the experimental data (Figure 2).

5 The Monte-Carlo Analysis Since the time high power computers entered into our lives, the Monte-Carlo method became one of the most powerful tools for researches in all fields of science. There are numerous types of the Monte-Carlo algorithms; however all of them follow the same pattern. 1. 2.

3.

4. 5.

Define the domain of possible inputs. In our case C1 and C2 should be greater than zero. Simulate the random variable with uniform distribution. The randDouble() function from standard C++ library has been used to simulate the uniform random variable in the interval from 0 to 1. Generate random inputs for specific distribution based on uniformly distributed random variable. Lognormal distribution has been used in the case of constant amplitude fatigue crack growth data scatter. Perform deterministic computations using random input. The UniGrow fatigue crack growth model has been used for calculating fatigue lives. Aggregate the results of individual runs into the final set of results.

In order to simulate random values corresponding to lognormal distribution the Box-Muller [10] method has been implemented. The probability density and the cumulative distribution functions for parameter C2 are shown in Figure 4. The sampled values of parameters C1 and C2 were subsequently inputted into the UniGrow model and deterministic calculations were carried out. The final sets of results are discussed in the following section.

Statistical Analysis of Fatigue Crack Growth Based on the Unigrow Model

421

Fig. 4 Cumulative probability distribution and probability density function for parameter C2.

6 Fatigue Life Predictions This section presents the deterministic and probabilistic results obtained using the UniGrow model and compares them with experimental data. It has been shown by a large number of investigators that an application of a single tensile overload (or multiple overloads) can result in a significant deceleration in fatigue crack growth. The same effect has been observed by Porter [2], and was analysed using the UniGrow model. Fatigue crack growth was first recorded under the constant amplitude loading spectrum. Next, high (approximately 90%) tensile overloads were applied after every 50 cycles of constant amplitude loading. In the next set of tests, the number of tensile overloads has been increased from 1 to 50 cycles (3, 6, 10, 25, 50). The summary of the experimental results and deterministic life predictions based on the mean values of C1 and C2 is shown in Table 1. Table 1 Comparison between experimental and theoretical results.

Name S1 S2 S3 S4 S5 S6 S7 S8

Description CA n=50, m=0 CA n=0, m=50 VA n=50, m=1 VA n=50, m=3 VA n=50, m=6 VA n=50, m=10 VA m=50, n=25 VA m=50, n=50

Experiment 75000 20000 211650 132600 103600 75000 49000 37500

UniGrow 66000 18800 226220 159900 113800 85000 49700 35000

Error % -12 -6 7 21 10 13 1 -6

422

S. Mikheevskiy, S. Bogdanov, and G. Glinka

The longest fatigue life was obtained using S3 loading spectrum and it gradually decreases as the number of overloads per 50 cycles goes up. Since overload cycles are applied quite often, local stresses and strains in the vicinity of a crack tip are mostly governed by the high tensile loading cycles. The addition of new high cycles therefore does not produce any extra residual stresses which can decelerate fatigue crack growth. Moreover, fatigue crack growth increases due to the larger magnitude of applied loading cycles. The probabilistic analysis described in the previous section was performed for each loading spectrum. Figure 5 shows the experimental data for the constant amplitude S1 spectrum together with the scatter band corresponded to the scatter shown in Figure 2. The UniGrow model was run 10000 times in order to simulate the distribution of lives with the final crack length a=60mm. Simulated distribution was fitted based on the Kolmogorov-Smirnov goodness parameter, and the best fit was obtained using Birnbaum-Saunders [11] two-parameter distribution with parameters α=0.45253 and β=65054 (Figure 6). The mean fatigue life estimated by the UniGrow model is relatively close to the experimental data (~12% error), but the possible scatter is rather large. As mentioned earlier, the same statistical analysis was performed for all of the variable amplitude loading spectra listed in Table I. Figure 7 and Figure 8 show experimental data, deterministic prediction, scatter band and probability density function of fatigue life for the S3 spectrum (which has the longest life). The distribution of final lives was obtained using the Birnbaum-Saunders three-parameter distribution with α=0.536, β=210000, and γ=-11761. The deterministic prediction based on the mean values of C1 and C2 parameters is very close to the experimental result. However, the deviation of fatigue lives obtained using the Monte-Carlo analysis is higher than it was in the case of constant amplitude loading.

Fig. 5 S1 FCG prediction, test and scatter.

Statistical Analysis of Fatigue Crack Growth Based on the Unigrow Model

Probability Density Function 1.4E-5 1.2E-5

f(x)

1E-5 8E-6 6E-6 4E-6 2E-6 0 0

50000

100000

150000

200000

Number of Cycles Fatigue Life (0.45253; 65054.0)

Fig. 6 Probability density function for fatigue life distribution under CA loading.

Fig. 7 S3 FCG prediction, test and scatter.

423

424

S. Mikheevskiy, S. Bogdanov, and G. Glinka Probability Density Function 4E-6 3.6E-6 3.2E-6

f(x)

2.8E-6 2.4E-6 2E-6 1.6E-6 1.2E-6 8E-7 4E-7 0 0

200000

400000

600000

x Fatigue Life (0.53609; 2.1010E+5; -11761.0)

Fig. 8 Probability density function for fatigue life distribution under S3.

This effect can be explained using the following argument. In the case of variable amplitude loading the uncertainties in the C1 and C2 parameters effect not only the instantaneous fatigue crack growth, but the retardation effect of tensile overloads as well. For example, in the case of high values of C1 and C2 fatigue crack grows much faster and can propagate out of the overload’s influence zone before the next overload is applied. Other words, the fatigue crack propagates so fast that material doesn’t have enough time to build up any sort of resistance ahead of the crack tip.

7 Conclusions In this paper the importance of the probabilistic approach to fatigue crack growth analysis was shown for both constant and variable amplitude cases. Fatigue crack growth was simulated using the Monte-Carlo method and the UniGrow fatigue crack growth model. The final fatigue life distribution was obtained and fitted using the BirnbaumSaunders distribution. The knowledge of the final life distribution allows for the set up of safe inspection intervals based on the desire probability of failure. The ability of the UniGrow model to account for the retardation effect of large repeated overloads was shown using the set of experiments performed by Porter [2].

References [1] Beden, S.M., Abdullah, S., Ariffin, A.K.: Euro. Jr. of Scientific Research 28(3), 364–397 (2009) [2] Porter, T.R.: Eng. Fract. Mech. 4, 717–736 (1972)

Statistical Analysis of Fatigue Crack Growth Based on the Unigrow Model [3] [4] [5] [6] [7] [8] [9]

425

Noroozi, A.H., Glinka, G.: Int. Jr. Fatigue 29, 1616–1634 (2007) Vasudevan, A.K., Sadananda, K.: Mat. Science and Eng. 188A, 1–22 (1994) Smith, K.N., Watson, P., Topper, T.H.: Jr. of Materials 5(4), 767 (1970) Walker, E.K.: ASTM STP 462, 1–14 (1970) Mikheevskiy, S., Glinka, G.: Int. Jr. of Fatigue 31(5), 1829–1836 (2009) Zhaoa, T., Zhanga, J., Jiang, Y.: Int. Jr. Fatigue 30(7), 1169–1180 (2008) Newman, J.C., Wu, X.R., Venneri, S.L., Li, G.G.: Small-crack effects on high strength aluminum alloys, Report No. A, NASA/CAE Cooperative Program, NASA Reference Publication 1309 (1994) [10] Box, G., Muller, M.E.: The Annals of Math. Statistics 29(2), 610 (1958) [11] Birnbaum, Z.W., Saunders, S.C.: Jr. Appl. Probability 6(2), 319 (1969)

26th ICAF Symposium – Montreal, 1-3 June 2011 Fatigue Life Estimation of Structures Subjected to Vibratory Loading M. Fressinet1, F. Fuchs2, and P. Madelpech1 1

2

DGA Aeronautical systems, Toulouse, France Helmut Schmidt Universität, Hamburg, Germany

Abstract. Fatigue life is commonly estimated by an analysis of the stress time history through a peak-valley counting method and the damage is calculated thanks to a damage summation method. Unfortunately, vibratory loadings are often random and this kind of calculation would be very time consuming. Such spectra require other methods and that’s why especially dedicated models have been developed. Known as spectral methods, they enable to calculate the fatigue damage in the frequency domain where the loading is expressed as a power spectral density function (PSD) of stresses [1]. Based on a large variety of PSD and fatigue spectrum, the objective of the study has been to test and compare the accuracy of different models found in the literature such as the one’s developed by Dirlick [2] or Tovo-Benasciutti [3]. The robustness and the sensitivity of different parameters such as the mean stress correction methods, the RMS value or the slope of the Wöhler curve, have been studied too. Finally an introduction to the safety factors to apply to these models is proposed in order to take account of their discrepancy.

1 Introduction Fatigue life is commonly estimated by an analysis of the stress time history through a peak-valley counting method and the damage is calculated thanks to a damage summation method. Unfortunately, vibratory loadings are often random and this kind of calculation would be very time consuming. In order to take account the randomness of the signal, the analysis is usually performed in the frequency domain where the loading is expressed as a power spectral density function (PSD) of stresses [1]. After a short introduction to the concept of power spectral density function (PSD) and to the principle of fatigue life estimation in the frequency domain, a brief review of the different studied models is made. The different sets of simulations are then exposed and the results are analysed. This article finally concludes on the accuracy of the different models and on an introduction to the safety factors to apply to these models.

2 Vibratory Loading and PSD Function As previously mentioned, vibratory loadings are usually random and therefore, the signal x(t) can only be described statistically. These time signals are generally

428

M. Fressinet, F. Fuchs, and P. Madelpech

considered as stationary, ergodic and Gaussian and thus can be statistically determined in the time domain univocally by their autocorrelation function Rx(τ) which is defined for a real process by:

R x (τ ) = E [x(t ) x(t − τ )] where E is the stochastic mean

(1)

In the frequency domain, the Fourier transform of the autocorrelation function is often used to characterize the signal. Denoted Sx(f), the one sided PSD usually used in mechanics can be expressed like that: ∞

S x ( f ) = 2∫ R x (τ )e − 2 jπfτ dτ −∞

Sx ( f ) = 0

f >0 f ≤0

(2)

The PSD is usually characterised through its spectral moments of different orders, the nth moment being expressed as follows: ∞

mn = ∫ f n S x ( f )df

(3)

0

Thanks to these spectral moments, the irregularity factor γ which is usually used to measure the bandwidth of the signal can be calculated:

γ=

m 22 m0 m 4

(4)

An irregularity factor close to 1 corresponds to a narrow band process whereas a wide band process is characterised by an irregularity factor approaching 0. An other very important result from Rice [4] lies in the fact that for a zero-mean process, the average expected rate of occurrence of peaks E[P] and of mean upcrossing E[0+] (i.e. zero crossing with positive slope) can be found from these spectral moments: E[ P] =

m4 m2

and E[0] = m2

(5-6)

m0

In the end, for a zero mean Gaussian process, the 0th moment of the PSD is equal to the well known RMS value of the signal.

3 Methodology to Assess Fatigue Damage in the Frequency Domain The damage is directly estimated from the stress power spectral density function. Due to the randomness of the signal the damage is an average damage per unit of time expressed by the following formula:

Fatigue Life Estimation of Structures Subjected to Vibratory Loading

E[ D ] = ∫

Sb .E[ P ]. f S ( S ).dS C

429

(7)

In fact this formula is nothing else but the rewriting of the Miner’s rule after having extracted different elementary cycles from the complex spectrum. Two terms can be distinguished in this expression: - E[ P]. f S ( S ).dS : f S (S ) represents the probability density function of having extracted from the signal an elementary cycle with an amplitude S. The term E[P ] is the average number of peaks of the signal per unit of time. Once multiplied together with dS, it comes the average number of cycles per unit of time with amplitude comprised between S and S+dS in the signal. In fact it corresponds to the results of a rainflow counting method. -

Sb represents the damage of an S amplitude cycle when the Basquin C

model is used to approximate the Wöhler‘s curve. Once integrated between 0 and +∞, the average damage per unit of time of the signal is obtained. Before going ahead, it must be emphasised that usually elementary cycles extracted from the signal are not only defined through their amplitudes but also through their mean values. As a matter of fact, fS(S) should be replaced by a joint probability density function fS,M(S,M). In fact for a Gaussian process, the mean stress effect can be neglected, a strong hypotheses which will be discussed later on. At last it must be noticed that in this article, the models are expressed in terms of stress range and not in terms of stress amplitude.

4 The Different Studied Models In fact it can be said that one model corresponds to one expression of the probability density function f S (S ) . The Narrow Band approach (NB) The first model found in literature has been proposed by Bendat [5] who assumed that the probability density function corresponds to a Rayleigh distribution. Some additional explanation about this model can be found in [1] but in this article we will just remind its expression: −S2

S 8m0 with m the 0th moment of the PSD f s (S ) = e 0 4m0

(8)

430

M. Fressinet, F. Fuchs, and P. Madelpech

In the formula (7), it is common to replace E[P] by E[0+] even if for a strictly narrow band process (i.e. γ=1) they are equal to each other. Finally, once integrated in the formula, the following expression can be used:

[ ]

E[ D NB ] = E 0 + with the gamma function Γ( x ) =

+∞

∫t

(

)

b ⎛ b⎞ C −1 2 2m0 Γ⎜1 + ⎟ ⎝ 2⎠

(9)

x −1 −t

e dt

0

Models based on the narrow band approach As it will be highlighted later, the NB approach is usually very conservative when applied to wide-band signals. Thus some authors proposed to modify the NB model through a correction factor leading to the following expression:

E[ DWB ] = λWB × E[ D NB ]

(10)

Wirshing and Light (WL) This is a purely empirical model [6] based on a large number of simulations and dependent on the irregularity factor and the slope of the Wöhler curve.

λWL = A(b) + [1 − A(b)]× [1 − ε ]

B (b )

⎧ A(b) = 0,926 − 0,033b ⎪⎪ with ⎨ B (b) = 1,587b − 2,323 ⎪ ⎪⎩ε = 1 − γ 2

(11)

Ortiz and Chen (OC) Proceeding with the same manner, Ortiz and Chen [7] developed the following correction factor:

λOC =

m2 m 2 βb b with β = γ m2 m2+ 2

(12) b

Tovo and Benasciutti (TB) The model developed by Tovo and Benasciutti [2] is a semi empirical factor based on Ryshlick results [8]. The NB correction factor is expressed as follows:

λTB = β + (1 − β )α 2b −1

(13)

Fatigue Life Estimation of Structures Subjected to Vibratory Loading

431

[

⎧ (α 1 − α 2 ) A(1 + α 1α 2 − (α 1 + α 2 ) )e Bα 2 + α 1 − α 2 β = ⎪ (α 2 − 1)2 ⎪ with ⎪ m1 m2 ⎨ α1 = α2 = ⎪ m0 m2 m0 m 4 ⎪ ⎪⎩ A = 1.112 B = 2.11

]

Direct models The models which are presented hereafter are directly linked to the probability density function of rainflow extracted cycles. Chaudhury and Dover (CD) The probability density function proposed initially by Chaudhury and Dover [9] is the sum of a Gaussian and a Rayleigh distribution.

f s (S ) =

⎡ ⎛ Z2 ⎢ D1 exp ⎜⎜ − m0 ⎣ ⎝ 21−γ

1

(

2

)

⎞ ⎛ Z 2 ⎞⎤ ⎟⎟ + D 2 ZQ exp ⎜⎜ − ⎟⎟ ⎥ ⎠ ⎝ 2 ⎠⎦

(14)

⎧ 1− γ 2 γ D2 = ⎪ D1 = 2 2π with ⎪ ⎨ ⎛ γ Z ⎞ ⎪ Q = 1 + erf ⎜ ⎟ ⎪ ⎜ 2 1− γ 2 ⎟ ⎝ ⎠ ⎩

(

)

and where erf is the error function expressed by:

erf ( x ) =

2

π

∫

x

0

e − t dt = 2

2

π

+∞

∑ (− 1) 0

n

x 2 n +1 n! (2 n + 1)

.

This model will be named “Chaundry and Dover Classique” to be distinguished from the one developed by Kam and Dover [10] which is the same as previously ⎞ ⎛ but where the term erf ⎜ γ Z ⎟ has been replaced by the following empirical 2 ⎜ 2 1− γ ⎟ ⎠ ⎝ function:

(

)

⎧ g (γ ) = 1 if γ ≥ 0,96 ⎨ 2 3 4 5 6 7 ⎩ g (γ ) = 0,3012γ + 0,4916γ + 0,9181γ − 2,3534γ − 3,3307γ + 15,6524γ − 10,7846γ else

This last model will be named “Chaundry and Dover modifié”.

432

M. Fressinet, F. Fuchs, and P. Madelpech

Dirlick (DK) Dirlick [2] proposed a model which is the sum of an exponential and two Rayleigh distributions: −Z

−Z 2

D1 Q D2 Z 2 R 2 e + 2 e + D3 Ze Q R f s (S ) = 2 m0

−Z 2 2

with Z =

S

(15)

2 m0

and several factors determined empirically

m1 m2 m2 m 4 1 − α 2 − D1 + D12 D2 = 1− R xm =

D1 =

2(x m − α 22 ) 1 + α 22

α 2 − xm − D12 1 − α 2 − D1 + D12 1.25(α 2 − D3 − D2 R )

R=

D3 = 1 − D1 − D2

Q=

D1

Zhao and Backer (ZB) The probability density function proposed by Zhao and Backer [11] is the sum of a Weibull and a Rayleigh distribution: β

f s (S ) =

wαβZ β −1e −αZ + (1 − w) Ze

−Z 2 2

2 m0

with Z =

S

(16)

2 m0

and several factors determined empirically

α = 8 − 7γ

⎧1,1 if γ < 0,9 w= ⎩1,1 + 9(γ − 0,9) if γ < 0,9

1− γ

β =⎨

1−

2

π

Γ(1 +

1

β

)α

−

1

β

Single moment (SM) This model proposed by Larsen and Lutes [12] holds its name from its dependency on only one spectral moment of the PSD. The integrated form of the formula can be expressed as follows:

( )

b ⎛1+ b ⎞ E[ DSM ] = C −1 2 2 ⎛⎜ m 2 ⎞⎟ Γ⎜ ⎟ b ⎠ ⎝ ⎝ 2 ⎠ b

(17)

Fatigue Life Estimation of Structures Subjected to Vibratory Loading

433

5 Comparison Methods In order to compare the different models, extensive sets of simulations were performed. The basic principle of the method is given hereafter: - More than 600 PSD of different types defined on the 0-100Hz band were used for the simulation. Special care was taken in order to use realistic PSD and it was chosen to work with multimodal PSD (see picture n°1).

n ⎛ ( f − f ci )2 ⎞ ⎟ with n = 1..6 S ( f ) = ∑ Ai × exp⎜⎜ − ⎟ Bi i =1 ⎠ ⎝

Picture 1 The different types of PSD used in this study.

- For each PSD, a set of time signals were generated thanks to the formula N

X (t ) = ∑ 2 S s ( f k )Δf cos(2πf k t + ϕ k ) where N is the number of points chosen k =0

to draw the PSD, Δf the frequency step between two points ( f k = f min + kΔf ) and φk a random phase angle defined independently for each frequency fk. Then the fatigue life of each signal was calculated thanks to the rainflow counting method, a certain mean stress correction method and the Miner rule. An average fatigue life was calculated and defined as the reference fatigue life for the studied PSD. - For each PSD the fatigue life was also directly calculated in the frequency domain by the different models presented previously. - A comparison between both fatigue lives was made for each PSD. - The following well known error indicator was calculated on the whole set of PSD in order to compare the different models:

1 IE = 100 × ntot

⎛ ⎛ Tspec ,i ⎜ log⎜ ∑ ⎜ ⎜ Tref ,i i =1 ⎝ ⎝ ntot

2

⎞⎞ ⎟ ⎟ where Tspec is the predicted fatigue life in ⎟⎟ ⎠⎠

the frequency domain and Tref the predicted fatigue life in the time domain for the PSD n°i.

6 The Different Studied Parameters The objective of the study was to determine the most influentiable parameters on the error committed by the different models. Thus it was decided to study the influence of:

434

M. Fressinet, F. Fuchs, and P. Madelpech

- the Basquin slope of the Wöhler curve. It was taken equal to 3, 5, 7, 9, 11 and 13 while the C factors was defined in order to have 1000 cycles to failure for an applied stress of 450 MPa, this value being defined as the tensile strength of the materials S_u. - the RMS value expressed in term of percentage of S_u. - the bandwidth of the signal in term of irregularity factor: thanks to well selected PSD, the irregularity factor covers the range between 0.1 an 1.0. - the mean stress correction method: the Goodman and the Gerber models were compared to a calculation without any mean stress correction method.

7 Parameters of the Time Signals The first step was to determine the number and length of time signals it was necessary to generate to get a right estimation of the average fatigue life of the signal. Thus a purely empirical study was carried out. It consisted in using a set of 46 unimodal PSD and then to generate 100 time signals of a duration of 500s, 1000s, 2000s, 4000s or 6000s. The fatigue life of each signal was calculated. Finally for each time duration, the evolution of the mean fatigue life and the associated standard deviation versus the number of time signals taken into account was studied. For these simulations no mean stress correction method was used and the b factor of the Basquin model was set to 13 due to the fact that a high b factor (i.e. a very horizontal curve)creates a large discrepancy. Average fatigue life evolution (b=13 ; duration = 500s)

Average fatigue life evolution (b=13 ; duration = 500s) 45

40

Relative standard deviation (%)

Normalized fatigue life

1

0,9

0,8

0,7

0,6

35

30

25

20

15

10

5

0,5

0

0

10

20

30

40

50

60

70

80

90

100

0

10

20

30

40

50

60

70

80

90

100

Number of time signals

Number of time signals

Average fatigue life evolution (b=13 ; duration = 6000s)

Average fatigue life evolution (b=13 ; duration = 6000s) 45 40

Relative standard deviation (%)

Normalized fatigue life

1

0,9

0,8

0,7

0,6

35 30 25 20 15 10 5 0 0

0,5 0

10

20

30

40

50

60

Number of time signals

70

80

90

100

10

20

30

40

50

60

70

-5

Number of time signals

Picture 2 Influence of the number of time signals on the fatigue life.

80

90

100

Fatigue Life Estimation of Structures Subjected to Vibratory Loading

435

The picture above shows the expected result that the longer is the time signal, the faster is the convergence. Moreover for a time duration of 6000s, the standard deviation falls under 10% with only 10 time signals. In this study, for time calculation reasons, it was decided to generate 40 time signals of 2000s, which enable to have a maximal error lower than 10%. Finally it must be noticed that due to the empirics of the method, these figures deeply depends on the frequency range on which the DSP are defined and must be used carefully for other DSP.

8 Influence of the Mean Stress Correction Method The following picture shows the influence of the mean stress correction method on the predicted fatigue life, each PSD being considered separately. Influence of the mean stress correction method (b=13)

Normalized fatigue life

1,05

1

0,95

0,9

0,85

0,8

Goodman

Gerber

Without correction

Picture 3 Influence of the mean stress correction method.

For a Gaussian process, the order between the different methods in term of predicted fatigue life is the same for each signal. Without any mean stress correction method, the fatigue life is always greater than with the Goodman and Gerber method but they are very close to each other, with in our case a maximum gap of 16%. To understand this tendency it must be stressed that for a gaussian process, the rainflow matrix is symmetrical. As a matter of fact, when a cycle with a certain mean stress and stress range is extracted, a cycle with the same stress range but with an opposite mean value exists statistically as shown in the following picture.

Mean stress

Picture 4 Example of a rainflow matrix for a gaussian process.

436

M. Fressinet, F. Fuchs, and P. Madelpech

At last it must be pointed out that in this article, the results compared to the different correction method are always presented in order to let the reader have an idea of the committed error on his commonly used method.

9 Influence of the RMS Value Influence on the commited error It can be easily demonstrated that the RMS value has no influence on the error committed by the different model compared to a calculation in the time domain without any mean stress correction method (the RMS value is actually only a scaling factor which can be extracted from both Tref and Tspec). Moreover, as the mean stress correction method has a low influence on the predicted fatigue life, it can be concluded that the RMS value has no real influence in every case. It is confirmed by the following table representing the committed error for RMS values between 0.05 and 0.25 times S_u: Table 1 Error indicator for various RMS value.

As expected, for RMS values between 0.05 and 0.20 times S_u, the committed error compared to a fatigue life calculation in the time domain without any mean stress correction method is always the same while the variation for other methods can be considered as insignificant. In all cases, the hierarchy between the different models does not change significantly. Concerning the calculation with a RMS value equals to 0.25 S_u, the discrepancy can be explained by the fact that for such high RMS values, the time domain calculation is not valid anymore because

Fatigue Life Estimation of Structures Subjected to Vibratory Loading

437

a none negligible number of points are above S_u, which is not acceptable. As a matter of fact, the RMS value was set to a value under 0.20 S_u and the results were considered as applicable to every possible RMS value. Finally the fact that the RMS value has no influence explains the reason of such a large range of b values. Indeed with a given RMS value, it will allow to assess the committed error of the different models in the low cycle fatigue domain (b between 3 and 7) and in the very high cycle life (b>7) too where the Wöhler curve is still decreasing but with a smoother slope. Influence on the fatigue life One other objective was to understand if this single RMS value of the signal was sufficient enough to determine the fatigue life of a structure subjected to vibratory loading. Indeed while the different models presented previously depend on more than one spectral moment, some people seems to use in some cases (see [14] for example) nothing more than a Wöhler curve expressed in terms of RMS stresses (i.e. the 0th spectral moment) versus the number of cycles to failure. In the following picture, the evolution of the fatigue life in function of the irregularity factor for a given RMS value can be discovered (results based on the whole set of PSD).

Fatigue life VS irregularity factor (b=3) 10000 9000

Fatigue life (s)

8000

Goodman 7000

NB with f0=100 6000 5000 4000 3000 2000 1000 0

0

0,2

0,4

0,6

0,8

1

Irregularity factor

Picture 5 Influence of the irregularity factor on the fatigue life.

Thus, it can be noticed that the fatigue life decreases when the irregularity factor tends to 1. Moreover it seems possible to define an asymptote. Indeed the Narrow band model which is exact for a strictly narrow band process (i.e. γ=1) leads to:

TNB =

k

[ ] (2

E 0+

2 m0

)

b

⎛ b⎞ Γ⎜1 + ⎟ ⎝ 2⎠

(18)

As E[0+] corresponds to the number of times that the time signal crosses the zero axis with a positive slope, it comes easily that E[0+]≤fmax. Thus the limit for the

438

M. Fressinet, F. Fuchs, and P. Madelpech

NB cases becomes

k

Tinf = f max

(2

)

b ⎛ b⎞ 2m0 Γ⎜1 + ⎟ , a limit, which seems to ⎝ 2⎠

be a limit for the whole set of bandwidth too and which corresponds to the horizontal (“NB with f0=100”) line in the previous picture. As a matter of fact it seems not inaccurate to use a Wöhler curve expressed in term of RMS value at the only condition that it has been obtained thanks to a narrow band signal (γ≥0.99).

10 Influence of the B Factor of the Basquin Model The following table presents the results of the simulations for the whole set of PSD in term of indicator error: Table 2 Error indicator for the different models for each b value.

First of all it can be noticed that the higher is the b coefficient of the Basquin model (i.e. the more horizontal the Wöhler curve is), the higher is the error committed by the different models. It can be easily explained by the fact that when an error is committed on the rainflow distribution (i.e. fS(S)), it is multiplied by the stress to the power of b. However, it could be different for some models whose probability density function depends on b and for which the trend might have been inverted. From this table, it can be found out that for b value between 3 and 7, the most accurate models are the Dirlick (DK), Tovo-Benasciuti (TB) and Ortiz & Chen (OC). For higher b value (i.e. for smoother slopes), the DK, Zaho-Backer(ZB) and Chaundry & Dover « modifié » (CD « mod ») are the most accurate. Moreover, if the error indicator is calculated on the whole set of b values, the DK one is the best.

Fatigue Life Estimation of Structures Subjected to Vibratory Loading

439

Table 3 Error indicator for the different models for the whole set of simulations.

11 Influence of the Bandwidth On the following picture, it can be observed that the accuracy of the different models deeply depends on the irregularity factor. Thus, the different models are much more accurate for narrow band processes. It is not surprising due to the fact that the models tend to the NB model when γ tends to 1. It must be reminded that the NB model is theoretically founded while unfortunately when the bandwidth increases, there is no more analytical link between the PSD and the fatigue life, which explains the increasing error. Influence of the bandwidth (b=3)

Influence of the bandwidth (b=3) 2,5

2 NB

1,8

WL 1,6

CD mod

TB 1,5

1,2

T spec / T ref

Tspec / Tref

2

OC

1,4

1 0,8

ZB

DL 1 OC

0,6 SM

0,5

0,4 0,2

0

0 0

0,1

0,2

0,3

0,4 0,5 0,6 Irregularity factor

0,7

0,8

0,9

0

1

0,2

0,4

0,6

0,8

1

1,2

Irregularity factor

Influence of the bandwidth (b=13)

Influence of the bandwidth (b=13)

2

2,5

1,8 CD mod

2

1,6

ZB

1,2

T spec / T ref

Tspec / Tref

1,4

1 0,8

1,5 DL

OC

1

SM

0,6 NB 0,4

0,5

WL OC

0,2

TB 0

0 0

0,1

0,2

0,3

0,4 0,5 0,6 Irregularity factor

0,7

0,8

0,9

1

0

0,2

0,4

0,6

0,8

1

1,2

Irregularity factor

Picture 6 Error indicator for the different models (b=3 and b=13).

More precisely, the different models give quite accurate prediction when γ is higher than 0.9 even for b value equal to 13. The only exception is the Wirshing & Light model (WL), which has been developed in a very empirical manner for a b value around 3. If the W&L model is excluded, the maximum error commited by

440

M. Fressinet, F. Fuchs, and P. Madelpech

the different models for γ>0,9 is below 35%. For γ≤0,9, the results are more mitigated. An intermediate value of 0,6 already seen by Bouyssy [15] can be identified. In such a way, the following table gives the value of the error indicator for a shorter band of irregularity factors. Table 4 Error indicator depending on the range on the irregularity factor.

For γ≥0,6, the single moment (SM) approximation becomes the most accurate. Nevertheless for smaller γ value, the DK model is still the most accurate. As a matter of fact it seems more convenient to use the DK model for a sizing when the irregularity factor at each point of the structure can not be checked.

12 Conservatism and Safety Factor Picture 6 highlights the conservatism of the NB model when the irregularity factor decreases. Nevertheless the discrepancy in this conservatism is not very high and it can explain why so many people tried to develop a correction factor of the NB approach. Some models such as the DK or TB ones are quite accurate but not conservative. Other models such as the OC one show no real tendency and both under or overestimate the fatigue life. In the end one interesting point is to try to define the safety factor to apply to the model to be safe compared to the fatigue life estimated in the time domain. In such a way, the following table gives the maximum value of the ratio

Tspec Tref

for

each model. Thus, it can be noticed that the models which have been previously defined as the most accurate needs the higher safety factor. On the one hand, the underconservatism of the DK approximation has been shown on picture 6 and for this model a factor around 3 has to be used to cover the discrepancy of the model. On the other hand, the NB and CD “class” models being always conservative, no safety factor has to be applied. Finally the following curves show that for a sizing, the DK model seems to be the most relevant for a b value of 5 while when b increases, the NB model might become the most appropriate.

Fatigue Life Estimation of Structures Subjected to Vibratory Loading

441

Table 5 Safety factor to apply to the different models.

Fatigue lives comparison (b=5)

Fatigue lives comparison (b=13) 1,00E+08

1,00E+05

1,00E+07

Tspec

Tspec

1,00E+06

1,00E+06

1,00E+04

Dirlick

Dirlick

Narrow-Band

Narrow-Band Coef. 2

Coef. 2 Coef. 3 1,00E+03 1,00E+03

1,00E+04

Tref

1,00E+05

Coef. 3 1,00E+06

1,00E+05 1,00E+05

1,00E+06

Tref

1,00E+07

1,00E+08

Picture 7 Evolution of the predicted fatigue life.

Last but not least it must be reminded that the calculated fatigue lives are average fatigue lives and only probabilistic approaches can lead to an accurate definition of the safety factor to apply on this model.

13 Conclusion In this paper some models to assess the fatigue life of structures subjected to vibratory loading were tested. In a first part, it was seen that the influence of the RMS value and of the used mean stress correction method was negligible. Then, the b parameter of the Basquin model and the irregularity factor, an indicator of the signal bandwidth, were identified as the most influent parameters on the error committed by the different models. Actually the studied models give very accurate fatigue life estimation for an irregularity factor above 0,9 but below, the results are far more mitigated. Finally the different simulations showed that the Dirlick models is the most accurate even if the non conservatism of the approach implies to

442

M. Fressinet, F. Fuchs, and P. Madelpech

use a large safety factor when applied in the domain where the Wöhler curve is quite horizontal. The next step will be now to evaluate the influence of the non gaussienity of the signal. At last, in order to assess the safety factor to apply on these methods, probabilistic approaches have to be undertaken.

References [1] Sherratt, F., Bishop, N., Dirlik, T.: Predicting fatigue life from frequency domain data: current methods, Part A (2004) [2] Dirlik, T.: Applications of computers in fatigue analysis, Ph.D. Thesis, University of Warwick (1985) [3] Benasciutti, D.: Fatigue analysis of random loadings, Ph.D. Thesis, University of Ferra (2004) [4] Rice, S.O.: Mathematical analysis of random noise. In: Wax, N. (ed.) Papers on noise and stochastic processes. Dover, New York (1954) [5] Bendat, J.S.: Probability functions for random responses, NASA report (1964) [6] Lalanne, C.: Dommage par fatigue, Hermès Science (1999) [7] Wirsching, P.H., Paez, T.L., Ortiz, K.: Random Vibrations. Dover Publications, New York (1995) [8] Frendahl, M., Rychlik, I.: Rainflow analysis: Markov method. lnternational Journal of Fatigue 15(4), 265–272 (1993) [9] Chaudhury, G.K., Dover, W.D.: Fatigue analysis of offshore platforms subject to sea wave loadings. International Journal of Fatigue 7, 13–19 (1985) [10] Kam, J.C.P.T., Dover, W.D.: Fast fatigue assessment for offshore structures under random stress history. In: Proc. of the Institution of Civil Engineers, Part 2, pp. 689–700 (1988) [11] Zhao, W., Baker, M.J.: A new stress-range distribution model for fatigue analysis under wave loading. Environmental forces on offshore structures and their prediction 26, 271–291 (1990) [12] Larsen, C.E., Lutes, C.D.: Predicting the fatigue life of offshore structures by the single-moment spectral method. Probabilistic Engineering Mechanics 6, 96–108 (1991) [13] Petrucci, G., Zucharello, B.: Fatigue life prediction under wide band random loading. Fatigue & Fracture of Engineering Materials and Structures 27, 1183–1195 (2004) [14] Falga, A.: Fatigue sonique à Airbus: état de l’art, challenges et voies d’améliorations, Présentation du séminaire Fatigue Vibratoire du CEAT, Toulouse (2009) [15] Bouyssy, V., Naboishikov, S.M., Rackwitz, R.: Comparison of analytical counting methods for Gaussian processes. Structural Safety 12, 35–57 (1993)

26th ICAF Symposium – Montreal, 1-3 June 2011 A Structural Defect Expansion Model Based on Physical Correlation S. Ito, S. Sugimoto, and T. Okada Japan Aerospace Exploration Agency, Tokyo, Japan

Abstract. The spatial correlation and time dependence parameters of aging structural defects are evaluated on the basis of a percolation model. Probabilistic model parameters are obtained by a Markov Chain Monte Carlo (MCMC) simulation using a Bayesian theory framework. The uncertainty in the progression of an aging structural defect can be appropriately evaluated by applying this model in such a way that the correlation of space and time is determined by the three-dimensional evaluation model. The generality of this model is verified on the basis of numerical simulation using the free and open-source software "R".

1 Introduction The probability of defect initiation and its expansion on a structural surface is modeled based on two-dimensional percolation[1]. In addition, a multilayer laminated structure of overlapping two-dimensional models is assumed to describe the space and time correlations of a three-dimensional structural defect. This model can also be applied to the uncertainty evaluation of the expansion of an aging defect by considering that the laminations represent the sample thickness. A structural part is divided into homogeneous unit elements at the beginning of the modeling. Within these elements, a Poisson process is assumed to govern the probability of a defect occurring in a unit of time. A Poisson model is also used for the correlation of defect expansion between elements. In addition, a simple limiting condition is assumed for expansion in the direction of defect depth. This method can be used to evaluate the expansion of corrosion defects. Because the distribution of the defect area and the corrosion depth directly influence structural strength and rigidity, it is important to comprehend the spatial pattern of defects in order to evaluate structural integrity[2]. In the calculated example, the initial parameters of the probabilistic model are set, and a virtual structural defect expansion model is created by numerical simulation. The unknown parameters of defect initiation and expansion are assigned, based on a virtual defect and using a Bayesian estimation method, and the validity of this model is verified. Because the posterior distribution of the unknown parameters is complex, the Metropolis method of MCMC[3] is applied here.

444

S. Ito, S. Sugimoto, and T. Okada

2 Structural Defect Evaluation First, the initiation and the expansion of a defect on a structural surface are modeled based on the two-dimensional percolation method. Next, a probabilistic model of the three-dimensional structural defect is composed as shown in Figure 1. The model assumes a laminated structure that overlies a two-dimensional plate model with multiple layers, and describes the space and time correlations in all laminated structures. This use of an assumed laminated structure can also be applied to uncertainty evaluation and to the expansion of an aging structural defect such as corrosion. A region of a structure is divided into similar unit elements, and a Poisson process with an average initiation rate is assumed as the probabilistic model describing the initiation of structural defects in these elements. The correlation of defect initiation and expansion is assumed by using a similar Poisson model that can incorporate the propagation of a defect from a damaged element to an adjacent undamaged one. Various conditions are actually considered concerning the through-thickness direction. For the sake of simplicity, the expansion of a defect in the depth direction is assumed to be limited only to the cases in which the upper element has received damage. The condition is a monotonic decrease of the probability of damage along the direction of depth. Evaluation appropriate to continuous defect expansion in the through-thickness direction is possible with such modeling. Probabilistic model formulation After a defect initiates in a structural element of the k-th layer shown in the laminated structure in Figure 1, it is assumed that the defect expands to the adjacent elements. When the initiation and expansion of the defect are expressed using the Poisson process described previously and the average occurrence rates are designated ak and bk, the probability of defect initiation and expansion are obtained as[1, 4]:

PI (t ) = a k exp( −a k t ), PP (t ) = b k exp( −bk t )

(1)

Using the conditional probability of a defect occurrence event in the upper laminate structure and the chain rule for the probability, the probability of all defect occurrences in the structure at a certain time tp can be expressed in the following equation: P ( L1 ,... L K ) =

K

∏ P( Lk k =1

| L1 ,..., L k −1 | tp )

(2)

The right hand side of Eqn.(2) shows the probability of a defect event occurrence in the k-th layer, and this probability is expressed as the conditional probability of an event in the layer above the k-th layer. This conditional probability and the compatibility requirement based on all defect data are expressed in Eqn.(3) by adding the expression in Eqn.(1) defined as the likelihood function of the unknown parameter vector of each layer.

A Structural Defect Expansion Model Based on Physical Correlation

445

Percolation model ・Defect initiation: Poisson process with average parameter a Number of elements: N

・Defect expansion: Poisson process with average parameter b

Surrounding element Modeling of three dimensional structure defect

The probabilistic model formulation

・All occurrence event probabilities in layer at time td K

P ( L1 ,...LK | t d ) = ∏ P( Lk | L1 ,..., Lk −1 | t d )

1st layer damaged element

k =1

・Conditional probability of k-th layer

…

2nd layer candidate element

P( Lk | L1 ,..., Lk −1 |t d ) = kk

m

m

∏ (a + ∑ b ) exp{−a t ( x )}∏ exp[ −b {t ( x k

k-th layer

k

i =1

∗

・Defect initiation and expansion condition of element of lower layer :

k

ik

n =1

k

ik

) − t ( x jnk )}]

n =1

Nk

m

∏ exp( −a t )∏ exp[ −b {t k d

i = k k +1

k

n =1

d

− t ( x jnk )}] / P ( L1 ,...Lk −1 )

xik ⊇ xik −1 , x jk ⊇ x jk −1 k k ≥ k k −1 ,

Defect exisitence in the element of upper layer

N k = k k −1 ≥ N k −1

Fig. 1 Structural defect model to consider spatial correlation and time dependence based on three dimensional percolation method.

P ( L k | L1 ,..., L k −1 : tp ) kk

= ∏ (a k + i =1

m

∑ b k ) exp{ − a k t ( x ik )}

n =1

m

∏ exp[ −b k {t ( x ik ) − t ( x jnk )}] ∗ n =1 m

∏ exp[ − b k {t d n =1

Nk

∏ exp( − a k t d )

(3)

i = k k +1

− t ( x jnk )}] / P ( L1 ,... L k −1 )

x ik ⊇ x ik −1 , x jk ⊇ x jk −1 k k ≥ k k −1 ,

N k = k k −1 ≥ N k −1

Here N is the total number of elements, tp is the latest time when information was obtained, (x1, ... , xk) is the element vector of k parts arranged in order of occurrence of defects, (xk+1,...,xN) is the no-defect element vector up to time tp, and m is the surrounding number of elements that influence defect expansion for element xi. Thus, this method describes defect initiation and expansion in the structural element using two kinds of Poisson processes, and models the time and spatial correlation of all of the defects.

446

S. Ito, S. Sugimoto, and T. Okada

Structural defect Three dimensional discretization Slice example (4 layers) Estimation of percolation model parameter (ak,bk: k=1 to 4) Bayesian analysis: MCMC Evaluation of reliability ( based on the estimated parameter) Fig. 2 Procedural outline of reliability assessment.

Prior distribution function (Uniformly distributed) Initial parameter values

f (0)(a), f (0)(b)

a(i=1),b(i=1)

Iteration: i Setting of the following value(for “a”) Comparison of the posterior density (Metropolis method) Setting of the estimate candidate value “b” is also estimated

p=

a(*)

L{a(*),b(i)}f (a(*)) f (b(i) ) L{a(i) ,b(i)}f (a(i) ) f (b(i) )

p ≥1 ⇒ a(i +1) = a(*) p <1 ⇒ a(i+1) = a(*) at prob. p a(i +1) = a(i ) at prob. 1− p

a,b, f(a),f(b) Fig. 3 Bayesian estimation by Metropolis method.

Unknown parameter estimation and reliability index

The discretization of a three-dimensional structural defect is expressed in Figure 1, and the process for estimating unknown parameters is outlined in Figures 2 and 3. The three-dimensional chart in Figure 1 shows an example of the virtual structural

A Structural Defect Expansion Model Based on Physical Correlation

447

defect model, and the complete defect inventory of the structure is separately expressed in each layer. The probabilistic model parameters, ak and bk (k = 1, ..., N), are estimated by Bayesian analysis based on the structural defect data. The unknown parameter vector (a, b) is calculated by the MCMC method, having the character of the detail balance condition and ergodicity, as shown in Figure 3, using the probability for the total event P(L1,...,Lk|a,b) that indicates the conditional with (a, b). The Bayesian theorem is expressed in the following equation : f

(t p )

(a, b | t p ) = P ( L1 ,..., L k | a, b ) f

(0)

(a, b )

(4)

/ ∫ ∫ ( Numerator ) dadb ba

A target reliability of the structure is evaluated based on the estimated result for unknown parameters. The reliability index defined here is shown in Figure 4. The damage rate of the k-th layer, Dk(td), estimated at a certain target time td is defined in Eqn. (5).

Dk (t d ) =

‐

Number of damaged elements in k th layer Number of total elements in k th layer

‐

(5)

When the modal values (aM, bM) in the posterior distribution at time td are used in the reliability calculation, the damage rate Dk(td) is expressed by the following equation: Dk = Dk (t d | a M , b M )

Posterior probability

f (tp)(a,b)

(6)

Reliability(td) time

tp

td

Fig. 4 Reliability at target time td.

3 Numerical Calculation Example A concrete numerical calculation example is shown in the following figures. Applying this model to a real structural defect requires calibration for the structural defect by means of a parametric study using MCMC. The excellent free software R is used for the above calculation.

448

S. Ito, S. Sugimoto, and T. Okada

Structural model and reliability evaluation

To begin with, the true parameter values (atrue, btrue) of the probabilistic model are set, and virtual structural defect initiation and expansion are produced in the numerical simulation. Parameter values (a, b) of an unknown probabilistic model are estimated for this virtual structural defect, and the validity of this method is verified by calculations using the software R. The defect model simulated up to time tp = 10 is shown in Figure 5, layer by layer. The structure is divided in length and breadth into 20 elements, each four layers deep. The true values (atrue, btrue) of the assumed model parameters are in bold-faced type in this figure. The unknown parameter (a, b) is estimated by treating the virtual values of these defect data as the field data of objective information, using Bayesian analysis based on the MCMC method. The numbers of iterations in the MCMC simulation process is 1,000 times. A part of the estimation process of an unknown parameter (a1,b1,a3,b3) is shown in the time series of Figure 6, where the horizontal lines in the figure indicate the true model values (atrue,k, btrue,k, k = 1, 3). The posterior distribution function f(ak, bk, k = 1, 3)(tp) at tp = 10 obtained by the simulation is shown in Figure 7. The time series of Figure 6 is shown in this figure, converted into a frequency distribution. Model of real damage (assumed true values: atrue, btrue)

Three dimensional percolation model (4 layer discretization)

(atrue, btrue)= (a 1=0.02, b1=0.05, a2=0.07, b2=0.07, a3=0.07, b3=0.07, a 4=0.07, b 4=0.07)

Fig. 5 Numerical example.

a1

b1

index

a3

index

b3

index

index

Fig. 6 Process of generation of MCMC unknown population parameter (defect initiation and expansion parameters).

A Structural Defect Expansion Model Based on Physical Correlation

a1

449

b1

a3

b3

: true values

Fig. 7 Posterior probability distribution function of defect parameters.

The average damage rate was defined previously in Eqn. (6) as an one of the reliability index. As an numerical example, the damage rate up to time td = 15 is estimated based on the information obtained at tp = 10. The damage rate obtained from the defect simulation result of using the true parameter values and the expected rate obtained from the modal values of the posterior distribution at tp = 10 are shown in Table 1. Comparing these with the damage rate based on the true value, the expected value Dk=4 of the bottom layer is conservatively estimated, on the safe side. However, the difference between the two damage rates over the entire layer is not large. On the other hand, a similar tendency is also expected for the value of the damage rate estimated from the following equation: Dk = ∫ ∫ D(t d | a, b) f

(t p )

(7)

(a, b | t p )dbda

ab

Dk in the Eqn. (7) shows so-called the Bayesian expected value of the damage rate, which is obtained from the posterior distribution f(tp)(a, b) in Eqn. (4). Table 1 Damage rate in each layer.

Time

True value : Dk

td td=11 td=12 td=13 td=14 td=15

Estimated value : Dk

1-st

Layer: k 2-nd 3-rd 4-th

1-st

Layer: k 2-nd 3-rd 4-th

0.60 0.67 0.74 0.77 0.81

0.28 0.33 0.42 0.49 0.55

0.59 0.66 0.73 0.78 0.82

0.26 0.35 0.43 0.51 0.59

0.06 0.10 0.15 0.20 0.27

0.02 0.02 0.03 0.04 0.08

0.07 0.13 0.19 0.26 0.32

0.02 0.02 0.05 0.09 0.17

Spatial pattern of corrosion defect estimated The results from the last investigation of fundamental properties were obtained in an aluminum alloy corrosion test, and the corrosion environment of the real defect is outlined in Table 2. The test piece was divided into 30x30 homogeneous elements in its plane, and the corrosion depth was divided into four layers. Five damage measurements were made at the time of j-th observation Tj(j=1,...,5). As

450

S. Ito, S. Sugimoto, and T. Okada

introduced in the previous figure 5, the real corrosion defect at time T5 was discredited into a three-dimensional structural defect shown in Figure 8. The simulation frequency in the MCMC is 2,000 times. Table 2 Corrosion environment of real corrosion defect.

Material 2024-T3 Al Clad 2mm thickness Exposure area 2mm * 2mm (surface) Environment 3.5% NaCl solution, Room Temp. Procedure Immersion*1 External load No load *1: Immersion was interrupted at the scheduled intervals to measure corrosion pit geometry.

1-st layer (depth=z1 )

2-nd layer (depth=z2 )

3-rd layer (depth=z3 )

4-th layer (depth=z4 )

Defect inititaion/expansion time : No defect,T1,T2,T3 ,T4 ,T5 :

Fig. 8 Numerical model of real corrosion defect (Refer to Table 2).

Index

Index

a1

b1

Index

a3

Index

b3

Index

a2

Index

b2

Index

Index

a4

b4

Fig. 9 Unknown population parameter for real defect.

Using these conditions, the unknown parameters (a, b) for corrosion defects were estimated by this probabilistic model. The result of applying this model to the corrosion defects is shown in the following figures. An unknown model parameter vector (a, b) and these posterior probabilities f(T5)(a, b) are shown in Figures 9 and 10, respectively. As no particular abnormality appears in these figures, the model should effectively depict the actual structural defect status, so that it is easily understood. However, the question remains whether such a satisfactory result can always be obtained, so more studies of defects must be made under a variety of conditions.

A Structural Defect Expansion Model Based on Physical Correlation

451

f(T5)(a 1 )

f(T5)(b1 )

f(T5)(a 2 )

f(T5)(b2 )

f(T5)(a 3 )

f(T5)(b3 )

f(T5)(a 4 )

f(T5)(b4 )

Presumption mode values:

a1M=0.24, b1M=0.18, a2M=0.25, b2M=0.17, a3M=0.55, b3M=0.18, a4M=0.38, b4M=0.18

Fig. 10 Posterior probability distribution of real defect parameters.

4 Conclusion Since the variables affecting an aging structural defect include significant uncertainty factors, the rigorous evaluation of defects is generally difficult. To overcome this, statistical calculations were introduced into the defect evaluation process. Using the probabilistic percolation model, the proposed method can be applied to the evaluation of the expansion of an aging structural defect, and also to defects in laminated materials in real structures. To apply the model to a real structural defect, the defect variables were evaluated in a parametric study using the MCMC method. The free software R can be expected to be effective here. The authors wish to thanks Dr. H. Itagaki a former president of Yokohama National University and Dr. H. Asada of Aviation & Railway Safety Promotion for their helpful advice.

References [1] Taku, K.: Ecological Research 59, 219–255 (2009) (in Japanese) [2] Makoto, O., Katashi, F., Makoto, T.: In: Proceedings of the Japan Society of Civil Engineers, vol. 672, pp. 109–116 (2001) (in Japanese) [3] Dani, G.: Markov Chain Monte Carlo, Text in Statistical Science. Chapman & Hall/CRC (1997) [4] Gibson, G., Otten, W., Filipe, J., Cook, A., Marison, G., Gilligan, C.: Statist Comput. 16, 391–402 (2006)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Link between Flight Maneuvers and Fatigue

Juha Jylhä1, Marja Ruotsalainen1, Tuomo Salonen2, Harri Janhunen3, Ilkka Venäläinen1, Aslak Siljander3, and Ari Visa1 1 Tampere University of Technology [email protected], [email protected], [email protected], [email protected] 2 Patria Aviation Oy [email protected] 3 Technical Research Centre of Finland [email protected], [email protected]

Abstract. Structural integrity management is in a key role when operating with an ageing fleet. Preventive actions aid in keeping structures healthy and ensuring the designed lifetime for the aircraft. Our research focuses on exploitation of collected usage data by identifying actual in-flight events that cause the major fatigue life expenditure of the fatigue-critical structural details. In order to build a link between the events and damage, we have developed software for flight maneuver identification. As a latest advancement, we have created data models for several flight maneuvers and constructed a model library, referred to as template library. The library instructs the software about what the interesting events look like in the data. Our software together with the template library allows us to perform maneuver-specific fatigue assessment and achieve knowledge concerning the fatiguecriticality of various flight maneuver types. This lays a foundation for detailed analysis of the identified, nominally similar, maneuvers and identification of small crucial actions within the maneuvers that are behind the fatigue. In this paper, we consider the issues related to maneuver-specific fatigue assessment and present analysis results for four structural details of F-18 aircraft with a template library of seven flight maneuvers. We summarize the requirements and prospect of our fatigue analysis approach and prove its applicability.

1 Introduction Traditionally aircraft fatigue management is based on structural inspections and primary load (such as G or high angle of attack) history monitoring techniques whereas the most advanced techniques are based on real-time monitoring of extensive strain gauge and flight parameter data collected during flights. Preventive actions within traditional fatigue management aim at ensuring the designed lifetime for the aircraft by conducting possible structural modifications or issuing usage recommendations based on human expertise. A more specific reasoning about the in-flight usage of the aircraft (i.e. its maneuvering) on the fatigue of the critical structural locations of the aircraft is possible. It would bring significant added *

Oral presentation.

454

J. Jylhä et al.

value for the preventive fatigue management. This would help in bringing the aircraft usage to a level at which any fatigue life expenditure (FLE), which is not justified by operational or training objectives, is avoided. To succeed in this, it is necessary to gather vast amounts of detailed information of the usage and loading of the structure in various missions and in various flight conditions therein. Exploitation of usage data has been studied, for example, concerning parameterbased fatigue monitoring [11], flight maneuver-specific damage estimation [7], and classification of helicopter maneuvers [6]. Concerning the Finnish Air Force F/A-18 Hornet fleet, two aircraft equipped with the Hornet Operational Loads Measurement (HOLM) system [9], [10], [12] collect the appropriate usage data (200 flight parameters at 20 Hz and 36 strain gauges at 640 or 1 280 Hz). The HOLM system enables estimation of FLE accumulation for each fatigue-critical structural detail even for events having duration of seconds. To be able to assess the contribution of the events to FLE, one requires tools for extracting the events, the flight maneuvers, from the recordings. Our earlier approach to studying the most damaging flight maneuvers was based on manual work of an experienced analyst. From the most damaging flights (typical duration less than 50 min), maneuvers (typical duration less than 1 min) producing most of the estimated damage were identified. This approach provided the first important conclusions about the most damaging maneuvers. Manual analysis of the bulk of data highlighted the need for a more intelligent data mining environment with automatic functionality. In [2], we introduced our first progress toward this automation. This paper advances the previous data mining environment by presenting a template library aiming to cover the most damaging flight maneuvers. Estimating the damage produced by the maneuvers gives an explanation for what kind of maneuvering causes the major FLE. A brief summary of the paper is as follows. The next section considers a fundamental question about how to estimate fatigue life expended during a short event in flight. Then we discuss data mining of the flight recordings. The presented experiments illustrate the idea of the data mining concept. Last the conclusions are drawn.

2 Flight-Maneuver-Specific Fatigue Estimation In this research, the fatigue damage is calculated applying the strain life method for the Rainflow cycle counted closed strain loops (cycles) and summing up these closed loop damages to form the total damage of the flight. The cycles’ peaks and valleys represent load reversals in the signal and they may be related to phenomena within a single maneuver (e.g. heavy vibration of structure during high-angleof-attack pull-up) or to totally different and separate maneuvers. The damage of the cycle should not be allocated entirely to the peak or to the valley because they both have contribution on the damage, but it should be shared between them, somehow. The “rule” for defining these shares is still an open issue and so equal shares have been used, so far. Finally, the total damage of a maneuver is produced by summing up all the peak and valley damages during the maneuver. That is the first method to calculate the damage of one flight maneuver.

Link between Flight Maneuvers and Fatigue

455

The second method excludes completely the effects of events before and after the maneuver on the fatigue damage. A flight maneuver happens during a short period of time on a flight, and the maneuver has its own short-term load history containing peaks and valleys. The case is straightforward and the fatigue damage is determined by feeding the load history within the maneuver into strain life analysis. It must be noticed that even if the whole flight would be covered with defined maneuvers the sum of the maneuver-wise calculated damages is usually not equal––or even close––compared with the damage calculated for the whole flight. We apply these two approaches in the research. The first method is suited for estimating an aggregate FLE caused by a set of in-flight events that represent the same type of maneuver. For example, certain amount of vertical tail stub FLE is expended during all the found split-S events together as illustrated by Figure 1. This method is reasonable when comparing aggregate FLEs of big amounts of maneuver events. The influence of variation in maneuver-specific FLE reduces when the number of aggregated events increases. Subsequently, this method is referred to as cumulated FLE. The second method is used when comparing similar flight events with each other; for example, when producing damage distributions for each type of maneuver. If one would analyze the contribution of only one of the split-S events in Figure 1 instead of their aggregate FLE, it would be reasonable to apply this method. FLE is calculated separately for each of the maneuver event ensuring that all the cycles open and close within the maneuver. This is due to producing comparable results for all the events separately––not results where the history and future of a single event contributes its FLE. Subsequently, this method is referred to as segmented FLE.

Fig. 1 The six found split-S events together represent about 90% of the vertical tail stub fatigue damage accumulated on this flight. There is a possibility to misunderstand the terms related to flight maneuver. In this paper, maneuver type refers to some class of maneuver such as split-S in general, and maneuver event implies one interesting period inside flight.

456

J. Jylhä et al.

3 Flight Data Driven Aircraft Usage Analysis Our maneuver-specific fatigue analysis can be seen as segmentation of the flight recordings, and then estimation of the FLE for each segment. It has been reasonable to choose the aim for segmentation so that the produced segments are flight maneuvers whose type is explicitly interpretable to such as loop or split-S. In [2], we presented our procedure for flight maneuver identification. First a representative maneuver was chosen, and a template for it was built by an experienced analyst. Then matching maneuvers were automatically identified from flight parameter data recordings using our template-based pattern recognition software [8]. At that time, the procedure was on its early stage and it was a straight process from the choosing of the maneuver to the identification of it. Automated flight maneuver identification enabled the calculation of the FLE distribution over all the found flight maneuvers of the same type. Data mining environment In this paper, we aim at comprehensive flight-maneuver-specific fatigue analysis. We have developed our flight maneuver identification procedure further as Figure 2 shows. Template related work, which consists of choosing and modeling flight maneuvers, requires knowledge about the usage, behavior, and structures of the aircraft, and it should be performed by an analyst familiar with the aircraft. Comprehensive fatigue analysis calls for systematic modeling of all the most damaging flight maneuvers for each fatigue-critical structural detail. In this paper, we propose the creation of the so-called template library containing the templates of these most damaging maneuvers. When the template library has been created, our fully automatic flight maneuver identification and analysis software, AMANA (Aerial Maneuver ANAlysis), utilizes the template library and finds all the matching maneuvers from the flight parameter data recordings. The resulting maneuver compilation associated with estimated FLE is presented in the result section of this paper. AMANA software consists of two computer programs: AMANA detector and AMANA analyzer. The described data mining environment provides a framework for detailed and sophisticated, flight data driven aircraft usage analysis. The framework assumes that flight parameter data recordings of good quality are available, because without them, the automatic analyses cannot be performed. Next we take a close look at the role of the template library in the procedure. Template library The template library is a collection of templates, i.e. modeled flight maneuvers. For every maneuver type, there is one template. As presented in [8], the template includes a set of parameters and a representative maneuver of the type in question. The representative maneuver is as a clip of flight parameter data stream that has been produced by the sensors of HOLM system. When establishing a new maneuver type to the library, the representative maneuver event is searched out of the flight recording data manually. After that, one has to define the template: select

Link between Flight Maneuvers and Fatigue

457

the relevant flight parameters; set their thresholds and weights; and determine the minimum and maximum length for the acceptable maneuver event. Based on the template parametrization, we can choose the generality of the events to be extracted from the flight recording data. Figure 3 illustrates a generic split-S template that we have used recently.

Fig. 2 The block diagram of our procedure for flight maneuver identification. The creation of the template library requires manual work of an analyst, but actual flight maneuver identification using our AMANA tool is fully automatic. The template library enables comprehensive analyses.

In our current template library, we have used generic maneuver templates. In other words, objective is to find all the maneuvers resembling the type in question, not only the most damaging ones. Therefore, parameters such as altitude, angular velocities, and accelerations have proven to be useful in the templates. The same parameter may also be used twice in the same template to include more features of the behavior of the parameter signal. For example, parameters in split-S template consist of roll (twice), roll rate, G, pitch (twice), and yaw rate. Most of the parameters in conjunction with the thresholds characterize the maneuver, and the rest of the parameters and thresholds are set to eliminate false detections. The selection of the flight parameters and the determination of the other template parameters are application-specific. When establishing this kind of data mining system, one has to import the knowledge about the application into the system. The characteristics of the flight monitoring data depend on the aircraft and its recording system so specific parametrization is not generally applicable. That is the reason why we do not discuss our choices in detail here but only give general guidelines. Created templates need to be validated. That is most easily done by utilizing AMANA tool and checking that found events truly represent the maneuver type that they are meant to. Based on the observations, one may have to adjust the template parametrization, run the program again, and then see if the new results are more expedient. However, after this iterative process for each template, the established template library lays a foundation for perusing the origin of fatigue for all the structural details, and it is available for future needs regarding the flight data driven fleet usage exploration.

458

J. Jylhä et al.

Fig. 3 The template of split-S maneuver consists in a representative split-S event chosen from HOLM recordings. Gray scale indicates the level of the used parameters throughout the template. Black refers to low parameter value; gray to moderate value; and white to high value. See articles [2] and [8] for more details about the creation of the templates.

AMANA software A template characterizes a maneuver so that the AMANA detector algorithm can run through the flight data recordings and extract all the maneuvers that resemble the template. In the field of artificial intelligence, this kind of data processing is generally called as pattern recognition [1]. Maneuver extraction calls for temporal pattern identification such as in [4] and [5]. AMANA detector applies a so-called approximate pattern matching algorithm [8]. It means that the found patterns resemble the template; in other words, they do not have to be exact matches. A pattern can be nonlinearly warped in time, and the signal values need only have similar behavior, not the exact shape. The extracted patterns are short clips of data stream that have been recorded during the corresponding maneuver events while flying. See more details about the templates and the detection algorithm in [2] and [8]. AMANA detector stores all the extracted patterns in the so-called flight maneuver database; see Figure 2. The role of AMANA analyzer is to process the flight maneuver database and provide decision support for the user. This paper considers a maneuver compilation which is clarified in the subsequent section. Further ideas to investigate the fatigue concern the reasons behind FLE scattering within the maneuvers of the same type. When the template library has once been created, the maneuver extraction can be performed over and over again, for example, when a new set of flight recordings is collected from the fleet. This kind of information extraction out of the huge amount of data is called data mining [3]. The creation of the template library expands the data mining capability from statistical maneuver-specific analysis to comprehensive analysis covering all the interesting maneuvers. The comprehensive analysis requires a few hours of computing for a data base of hundreds of flight hours. This data mining environment enables finding out the most critical flight maneuvers from the FLE point of view and analyzing the diverse usage of the fleet.

Link between Flight Maneuvers and Fatigue

459

4 Results and Discussion This section discusses the results and illustrates the potential of our approach for preventive aircraft fatigue management. All the results are based on HOLM recordings of the Finnish Air Force F-18 aircraft. We have studied the role of a set of maneuvers in the fatigue of certain critical structural details. First, we demonstrate the data mining concept by presenting a maneuver compilation table produced using a library of seven maneuver types and four structural details. The compilation is based on 297 HOLM flights. Then we discuss vertical tail root in more detail via maneuver-specific FLE distributions. Table 1 A compilation of the found maneuvers and their share of the total fatigue damage accumulation in 297 HOLM flights. The share is based on the cumulated FLE. Vertical tail root Proportion of detected maneuvers in flight set total damage

Inner wing shear tie Proportion of detected maneuvers in flight set total damage

Fuselage bulkhead Proportion of detected maneuvers in flight set total damage

Elevator spindle Proportion of detected maneuvers in flight set total damage

Damage Damage Damage Damage Damage Damage Damage Damage allocated allocated to allocated allocated to allocated allocated to allocated allocated to In total In total to peaks valleys In total to peaks valleys In total to peaks valleys Structural detail to peaks valleys 28 % 29 % 28 % 2% 2% 2% 0% 1% 0% 0% 0% 0% Split S 0 % 1 % 1 % 8 % 5 % 7 % 0 % 1 % 1 % 9 % 10 % 9% Turn→roll→turn 12 % 12 % 12 % 30 % 17 % 24 % 58 % 0% 29 % 17 % 13 % 15 % Turn 7% 6% 6% 2% 1% 1% 3% 1% 2% 0% 2% 1% Loop 0% 0% 0% 0% 0% 0% 0% 3% 1% 0% 0% 0% Push 3% 1% 2% 3% 5% 4% 0% 0% 0% 66 % 46 % 56 % Roll 2% 2% 2% 1% 1% 1% 2% 0% 1% 0% 0% 0% Oblique loop Sum 51 % 50 % 51 % 45 % 30 % 37 % 63 % 6% 34 % 91 % 72 % 81 %

Flight maneuver FLE compilation The compilation of maneuvers is presented in Table 1. It gives the relative FLE of each maneuver type, i.e. maneuver type’s share of the total FLE over the flight set for each structural detail. The results are calculated with the cumulated FLE method. The damage allocated to peaks and valleys is presented separately, and the third field provides the average of the peak and valley damages. The FLE of vertical tail root is similar when comparing peaks and valleys because the oscillating loads typically cause the relevant cycles which open and close within the maneuver. For fuselage bulkhead, the peak damage is much higher than the valley damage because cycles’ peaks are usually found on the identified maneuvers but cycles’ valleys usually occur on level flight. We have not built a template for level flight in the used template library. The results show that split-S, turn, and loop are the most severe maneuvers for vertical tail root, and they cover almost half of the total damage. The rest of the considered maneuvers are rather insignificant. Turns are the most severe maneuvers for center fuselage bulkhead and wing fold inner wing shear tie. On the contrary, roll maneuvers cover over 50 % of elevator spindle damage. Based on the compilation, the significantly damaging maneuvers for each structural detail can be chosen for the further analysis.

460

J. Jylhä et al.

FLE distributions This part of the experiments provides more detailed statistical view of the vertical tail root FLE. Figure 4 presents FLE distribution of each maneuver type in the library. Histograms have only five bins, and the distributions are rather coarse. To produce smoother distributions, more flights (and more maneuvers) should be taken into consideration. Nonetheless, one can see the general statistical properties of the presented seven maneuvers here. The histogram of split-S is heavily weighted to the severely damaging side. Interestingly, two of the fatigue-relevant maneuvers, split-S and loop, have notable variation in their distribution. Notice in Figure 4 that FLE is normalized by the most damaging maneuver event, and the xaxis is logarithmic. So the bar in “zero” damage indicates number of the events whose FLE is close to the most damaging event, and -2 implies the normalized FLE of 10-2. Note also the right tail in the distributions that is produced by the most damaging events of the maneuver type in question. If the tail reaches longenough, there can be a few very badly damaging events that should be investigated although the main part of the distribution is on lightly damaging side. Discovering the reasons behind the variance and the mostly damaging events may provide relevant information for assessing the cost of the diverse use of the aircraft.

Fig. 4 FLE distributions for vertical tail root. One histogram is presented for each maneuver in the library. The segmented FLE is normalized by the most damaging maneuver event, so the histograms are in the equivalent FLE scale. Notice the varied expectation value, the spreading, and the tails of the distributions. When reviewing the histograms, consider also the FLE shares in Table 1 giving the aggregate FLE of all the events together for every maneuver type.

Link between Flight Maneuvers and Fatigue

461

Figure 5 presents an accumulation plot of vertical tail root damage for split-S. To produce the curve, the split-S maneuver events are arranged in descending order by their segmented FLE. Then the values are cumulated and finally normalized by the aggregate FLE. For this kind of curve where the contribution of each maneuver event is considered, it is reasonable to use the method of segmented FLE. One can see that 50 % of the aggregate FLE is produced by only 13 % of the found split-S maneuvers. Notice that when analyzing what causes the bias of the damage toward certain maneuver events, one has to extract, besides the most damaging events, certain amount of events at low FLE level. Workload on manually examining the maneuvers of all types relevant for fatigue production would be enormous when we aim to assess a significant part of the FLE history. This calls for further development of AMANA analyzer to provide more detailed reasoning of the origin of fatigue, even concerning the short-term actions performed within maneuvers. Last we discuss a few more observations concerning vertical tail root FLE caused by split-S maneuver events. For these observations, we have used a different split-S template which involves requirement for high angle of attack during the maneuver. The numbers below are calculated from 110 HOLM flights. The found maneuvers cause • • • •

31.7 % of the total FLE for the left vertical tail with the cumulated FLE; 29.0 % of the total FLE for the left vertical tail with the segmented FLE; 23.3 % of the total FLE for the right vertical tail with the cumulated FLE; 21.1 % of the total FLE for the right vertical tail with the segmented FLE.

Fig. 5 Damage accumulation over the split-S maneuver events extracted from the HOLM data. The events are sorted by the segmented FLE they produced to vertical tail root. The summed FLE of all the found split-S maneuvers is seen as 100 % accumulation in this curve. This summed FLE corresponds to the 28 % of the total FLE presented in Table 1 that is based on method of cumulated FLE. The relative maneuver amount of 100% implies all the found split-S maneuver events. The main point here is the growth rate of FLE in the beginning of the curve. Compare this curve to the split-S histogram in Figure 4 illustrating the distribution of segmented FLE.

462

J. Jylhä et al.

The methods of cumulated and segmented FLE produce congruent result for split-S owing to the oscillating behavior of the vertical tail root loads. It causes the cycles open and close typically within the maneuver, so there is no particular difference between cumulated and segmented FLE methods. This is not the case with structural details, such as fuselage bulkhead, where the loads are not symmetric. The FLE share of the left vertical tail is in line with the corresponding value in Table 1 that is calculated by using 297 flights and the different template aimed to extract split-S events performed in a more general way. This indicates that the angle-of-attack restriction in the template leads to smaller number of extracted splitS events but to rather congruent total damage; in other words, those high-angle-ofattack events have the major contribution to aggregate split-S FLE while the rest events are of smaller contribution. Note also the different result for the left and right vertical tail. Within this set of split-S, the damage seems to be moderately biased toward the left vertical tail due to one specific maneuver event that calls for more investigations. In the near future, we are striving to develop a method that is able to discover knowledge about the causes of FLE inside the maneuvers. The basic idea is to establish a template library whose maneuver types represent the major share of the total FLE for all the fatigue-critical structural details. The new method would then process all the thousands of the found maneuvers in the maneuver database and return the sophisticated information about causes of diverse segmented FLE produced by nominally similar maneuvers.

5 Conclusion This paper discusses a data mining approach to estimating fatigue-severity of different flight maneuvers. The flight maneuver identification method and collected template library are the major steps toward understanding the effects of the aircraft maneuvering on the fatigue. The presented compilation results of hundreds of the Finnish Air Force HOLM flights consider seven maneuver types and four structural details. When calculated for all the fatigue-critical structural details and for the most damaging flight maneuvers, the analysis will provide statistical knowledge of the reasons behind the structural damage, based on real-world flight monitoring data. That knowledge supports the decision making and may provide valuable guidance for adjustments to the content of the flight training syllabi in order to increase the lifetime of the fleet and reduce operating costs.

Acknowledgment The authors would like to thank the Finnish Air Force and the Nokia Foundation for funding and support.

Link between Flight Maneuvers and Fatigue

463

References [1] Duda, R., Hart, P., Stork, D.: Pattern classification, 2nd edn. Wiley, New York (2001) [2] Jylhä, J., Ruotsalainen, M., Salonen, T., Janhunen, H., Viitanen, T., Vihonen, J., Visa, A.: An article Towards automated flight-maneuver-specific fatigue analysis, in: Bridging the Gap between Theory and Operational Practice. In: Bos, M.J. (ed.) Proceedings of the 25th ICAF Symposium, Rotterdam, pp. 1121–1134 (2009) [3] Han, J., Kamber, M.: Data mining: Concepts and techniques, 2nd edn. Morgan Kaufmann, San Francisco (2006) [4] Harada, L.: An article Detection of complex temporal patterns over data streams. Information Systems 29(6), 439–459 (2004) [5] Mitrovic, D.: An article Reliable method for driving events recognition. IEEE Transactions on Intelligent Transportation Systems 6(2), 198–205 (2005) [6] Oza, N.C., Tumer, K., Tumer, I.Y., Huff, E.M.: An article Classification of Aircraft Maneuvers for Fault Detection. In: Windeatt, T., Roli, F. (eds.) MCS 2003. LNCS, vol. 2709, p. 160. Springer, Heidelberg (2003) [7] RTO-TR-045, Design Loads for Future Aircraft, Sections 3.2.2–3.2.4, RTO/NATO, France (2002) [8] Ruotsalainen, M., Jylhä, J., Vihonen, J., Visa, A.: An article A novel algorithm for identifying patterns from multisensor time series. In: Proceedings of the 2009 WRI World Congress on Computer Science and Information Engineering (CSIE 2009), Los Angeles, vol. 5, pp. 100–105. IEEE, Los Alamitos (2009) [9] Siljander, A.: A report A Review of aeronautical fatigue investigations in Finland. In: Siljander, A. (ed.) Presented at the 31st Conference of the International Committee on Aeronautical Fatigue (ICAF), Rotterdam, ICAF Doc. 2418 (May 2007 – April 2009) [10] Siljander, A.: A report A Review of aeronautical fatigue investigations in Finland. In: Siljander, A. (ed.) To be published at the 32st Conference of the International Committee on Aeronautical Fatigue (ICAF), Montreal (May 2009 – April 2011) [11] Tikka, J., Salonen, T.: An article Parameter based fatigue life analysis for F-18 aircraft in: Durability and Damage Tolerance of Aircraft Structures: Metals vs. Composites. In: Lazzeri, L., Salvetti, A. (eds.) Proceedings of the 24th ICAF Symposium, Naples, vol. I, pp. 412–426 [12] Viitanen, T., Koski, K., Bäckström, M., Voutilainen, E., Lahtinen, R., Siljander, A.: An article The OLM database as a tool to sort particular data sets from a bulk of data. In: Proceedings of the 23th ICAF Symposium, vol. II, pp. 513–540. EMAS Ltd, Hamburg (2005)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Health and Usage Monitoring of Unmanned Aerial Vehicles Using Fiber-Optic Sensors I. Kressel1, A. Handelman2, Y. Botsev2, J. Balter1, P. Gud’s1, M. Tur2, S. Gali5, A.C.R. Pillai3, M.H. Prasad3, A.K. Yadav3, N. Gupta4, S. Sathya4, and R. Sundaram4 1

2

IAI Engineering Division Ben Gurion International Airport, Israel School of Electrical Engineering, Tel-Aviv University, Tel-Aviv, Israel 3 Aeronautical Development Establishment, Bangalore India 4 National Aerospace Laboratories, Bangalore India 5 Consultant, Tel-Aviv, Israel

Abstract. An airborne, high resolution, load tracking and structural health monitoring system for unmanned aerial vehicles composite structure is presented. The system is based on embedded optical fibre Bragg sensors, interrogated in real time during flight. The vibration signature during ground testing has been recorded and analyzed, making it possible to identify and trace the dynamic response of the airborne structure and engine during flight. Tracking the structural behaviour over time can be used for Condition Based Maintenance (CBM), reducing maintenance cost.

1 Introduction The high manoeuvrability and harsh launch and landing conditions of modern Unmanned Aerial Vehicles (UAVs), demand constant monitoring of their structural airworthiness. The recently introduced Health and Usage Monitoring Systems (HUMS) concepts, aim towards effective real-time assessment of the structural integrity of flying vehicles, should provide practical means of maintaining structural airworthiness at minimal cost. This is highly important for composite-made UAVs, where conventional inspection methods of critical structural components are stymied by limited accessibility. Fiber optic sensors, in particular Fiber Bragg Grating sensors (FBG), appear to be excellent candidates to be used in HUMS applications due to their high sensitivity to mechanical strain, small size, immunity to electrical interference, low weight, long life, durability under extreme environmental conditions, and capability of high speed sensing. Moreover, multiplexing techniques have been devised, where quite a few sensors, longitudinally spaced on the same fiber, can be individually addressed to spatially cover strain and temperature fields. For composite structures, these sensors can be easily embedded into the structure during manufacturing, eliminating the need for sensor protection. *

Oral presentation.

466

I. Kressel et al.

This work presents an advanced smart loadstrain monitoring, airworthy system, for the composite tail booms of a UAV, based on an array of FBG sensors, embedded in the tail boom during manufacturing. The FBG sensor net, comprising four fibres, each with four sensors per boom, was tailored to monitor critical locations in the boom, based on a detailed finite element analysis. The system was tested on ground in order to verify its ability to track both static and dynamic boom loading. Structural characteristics like strain distribution under static loading, impact response, and normal modes were successfully traced by the system. As a final proof of concept the system was integrated in the UAV and was successfully flown for approximately two hours.

2 Principle of an Embeded Bragg Grating Sensor The reflection spectrum from a fiber Bragg grating is shifted when the periodic grating in the fiber is either mechanically stressed or heated/cooled [1]. For a free (not embedded) grating, the relative shift in the Bragg wavelength, ΔλB/λB, due to an applied strain along the fiber (ε) and a change in temperature (ΔT) is approximately given by:

ΔλB

λB

= (1 − pe ) ε + (α Λ + α n ) ΔT ,

(1)

where ε is the mechanical strain along the fiber direction, ΔT is the temperature change, pe is an effective strain-optic constant, αΛ is the thermal expansion coefficient for the fiber and αn represents the thermo-optic coefficient [1]. An embedded fiber, on the other hand, is also affected by the thermally induced mechanical strains in the composite substrate. Therefore, for an embedded fiber, where the transverse mechanical coupling between the fiber and the composite substrate is small, Eq. (1) should be rewritten as:

ΔλB

λB

= (1 − pe ) ε + (α Λ + α n ) ΔT + (1 − pe )(α substrate − α Λ ) ΔT , (2)

where αsubstrate is the thermal expansion coefficient of the composite substrate. Since the UAV is exposed to temperature changes during flight, temperature compensation should be considered. Boom heating test was conducted in order to evaluate the total thermal effect on the embedded Bragg grating sensor wavelength shift. This thermal correction factor was applied using the standard temperature- altitude profile.

3 The Implemented Hums Concept The Nishant UAV, designed and manufactured in India by ADE was selected as the test-bed for the evaluation of this HUMS concept (Figure 1). Each of the two tail booms of the Nishant is a composite structure made of two thin wall “C”

Health and Usage Monitoring of Unmanned Aerial Vehicles

467

section channels, riveted together to form a close rectangular beam. Each boom is connected to the wing by two removable bolts and at the back the booms hold the empennage comprising horizontal tail (with elevator), vertical tails (with rudder). The thickness of the composite boom walls is optimized for both strength and stiffness [2]. The boom is basically a cantilever beam with a relatively large mass at the back end, composed of the rudder and elevator actuators and the horizontal tail. The main boom loading conditions are vertical and horizontal bending. In order to track such loading condition two fiber were embedded at the center of the boom ("center fibers", CH2, CH3 in Figure 2) and another two fibers were embedded near the corners ("side fibers", CH1, CH4 in Figure 2). On each fiber, four FBG sensors were placed at the same distance from the boom end (Figure 2). For such sensing net arrangement, the two centre fibers are only sensitive to the vertical bending.

FBG were embedded in the top and bottom parts of the two booms, see arrows.

Fig. 1 The Nishant UAV on the launcher.

The side fibers are sensitive to the vertical bending in a similar manner as the central ones, but will also react to the horizontal bending. Since no tension loading is applied on the boom, the vertical bending will introduce similar but opposite strains in the top and bottom fibers. The two side fibers are on the same side with respect to the center line. Hence, the horizontal bending will induces similar strains on both side fibers, in addition to the vertical bending contribution. A solid state, high sampling rate (>2kHz) FBG interrogation unit is used, capable of tracking multiple fibers, having multiple FBGs on each. The optical fibers are polyimide-coated to assure good bonding to the composite structure during embedment. The interrogation and data logging systems were placed in the UAV payload bay. Optical fibers were routed in the UAV from each boom to the interrogation unit in the payload bay.

468

I. Kressel et al.

FBG 4 1FBG 3 FBG 2 CH1 CH2

FBG 1

CH4

C/L

CH3

Top

Bottom

Fig. 2 Boom general layout and embedded optical fibers location inside the boom.

4 System Validation Static calibration tests were performed in order to correlate the embedded FBG reading with boom loading. The boom was fixed to a rigid support through the same attachment points used to attach it to the wing. Tail loading in both the vertical and horizontal direction was applied at the boom end up to the maximum design limit load. FBG readings were taken at 10% increments. The static test set-up is shown in Figure 3. In order to verify the quality of the embedding process,

Loading Support

Boom Fig. 3 Boom calibration test set-Up.

Health and Usage Monitoring of Unmanned Aerial Vehicles

469

electrical strain gauges were mounted on the boom outer surface on top of each FBG sensor. Typical calibration results are presented in Figure 4. It is seen that the FBG readings are linear with respect to the applied load and are also in good agreement with the electrical strain gauges measurements. FBG4 (Ch3) and SG17 Strains in function of load (area corresponding to Bottom Side of Section 1020mm)

FBG4 (Ch2) and SG18 Strains in function of load (area corresponding to Top Side of Section 1020mm) 1600

1600

1400

1400

FBG4 SG18

FBG4 SG17

1200 STRAIN [ME]

STRAIN [ME]

1200 1000 800 600

1000 800 600 400

400

200

200

0 0 0

25

50

75

100

0

125

25

50

FBG4 (Ch1) and SG20 Strains in function of load (area corresponding to Top Center of Section 1020mm)

100

125

FBG4 (Ch2) and SG18 Strains in function of load (area corresponding to Top Side of Section 1020mm) 1600

1600

1400

FBG4 SG20

1400

FBG4 SG18

1200 STRAIN [ME]

1200 STRAIN [ME]

75 LOAD[KG]

LOAD[KG]

1000 800 600

1000 800 600

400

400

200

200

0

0

0

25

50

75

100

125

0

25

50

LOAD[KG]

75

100

125

LOAD[KG]

Fig. 4 Comparison between FBG reading and electrical strain gauges during ground calibration test.

In order to evaluate the ability of the embedded sensor to track the dynamic behavior of the boom, an impact test was performed. A weight of 60Kg was attached to the boom at the end and released by cutting its attachment to the boom. The ch1

X: 3.998 Y : 466.1

400

X: 3.986 Y : 424.5

strain (microstrain)

300

200

100

0

-100

-200

-300

-400

4

.

4.1

4.2

4.3

4.4

4.5

time (sec)

Fig. 5 FBG reading during impact test.

4.6

470

I. Kressel et al.

FBG readings during the impact test are shown in Figure 5. The first bending mode frequency, as obtained by FFT analysis (Figure 6), of the cantilever boom was found to be in good agreement with sinusoidal vibration sweep test. Ch#=1 FBG#=2 40

20

20

Amplitude(dB)

Amplitude(dB)

Ch#=1 FBG#=1 40

0 -20 -40

-20 -40

0

50

100 150 Freq (Hz) Ch#=2 FBG#=1

200

40

40

20

20

Amplitude(dB)

Amplitude(dB)

0

0 -20 -40

0

50

100 150 Freq (Hz) Ch#=2 FBG#=2

0

50

200

0 -20 -40

0

50

100 Freq (Hz)

150

200

100 Freq (Hz)

150

200

Fig. 6 FFT analysis on the impact test.

The Nishant UAV (Figure 1), Equipped with the airworthy HUMS system, was flown at Kolar air field near Bangalore, India. It was demonstrated that the flight worthy interrogation system, integrated into the Nishant UAV, withstood all flight conditions including 9g launch, flight maneuvers and parachute landing. FBG readings of ΔλB/λB (optical strain) during launch and parachute recovery are given in Figure 7. Optical strain

Optical strain 900.0 F1 FBG1

700.0 500.0 300.0 100.0 -100.0 -300.0 -500.0

125.20 125.22 125.24 125.26 125.28 125.30

Time [min.]

(a)

Time [min.]

(b)

Fig. 7 Typical FBG sensors reading. (a) Launch; (b) Parachute recovery.

Health and Usage Monitoring of Unmanned Aerial Vehicles

471

5 Summary It is concluded that the readings of the FBG sensors tracked the structural static and dynamic behavior of the UAV booms. Vibration modes and boom loads can now be identified and monitored. Usage of such highly promising, fibre optic sensors based interrogation techniques will avoid periodic grounding of the aircraft and make the maintenance schedules to be more like "on condition maintenance".

References [1] Andreas, O., Kyriacos, K.: Fiber Bragg Grating. Fundamentals and Applications in Telecommunications and Sensing. Artech House, Boston (1999) [2] Tur, M., Kressel, I., Botsev, Y., Handelman, A., Gud’s, P., Ronen, S., Cohen, S., Pillai, A.C.R., Hari Prasad, M., Kamath, G.M., Gupta, N., Khatkhate, A., Gali, S.: The use of Optical Fiber Sensors for Load Tracking and Structural Health Monitoring of UAV Composite Structures. In: ICAUV, Bangalore, India, April 3-4 (2009)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Airframe Loads and Usage Monitoring of the CH-47D “Chinook” Helicopter of the Royal Netherlands Air Force A. Oldersma and M.J. Bos National Aerospace Laboratory NLR, the Netherlands

Abstract. Prompted by severe structural maintenance issues, the Royal Netherlands Air Force has tasked the National Aerospace Laboratory NLR to develop an airframe loads & usage monitoring programme for their CH-47D helicopter fleet. After an initial pilot phase during which the technical and operational possibilities were explored, a routine programme named “CHAMP” (CHinook Airframe Monitoring Programme) was started in 2007. In addition to a fleet wide installation of a Cockpit Voice & Flight Data Recorder for the collection of the relevant parameters from the digital avionics data bus, two airframes have been equipped with a state-of-the-art data acquisition system and nine strain gauges each, which are recorded at a high sample rate. All data processing is performed off-board; no onboard data reduction is done. This has led to a vast and ever-growing database that can be used to conduct analyses that go beyond those traditionally performed within a loads & usage monitoring programme. This paper gives an overview of CHAMP and the underlying structural integrity concept that has been dubbed the “stethoscope method”. This method centres around the development of Artificial Neural Networks that use the recorded data bus parameters to predict internal loads at the strain gauge locations. After the creation of such a “virtual strain gauge”, the actual strain gauge can be relocated to monitor other key structural locations. Successive relocation of strain gauges finally results in a usage monitoring system that, in the long run, will be invaluable for structural life cycle management. Attention is paid to the acquisition and the off-board storage of the large sets of collected data, and an explanation is given of the analysis tools and methods that have been developed, and the results that have been achieved so far.

1 Introduction In the 1990s, after the end of the Cold War, the territorial defence forces of the Netherlands were transformed into a mobile army that can be deployed worldwide to protect the integrity of national and allied territory, and to promote the stability and the international rule of law. To this end, the transport capacity of the Royal Netherlands Air Force (RNLAF) has been significantly increased and now includes various types of fixed-wing aircraft and transport helicopters, among which the CH-47D “Chinook”. Initially 13 helicopters of this type were acquired. Six of these were new and featured a digital glass cockpit and a preliminary version of the machined frames that *

Oral presentation.

474

A. Oldersma and M.J. Bos

were to be introduced on the CH-47F model. The remaining seven, with classical built-up sheet metal frames, had already served in the Canadian Forces prior to being refurbished and upgraded with a glass cockpit by Boeing. Both versions are equipped with the T55-L-714A engine, which has 17% more power than the legacy T55-L-712 engine and enables operating in the hot and high conditions as encountered in for instance Afghanistan. In 2007 the Defence Materiel Organization of the Dutch Ministry of Defense ordered an additional batch of six new-build CH-47F (NL) Chinook helicopters from the Boeing Company. These Netherlands-unique version helicopters will offer advanced avionics, improved situational awareness and survivability features and special operations equipment. The aircraft will be delivered between November 2011 and June 2012. Starting in 2015, the existing RNLAF Chinook fleet will be upgraded to this latest standard. Fairly soon after entry into RNLAF service it became clear that the airframe of the CH-47D is prone to fatigue cracking, despite the fact that it has originally been designed for an infinite life. This may partly be attributed to the use of the more powerful -714 engine, but other operators have reported similar findings for fleets with the -712 engine. Most of the airframe cracking in the RNLAF fleet is found in secondary structure of the aft fuselage and aft pylon and is usually referred to as “nuisance cracking”. There is no obvious correlation with flight hours, and cracking occurs in both versions of the CH-47D that are currently operated by the RNLAF (i.e. with machined frames and with built-up frames). Although the cracks usually do not affect flight safety, they entail a tremendous amount of maintenance work and a reduction of the fleet operational availability. Because of this, and considering that it is unclear whether primary airframe structure will be affected in the long run, the RNLAF has tasked the National Aerospace Laboratory NLR to develop, implement and conduct an airframe loads & usage monitoring programme in order to keep track of the current and future operational usage of the CH-47D fleet and of the individual helicopters in the fleet, and to correlate this usage to the accrual of fatigue damage in the airframe. As a start, to explore the technical and operational possibilities and to demonstrate the benefits of a routine loads & usage monitoring programme, a pilot programme was initiated in 2001. Within this programme, one helicopter was instrumented with a simple data recorder and four strain gauges to collect airframe loads data. Additionally, a usage monitoring concept based on Flight Regime Recognition (FRR) was developed. In 2007, after successful conclusion of the pilot programme, a routine loads & usage monitoring programme named “CHAMP” (CHinook Airframe Monitoring Programme) was started in which the flight regimes were re-evaluated and the FRR algorithms were refined. In addition, two airframes were equipped with a state-of-the-art data acquisition unit and nine strain gauges each. A relatively high sample rate was selected to enable detailed vibration analyses and the development of so-called “virtual strain gauges”. The following sections provide an overview of CHAMP and the underlying structural integrity concept. Attention is paid to the acquisition and the off-board storage of the large sets of collected data, and an explanation is given of the analysis tools and methods that have been developed, and the results that have been achieved so far.

Airframe Loads and Usage Monitoring of the CH-47D “Chinook” Helicopter

475

2 Instrumentation In the pilot programme a simple four-channel data acquisition and recording unit (DARU) was installed in one helicopter (tail number D-101, with machined frames) to collect the data from four different strain gauges. This system, the Spectrapot-4C from Swiss Aircraft & Systems Enterprise, had previously been used in the F-16 loads & usage monitoring programme of the RNLAF, and had become redundant after the introduction of a more advanced DARU in the F-16 fleet. In addition, use was made of the FA2100-3073-00 Cockpit Voice & Flight Data Recorder (CVFDR) from L-3 Communications for the collection of flight data from the ARINC-429 avionics data bus. This system had recently been acquired for fleet wide installation to enable Military Flight Operations Quality Assurance (MFOQA) and mishap investigations. It is capable of recording 50 flight hours of data with 128 words per second. Since the recorded data can easily be downloaded on a portable PC with a PCMCIA card, the system is very well suited for usage monitoring as well. Both data recorders were installed in the avionics rack behind the cockpit. The CVFDR for usage monitoring and the Spectrapot for load monitoring were completely separate systems, i.e. there was no onboard connection in the form of wiring. The data from both systems were linked afterwards during ground station post-processing. This turned out to be a tedious task and the results were not always very accurate due to the poor synchronization of the two systems. In addition, the memory size of the Spectrapot cartridges was limited to 2 MB. Using the onboard processing capability (i.e. peak-valley-peak counting with time stamp retention) the amount of collected strain data could be sufficiently reduced as to allow an acceptable number of flight hours before retrieving the data from the helicopter. This could only be achieved by applying a relatively large range filter, however, which effectively removed the very damaging 3/rev and other low amplitude cycles. As a consequence the damage rates as developed in the pilot programme were incomplete and mainly pertained to manoeuvre loading. For this reason it was decided at the start of CHAMP to replace the Spectrapot by the modular state-of-the-art ACRA KAM-500 data acquisition unit and a SES S3DR-C solid state data recorder, which uses an industry standard PCMCIA ATA-Flash solid state memory card. The combined system was installed in the avionics compartment of two helicopters (the D-103, with machined frames, and the D-664, with built-up frames) – see Figure 1. Initially the signals of five strain gauges were recorded. In 2008 an additional A/D converter card was placed in each of the data acquisition units to monitor four more strain gauges. In order to avoid the synchronization problems that were experienced in the pilot program, a bus monitor card was also included to redundantly collect the relevant parameters from the ARINC-429 bus, in parallel with the CVFDR. Since the acceleration data are not present on this data bus, an additional triaxial accelerometer was installed in each of the two selected airframes; from a certification point of view it was undesirable to tap into the signals of the existing accelerometers, since they are used by the flight control system.

476

A. Oldersma and M.J. Bos

Fig. 1 Installation of the ACRA KAM-500 system with the SES solid state recorder in the avionics compartment of the D-103.

Appendix A provides a listing of the parameters that are collected from the ARINC-429 avionics data bus, including their sample rates. Figure 2 provides an overview of the strain gauge locations. Two of these strain gauges, viz. SG01 and SG05, were already used in the pilot program and are meant to enable a ompareson between the two programmes. SG01 is located on the web of the Left-hand Butt Line LBL 20.0 longeron in the cabin fuselage roof, at Frame Station FS 331. This gauge is primarily excited by bending of the fuselage around the lateral axis and is assumed to provide a general indication of the severity of fuselage loading. Strain gauge SG05 was also used in the pilot programme. It is positioned on the outer cap of the crown frame on FS 534, at LBL 26.5, which is the same location as the Boeing strain gauge 54060 as used during a strain survey on a CH-47D model for the Royal Air Force [1]. This location has a crack history on older Chinook models. It is also one of the locations where the stress level of the machined frames has increased with regard to that of the built-up frames as used in the older D-version. The other strain gauges are placed at highly stressed areas in the aft fuselage, usually at locations that were used in the Boeing strain survey for the RAF. The selection of the SG09 location was based on the presence of a crack in the FS 440 frame in one of the RNLAF Chinooks. Initially the strain data were sampled at 1024 Hz, but after analysis the sample rate was reduced to 512 Hz. To further limit the storage of irrelevant data as much as possible, the ACRA system uses a built-in trigger to start and end recording at values above resp. below 15% of the nominal rotor speed. This prevents the recording of data during maintenance related power-on periods. No on-board data reduction such as peak-valley-peak filtering or rainflow counting is applied. The retention of all strain data (together with all relevant parameters from the avionics data bus) allows more detailed analysis, which not only has proved to be useful for the development of virtual strain gauges but also for the evaluation of practical maintenance questions related to the rotor track & balancing (RTB) process.

Airframe Loads and Usage Monitoring of the CH-47D “Chinook” Helicopter

477

Fig. 2 Location of the nine strain gauges used in CHAMP.

Figure 3 provides an example of the recorded strain data, for strain gauge SG01. This graph contains two range exceedance curves, one for a reference batch of 96 flights for the D-103, with a cumulative flight duration of 159.4 hours, and the other one for a reference batch of 93 flights for the D-664, with a cumulative flight duration of 157.2 hours. The D-103 data have been collected during an outof-area operation whereas the D-664 data are representative of training missions in the Netherlands. The two exceedance curves are very similar. They are typical for helicopter airframe loading, which usually consists of a relatively small number of large load cycles (ground-air-ground cycles and manoeuvre loading) and a large number of smaller cycles due to vibratory loads that are induced by blade stall, rotor wake impingement, rotor and drive train imbalances and other causes. It is noted that the strain data in Figure 3 have been scaled to meaningful stress levels to account for the fact that the strain gauges are not located at the hot spots where fatigue cracks may develop. These cracks usually form at fastener holes or at sharp corners, notches, etc., where the presence of the fastener or the high stress gradient precludes the installation of a strain gauge. To convert the measured stress level of a particular strain gauge to a meaningful level that is representative for a hot spot (i.e. a level that gives a finite fatigue life), a spectrum stress scale factor has been determined such that the total fatigue life of the location that is covered by the strain gauge is equal to 5,000 unfactored flight hours, based on a simple fatigue damage model (i.e. S-N curve) and assuming a representative stress concentration Kt of 3.0. The material that has been considered was Aluminium 7050-T7451, which is used in the Chinook. No fatigue limit has been assumed. The measured data have been rainflow counted on a flight-by-flight basis.

478

A. Oldersma and M.J. Bos

Fig. 3 Range exceedance plots of the recorded SG01 strain values.

3 Data Storage and Data Processing For the storage of the vast amount of loads and usage data from CHAMP and similar programmes for other helicopter types, the NLR has developed and operates the dedicated yet flexible HELIUM (HELIcopter Usage Monitoring) database. This secure database is capable of handling data from any type of data source, including flight administrative data, maintenance data, raw and processed data from Health and Usage Monitoring Systems (HUMS) and other flight data recorders, etc. HELIUM is an XML database, with XPath and Xquery access to the stored data. Using XML documents as input for the data storage provides many advantages, such as easy converting to and from XML documents, a well defined syntax using XML schemas, excellent support with regard to Java, and a human readable format. The relevant flight administrative data (e.g. date, mission type, flight duration, take-off and landing location, etc.) is obtained from the centralized Integrated Maintenance Database System (IMDS) that is used by the RNLAF. This is done on a monthly basis through a remote access link. Whenever needed the frequency of this action is adjusted. The measured data are physically transported to the NLR by means of PCMCIA cards. After data validation and conversion into engineering units, they are uploaded into HELIUM. For the processing and use of the HELIUM data, a dedicated Graphical User Interface named “Sustain” has been developed. It is an IT-facility for the military operator with a fully integrated toolbox for the analysis of usage, loads and

Airframe Loads and Usage Monitoring of the CH-47D “Chinook” Helicopter

479

maintenance data in a web-based application of acquisition, processing, storage, visualisation and reporting of data. The Sustain tools can also be applied on the contents of other databases – see Figure 4.

C-130 Database

HEL IUM Database

F-16 Database

Tools Algorithms Other sources

SUSTAIN – reporting, analysis and visualisation n i tc o d e 'Q ' u

1 0 2 ur e n l a n d in g n e 1 0

Total D amag eper Fl g i ht Regi me 0.000014

#

0 8

0 6

0.000012 wt 1

wt 2

wt 3

w t4

0 4

0.00001

D amag

0 2

0 A 1 D01 Q

A D6 1 Q

AD 6 4 6Q

A D66 Q

0.000008

0.000006

/ c a

0.000004

0.000002

0 G round:

Hover :

Ascent:

Level:

D es c ent:

Aut orot ation:

Si g nl e eng i ne:

Napof Ear h t:

Ot e hr :

Fig. 4 Schematic of Sustain.

4 Flight Regime Recognition In the design and development phase of a helicopter, the analysis and the demonstration of the safe lives of the airframe and the various components of the dynamic system are based on an assumed usage. This design usage spectrum usually is a composite worst case spectrum that is to cover all possible missions by all anticipated operators in a conservative way. It is composed of the expected flight conditions, manoeuvres and other specific events, together with their relative durations (as a percentage of the total life) or amount of occurrences per flight hour. Together with the component strength characteristics and the loads measured during a flight loads survey with specially instrumented aircraft, it allows the original equipment manufacturer to assess the fatigue lives of the safety critical components, and to establish appropriate maintenance intervals. In practice the actual usage of a particular operator will be different from the design usage. From a fatigue point of view the usage will often be lighter, considering the conservative approach that is usually followed in the construction of the design usage spectrum. In this case life-limited components could be kept in service for a longer period than prescribed in the maintenance manuals and maintenance intervals could be increased. Sometimes, however, the design spectrum underestimates the actual usage [2-4], which then becomes a safety issue. In any

480

A. Oldersma and M.J. Bos

case it is advisable to monitor the actual usage of a helicopter fleet and of the individual aircraft in this fleet, in order to be able to anticipate any maintenance and safety issues and/or benefit from possible maintenance credits. To accomplish this, an automatic Flight Regime Recognition (FRR) algorithm has been developed for the CH-47D of the RNLAF [5]. ‘Regime’ in this respect can be a flight condition, a manoeuvre or a specific event such as autorotation, single engine operation or landing. The flight regime definitions used in CHAMP are similar to the basic mission profile for fatigue analysis as composed by Boeing for the CH-47D. Six main regimes are distinguished, viz.: Ground, Hover, Ascent, Level, Descent and Autorotation. A further subdivision gives a total of 127 fine flight regimes. The input to the FRR algorithm is formed by the avionics bus parameters as recorded with the CVFDR – see Appendix A. The regime data are extracted with ground-based software, similar to what has been done by other authors [6-8]. Advantages of such an approach are that (i) there is no need to integrate complex real-time airborne software and additional equipment in the current avionics suite, (ii) helicopter modifications and the effect on operations are minimized and (iii) there are more possibilities for checking the integrity of the data. To determine the flight regimes from measured CVFDR data, the following deterministic routines have been developed: • • • •

parameter processing routines to smooth and clean the measured data, basic ‘state identification’ routines that handle one parameter, additional ‘state identification’ routines that handle two or more basic states, and regime identification routines that combine states to extract regimes.

The FRR algorithms have been validated on the basis of well-defined and welldocumented sorties, by comparing the recognized flight regimes as determined from the CVFDR data with those deduced from the pilot cards. Because of the deterministic nature of the FRR algorithms, only a relatively limited number of validation flights were needed for this purpose; this would not have been the case for neural network based algorithms. Continuous improvements are made to the FRR software following analyses of new sets of data. This applies to the state identification routines for which other test flights with specified manoeuvres are used, but also to the cleaning algorithms. In this respect it is noted that the CVFDR and the KAM-500 data acquisition units have their own characteristics which sometimes require tailored data cleaning. The usage in terms of the flight regime distribution is routinely reported to the weapon system manager, together with other usage statistics such as altitude, weight and air speed distributions and exceedance plots. An example is provided in the graph below.

Airframe Loads and Usage Monitoring of the CH-47D “Chinook” Helicopter

481

Time spent in regime (%)

60%

NL training

50%

Out-of-area deployment 40%

30%

20%

10%

0%

Power On

Ground

Hover

Ascent

Level

Descent

Autorotation Other

Fig. 5 Example of usage statistics: out-of-area operations versus training missions in the Netherlands (2007).

By combining the measured strain data (scaled to a meaningful stress level) with the identified flight regimes, a relative fatigue damage rate per flight regime for each of the strain gauges could be established, using a simple fatigue damage model. A novel method has been used for allocating the appropriate fatigue damage rates to the various flight regimes, based on a modified rainflow counting method to create series of stress cycles in which the sequence of the peaks is retained. Each rainflow counted cycle is then allocated to a flight regime on the basis of the time stamp of the peak stress. In this way the transient cycles are also allocated to a flight regime, rather than collecting them in a “super Ground-AirGround (GAG) cycle” as is usually done. It is noted that the CHAMP definition of flight regimes does include the GAG condition, however; on the basis of the reported number of landings N in a sortie, the N largest rainflow counted stress cycles are allocated to this condition. The allocated damage rates are used in conjunction with the FRR algorithm to routinely keep track of the cumulative time spent in the various flight regimes and, in a relative way, of the fatigue damage that is accrued for each helicopter in the CH-47D fleet.

5 Virtual Strain Gauges The fatigue damage rates that are coupled to the flight regimes are average values for only a limited number of altitude, air speed and weight classes. They do not reflect the behaviour of individual pilots, which is recognized to be an important cause for the variability in helicopter loads, or rotor smoothing. Moreover, the allocation of the damages associated with the transient manoeuvre-to-manoeuvre cycles and the ground-air-ground cycles to the various flight regimes depends on the operational flying doctrine or flying conditions at the time that the operational loads measurement campaign (OLM) was conducted; any changes in the way that the helicopter fleet is operated (e.g. due to an out-of-area deployment) will therefore not be accommodated for unless a new OLM is carried out.

482

A. Oldersma and M.J. Bos

Within CHAMP an important next step has therefore been made. Much effort has been spent to develop an intelligent loads monitoring (ILM) technique to synthesize helicopter airframe loads from the flight parameters (air speed, altitude, bank angle, etc.), engine parameters, discretes (such as the weight-on-wheels switch) and pilot inputs that are collected with the CVFDR from the avionics data bus. For fixed-winged aircraft this technique has been around for some time, but for the complex vibratory helicopter loads the development of these so-called ‘virtual strain gauges’ is much more challenging. In a dedicated literature study it was concluded that load prediction techniques based on artificial neural network (ANN) methods are in general more accurate than prediction methods based on multiple regression methods [9-12]. Although computationally more intensive during training than regression methods, neural network methods have demonstrated better generalization properties, better representation of non-linearities in the data and slightly better results. It was therefore decided to use neural network methods to correlate the measured strains to the bus parameters. The digital bus data are recorded at relatively low sampling rates of 1 to 8 Hz, depending on the parameter that is considered. This is much lower than the 512 Hz sample rate that is used for the strain data. This disparity in sampling rate necessitated an intermediate processing step before correlation of the strain data to the digital bus data. A simple interpolation scheme did not suffice to map the strain data to the coarser time grid of the bus parameters; since fatigue damage in helicopter airframe components can usually be attributed to the vibrational loading components this would have led to the loss of essential information. On the other hand, the mapping of the bus parameters to the 512 Hz time grid of the strain data would have resulted in huge data sets, which is extremely impractical from a computational point of view. Some meaningful condensation of the strain data was thus required. The strategy that has been followed was to take the CH-47D rotor frequency R of 3.75 Hz (≈0.267 s period of revolution) as the common sampling rate for the bus parameters and the strain data. For this purpose the frequency of most of the bus parameters had to be increased slightly by interpolation. For each 0.267 s interval the average or quasi-static part of the strain signal was determined, together with the maximum and minimum of the signal within that time frame, while the deviation of this average – the dynamic part – was summarized by means of equivalent amplitudes for the frequencies 1R to/incl. 15R; in the frequency domain these multiples of the rotor frequency dominate the amplitude spectra of the strain signals, with the 3R and 6R frequencies being the most important ones1. The computation of the equivalent strain amplitudes was done on the basis of equivalence of fatigue damage potential, using a simple damage model. In the following, yavg, ymin and ymax represent one sample of the average, minimum and maximum strain during a rotor revolution, whereas yavg, ymin and ymax indicate a time sequence of the average, minimum and maximum values. In the same way f denotes one time instance of the 15 equivalent strain amplitudes and

1

That is, for the CH-47D tandem rotor helicopter, with three blades per rotor.

Airframe Loads and Usage Monitoring of the CH-47D “Chinook” Helicopter

483

F represents a time sequence of these equivalent amplitudes. For n samples F is an n × 15 matrix. Comparison of the fatigue damage potential as calculated for the original strain signals and for the signals reconstructed from yavg and F yielded an average error of less than 1%, with occasionally values up to 3%. In other words, from a fatigue damage point of view it is sufficient that the correlation model is able to accurately predict yavg and F. A further simplification could be made by accounting for the fact that the signals measured with the strain gauges are dominated by a single component of the loading spectrum. Depending on the location, this is either the 3R component (blade passing frequency) or the 6R component. In general, assuming that the fatigue damage rate is proportional to the third power of the stress or strain amplitude - which is a reasonable approximation in case of crack growth based damage - the 15 equivalent strain amplitudes f at a time instance i can be mapped to one equivalent amplitude a for frequency kR by the following equation: 15

a=

3

∑

1 3 mf m k m =1

(1)

where k=3 (most gauges), or k=6 (e.g. strain gauge SG09). The damage that can be computed from yavg and a is a good approximation of the damage that can be obtained from yavg and F. While initially the approach was to build models to estimate yavg and a, it was later on decided that the models should also predict the largest cycle per time frame, i.e. ymax-ymin, in order to be able to accurately reconstruct the transient behaviour of the strain signal when going from one flight condition to another. Assuming that the largest cycle is part of the kR content of the signal, the equivalent amplitude of Eqn.1 was adapted in the following way:

a=

3

15 − y min ⎛y 1 3 mf m − ⎜⎜ max k − 1 m=1 2 ⎝

∑

⎞ ⎟⎟ ⎠

3

(2)

So finally three ANN-based correlation models had to be developed for each of the strain gauges to predict yavg, (ymax-ymin) and a from the recorded bus parameters. From these predicted parameters a stress or strain history can be constructed and the associated fatigue damage or crack growth rate can be calculated with any desired model. For correlating yavg, (ymax-ymin) and a to the avionics bus data, feed-forward models with one hidden layer have been used. Each of the three models had 30 hidden nodes. The activation function that was applied in the layer with the hidden nodes was either the ‘tanh’ function – for (ymax-ymin) and a – or a linear activation function – for yavg; the latter function does not change the value of the input. The selection of the subset of model input parameters from the complete set of available bus parameters of Appendix A has been performed such that the virtual

484

A. Oldersma and M.J. Bos

strain gauge models generalize well on new data. Two parameters turned out to be unreliable and had to be discarded; they were the helicopter gross weight (which is manually input by the pilot) and the measured accelerations (affected by drift). The final set of selected model input variables is provided in Appendix B. Two large data sets have been assembled for the training and validation of the models, one for the D-664 with built-up frames and one for the D-103 with machined frames. From the available data files about 1/6th with the largest damage per flight hour and about 1/6th with the smallest damage per flight hour were used to build the models. From the selected files every twentieth data point has been used, yielding 20,000-25,000 data points with which the models have been trained.

Fig. 6 Virtual fatigue damage against actual fatigue damage for two locations in the airframe of the D-103. Each circle represents one flight.

The models have been validated against newly obtained data, which were not used in the training. The predictive capability has been assessed by comparing the fatigue damage content of the predicted strain sequences with the actual strain sequences, on a flight-by-flight basis. The accuracy turned out to be very satisfactory, although some variation was observed and some issues were identified. Examples of the validation results are provided in Figure 6, which shows the fatigue damage as computed on the basis of the predicted loads (virtual strain gauges) against the damage on the basis of the actual loads (real strain gauge data) for two locations in the airframe of the D-103. For perfect virtual strain gauges the predictions in above graphs would all lie on the diagonal lines. It is noted that the accuracy is expected to improve even more once larger and better sets of input data become available. An example of a predicted strain sequence in comparison with a measured strain sequence (both scaled to meaningful stress levels by the same factor) is given in Figure 7, for strain gauge SG06. The data pertain to a 5.6 hr flight of the D-664 in July 2007. The computed fatigue lives for the predicted and measured sequences differ by only 2% to 3%, depending on the adopted damage model.

Airframe Loads and Usage Monitoring of the CH-47D “Chinook” Helicopter

485

Fig. 7 Comparison of a predicted and a measured load sequence.

5 Fleet Life Management The intelligent loads synthesis technique described in the previous section has been incorporated in the Sustain tool box. It can easily be extended to other areas in the airframe of the Chinook helicopter or other helicopter types with a digital data bus, as long as measured loads data are available to train the underlying neural networks. In this respect it is noteworthy that much work is going on worldwide in the development of wireless sensors that can be used to monitor rotating components – see for instance [13]; this eventually will enable the application of the virtual strain gauge technology to the dynamic system of a helicopter as well. In combination with the HELIUM database, virtual strain gauges are the key to the rational management of the life of any structural component of any helicopter in the inventory of the RNLAF. Based on this technique, a new fleet life management (FLM) concept has been developed called the “Stethoscope Method”, which to a large extent is already operational for the CH-47D fleet of the RNLAF. This method is outlined in Figure 8. It involves the fleet wide collection of all relevant flight, engine and control parameters that are available from the digital data bus, plus the simultaneous collection of loads data in one or more dedicated OLM helicopters. For this purpose all helicopters in the fleet need to be equipped with a digital flight data recorder (FDR; for the Chinook this is the CVFDR). The OLM helicopters will need an additional structural data recorder (SDR; for the Chinook this is the ACRA KAM-500 unit) or, alternatively, an additional FDR loads monitoring functionality. Depending on the need, the SDR may consist of a fully configurable multichannel system or a simple to install stand-alone recorder such as the Stand ALone Structural data Acquisition system (SALSA) that has recently been developed by the NLR.

486

A. Oldersma and M.J. Bos

FLEET

OLM h/c

FDR Tail 1 FDR + SDR FDR Tail 2

FDR Tail 3

FDR Tail n

SUSTAIN tool box

MatDat

SG SG SG

Other Tool FRR

CDI

ILM

Virtual Virtual Virtual

training module ILM tool box

Damage Model

Other Tool

Fleet Life Management Fig. 8 The Stethoscope Method for Fleet Life Management.

All measured data are or will be made accessible through the HELIUM database system that serves all helicopter types that are used by the RNLAF. This opens the door to the use of virtual strain gauges for load monitoring purposes, where the strain or stress sequence at a particular location in the airframe or dynamic system is derived from the sequence of digital bus parameters. The loads data from the SDR in the OLM helicopters will be used to continually train and improve the ANN-based virtual strain gauge models. By moving the strain gauges in the OLM helicopters around on a regular basis2, models will be obtained for more and more points in the airframe or dynamic system that are relevant from a fatigue point of view. This will allow the establishment of the safe life consumed so far and the assessment of the severity of in-service developed fatigue cracks for each critical point. In this respect it is essential to start collecting the digital bus data from the very first moment that a helicopter is commissioned into service. For the virtual strain gauge models this is less crucial; when developed at a later stage these models can use the historic bus data in the HELIUM database to ‘roll back’ to day one.

2

In the CH-47D the wiring has been installed such that the gauges can easily be moved around once a sufficient set of data has been collected (typically a few hundred flight hours of data acquisition).

Airframe Loads and Usage Monitoring of the CH-47D “Chinook” Helicopter

487

6 Typical Programme Results Vibration analysis Owing to the high sample rate of the strain gauges it was possible to investigate the relevance of the various vibration components with respect to the airframe loads. As a first step in the analysis, the strain signals were scaled to obtain meaningful stress levels (i.e. levels that yield a finite component life). After that the frequency content of the resulting stress sequences was determined by means of a Fast Fourier Transform (FFT). The first twelve harmonics of the 3.75 Hz rotor frequency (in other words the 1R to/incl. 12R components) were extracted from the FFT spectrum using a bandwidth of plus and minus 5%. After an inverse FFT this band-pass filtered spectrum was processed to assess its damage content. This involved the extraction of the stress peaks and valleys and rainflow counting. Per frequency band the resulting stress ranges were transformed to a equivalent stress cycle that yields the same fatigue damage. Comparison of the equivalent stress cycles (and their frequency!) and the application of a simple damage model finally provided the relative contributions of the frequency components to the accrual of fatigue damage. Normalising to a total sum of 100 % gave results like those shown in Figure 9 for strain gauge SG06 in the D-664. For this gauge the dominant frequency turned out to be the 3R, which is the blade-passing frequency. This observation is valid for most of the gauges in the aft fuselages of the D-664 and D-103. An exception is strain gauge SG09, for which the 6R frequency is the dominant frequency from a airframe fatigue point of view.

Fig. 9 The relative importance of the various frequency components for strain gauge SG06 in the D-664, based on a batch of 88 flights.

In a different approach the stress sequences were manipulated by removing the contribution of the 3R-vibration (+/- 10%), the 1R-vibration (caused by rotor imbalance), the 6R-vibration, all vibrations (keeping the quasi-static manoeuvre loading only) and all vibrations except the 3R. It was assumed in this respect that manoeuvres take place at frequencies below 3 Hz. For the manoeuvre-only loading all frequencies exceeding 3 Hz were filtered out. With the same stress scale factors used in the previous analysis, the lives under the manipulated spectra were

488

A. Oldersma and M.J. Bos

calculated and the relative importance of the various frequency components was determined. The relative importance was defined as: (3)

Relative Importance = (1 --- Original Life / Life)*100 %

where “Original Life” is the life under the unmodified (but scaled) stress spectrum and “Life” is the life under the spectrum from which the frequency component in question has been removed. The results for strain gauges SG01, SG06 and SG09 are shown in Figure 10. They confirm the findings from the previous analysis. It is clear that manoeuvre loading does not significantly contribute to the accrual of fatigue damage at the stress levels that lead to a finite life under the unmodified but scaled load spectrum. These results also anwer the question whether more frequent rotor balancing would be benificial in order to reduce the accrual of fatigue damage. Since rotor balancing primarily reduces the 1R vibration, it is concluded from Figures 9 and 10 that the reduction in fatigue damage will be negligible. In the analysis no distinction has been made between flights just after an RTB maintenance action and flights just before that, but apparently the current maintenance frequency is sufficient to keep the 1R vibration levels at an acceptable level.

relative importance w.r.t. fatigue damage (%)

120 SG01 100

SG06 SG09

80

60

40

20

0 1R vibration

3R vibration

6R vibration

all vibrational components, except 3R

all vibrational components

Fig. 10 The relative importance of the various frequency components for strain gauge SG06 in the D-664, based on a batch of 88 flights and 214 flight hours.

It should be noted that it is a theoretical impossibility to truly separate the contributions of the various frequency components to the accrual of fatigue damage. The reasons are that (i) the fatigue content of a stress or strain signal in the time domain is best characterized with the rainflow counting method, which couples stress peaks and valleys that can be spaced apart quite significantly, (ii) the peaks

Airframe Loads and Usage Monitoring of the CH-47D “Chinook” Helicopter

489

and valleys in a stress or strain signal in the time domain are usually constituted by more than one frequency component, and (iii) the effects of load interaction cannot be accounted for in a frequency analysis. The relative fatigue damage values in above histogram are therefore indicative only. Operational speed limit In a separate analysis of the accrual of fatigue damage per flight regime it was shown on the basis of strain measurements that Straight & Level (S&L) flight is the main contributor to the severe fatigue related maintenance issues in the aft fuselage of the CH_47D fleet of the RNLAF. Although many other flight conditions are more severe in terms of accrued fatigue damage per flying hour (or damage rate), the helicopters in the fleet simply spend most of their time flying S&L. Further analysis indicated a strong dependency of the damage rate on the air speed. An example is give in Figure 11, which shows the normalized damage rate for strain gauge SG09 versus the indicated air speed (IAS) for a number of selected S&L flight segments that were flown in the Netherlands (i.e. at relatively low pressure altitudes). Above 80 KIAS the damage rate grows exponentially with increasing air speed.

Fig. 11 Fatigue damage rate for S&L flight segments, relative to the fatigue damage rate at 120 KIAS (strain gauge SG09 in the D-664).

In view of these results an operational limit of 120 KIAS has been proposed, unless prohibited by operational circumstances. Backed by the Operations Directorate, the CH-47D flying community was briefed in December 2009 to obtain support and understanding. The actual limit was imposed in January 2010. Based on peace-time usage in the Netherlands, a reduction of at least 22% in fatigue damage related maintenance man hours is expected. This will be evaluated in due course.

490

A. Oldersma and M.J. Bos

At present it is not clear whether the fatigue loads are governed by the dynamic pressure (in other words, IAS) or by the true air speed (TAS), for instance due to retreating blade stall and/or Mach effects. For usage in the Netherlands this does not really matter. For out-of-area operations in mountainous areas like Afghanistan, the TAS is typically 110-120% of the IAS, however. The effect of this remains to be analysed.

7 Conclusion and Outlook CHAMP is a well established programme that is used in support of the sustainment of the CH-47D fleet of the RNLAF. The use of two dedicated OLM helicopters that are equipped with a structural data recorder and strain gauges has proven to be invaluable. The high sample rate of the strain data enables detailed analyses that are beyond the scope of traditional loads & usage monitoring programmes. The so-called ‘virtual strain gauges’ that have been developed for a number of locations in the airframe of the CH-47D are capable of predicting both the quasistatic loads due to manoeuvres, ground-air-ground cycles, etc., and the vibratory loads that are induced by blade stall, rotor wake impingement, rotor and drive train imbalances and other causes. Some tuning is still needed. Although already satisfactory, the accuracy is expected to improve once larger and better sets of input data (incl. all-up weight and accelerations) become available. The tools, models and methods that have been derived are very powerful as they can easily be extended to other areas in the airframe or drive train of the Chinook helicopter or other helicopter types.

Acknowledgement The research described in this paper has been performed under various contracts from the Defence Materiel Organization of the Netherlands. The support from the Defence Research & Development section and the Weapon System Management Office is gratefully acknowledged.

References [1] Helicopters, B.: RAF HCMk2A M4454 Ground and Flight Test Report - Appendix F: Flight Strain Survey D352-10077-1, Philadelphia (1998) [2] Robeson, E.: MH-47E Structural Usage Monitoring System (SUMS) Fleet Demonstration Results. In: American Helicopter Society 56th Annual Forum, Virginia Beach, USA (2000) [3] Harris, W.D., Larchuk, T., Zanoni, E., Zion, L.: Application of probabilistic methodology in the development of retirement lives of critical dynamic components in rotorcraft. In: American Helicopter Society 55th Annual Forum, Montreal, Canada (1999)

Airframe Loads and Usage Monitoring of the CH-47D “Chinook” Helicopter

491

[4] Adams, D.O., Kershner, S.D., Thielges, J.: Economical and reliable methods of processing HUMS data for maintenance credits. In: American Helicopter Society 55th Annual Forum, Montreal, Canada (1999) [5] ten Have, A.A., de Witte, P.J.H.H.: Enhanced platform availability through new FLM concepts, NLR Technical Publication TP-2008-415, Amsterdam, The Netherlands (2008) [6] Dickson, B., Cronkhite, J.D., Summers, H.: Usage and structural life monitoring with HUMS. In: American Helicopter Society 52nd Annual Forum, Washington D.C., USA (1996) [7] Teal, R.S., Evernham, J.T., Larchuk, T.J., Miller, D.G., Marquith, D.E., White, F., Deibler, D.T.: Regime Recognition for MH-47E Structural Usage Monitoring. In: American Helicopter Society 53rd Annual Forum, Virginia Beach, USA (1997) [8] Lu, Y., Christ, R.A., Puckett, T.A., Teal, R.S., Thompson, B.: AH-64D Apache Longbow Structural Usage Monitoring System (ALB SUMS). In: American Helicopter Society 58th Annual Forum, Montreal, Canada (2002) [9] Haas, D.J., Milano, J., Flitter, L.: Prediction of Helicopter Component Loads using Neural Networks. Journal of the American Helicopter Society 40(1) (1995) [10] Haas, D.J., Flitter, L., Milano, J.: Helicopter Flight Data Feature Extraction/ Component Load Monitoring. Journal of Aircraft 33(1) (1996) [11] Cabell, R.H., Fuller, C.R.: Neural Network Modelling of Oscillatory Loads and Fatigue Damage Estimation of Helicopter Components. Journal of Sound and Vibration 209(2) (1998) [12] Allen, M.J., Dibley, R.P.: Modeling Aircraft Wing Loads from Flight Data Using Neural Networks, NASA report NASA/TM-2003-212032 (2003) [13] Arms, S.W., Townsend, C.P., Galbreath, J.H., Liebschutz, D., Phan, N., Jones, A., Baker, T.: Flight testing of a wireless sensing system for rotorcraft CBM, Combined AHS and AIAA specialists’ meeting on Airworthiness, Condition Based Maintenance and Health/Usage Monitoring Systems, Huntsville, AL, USA (2011) [14] Brown, W.P., Steinmann, H.H.: The CH-47 Cruise Guide Indicator. Presented at the 26th Annual National Forum of the American Helicopter Society, Washington D.C., USA (1970)

492

A. Oldersma and M.J. Bos

Appendix A – Recorded Data Bus Parameters An overview of the flight parameters, engine parameters and discretes that are collected from the ARINC-429 avionics data bus is given in the table below. The signals from the triaxial accelerometer that has been specifically installed for the CHAMP programme are labelled with ‘_NLR’. Long name

Short name Unit

Sample freq. [Hz] CVFDR ACRA

Aircraft type and sequence code Pitch Angle Roll Angle Wind Angle Longitudinal CG Acceleration Lateral CG Acceleration Normal CG Acceleration Longitudinal CG Acceleration Lateral CG Acceleration Normal CG Acceleration Calibrated Air Speed Cruise Guide Indicator Date Day Collective Control Position Pitch Stick Position Roll Stick Position Yaw Pedal Position Engine #1 Fuel Flow Engine #2 Fuel Flow LH aft tank fuel qty RH aft tank fuel qty LH fwd tank fuel qty RH fwd tank fuel qty LH main tank fuel qty RH main tank fuel qty FMS Ground Speed Magnetic Heading Calculated Present Position Lattitude Hook Load Calculated Present Position Longitude Engine #1 Gas Generator Speed Engine #2 Gas Generator Speed Engine #1 Power Turbine Speed Engine #2 Power Turbine Speed Calculated Pressure Altitude Engine #1 Oil Pressure Engine #2 Oil Pressure Engine #1 PTIT Engine #2 PTIT

AC AP AR AWND AXB AYB AZB AX_NLR AY_NLR AZ_NLR CAS CRGD DAY DCOL DSP DSR DYP FFEN1 FFEN2 FQTA1 FQTA2 FQTF1 FQTF2 FQTM1 FQTM2 GSFMS HDGM LATC LHOOK LONC N1EN1 N1EN2 N2EN1 N2EN2 PAC POEN1 POEN2 PTITEN1 PTITEN2

0.015625 2 2 0.25 4 4 8 n.a. n.a. n.a. 1 1 0.015625 2 2 2 2 1 1 0.015625 0.015625 0.015625 0.015625 0.015625 0.015625 1 1 1 2 1 1 1 1 1 1 1 1 1 1

ctd deg deg deg g g g g g g kts % ctd inch inch inch inch lbs/hr lbs/hr lbs lbs lbs lbs lbs lbs kts deg deg lbs deg % % % % ft psi psi ºC ºC

n.a. 2 2 n.a. n.a. n.a. n.a. 4 4 8 1 1 n.a. 2 2 2 2 n.a. n.a. 1 1 1 1 1 1 1 1 1 2 1 1 1 1 1 1 n.a. n.a. n.a. n.a.

Airframe Loads and Usage Monitoring of the CH-47D “Chinook” Helicopter

493

Long name

Short name Unit

Sample freq. [Hz] CVFDR ACRA

Radio Altitude Altitude Rate Static Air Temperature Engine #1 Torque Engine #2 Torque Rotor Speed Wind Speed Gross weight Weight on wheels, left rear Weight on wheels, right rear

RA RALT SAT TQEN1 TQEN2 VRTR VWND WEIGHT WOW1 WOW2

1 2 0.5 1 1 2 0.25 0.015625 1 1

ft ft/min ºC % % % kts lbs ctd ctd

1 2 1 1 1 2 n.a. 1 1 1

Appendix B – Ann Model Input Parameters The parameters that are listed in the table below and their cross-terms (for instance TQEN1*CRGD) are the inputs for the ANN models that constitute the virtual strain gauges. They are a subset of the parameters described in Appendix A. Variable/equation

Comment

1-(WOW1 or WOW2) AP AR TQEN1 Dynamic Pressure DSP DYP DCOL CRGD CRGD2 log(CRGD) 1/ CRGD2 1/ CRGD3

Weight-On-Wheels switch Pitch angle Roll angle Engine #1 torque computed from PAC and CAS Pitch stick position Yaw pedal position Collective control position Value of Cruise Guide Indicator on cockpit display; 0-100% indicates acceptable operation; flight operations for CGI >100% should be minimized to avoid excessive build-up of fatigue damage in drive train

It is noted that the CRGD parameter is specific to the Chinook helicopter. It is derived from the signals of strain gauge bridges that are installed on the aft pivoting actuator and fixed link of the aft rotating swash plate [14]. A narrow-band filter is applied around the 3R frequency component. This means that for other helicopter types and for locations that are governed by frequency components other than the 3R a different set of input parameters will be needed, possibly including the triaxial accelerations and the helicopter weight characteristics. This needs to be established case by case. For the Chinook helicopter it seems unnecessary to include measured RTB data in the input set. The 1R vibration is judged to be irrelevant for the loading of the airframe, provided that the RTB limits are properly maintained, whereas the 3R vibrations are implicitly accounted for by the CRGD parameter. This may not be the case for other helicopter types, however.

26th ICAF Symposium – Montreal, 1-3 June 2011 Damage Detection System for Automated Hot Spot Monitoring Based on Different Technologies Used in Component Testing for Shot Peening Validation C. Stolz, M. Neumair, and L. Benassi CASSIDIAN, Structural Integrity Management & SHM, Manching, Germany

Abstract. Structural Health Monitoring (SHM) is an important requirement to handle military aircraft safety. In contrast with civil aircraft, missions, configurations and environment are changing frequently and therefore also the load spectra, which lead to various life consumptions. Event and fatigue monitoring, remaining life assessment and damage detection monitoring are the basic functions for economic and safe in-service operations of a flying weapon system. Hot spot monitoring for damage detection is an important part of modern SHM systems which can be installed in new aircraft as well as in existing aging aircraft structures. The availability of information on damage dimension is essential for a risk evaluation. Information on the damage growth delivers the input for degradation prognostics. Together with data from the individual usage and loads monitoring, a remaining life assessment is possible. This assessment is part of an integrated health monitoring system. Basic functionalities and an approach for the architecture will be described. This paper describes the results of complex landing gear component tests conducted for the validation of shot peening life enhancements. Details on life benefit due to the residual stresses of the high-strength aluminium alloy design will be presented. One part of the testing was to evaluate the performance of hot spot monitoring technologies using imaging ultrasonic and acoustic emission sensors. Initial objectives of the tests were to obtain information on durability, reliability of the sensor system and system validation. Automatic signal processing and damage size quantification were further objectives of the tests. Referring to this test experience, new ideas for hot spot monitoring, combining different technologies in one system will be presented and the requirements for further development on a modern SHM-System and its implementation into an integrated health monitoring system will be discussed.

1 Structural Health Monitoring The aim of Structural Health Monitoring Systems is to monitor the structural condition of an aircraft or aircraft structure. To ensure the structural integrity of the airframe and structural systems by modern structural health monitoring systems the four main functions, shown in Figure 1, are essential: event and fatigue life monitoring, including remaining life assessment with interface to logistic support and damage monitoring. [1]

496

C. Stolz, M. Neumair, and L. Benassi

Fig. 1 Main parts of a modern SHM-System [1].

In detail, the first function, monitoring of selected structural events, includes real time monitoring of every structural load exceedances, resulting from a usage outside the cleared envelope (for military fighter aircraft mainly due to manoeuvres). This function is related to the requirements of airframe safety and certification. Monitoring of heavy impacts may become a requirement for future UAV's. The second function, fatigue life monitoring, includes the usage monitoring and the monitoring of the life consumption of all fatigue relevant and critical locations. The monitoring of the life consumption against the certified life of the structure is the main task here, in order to fulfil the airframe certification requirements. In the last years the importance of the third function, damage monitoring, is growing. It is the area of continuous and automated inspections at fatigue relevant locations and Structural Significant Items (SSI’s) including all areas which are exposed to a high risk of impacts causing structural damage or an operational impact due to damage (e.g. impact damage on a radome structure). In addition, the monitoring of the long term damage propagation is a special task here, which requires sophisticated diagnostic and prognostic capabilities. This function is again related to the requirements of airframe safety and certification and to the requirements of mission assurance. The fourth function, logistic support, is based on the capabilities of all three previous functions and uses its results to monitor life time and fatigue life exceedances in real time. It provides the essential life parameter like remaining useful life, to the logistic and maintenance support system. Sophisticated prognostic capabilities for aircraft usage and damage assessment are required for this task. In addition to the requirements of airframe safety and certification, the main requirements for this function are related to mission assurance and operational philosophy, in connection to logistic concepts like condition based maintenance (CBM).

Damage Detection System for Automated Hot Spot Monitoring

497

To reduce the amount of periodic inspections (one major SHM goal), modern SHM systems should have the capability to integrate damage detection functionalities for hot spot monitoring of specific components. The aim of hot spot monitoring is to monitor a specific area of a component with a high confidence level and to quantify occurring damage. This paper focuses only on the damage monitoring function, describing the integration into a health management system and special functionalities used for component testing. Elements of a Damage Monitoring System For a robust and efficient monitoring, the monitoring system consists of several design elements with defined interfaces. The raw monitoring signals are generated by sensors with are directly connected or integrated into the structure. The sensors are connected by wires or wireless to the interrogation unit. The interrogation unit collects the raw data from several sensors. Depending on the technologies, a first data manipulation and, if needed, data conversion is performed in order to provide the relevant monitoring data to an aircraft digital bus system. The monitoring data from the bus system are transferred to the next design element, where an onboard processing and data storage are performed. Modern systems will have an integrated health management (HM) system with a central control and processing unit and data storage for all connected aircraft systems. The onboard data are transferred to the ground element of the health management system. The main data processing activities and the detailed data analyses are performed in this design element with sophisticated diagnostic and prognostic features. Finally, for data and analysis interpretation the results are provided to the engineering and maintenance staff. [1] System Architecture The proposed strategy for an SHM system is to integrate both usage and damage detection in order to emphasise that both aspects are necessary and they complement each other. For this approach a global system architecture is necessary which allows direct interfaces to other aircraft system. The core of the proposed integrated health monitoring system is based on an Open System Architecture for Condition-based Maintenance (OSA-CBM)[2, 3], as shown in Figure 2. Each sub-system contains the Data Acquisition (DA), Data Manipulation (DM) and State Detection (SD) layers locally, whilst the Health Assessment (HA), Prognostic Assessment (PA) and Advisory Generation (AG) layers are centralised within the health management core which correlates the health-related messages arriving from each sub-system. After landing, the PA interacts with the aircraft Configuration Management database and the Mission Planner in order to produce an appropriate AG for each aircraft. Then, the AG routine instructs the Command and Control, Mission and Maintenance crews on the combined diagnostic and prognostic assessment of the systems managed within the IVHM such that maintenance activities are planned with minimal interference to the mission schedule. Thus, the implementation of such

498

C. Stolz, M. Neumair, and L. Benassi

Condition-based Architecture enhances the Mission Capability Rate by increasing the fleet availability. [4]

Fig. 2 OSA – CBM System Architecture for SHM [4].

2 Component Tests for Shot Peening Validation The use case for damage monitoring described here is a testing campaign for the validation of shot peening fatigue enhancement. The tested component is a main landing gear back-up structure of a fighter aircraft. It consists of a complex machined high-strength aluminium (Al 7050-T7651) structure with two lugs as shown in Figure 3. The component is loaded through a hydraulic actuator which is attached to the lugs by a bolt. The complete test structure is fixed on a rigid steel test rig. In the area of the run-out of one lug, the stresses due to fatigue load are relatively high as shown in the FEM results of Figure 3. During fatigue testing cracks initiated at the lug surface and grew through the lug as shown in the dyepenetrant testing image of figure 4. In order to extent the time until a crack initiates in the high stressed area shot peening of the critical area has been performed. The objective of this component test is to validate the shot peening life benefit by comparing the results of pristine and spot peened components. The fatigue tests were performed with a load sequence simulating the loads during retraction and lowering cycles (R/L cycles) of the landing gear. The maximum load was 17,8 kN (actuator in tension) and the minimum load was -34 kN (actuator in compression). The resulting maximum strain at the sensor position were next to 1000 µm/m. Values in the high stressed area were significantly higher.

Damage Detection System for Automated Hot Spot Monitoring

Geometry model Sensor

499

FEM result

Fig. 3 Geometry model and FEM stress results of the tested component.

Crack Fig. 4 Dye-penetrant testing image of the cracked component.

Several test components were equipped with damage monitoring sensors to detect the crack, monitor its growth and to investigate the durability and reliability of the sensor system. A picture of the test component installed in the steel test rig and the applied sensor is shown in Figure 5.

Sensor

Fig. 5 Test component with bonded SHM sensor.

500

C. Stolz, M. Neumair, and L. Benassi

Effect of Shot Peening Different component configurations (pristine, initially shot peened and shot peened after a pre-damage of 4000 R/L cycles) were tested to assess the benefit of shot peening on the component fatigue life. In the untreated components, initial fatigue cracks developed shortly after 5000 R/L cycles. The initial cracks of the shot peened components without pre-damage were detected after 15800 cycles and 20000 cycles. This means, the machined and initially shot peened components (i.e. no consideration of pre-damage) have shown an approximately three to four times higher fatigue life than the untreated ones. Initial cracks in the pre-damaged shot peened components developed in the period between 15800 cycles and 20600 cycles after the shot peening. These test results, shown in Figure 6, have revealed that shot peening does have the capability to "reset" the pre-damage to "zero". The observed fatigue life up to initial crack is in the range of the initial shot peened components. All shot peened components have been cracked (crack length greater than 20 mm) before 30000 R/L test cycles.

R/L cycles up to initial crack (post shot peening)

25000 20000 15000 10000 5000 0 Pristine

Shot-peened

Pre-damaged & Shot peened

Fig. 6 Fatigue test results.

3 Combined Acoustic Emmission and Imaging Ultrasonic Sensor System for Damage Detection In order to explore and validate the damage detection system functionalities and their interfaces to an IVHM system, an ultrasonic sensor system was used for several component tests. The system KTS221 [5] provides a low-voltage, broadband multichannel sensor network, which fulfils the requirements for the hot-spot monitoring task. Here, a 20-channel piezo-transducer network with 2 MHz bandwidth, chained up as a linear array in a bus line is bonded to the metallic

Damage Detection System for Automated Hot Spot Monitoring

501

component surface of a lug by 2K-epoxid adhesive (Figure 5, sensor width: ~6mm). Additionally on some components, two single channel sensors with 200 kHz bandwith where bonded on the surface and connected to the bus line to monitor the acoustic emissions. Acoustic Emission

AE counts per interval

All three sensors (one 20-channel-2MHz- and two single-channel-500kHzsensors) were used to detect acoustic emissions which originate from the initiating and growing fatigue crack. The acoustic events received by the sensors were counted periodically over small time periods during the complete fatigue test. Analysing these data together with the optically measured crack growth, shown in Figure 7, reveals a correlation in the early crack growth phase (crack length smaller than 2 mm). The number of emission events is very high in this phase. In the later crack growth phases the system showed only very limited effects. Therefore the system philosophy needs reviewing and acoustic emission analysis algorithms need enhancement. 300000 250000 200000 150000 100000 50000

Optical crack length [mm]

0 0

5000

10000

15000 20000 R/L cycles

25000

30000

35000

0

5000

10000

15000 20000 R/L cycles

25000

30000

35000

16 14 12 10 8 6 4 2 0

Fig. 7 Acoustic emissions measurements and optical crack length test results.

Imaging Ultrasonic The Imaging Ultrasonic System comprising the 2 MHz sensor with 20 channels arranged in an array, was triggered and guided signals were sent through the structure. Reflections from geometric faces (like back walls, edges and the crack surface) were received by the same sensor and the data were analysed, processed and the results were shown as an image. The complex scanning, processing and imaging algorithms need several seconds until the result is available for the user. Therefore a periodical scanning was used.

502

C. Stolz, M. Neumair, and L. Benassi

A schematic overview of the imaging scanning area and an imaging ultrasonic picture of a pristine component is shown in Figure 8 and 9. Tests of the KTSSensor showed good results with an initial crack detection sensitivity of 2-5mm at 85 degrees angle w.r.t. to the array normal due to a partial masking of the back wall echo. Imaging ultrasonic pictures with masked back wall echos and an increasing crack are shown in Figures 10, 11 and 12. The further the crack propagates into the core field of the imaging aperture the better becomes the azimuth resolution towards the physical limit: about 2-3mm for a crack length of approx. 9mm.

Fig. 8 Schematic overview of the imaging scanning area of component.

Echo of bolt hole

Echo of complex backwall

Fig. 9 Imaging scan of undamaged component.

Damage Detection System for Automated Hot Spot Monitoring

503

Echo of bolt hole

Echo of complex backwall

Fig. 10 Imaging scan without a crack.

Signs of crack echo

Echo of bolt hole

Fig. 11 Imaging scan with a crack length of approx. 5 mm.

Echo of partially masked backwall

Fig. 12 Imaging scan with a crack length of approx. 11 mm.

Echo of bolt hole

504

C. Stolz, M. Neumair, and L. Benassi

The tested durability of the sensor and its bonding to the metallic surface is good and satisfied the expectations. The sensor application lasted for more than 35000 R/L cycles. Automated signal processing For an automated signal processing, the imaging data must be transferred to "upper layers" of the OSA-CBM architecture in a digital way using Built-In Test (BIT) messages. The BIT messaging architecture was derived from an existing EADS aircraft type and applied to all sub-systems. The BIT message of an SHM damage detection sensor then contains the definition of an area of concern in which damage occurs in terms of a box, defined by two diametrically opposed vertices A and B and their respective spatial Cartesian coordinates (Ax,Ay,Az,Bx,By,Bz) to give a coarse estimate of damage location and its extent continuously at relatively short time intervals. These six coordinate values are determined as follows: 1. For all grid points (resolution d << wavelength l) in the area of interest the local echo distribution is compared to a previously recorded local echo distribution: for a required confidence level (e.g. 99,5%) the decision is made, whether the actual distribution is statistically sufficiently equal to the original echo representation. This leads to a set of points that represent potentially damaged points Di. Only if the accumulated number S of such points times the grid`s unit volume d3 substantially exceeds (l/2)3 it is necessary and reasonable to indicate damage, i.e. 2. For S*d3 < (l/2)3 the BIT message replies with “ok”. 3. For S*d3 > (l/2)3 a simple Min/Max operation on the potentially damaged point’s coordinates Di gives the six coordinate values of the vertices of the “crack box” to be sent as part of the BIT message “damage” to coarsely indicate damage location and its extent. (Ax,Ay,Az)=(Min(Di x),Min(Di y), Min(Di z) (Bx,By,Bz)=(Max(Di x),Max(Di y), Max(Di z) If “damage” is indicated, the underlying raw data set is automatically kept to assist a detailed diagnosis, prognosis, assessment and maintenance process that is triggered based on BIT messages. Combined system To enhance the applicability in an aircraft scenario, the idea is to combine these two SHM technologies into one system process. The permanent or short interval scanning with the acoustic emission technique will be used to detect the initial crack and in case of a crack, it will trigger an examination with Imaging Ultra Sonics to monitor the crack growth. This process will use the benefits of both technologies, lower false alarm rates, enhance the robustness of the SHM system and reduce the scanning effort.

Damage Detection System for Automated Hot Spot Monitoring

505

Further a sophisticated combination of usage monitoring with damage monitoring is a major step for improved prognostics. Important issues for the integration of the SHM systems into health monitoring systems are the increase of confidence in damage detection diagnostics and prognostics. These functionalities will need further enhancements in order to satisfy the challenges of future SHM systems.

References [1] Stolz, C., Neumair, M.: Int. Journal of Structural Health Monitoring 9(3), 209–217 (2010) [2] Puri, G.V., Boylan, D.C., Walter IV, R.L., Lebold, M.S.: Building an OSA-CBM (Open System Architecture for Condition-based Maintenance) System (2006), http://www.mimosa.org/downloads/types/whitepapers/ index.aspx [3] Bever, K.D.: Understanding MIMOSA’s Open Standards for On-board Health Assessment and Enterprise Application Integration. Presented at the AIAA 2007 Conference and Exhibit, Rohnert Park, CA, May 7-10 (2007) [4] Benassi, L., Neumair, M., Buderath, M.: From System Integration to Autonimous Systems. In: Chang, F.-K. (ed.) Proceedings of the 7th International Workshop on Structural Health Monitoring, vol. I, pp. 410–417. DEStech Publications, Lancaster (2009) [5] Kress, K.P., Bach, M., Saby, M.: From System Integration to Autonimous Systems. In: Chang, F.-K. (ed.) Proceedings of the 7th International Workshop on Structural Health Monitoring, vol. I, pp. 846–853. DEStech Publications, Lancaster (2009)

26th ICAF Symposium – Montreal, 1-3 June 2011 Memorization and Detection of an Arrested Crack in Foam-Core Sandwich Structures Using Embedded Metal Wires and Fiber-Optic Sensors Shu Minakuchi1, Nobuo Takeda1, and Yasuo Hirose2 1 Department of Advanced Energy, Graduate School of Frontier Sciences, The University of Tokyo, 5-1-5 Kashiwanoha, Kashiwa-shi, Chiba, 277-8561, Japan 2 Commercial Aircraft Project Engineering Division, Aerospace Company, Kawasaki Heavy Industries, Ltd., 1 Kawasaki-cho, Kakamigahara-shi, Gifu, 504-8710, Japan

Abstract. A crack arrester has been recently developed to suppress crack propagation along the interface between the facesheet and the core of foam core sandwich structures. The crack arrester is a semi-cylindrical stiff material inserted into the interface. The crack arrester decreases an energy release rate at the crack tip by suppressing local deformation around the crack. If the arrested crack can be instantaneously detected, damage tolerance of foam core sandwich structures is dramatically improved. This study establishes an innovative crack detection technique using metal wires and fiber Bragg grating (FBG) sensors embedded at both edges of the arrester. Specific strain distribution induced by arresting the interface crack is first memorized by the metal wire and the consequent residual strain is then picked up by the FBG sensor as a damage signal. This study began by simulating sensor response to evaluate the feasibility of the proposed technique. A verification test was then conducted, confirming the spectral change of the FBG can indicate propagation direction and tip location of the arrested crack.

1 Introduction Carbon fiber reinforced plastic (CFRP) is used for almost all modern commercial aircraft as a primary structural material. However, the potential capability of CFRP cannot be maximized under the conventional structural design concept, which consists of skins, stringers, and frames. One innovative structural concept is a foam-core sandwich structure [1-3]. The integral construction consists of two thin facesheets and a lightweight foam core, which can considerably reduce the weight and the number of parts compared to conventional structures. However, crack propagation along the interface between the facesheet and the core is a critical issue [4-6]. An interface crack originates from manufacturing defects, impact damage, or fatigue shear cracks in the foam core. A crack below the facesheet is difficult to detect using conventional non-destructive inspection techniques. However, the interface crack seriously degrades structural integrity. Thus, Hirose et al.

508

S. Minakuchi, N. Takeda, and Y. Hirose

developed a crack arrester (Fig. 1) [7-9], which is a semi-cylindrical stiff material inserted into the interface. When a crack approaches the arrester, the arrester decreases the energy release rate at the crack tip by suppressing local deformation around the crack. Figure 2 illustrates the distribution of the normal stress in the vertical direction, calculated in finite-element analysis. As the crack approaches the crack arrester, the stress at the crack-side edge of the arrester gradually increases; consequently, stress concentration at the crack tip is reduced. In practical applications, the arrester is arranged in a grid pattern, and the interface crack is trapped inside the grid (Fig. 1). For practical use, however, the arrested crack should be instantaneously detected, and appropriate countermeasures must be taken against the damaged area. Arrested but undetected cracks induced from different damage in neighboring grids or, more critically, adjoining grids (A in Fig. 1) significantly degrade mechanical properties and increase the stress around these grids. As a result, an area without a crack may fail and/or undetected cracks may penetrate the arresters, leading to catastrophic failure of the entire structure. In this context, the authors developed a crack detection technique using fiber Bragg grating (FBG) sensors embedded in a crack arrester to locate grids with an arrested crack immediately after its occurrence [10]. The change in strain distribution in the crack arrester induced by arresting the crack is evaluated using reflection spectra from the FBG sensors. However, since the developed technique utilizes “elastic” strain change during the high-speed crack propagation, the established system cannot detect the crack when the crack is closed after unloading, and thus the system requires real-time, high-speed measurement of FBG spectral shape using an on-board system. In a practical large-scale aircraft structure, hundreds of FBG sensors are deployed in a wide structural area, and enormous volume of data obtained from those sensors needs to be analyzed and/or stored in a high cost processing system. Furthermore, during a flight, the airframe structure is loaded both statically and dynamically, and the FBG spectra change continuously. Hence spectral deformation induced by arresting a crack could easily be obscured, overlooking a fatal interface crack in the worst-case scenario.

Fig. 1 Crack arrester.

Memorization and Detection of an Arrested Crack

509

Fig. 2 Distribution of normal stress in vertical direction depending on crack tip position.

This study establishes a more easily-applicable and reliable crack detection technique by advancing the previously developed method. Metal wires are additionally embedded at both edges of the arrester, and an arrested crack is statically detected on the ground after a flight. This study begins by proposing the crack detection technique, and then conducts numerical analysis to predict the sensor response due to crack propagation and to validate the feasibility of the proposed technique. Finally, the technique is verified by an experiment.

2 Crack Detection Technique Figure 3 schematically illustrates the crack detection technique. Two metal wires and two optical fibers with FBG sensors are aligned parallel to the arrester grid line and embedded at both side edges of the crack arrester. When a crack approaches the crack arrester and the arrester suppresses crack propagation, shear deformation of the crack-side edge of the arrester is introduced [10], inducing plastic deformation of the embedded metal wire. Consequently, the high-speed crack propagation is memorized in the crack arrester, and, even when the crack is closed after unloading, residual strain remains at the crack-side edge of the arrester. This strain change is then statically evaluated using the embedded FBG sensor. The FBG sensor has a periodic variation in the refractive index along the length of a single mode optical fiber [11]. When broadband light is launched into the FBG sensor, a narrow spectral component is reflected back, and the reflection spectrum gives the measure of strain and/or temperature. When a nonaxisymmetric strain state arises at the core of the FBG sensor (Fig. 3), the reflection spectrum from the sensor splits into two peaks due to a birefringence effect [12, 13]. The difference between the central wavelength Δλ of the two peaks, λp and λq (λp > λq), is calculated using the following equation:

Δλ = λ p − λ q =

n02λ 0 ( p12 − p11 )(ε1 − ε 2 ) 2

(1)

where λ0 is the center wavelength of the initial reflection spectrum, n0 is the initial refractive index of the optical fiber core, p11 and p12 are the photoelastic constants, and ε1 and ε2 are the maximum and minimum principal strains at the core in the cross-sectional direction of the FBG sensor (Fig. 3). This equation indicates that the difference between the center wavelengths of the two peaks is proportional to

510

S. Minakuchi, N. Takeda, and Y. Hirose

the difference between the maximum and minimum principal strains at the core. Since the plastic shear deformation and, thus, spectrum change are induced only at the crack-side edge of the arrester, the crack propagation direction can be determined from reflection spectra obtained from FBG sensors. This technique requires no real-time, high-speed measurement with high cost processing systems, and the crack can be reliably detected on the ground after a flight.

Fig. 3 Schematic of crack detection technique.

In this study, Mode I type crack was evaluated using a beam-type specimen. First, finite element analysis (FEA) was conducted on a double-cantilever beam (DCB) to simulate plastic deformation of the metal wire and to calculate consequent residual strain change in the FBG sensor after unloading. The reflection spectrum is simulated based on the calculated values of principal strain ε1 and ε2.

3 Finite-Element Analysis Finite-element model Figure 4 presents a 2-D finite-element model with interface cracks modeled in ABAQUS 6.9. Plane strain was assumed, and thermal deformation was considered (curing temperature 130oC). The DCB specimen consisted of CFRP facesheets (UT500/#135, Toho Tenax Co., Ltd., [0/90]2S, thickness 1.44mm), a foam core (PMI Rohacell WF-110, Evonik Rohm GmbH, thickness 35mm), and a semicylindrical crack arrester (HC 9872 SynCore, Hysol Aerospace Products., radius 10mm). In Ref. 7, a microscopic observation of foam-core sandwich specimens revealed that the resin from the facesheet impregnated into cells of the foam core

Memorization and Detection of an Arrested Crack

511

adjacent to the facesheet, and a 0.34mm-thick resin layer was formed. An interfacial crack propagated between this resin-impregnated layer and the original foam core. In this study, finite-element models were developed based on this observation. The distance between the crack tip and the crack arrester, L, was set as 20, 10, or 0mm to investigate the influence of crack propagation. The load applied to each specimen was determined based on fracture loads obtained in preliminary tests: 2.0N/mm was applied. Two lead wires and two polyimide-coated FBG sensors were embedded at both side edges of the crack arrester along the specimen width direction, to be parallel to the arrester grid line. The lead wire with 1mm diameter was selected for the plastic material, since lead is relatively soft, easily yields, and thus can sensitively memorize crack. The optical fiber was embedded between the lead wire and the foam core at the upper left (or right) corner of the crack arrester (Fig. 4). This position is most suitable to sensitively detect plastic shear deformation of the lead wire. The cladding diameter of the FBG sensor was 125μm, and its outside diameter was 150μm. The cladding and the core of the FBG sensor were modeled as an isotropic material with the property of fused silica glass, and the strain at the center of the optical fiber was defined as the strain at the core. The residual strain state after unloading obtained from the center of the optical fiber was used to calculate the maximum and minimum principal strains at the core (i.e., ε1 and ε2 in Eqn. 1), and these principal strain values were then utilized to simulate the spectral response of the FBG sensor, as will be presented in the next section. “Sensor A” is the one embedded at the crack side, and “Sensor B” is the one at the opposite side.

Fig. 4 Finite-element model.

512

S. Minakuchi, N. Takeda, and Y. Hirose

Results Figure 5 presents magnified deformed shape of the arrester edge with calculated equivalent plastic strain distribution within the lead wire (L: 5mm, load value: 2.0N/mm, displacement magnification factor: 20). The shear deformation of the arrester edge, which was induced by arresting the crack, produced about 1 percent equivalent plastic strain in the lead wire. Figure 6 illustrates deformation of the FBG sensor demonstrated in the FEA. Before crack propagation (Fig. 6 (a)), thermal residual strain in the shear direction was induced due to the difference in coefficient of thermal expansion (CTE) of each component, and thus the sensor core was in non-axisymmetric strain state. When the crack approached the arrester (Fig. 6 (b)), shear deformation of the arrester edge was induced, and the lead wire was plastically deformed (Fig. 5). Finally, after unloading (Fig. 6 (c)), the plastic shear deformation of the lead wire counteracted the non-axisymmetric thermal residual strain. Figure 7 plots the obtained principal strains ε1 and ε2 in the FBG sensors after unloading. As L became smaller and the crack approached the arrester, the difference between the maximum and minimum principal strains decreased in Sensor A at the crack side. In the sensor at the opposite side of the crack, however, both the maximum and minimum principal strains were almost constant. These results suggest that only the arrester edge at the crack side contributes to suppressing crack propagation.

Fig. 5 Obtained deformation shape of arrester edge with equivalent plastic strain distribution within lead wire during loading.

ï

(a) Before loading.

(b) During loading.

(c) After unloading.

Fig. 6 Schematics of sensor deformation demonstrated in FEA.

Memorization and Detection of an Arrested Crack

513

Next, the reflection spectra were simulated. First, the difference between the two peak wavelengths Δλ was calculated from Eqn. (1) with n0 = 1.449, λ0 = 1550 [nm], p12 = 0.252, and p11 = 0.113. The reflection spectrum disturbed by the birefringence effect was then obtained by simply superpositioning the two power spectra Δλ away from each other [13, 14]. The initial power spectra of the FBG sensors used in the verification tests (next section) were utilized for the calculation. Figure 8 presents the simulated spectra. The intensity of each spectrum is normalized by the intensity of the highest component in the initial spectrum before embedding. In addition, to clearly compare the spectrum shapes, the wavelength is expressed by the detuning from the peak wavelength of each spectrum. In the reflection spectrum from the FBG sensor at the crack side (Sensor A), two peaks, which was induced by the non-axisymmetric thermal residual strain, gradually unites into one, as the crack approaches the arrester and thus the difference between the principal strains decreases (Fig. 7). However, the reflection spectra from

Fig. 7 Principal strains in sensor core after unloading depending on L.

(a) Sensor A.

(b) Sensor B.

Fig. 8 Simulated reflection spectra.

514

S. Minakuchi, N. Takeda, and Y. Hirose

the FBG sensor opposite the crack (Sensor B) hardly changes. Hence, it is expected that crack propagation direction and crack tip position can be easily determined from the spectra of the FBG sensors embedded at both edges of the arrester. The next session discusses a verification test using the DCB specimen.

4 Verification Test Materials and methods The experiment setup is depicted in Fig. 9. The configuration of the specimen was the same as the one used in the FEA. The specimen was 5cm wide. The facesheets and the arrester were co-cured with the foam core in an autoclave. Two lead wires (1mm diameter) and two FBG sensors (grating length 15mm) were embedded in the arrester. The initial crack was introduced by inserting a piece of 0.01mm-thick polyimide film. The length of the initial crack was 80mm, and the distance between the crack tip and the crack arrester, L, was 20mm. The specimen was loaded at a constant crosshead speed of 2.0mm/min using a material testing system (AG-50kNI, Shimazu Co.). Once the crack propagated, the specimen was unloaded, and the reflection spectra from the two FBG sensors were recorded. The test was then resumed, and this procedure was repeated until the crack reached the arrester edge. The optical fiber was illuminated by an amplified spontaneous emission (ASE) light source (AQ4310(155), Ando Electric Co., Ltd.), and the reflection spectra from the FBGs were measured by an optical spectrum analyzer (AQ6317, Ando Electric Co., Ltd).

Fig. 9 Experimental setup.

Memorization and Detection of an Arrested Crack

515

Results and discussions Figure 10 presents reflection spectra obtained after unloading. The intensity of the spectra is normalized by the maximum intensity of the initial spectrum before embedding. For a clear comparison, the wavelength is expressed by detuning from the peak wavelength of each spectrum. Before the test, the spectrum from the FBG sensor at the crack side (Sensor A) had two clear peaks, due to the thermal residual strain. However, several peaks were observed at the longer wavelength side of the spectrum from the sensor opposite the crack (Sensor B), probably due to misalignment of the FBG sensor. Since the optical fiber was severely deflected near the side edge of the beam-type specimen (Fig. 9), the position of the optical fiber partially deviated from the desired position (i.e., the edge of the crack arrester). Consequently, non-uniform strain distribution was induced along the entire length of the FBG sensor, introducing several peaks in the reflection spectrum [15]. It is important to note that misalignment of the sensor is a problem that is unique to the beam-type specimen and thus will not occur in a practical plate-type structure. As the crack approached the arrester, two peaks in the spectrum from Sensor A gradually came closer and, finally, united into one. However, the spectrum from Sensor B changed less. These changes in the two spectra are consistent with the result of the FEA, confirming that crack propagation direction can be determined from the spectrum change, and the crack tip location can be estimated from the distance between the two peaks. To summarize this study, easily-applicable and reliable crack detection technique was developed by advancing the previously developed method. The developed technique enables effective application of the crack arrester and thereby results in significant improvement of the damage tolerance of foam-cored sandwich structures. This study utilized lead wire to memorize crack propagation. For practical application, however, more environmentally-friendly alternative material needs to be utilized, and thus future work will address the material selection issue. Furthermore, the technique will be extended to more practically relevant 3-D plate structures by utilizing fiber-optic-based distributed strain measurement systems [16]. The validity of the technique under more complicated 3-D strain state should be verified through comprehensive tests using sandwich panel specimens with an arrester grid (Fig. 1). As foreseen by the authors, this technique is also valid with plate structures, since the technique is based on the particular strain state induced when the crack arrester suppresses interfacial crack, and thus is independent of global deformation of the structures.

516

S. Minakuchi, N. Takeda, and Y. Hirose

(a) L = 18mm.

(b) L = 6mm.

(C) L = 3mm. Fig. 10 Reflection spectra measured after unloading.

5 Conclusions An advanced technique for detecting arrested cracks in foam-core sandwich structures was developed. The technique utilized plastic materials to memorize crack

Memorization and Detection of an Arrested Crack

517

propagation and fiber-optic sensors to evaluate the consequent residual strain change. First, FEA was conducted to simulate plastic deformation of the plastic material and then predict reflection spectra from the FBG sensors, validating the feasibility of the proposed technique. Verification tests clearly demonstrated that the spectral change of the FBG can indicate the propagation direction and the tip location of the arrested crack. The proposed technique enables effective application of the crack arrester and thereby enables significant improvement of the damage tolerance of foam-core sandwich structures.

References [1] Zenkert, D. (ed.): The Handbook of Sandwich Construction. EMAS Publishing, Warrington (1997) [2] Hirose, Y., Kosugi, K., Nishitani, M., Sashikuma, H., Imuta, M., Fukagawa, H., Kikukawa, H.: In: Proceedings 23rd International Congress of Aeronautical Sciences, pp. 343.1–343.10 (2002) [3] Herrmann, A.S., Zahlen, P.C., Zuardy, I.: In: Proceedings of the 7th International Conference on Sandwich Structures (ICSS-7), pp. 13–26 (2005) [4] Burman, M., Zenkert, D.: International Journal of Fatigue 19(7), 551 (1997) [5] Shipsha, A., Hallstrom, S., Zenkert, D.: Journal of Sandwich Structures & Materials 5(1), 7 (2003) [6] Jakobsen, J., Bozhevolnaya, E., Thomsen, O.: Composites Science and Technology 67(15-16), 3378 (2007) [7] Hirose, Y., Hojo, M., Fujiyoshi, A., Matsubara, G.: Advanced Composite Materials 16(1), 11 (2007) [8] Matsuda, H., Matsubara, G., Hirose, Y., Hojo, M.: In: Proceedings of 16th International Conference on Composite Materials, FrKA1-02 (2007) [9] Hirose, Y., Matsuda, H., Matsubara, G., Inamura, F., Hojo, M.: Journal of Sandwich Structures & Materials 11(6), 451 (2009) [10] Minakuchi, S., Yamauchi, I., Takeda, N., Hirose, Y.: In: Proceedings of the 25th Symposium of the International Committee on Aeronautical Fatigue, p. 209 (2009) [11] Othonos, A., Kalli, K.: Fiber Bragg Gratings: Fundamentals and Applications in Telecommunications and Sensing. Artech House Publishers, Norwood (1999) [12] Gafsi, R., El-Sherif, M.: Optical Fiber Technology 6(3), 299 (2000) [13] Zhang, A.P., Guan, B.O., Tao, X.M., Tam, H.Y.: Optics Communications 206(1-3), 81 (2002) [14] Okabe, Y., Shigeki, Y., Tsuji, R., Mizutani, T., Takeda, N.: Composites Part A: Applied Science and Manufacturing 33(7), 991 (2002) [15] Peters, K., Pattis, P., Botsis, J., Giaccari, P.: Optics and Lasers in Engineering 33(2), 107 (2000) [16] Kishida, K., Li, C.: In: Ou, J., Li, H., Daun, Z. (eds.) Structural Health Monitoring and Intelligent Infrastructure, pp. 471–477 (2006)

26th ICAF Symposium – Montreal, 1-3 June 2011 Structural Load Monitoring Systems for Military Aircraft in the Polish Armed Forces with Examples of Selected Activities M. Kurdelski, A. Leski, S. Klimaszewski, and M. Stefaniuk Air Force Institute of Technology Księcia Bolesława 6 Str., Warsaw, Poland

Abstract. Airplanes and helicopters currently operated by the Polish Armed Forces have been introduced into service without any load monitoring systems. The only exception is the F-16 and its Aircraft Structural Integrity Program. In spite of the fact that the majority of the Polish military airplanes and helicopters are operated according to the safe life principle, a structural load monitoring programs have been developed for a sizeable population thereof. These programs have been originated and implemented by the Polish Air Force Institute of Technology. Most often they accompany the process of equipping the airplanes and helicopters with digital flight data recorders. The majority of currently operated systems are based on the collection and analyses of the nz-signal data. Depending on the type of an airplane or helicopter, the data may be used in different ways. In some cases, reports on aircraft affecting loads with respective analyses are periodically delivered to aircraft operators. In some other instances such reports are created on demand, or when some current, aircraft-structure affecting problem has to be solved.

1 Introduction Polish Air Force (PAF) is an organization which aims to carry out its assignments professionally. Because the fleet numbers are diminishing, the aircraft are used more intensively and stay in service longer than it was assumed in design and the initial stages of service. Most of the aircraft currently in use were manufactured more than twenty years ago by the Soviet Union or polish manufacturers. These aircraft were designed according to the safe-life philosophy, in which a certain safe operation period is specified for the aircraft and its subsystems. This safe life period is commonly specified in flight hours or service years, whereas structural elements such as aircraft landing gears may also have the service life defined in number of take off – landing cycles. The safe-life approach implies the assumption, that in the specified service period - provided the allowable design loads are not exceeded - structural fatigue symptoms will not occur. Basic aircraft types in Polish Air Force’s service are: F-16 (multirole fighter), MiG-29 (fighter), Su-22 (fighter-bomber), PZL TS-11 (trainer jet), PZL-130 (turboprop trainer), EADS CASA 295M (turboprop transport), the helicopters:

520

M. Kurdelski et al.

Mi-8/17, Mi-14, Mi-24, PZL W-3. Among the mentioned aircraft, only the most modern F-16 are equipped with manufacturer-provided maintenance data monitoring and recording systems. The other airplanes and helicopters, even if were fitted with flight data recorder systems by the manufacturer, were done so only with the purpose of enabling the continuous assessment of pilot performance or as an aid in investigating aircraft accidents. The progress in computer technology, as well as in the field of digital data storage, enabled the acquisition and long-term storage of sizeable flight parameter records. This led to a shift in the way the on-board flight data recorders have been used. No longer were they only the popular “black boxes” witnessing the aircraft crashes. The Flight data recorders became a means of assessing the pilot’s performance, as well as an aid to the training programs. The onboard recorders were also presented with a new task - the monitoring of service loads, a step necessary in determining the fatigue life consumption of the airframe. In Poland, the institution responsible for research and development duties in the field of military aviation Is the Air Force Institute of Technology which, in its current form, exists since 1953. AFIT is a scientific and research organisation, which is supervised by the Ministry of National Defence. Its mission is scientific support and research into problems of operating of products of aeronautical engineering. Owing to the studies in the field of reliability and broadly understood flight safety, the Institute has significantly contributed to the development of Polish aviation. The Institute's output comprises hundreds of elaborations - effects of research and experimental works, design efforts, and technical/servicing activities - applied in the Air Force of the Armed Forces of the Republic of Poland. Several years of experience of the institute’s personnel contributed to the creation of flight data storage systems as well as to the implementation of modern, digital flight data recording and flight record processing systems. Several flight data recorders designed and manufactured by the AFIT are in use onboard the PAF aircraft. These are: -

-

IP-8 maintenance data recorder, onboard the Su-22; IP-16 maintenance data recorder, onboard PZL M-28; S2-3a maintenance and crash data recorder, onboard PZL TS-11, PZL M-28, PZL-130 TC-II and the PZL W-3, PZL SW-4, Mi-8/17, Mi-14 and Mi-24 helicopters S2-3 and S3-1a maintenance and crash data recorder, onboard PZL-130 TC-I.

Additional systems required for flight data analysis are the data processing systems. Those enable: data download, measurement error correction, removal of invalid records, as well as the creation and management of flight record databases. AFIT, owing to its’ staff experience as well as to its IT resources and databases accumulated throughout the years, strives to impact the manner in which the aircraft are operated and maintained by the air force, and also to shape the training and instruction of operator’s aircrews and technical personnel. Continuing improvement of qualifications significantly improves the quality of maintenance work, as well as it limits its adverse impact on the durability and fatigue of the fleet’s aircraft. Limiting of the factors which adversely influence the fatigue life is particularly important when the operated aircraft approach the limit of their designed period of operation - even more so, if they have exceeded it.

Structural Load Monitoring Systems for Military Aircraft in the PAF

521

Fig. 1 S2-3a flight data recording system. A – acquisition unit S3-1a-2, B – protected unit S2-3a-K, C – pilot index feeding unit S2-3a-P, D – recorded data reader S3-1c-0, E – computer tester WTS-4, F – tester WTS-4/AP702C

2 Selected Structural Load Monitoring Activities Structural load monitoring of the MiG-29 fleet The first aircraft to be subjected to structural load monitoring in the polish air force were the MiG-29s acquired from the German Federal Republic’s military. 22 MiG-29 airplanes which had previously been used by the East German air-force, and, after the German unification, by the united Federal Republic’s Air force, were delivered to Poland in 2005. Usage severity factor for these aircraft was calculated by the EABG company during their service in Germany. EABG conducted analyses using the LEDA life management system, unified for the German Air Force. Along with the acquisition of the aircraft, Polish side also received the data regarding their operational usage, including digital parameter records of flights conducted. In the same period, the AFIT have been conducting extensive research programs aimed at implementing usage severity monitoring for the Su-22 and MiG-29 fighter jets. The purpose of the research was to devise new rules that would regulate the aircraft’s maintenance and operational usage, and particularly to implement the “technical condition based maintenance system”. Because necessary laws and standards were absent in Poland, the polish system was based on the

522

M. Kurdelski et al.

North American ASIP program – which is run in accordance to the MIL-STD1530 [5] standard and British Defense Standard 00-970 [6]. In the field of service load monitoring, a program comparable to the German LEDA system was introduced. This system encompassed all of the MiG-29 aircraft that remain in polish service. In the system, the flight data is analyzed in 3 month intervals [3]. Main result of these data analyses is the actual usage severity factor – “k”, computed for each fighter. The k factor value is calculated based on the recorded nz (vertical load factor) signal data. Simultaneously, the equivalent flight hours Neq are calculated - those may be considered to be an objective measure of airframe fatigue life consumption. Every quarter-year, reports containing the analyses of actual k factor are submitted to the operator, along with the Equivalent Flight Hour report. Based on the analyses performed, it has been determined that significant differences were present between the nature of operational usage in the Polish and German air forces. The most prominent differences were between the range and intensity of the vertical load factor nz and, in consequence, between the computed usage severity factor - k. The differences between usage severity observed in the Polish and German air forces were so profound, that to this day there is a significant gap between the levels of fatigue life consumed in aircraft received from the GAF and the aircraft that served in the PAF from the beginning. Polish activities in the F-16 ASIP The most advanced condition based maintenance program in the Polish air force is the one implemented on the F-16 C/D airplanes [4]. With the introduction of the F-16 into service the ASIP [5] regulations were also implemented. ASIP Program tasks IV and V are concerned with continuous maintenance. Task IV is carried out by the aircraft manufacturer, Lockheed Martin, whereas Task V is realized by the Polish Air Force - the operator - in cooperation with AFIT. Condition monitoring activities amount to: -

storing operational usage data, performing L/ESS analyses, performing IAT analyses, individual airplane scheduled maintenance actions, structural maintenance records, weight and balance records.

The AFIT serves the role of the Data Processing Center (DPC), which duties are: -

maintaining flight data records, maintaining of databases required for the continuous aircraft condition monitoring, devising periodical IAT and L/ESS reports, working out Interim Operating Supplement, working System Maintenance Reports, other DPC duties.

Structural Load Monitoring Systems for Military Aircraft in the PAF

523

However, it should be noted that Poland only recently became an operator of the F-16 C/D Block 52+ variant aircraft. The planes were delivered as newly manufactured, so, in the first years of operation, the operator (Polish Air Force) as well as the AFIT will be accumulating the experience necessary for the further years of service. Structural load monitoring for Mi-8/17, Mi-14 and Mi-24 helicopters Mi-8/17 family of helicopters, and the related Mi-14s constitute the basic equipment of the ground forces (Army) and the Navy in Poland. These are heavy transport and attack/assault helicopters. The approximate numbers used in the Polish Armed Forces are as follows: 41 Mi-8/17s, 32 Mi-24s and 12 Mi-14s. The age of most of those helicopters exceeds 25 years. However, because of the cost of introducing new aircraft, as well as because of the relatively low number of flight hours accumulated by the helicopters in service, research and testing are performed by the AFIT with the aim of extending the helicopters’ operational service. Examples of such research are the Helicopter Structural Integrity Programs of Polish Mi-24 Hind and Mi-14 Haze helicopters [7, 8]. A crucial element of these programs has been the fitting of AFIT’s S2-3a flight digital data recording system onboard the aircraft. In case of the helicopters, the continuous service load reporting is not performed. The recorded flight data are only stored and the load analyses are performed at the operator’s request, during health condition checks and testing activities. As a result of such analyses, cases of pilot conduct causing severe loading on the airframe were revealed. Subsequent geometry (structural control point) measurements demonstrated that a change occurred in structural dimensions specified in the technical conditions. The source of such a situation might have been the fact that some aircrews performed flight maneuvers that caused exceedances of allowable load limits. In such a situation, the introduction of appropriate operational usage recommendations and regulations along with the control of their implementation might bring a significant improvement of the structural durability. PZL-130 load spectrum development PZL-130 Orlik is a turboprop military trainer aircraft with a metal semimonococque airframe. It was designed in the early 1980’s in the Warsaw-based PZL-Okecie company. It has been utilized by the Polish Air Force since 1994 for basic pilot training. Currently, a shift in the aircraft’s maintenance system is taking place. The costly overhauls performed each 1000 FH are to be replaced with airframe inspections in line with the “condition based maintenance” program. An important point in the program’s implementation is the Full Scale Fatigue Test. Flight data from the 15-year operational history of PZL-130 Orlik was employed in formulating the load spectrum for the Test. Each of the aircraft in service in the Polish Air Force has been equipped with a digital flight data recorder. The main purpose of installing the recorders was to support the pilot training assessment

524

M. Kurdelski et al.

and the debriefing process. The recorded data, in the form of files (with each file corresponding to one flight) was saved and stored, and therefore could be used in devising the operational profile. First problem that needed solving during the analysis of recorded data was to distinguish the take-off and landing phases in the nz signal. The flight data recorder didn’t register directly any signal that specifically defined whether the aircraft was on the ground or in the air. The issue was solved with a software analysis of other parameter records such as velocity, altitude, rate of engine revolutions and other. Because of that a separation of on-ground and in-air states was made possible. Main parameter taken into account in the analysis was the nz (vertical load factor). A statistic histogram of nz cycles has been worked out. Because the PZL-130 Orlik airplanes are used for training, most of the flights are a realization of one of the training missions specified in pilot training program. The index number of a mission is entered in the record data file during the pilot debriefing. This index constitutes another information that aids identification of the operational profile. During analysis it turned out that a portion (around 30%) of the recorded flights had the wrong index number attributed. Majority of those flights were the flights conducted outside the training program, which included the aerobatic team’s activities (those contained a very high proportion of high-intensity maneuvers) . A representation of the flight profile devised from the recorded flight numbers is shown in Table 1. Table 1 Percentage distribution of flight categories. Basic/simple pilotage (basic maneuvers) 21%

Medium pilotage (medium-difficulty maneuvers) 13%

Circuits

Spins

Routes

Others

8%

4%

23%

31%

The average number of landings per flight hour has also been determined eqn. 1:

nl = 2.31

landings fth

(1)

More take-off-landings than stop landings were recorded. The spins and stalls impose significant adverse loading on the airframe. These maneuvers are not easily distinguished in the flight data records through the analysis of the nz signal, therefore to determine their frequency in the flight profile the missions’ description along with the mission statistics were employed. On this basis, the average number of spins per flight hour was established to be 0.16.

Structural Load Monitoring Systems for Military Aircraft in the PAF

525

3 Selected Structural Loads Monitoring Activities – Direct Measurements AFIT’s methodology for direct load measurement Direct load measurement programs supplement the service load monitoring systems. Such measurements are performed by the AFIT. Most often there are a part of various research and testing programs needed for airframe modernization, establishment of upgrade projects or for maintenance system modifications. AFIT is the only scientific/research center in Poland permitted to perform research and testing on polish military aircraft. The research encompasses: extensive flight tests, military technology acceptance tests, special research and studies, simulation, as well as the tests of AFIT technology and others. For the purpose of service load monitoring the following aircraft types have been subjected to tests by the AFIT in the recent years: -

PZL 130 Orlik TC-I Mi-14 HAZE Su-22M4 FITTER MiG-29 FULCRUM Mi-24 HIND PZL 130 Orlik TC-II

(2001) (2002) (2004) (2006, 2007) (2009) (2010)

Airplanes and helicopters intended for the flight tests, are subjected to the preparatory process during which a strain gauge sensor instrumentation is fitted on the airframe. Fitting of the sensor suite is most often carried out during the aircraft’s stay at the overhaul and repair facilities. This provides a better access to the airframe’s structural elements. The sensor system design, measurement methodology and the flight test plan are devised by the AFIT. The instrumentation fitting is carried out jointly by the repair facility staff and the AFIT personnel. The basic configuration of the measurement system is composed of the strain gauge suite and the test data recorder (ACRA KAM-500). When specific needs arise additional measurement equipment is installed, e.g laser distance/displacement meters, landing gear deflection sensors etc. [14, 15] After the instrumentation process is complete a preliminary test flight is carried out. Subsequently, physical and electrical strain gauge calibration is performed by applying known loads on various components: wings, empennage and the de-installable components such as pull-rods, fittings etc. Flight test program consists of a number of test flights. During the tests one of the measurement channels is reserved to transmit one-time commands (flags) that flag the beginning and the end of a flight maneuver. After a testing cycle has been concluded the instrumentation is de-installed and the test aircraft returns to service.

526

M. Kurdelski et al.

Flight test program for PZL-130 TC-II Orlik The implementation of the “Service Life Assessment Program” (Polish: SEWST – System Eksploatacji Według Stanu Technicznego) is the most sizeable Polish research project in the field of direct monitoring of aircraft service loads in the recent years. One of the steps of the SEWST program was the formulation of the load sequence for the full scale fatigue test. The load sequence was worked out based on the operational usage monitoring data records, which were supplemented with the results data of the direct load measurement phase. [15, 16] The test aircraft was equipped w the KAM-500 recorder, along with the strain gauge instrumentation. Sensor fitting was carried out by the AFIT personnel in cooperation with the manufacturer. The strain gauge system consisted of over 100 measurement sensors in 86 measurement points and 13 measurement sections. The key measurement channels were secured by installing additional redundant sensors in adjacent locations. A multi-point measurement of complex loading states was facilitated. The aircraft, after having been equipped with this instrumentation was transferred to flight testing. The testing included all of the flight stages present in the standard training mission programs. After a series of flight tests, portion of the equipment has been uninstalled from the airframe, along with the selected strain gauge sensors. In this configuration the aircraft has been delivered to the military unit to perform its routine flight duties. At that stage load signals were also recorded. The purpose of collecting that additional load signal data was to supplement the records gathered throughout the flight testing. In this stage a sample of the typical training load histories is recorded.

4 Summary and Conclusion Service load monitoring programs of military aircraft have come a long way in Poland. The team carrying out those programs had started with a limited experience on the subject. The research personnel enhanced its experience and expertise through its own failures and by accumulating the know-how of other researchers as well as by consulting the subject literature. Own methodologies have been developed. Experiences amassed through the various programs confirm that load monitoring is a crucial element in the maintenance and operation of modern aircraft equally those operated on the principles of safe-life as those designed and operated according to the damage tolerance approach. Basing on the expertise collected, the AFIT implements direct service load monitoring systems that employ airframe-fitted strain gauge instrumentation. Equipping further aircraft with strain gauge instrumentation is planned, to eventually encompass the majority of each type’s population. In the coming years the activities undertaken by the AFIT will be concerned with implementing SHM for the aircraft. For this, installation of universal digital flight data recorders on board the aircraft in service will be essential. Equipping the aircraft with data recorders will greatly ease the implementation of SHM and lower its costs.

Structural Load Monitoring Systems for Military Aircraft in the PAF

527

References [1] Malinowski, L.: z zespołem. Algorytmy działania systemu analizy obciążeń eksploatacyjnych samolotów Su-22 na podstawie parametrów rejestrowanych przez rejestrator pokładowy TESTER. ITWL, Warszawa (2001) [2] Żurek, J.: z zespołem. Sprawozdanie nr 95/31/2005. Analiza obciążeń eksploatacyjnych samolotów MiG-29 i MiG-29UB na podstawie analizy wartości przeciążeń pionowych (współczynników obciążenia) nz. ITWL, Warszawa (2005) [3] Biuletyn eksploatacyjny nr P/O/R/U/4799/2005. ITWL, Warszawa (2005) [4] Biuletyn eksploatacyjny nr P/5062/E/2008. ITWL, Warszawa (2008) [5] MIL-STD-1530C Aircraft Structural Integrity Program [6] Defense Standard 00-970, Design and Airworthiness Requirements for Service Aircraft. Ministry of Defense (2) (December 1999) [7] Klimaszewski, S., Leski, S., Dragan, K., Kurdelski, M.: Structural Integrity Program of Polish Mi-24 Hind Helicopters. In: Proceedings of the 25th Symposium of the International Committee on Aeronautical Fatigue, Rotterdam, The Netherlands, May 27-29 (2009) [8] Klimaszewski, S., Leski, A., Żurek, J.: Helicopter Structural Integrity Program of Polish Navy Mi-14 Haze Helicopters. In: Proceedings of ASIP, Conference. Memphis, pp. 29.11–3.12 (2004) [9] Dragan, K.: NDE activities connected with Service Life Monitoring Extension of Main Rotor Blades of Helicopter used in Polish Armed Forces. In: Proceedings of 7th Australian Pacific Vertiflite Conference on Helicopter Technology, March 9-12 (2009) [10] Dragan, K., Klimaszewski, S.: Multimode NDE for structural integrity monitoring of helicopter main rotor blades. In: International Workshop on Structural Health Monitoring, Stanford University, Stanford, pp. 09.09.-11.09.2009 r [11] Dragan, K., Klimaszewski, S., Kudela, P., Malinowski, P., Wandowski, T.: Health Monitoring of the helicopter main rotor blades with the structure integrated sensors. In: EWSHM, Sorrento, Italy, pp. 29.06 – 02.07 (2010) [12] Leski, A., Obrycki, Ł.: i inni. Wyznaczanie profilu eksploatacji samolotów PZL-130 Orlik używanych w Siłach Powietrznych RP. ITWL, Warszawa (2010) [13] de Jonge, J.B.: The Crack Severity Index of Monitored Load Spectra. An Assessment of Fatigue Damage and Crack Growth Prediction Techniques. AGARD Report 797 (1993) [14] Jenkins, J.M., DeAngelis, V.M.: A summary of Numerous Strain-Gage Load Calibrations on Aircraft Wings and Tails in Technology Format. NASA Technical Memorandum 4804. Dryden Flight Research Center Edwards, California (1997) [15] Podskarbi, S.: i inni. Sprawozdanie 89/31/2010. Dokumentacja techniczna z zabudowy czujników tensometrycznych na płatowcu PZL-130 Orlik TC-II nr boczny 037. ITWL, Warszawa 2010. [16] Parker, R.: A full-scale fatigue test of a Pilatus PC9/A trainer aircraft, DSTO-TR1107 (April 2001)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

A320 ESG Full Scale Fatigue Test - Lessons Learned G. Hilfer1, N. Rößler1, C. Peters1, and C. Herrmann2 1

IABG mbh, Ottobrunn, Germany 2 Airbus Deutschland GmbH, Hamburg, Germany

Abstract. In 2007, IABG was contracted by Airbus with the set-up and performance of two of the major structure fatigue tests related to the Extension of Service Goal (ESG) of the Airbus Single Aisle family, namely the NEF2 (centre fuselage with both wings) and NEF3 (rear fuselage) tests. IABG managed the development of both test set-ups within the condensed timing for the test structure deliveries in January 2008. Taking advantage of IABG’s experience from previous full-scale test programs, the tests could be commissioned until mid of 2008 and afterwards were taken into full operation. Both tests had the objective to validate the structures and the maintenance program for the extended service life goal (of 90.000 operational flight cycles) – thus requiring the simulation of 180.000 flights. Both test programs have made very good progress, and have been completed in summer/autumn 2010. This paper summarizes the contributing facts of this high speed test program and presents some conclusions. Here, emphasis is put on the test speed and the loading accuracy. It is shown that dedicated technical and logistics developments of IABG to increase the test speed have been applied to this test.

1 Introduction Starting with its foundation in 1961, IABG today looks back on 50 years of fullscale fatigue testing. Each project realized since then had its own challenges and brought forth some very specific cognition. While the requirements on the technical performance of such tests have continuously scaled up analogously to the advancements in structures, there is an ever growing need also for improvements and optimizations regarding quality, documentation, organization and communication in order to fulfil today’s need for extensive testing while matching the related expenditures with budgetary and schedule constraints. The requirements for the A320 ESG test program were specific because test specimen taken out of the production had to undergo fatigue cycling over the already known range of service life before entering the life extension phase as the intrinsic purpose of the test. The rationale for this test as well as test requirements and facility characteristics can be found in [1]. *

Oral presentation.

530

G. Hilfer et al.

In view of the program schedule constraints, it was necessary to pass the test phase up to 120,000 simulated flights as quickly as possible without compromising the validity of fatigue results gained. The pressure on the program invoked the development of a number of measures which were implemented into the test performance. Some of these measures made use of the specific nature of this program, others will be of interest also for future full-scale fatigue test programs.

A320 NEF2

A320 NEF3

Fig. 1 Airbus ESG Tests at IABG.

2 Timeline of the A320 Nef2/ Nef3 The timeline requirements in the A320 ESG program of the preparation phase for the test set-up and fatigue testing phase posed a challenge on IABG by the short lead time. After the project start in March 2007 with the engineering, design and manufacturing phase the first major project milestone was reached as planned with the delivery of both test specimen beginning of January 2008 by Beluga transportation to Munich Airport. After the follow-on works for the final assembly of the test specimen the two test set-ups were completed and the commissioning phase was started as scheduled beginning in July 2008 for NEF3 and September 2008 for NEF2, respectively. The NEF3 fatigue test started in August 2008, followed by the NEF2 fatigue test in November 2008 both to reach the ESG1 milestone of 120,000 simulated flights in July 2009 as well. The ESG2 milestones of 180,000 simulated flights have been completed on August 2010 in the case of NEF3 and in October 2010 in the case of NEF2. The test progresses of NEF2 and NEF3 are given in Figure 2 and Figure 3, respectively. The utilization of long-term experience in major airframe fatigue tests for all Airbus types by IABG, but namely the experience gained during the A320 EF2 fatigue test 20 years ago and recent improvements in test technology and test management, have contributed to this extremely fast test performance.

03.08.09

06.07.09

08.06.09

11.05.09

13.04.09

16.03.09

16.02.09

19.01.09

22.12.08

Time

Fig. 3 Test Progress of NEF3.

30.08.10

02.08.10

05.07.10

07.06.10

10.05.10

12.04.10

15.03.10

15.02.10

18.01.10

21.12.09

23.11.09

26.10.09

D-Inspection

19th B-Inspection

11th C-Inspection 18th B-Inspection 12th C-Inspection

17th B-Inspection

10th C-Inspection

16th B-Inspection

15th B-Inspection 9th C-Inspection

8th C-Inspection

14th B-Inspection

7th C-Inspection

13th B-Inspection

6th C-Inspection

8th B-Inspection 9th B-Inspection 3rd C-Inspection 10th B-Inspection 4th C-Inspection 11th B-Inspection 5th C-Inspection 12th B-Inspection

7th B-Inspection 2nd C-Inspection

5th B-Inspection 1st C-Inspection 6th B-Inspection

4th B-Inspection

50.000

40.000

30.000

20.000

10.000

0 0

1st B-Inspection

26.10.09

28.09.09

31.08.09

03.08.09

06.07.09

08.06.09

11.05.09

13.04.09

16.03.09

25.10.10

27.09.10

30.08.10

02.08.10

05.07.10

07.06.10

10.05.10

12.04.10

15.03.10

15.02.10

18.01.10

21.12.09

23.11.09

D-Inspection

18th B-Inspection

11th C-Inspection

17th B-Inspection

10th C-Inspection

16th B-Inspection

9th C-Inspection

15th B-Inspection

8th C-Inspection

14th B-Inspection

7th C-Inspection

13th B-Inspection

6th C-Inspection

12th B-Inspection

11th B-Inspection 5th C-Inspection

10th B-Inspection 4th C-Inspection

3rd C-Inspection

9th B-Inspection

8th B-Inspection

7th B-Inspection 2nd C-Inspection

6th B-Inspection

5th B-Inspection 1st C-Inspection

4th B-Inspection

60.000

16.02.09

70.000 2nd B-Inspection 3rd B-Inspection

80.000

19.01.09

22.12.08

Time

28.09.09

31.08.09

90.000

24.11.08

100.000

27.10.08

110.000

1st B-Inspection

90.000

2nd B-Inspection 3rd B-Inspection

100.000

29.09.08

24.11.08

simulated Flights 110.000

01.09.08

04.08.08

simulated Flights

A320 ESG Full Scale Fatigue Test - Lessons Learned 531

200.000

190.000 1

180.000

170.000

160.000

150.000

140.000

130.000

120.000

Fig. 2 Test Progress of NEF2.

200.000

190.000 1

180.000

170.000

160.000

150.000

140.000

130.000

120.000

80.000

70.000

60.000

50.000

40.000

30.000

20.000

10.000

0

0

532

G. Hilfer et al.

3 Factors Contributing to the Good Test Progress Project Management Due to the challenging time constraints of the program, it was very important to establish an efficient and synergic project management enabling a smooth test organization of the two major airframe tests. One of the challenges in this respect was that the tests were operated in parallel using to the maximum extent the same test infrastructure and test operation staff resources. In the first instance the personnel planning, the shift operation planning, the stand-by team and the internal coordination using status reviews of system operation and facility resources had to be optimized for parallel operation of the two tests. Daily progress reports to Airbus were installed to keep Airbus constantly informed about work progress. Continuous updates of the time schedule were effected to track and maintain the major milestone dates in the ESG program and to effectively organize trouble shooting events. An Airbus representative located at IABG site was continuously present during the whole test time. This enabled quick decision making and provided flexibility for planning and execution of work. Periodical status reviews were held with the customer in order to track progress, achievements and actions and to anticipate major challenges of the forthcoming test leg. In addition, working parties from Airbus-UK and from RUAG Aerospace were contracted by Airbus to perform quick repairs and modifications. By using the IABG machine shop, minor adjustments of spare parts were performed in a short time. The access to the inspection areas, e.g. removal and reinstallation of manhole covers, brackets, windows, etc., and some repairs, e.g. cold-working of bore holes and installation of oversize bolts, placement of crack stop drilling holes, change and readjustment of parts, were performed by IABG aircraft mechanics to reduce travel activities of Airbus personnel and the related efforts. Thus downtimes were minimized. Normally, fixed inspection intervals are prescribed by the inspection program. Due to the available knowledge and hence anticipation of the fatigue behaviour of the airframe, a flexible inspection stop decision was implemented by Airbus to minimize the downtime. Therefore in some cases test operation was continued overnight although being close to an inspection interval in order to enable start of a major inspection campaign on the next working day (see also section 3.). The proven IABG proprietary damage documentation (“DamDoc V3”) process ([1], [2] & [3]) was adapted to fit the tight schedule of daily inspections and the short inspection downtimes. Due to close collaboration between the inspection team and the “DamDoc” team, the damage reports could be delivered to Airbus within short time. This enabled Airbus to respond quickly to findings and to avoid otherwise emerging downtime. Technical Engineering Using extensive know-how and experience from former Airbus tests and complementing it with the adoption of new technologies, the development time for the new technical concept for A320 ESG, the commissioning phase and the initial speed-up phase of the NEF2 and NEF3 tests could be reduced to a minimum.

A320 ESG Full Scale Fatigue Test - Lessons Learned

533

IABG used complex FE models of the NEF2 and NEF3 structure provided by Airbus to generate reduced structural models, i.e. stiffness and mass matrices. The reduced structural models were integrated into overall models representing the entire dynamics of the hydraulic loading system. These overall models comprise accurate nonlinear models of the hydraulic jacks as well as of the pneumatic pressurization system. Based on these sophisticated models and the loading program, IABG analysis specialists calculated the oil consumption of each jack as an important prerequisite for the design of the hydraulic system. Last but not least all throttle valve pre-adjustments of the jacks – the throttle valve is an essential and crucial device to unload the test article safely in case of a test shut down - are pre-calculated by this analysis. Based on a quasi-static inverse model of the plant a feed forward control of the valve flows was introduced for optimizing and speeding up the control system behaviour. Making use of controller parameters derived from previous tests, the duration of the dynamic commissioning phase was reduced significantly. After the commissioning phase the Static Test Campaign prior to 1st Flight was performed. IABG implemented a trigger function between the control and monitoring system and the data acquisition system, which allows fully automatic data acquisition of the pre-defined instrumentation channels. Approximately 100 load cases with several loading steps could be measured per day using this method. The same data acquisition technique was used for structural health monitoring campaigns and for continuous measurements for the verification of the test accuracy or for the monitoring of damages. IABG has developed a high-performance control and monitoring system based on commercial off-the-shelf hard- and software in order to meet future test requirements of the customer always calling for higher test speeds. Figure 4 shows the occurrences of load case durations for NEF2 and NEF3. An overall average test speed of approximately 1.1 to 1.4 seconds per load case was reached. This test speed value, however, includes the pressurisation and depressurisation load cases of the fuselage. Without the pressurization and depressurization cycles which affect the overall speed considerably, the average load cycle time is reduced to 1.0 and 0.87 seconds, respectively. These values give an impression of how fast the mechanical load cycles have been applied considering the size of the structures, the significant number of control channels involved and the large displacements particularly at the wing tips. A number of optimizations across all test systems in use were implemented to increase test speed. Fundamentally, a fine tuning of all control parameters was performed for each hydraulic actuator based on the settings of the commissioning phase. In close cooperation with Airbus, an optimization of the loading program was performed with respect to the capabilities of the control system. The differential pressure characteristics were matched with the control loop resources of the pneumatic system thus improving speed and overall performance. Finally, the synchronization of actuators was assessed leading to further improvements in some cases. Summarising the effect of all of the above mentioned technologies and processes, some of a more generic kind, others rather specific for this test project, an improvement in test speed of more than 30% with respect to the planning was achieved. At this stage, the analysis of the loading accuracy proved to be fully

534

G. Hilfer et al.

within the requirements. In Figure 5 the stochastic distribution of the force control error for the A320 NEF2 and NEF3 fatigue test is depicted being typical for large scale, multichannel fatigue tests.

60%

50%

Occurrence

40% NEF2: all LC's considered; avg. Speed = 1,12 s/LC NEF2: LC's during Differential Pressure Change not considered; avg. Speed = 1,00 s/LC NEF3: all LC's considered; avg. Speed = 1,42 s/LC NEF3: LC's during Differential Pressure Change not considered; avg. Speed = 0,87 s/LC

30%

20%

10%

> 4,1

3,9

3,7

3,5

3,3

3,1

2,9

2,7

2,5

2,3

2,1

1,9

1,7

1,5

1,3

1,1

0,9

0,7

0,5

0%

Time Durations per Load Case [s]

Fig. 4 Test Speed

16%

14%

12%

8%

6%

4%

2%

3,0%

2,8%

2,6%

2,4%

2,2%

2,0%

1,8%

1,6%

1,4%

Deviations of Jack Loads within contracted Tolerance

1,2%

1,0%

0,8%

0,6%

0,4%

0,2%

0%

0,0%

-0,2%

-0,4%

-0,6%

-0,8%

-1,0%

-1,2%

-1,4%

-1,6%

-1,8%

-2,0%

-2,2%

-2,4%

-2,6%

-2,8%

0% -3,0%

Occurrence

10%

Fig. 5 Typical Distribution of the Force Control Error for a Large Scale Fatigue Test

A320 ESG Full Scale Fatigue Test - Lessons Learned

535

IABG made use of modern NDT equipment in order to perform the inspections effectively, fast and accurately. The following inspection equipment was used for the NEF2 and NEF3: Table 1 IABG Inspection Equipment.

Means B588; M2 V2; B300 USM35 Endoscope

Main feature EC Rototest; EC Multifunctional devices US single element inspections visual

MT, PT EddyMax 4U II

MT- & PT-Inspection 4f EC for lap joint & bore hole inspections

Logistics Engineering Compared to former tests the facility availability could again be increased to more than 98.2%, due to consequent periodical maintenance of the test-set up and due to redundant test facility components. E.g., only three of four existing compressors (with a nominal rating of 1,000 kW in total) for the fuselage pressurisation and part of the hydraulic power supply were required for the parallel operation of both tests. In terms of spare part inventory, IABG has essential elements like load cells, valves or manifold blocks on stock for quick replacement. With the huge number of more than 700 actuators available at IABG, even actuators can be overhauled or eventually replaced in most cases. The effectiveness of this logistics concept, i.e. the availability of the test facility, can be seen in particular in the phase up to 80,000 simulated flights (NEF2; see Figure 2) and 110,000 simulated flights (NEF3, see Figure 3), respectively, where hardly any test specimen related interruptions occurred. Test operation was performed automatically and monitored in three work shifts for 24 hours per day and 7 days per week, including some public holidays. Beyond the regular working time and especially on week ends a standby task force for the test facility and for the airframe was established. At the end of a major B- or C-inspection day the test article was prepared for test continuation during the night. In the morning, test operation was stopped again to resume inspection work. Daily walk around inspection stops could be curtailed by temporarily increasing inspection man power. The NDT process itself was improved and thus accelerated. E.g., by implementing enhanced administration procedures, each inspector was able to process several inspection job cards in one run. The web based DMCT data base of Airbus allowed to change or create new inspection items online in it without having to create a new edition of the data base for every change or newly implemented item and exchange it with Airbus in a separate file. This reduces significantly the potential of misunderstandings and

536

G. Hilfer et al.

clarification issues with Airbus. Furthermore, a tracking list of all changed or newly created items is provided by the data base. The results of each inspection were thus available in the same data base. The subsequent inspection then profited from the fact that all the damage information and remarks were online available from the previously performed inspection. The DMCT data base was also used to support inspection organization. The inspection administration was able to adapt inspection related tasks, e.g. the deployment of inspectors, from previous inspections to the current inspection.

4 Conclusions IABG pursues a continuous improvement programme of its test technologies – which led to significant test speed gains for these A320 ESG tests. The performance of two major airframe tests at IABG at the same time in the same hangar allowed for synergies in the project management, the assignment of staff, and the coordination effort on Airbus side for the two tests. A more flexible scheduling of tasks like the performance of non-destructive inspections led to an optimized workflow and a minimization of otherwise inevitable downtimes. The performance of comprehensive systems simulations proved again to be a valuable tool not only to gain design data for the hydraulic system but also to shorten commissioning and the subsequent optimisation of control parameters significantly. The repair support by IABG using qualified personnel helped to reduce the coordination effort of external repair teams, travel cost and overall processing time of repairs. A detailed analysis of all test systems based on the specific requirements of these two tests brought forth further improvements for the loading program on Airbus side, control parameters and pneumatic system. Efficient maintenance procedures as well as a tailored redundancy and spare parts policy enhanced facility availability up to 98.2%. The entire package of implemented improvements resulted in a test speed increase of more than 30% with respect to the initial value while maintaining measurement accuracy and facility availability.

Acknowledgement The authors would like to thank Mr. Th. Laudan of Airbus Operations Hamburg and the entire team involved in the tests, for the excellent co-operation within the project.

A320 ESG Full Scale Fatigue Test - Lessons Learned

537

References [1] Rößler, N., Peters, C., Tusch, O., Hilfer, G., Herrmann, C.: IABG, Germany Airbus Operations, Germany, Concept of the New A320 Fatigue Test. In: Proceedings of the 25th ICAF Symposium, Rotterdam, The Netherlands (2009) [2] Tusch, O., Woithe, K.: A340-600 Full Scale Fatigue Test: A Further step forward into a efficient structure qualification. In: Proceedings of the 21st ICAF Symposium, Toulouse, France (2001) [3] Schwarberg, F., Eichelbaum, F.: An efficient Load Introduction Concept for the A380 Full Scale Fatigue Test. In: Proceedings of the 23rd ICAF Symposium, Hamburg, Germany (2005)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

New Connection Strap Concepts for A320 Wheel Well Area Tested during the Airbus A320 Extended Service Goal Full Scale Fatigue Test Nikola Cenic, Bernd Zapf, and Till Haberle RUAG Aerospace Structures GmbH, 82205 Gilching, Germany

Abstract. This paper describes the test campaign performed at the end of the A320 ESG (Extended Service Goal) full scale fatigue test with the goal of evaluating the new connection strap concepts and their effect on the airworthiness of the A320 MLG Bay Area. Instead of reinforcing locations with low fatigue reserves on damage tolerant Principal Structural Elements (PSE) an approach was taken to modify the load distribution within the MLG Bay Area by replacing the rubber connection straps with a more effective load carrying solution. Five different connection strap concepts were tested by installing them in the test airframe. Two concepts were based on 2024 Aluminum, two on CFRP and one was a fiber-metal laminate. The test loading was based on the most damaging fatigue load cases. The effect of different connection straps on the structural behavior was monitored using strain gauges, displacement transducers and an ARAMIS deformation measurement system. The prediction of the crack growth rate in the Side Box Beam flange was accomplished using Forman's formulation and an edge crack stress intensity function calibrated with the full scale fatigue test evidence. The effects and trade-offs between different connection strap designs are presented with respect to fatigue, damage tolerance and weight saving opportunities.

1 Introduction The ESG full scale fatigue test of Airbus A320 took place at the iABG test site in Ottobrunn, Germany in 2009 and 2010 [1]. The end of the test provided a possibility to test new connection strap concepts in the wheel well area of the main landing gear (Figure 1) under simulated flight conditions. This paper describes the test campaign for the new connection strap concepts, different measurements taken and challenges faced in evaluating the effect the new designs would have on the airworthiness of the Main Landing Gear bay area of A320. The goal of the test campaign was to understand the behavior of the MLG bay area at the location marked in Figure 2 under normal flight and ground loading conditions. Five structural concepts were developed and tested to optimize the current design with an intention of decreasing the stress levels and improving the low fatigue life reserve in the Side Box Beam flange (Figure 3) classified as a Damage Tolerant Principal Structural Element. Achieving lower fatigue stress levels at *

Oral presentation.

540

N. Cenic, B. Zapf, and T. Haberle

Fig. 1 Floor Section of A320 Wheel Well Area with existing connection strap concept at Frame C43.

Fig. 2 Structural elements of interest for the test campaign.

New Connection Strap Concepts for A320 Wheel Well Area

541

this location would extend the mandatory inspection intervals and relax the current inspection procedure and maintenance requirements. Instead of modifying and reinforcing the Side Box Beam inboard flange, an approach was taken to optimize the load distribution to the flange by modifying the neighboring structure. The structural improvement was based on replacing the rubber connecting straps (Figure 1) with a more versatile part that can carry the flight loads with greater efficiency. The current design utilizes the rubber connecting straps which are found inbetween the System Plate and the Pressure Membrane in the longitudinal direction and the Side Box Beam and the Longitudinal Floor Beam in the circumferential (hoop) direction (Figure 2). The currently implemented rubber connecting straps provide a feasible solution in providing structural continuation in longitudinal direction, however they lack the longitudinal and shear load carrying capability between the System Plate and the Pressure Membrane. This characteristic causes a higher load concentration in the connecting flanges of the Side Box Beam thus leading to a low fatigue life reserve.

Fig. 3 Fatigue Critical Location - Side Box Beam flange.

2 New Design Concepts In total five different concepts were developed for testing in the A320 Extended Service Goal Full Scale Fatigue Test. Concept No.1 (Figure 4) is a 2.0mm thick metal solution (Aluminum Alloy 2024 T42) that can be positioned in the location of the rubber connecting straps without any additional structural modification requirements. The concept can be easily introduced as a simple modification for the current A320 fleet. The advantages are the longitudinal, circumferential and shear load carrying capacity, and a simple installation procedure. The disadvantage in comparison to the other concepts is the low expected fatigue life thus leading to possible mandatory replacements during the aircraft's service life.

542

N. Cenic, B. Zapf, and T. Haberle

Flight Direction

Fig. 4 Concept No. 1 (Aluminum 2024 T42).

Concept No.2 (Figure 5) is similarly designed to concept No.1 except that it contains a cut-out in the middle giving it flexibility in the flight direction. The idea was to create a continuous load carrying path to take over a portion of the 1g cruise and vertical gusts loads that are currently transferred through the Side Box Beam and the Longitudinal Beam flanges. The advantages of Concept No.2 are the same as for Concept No.1 with the addition of weight minimization. The shear load carrying capacity and to some extent the longitudinal load carrying capacity are degraded.

Flight Direction Fig. 5 Concept No. 2 (Aluminum 2024 T42).

Concept No.3 (Figure 6) is a monolithic composite part made out of Carbon Fiber Reinforced Plastic (CFRP) with an oval indentation located in the middle for providing lower stiffness and greater movement flexibility in the flight direction. The part was designed to provide a shear carrying capability by placing fibers at ±45° in relation to the flight direction. The advantage is the elimination of the material fatigue effect in the part itself, apart from delamination and debonding that need to be considered. The design requires a modification of the surrounding structure, specifically the Pressure Membrane located between frames C43 and C46. An issue that was not addressed in the test campaign is the effect of temperature differences in the flight spectrum. The temperature difference would induce unwanted thermal stresses that could lead to a degrading fatigue life of the surrounding structure parts including the critical location on the Side Box Beam. These thermal effects could not be tested during the A320 full scale fatigue test because the test was performed at the constant indoor temperature.

New Connection Strap Concepts for A320 Wheel Well Area

543

Flight Direction Fig. 6 Concept No.3 (CFRP).

Concept No.4 (Figure 7) is similar to Concept No.3 with the addition of elongated “ears” in which carbon fibers are oriented in the flight direction. The reason for the addition of “ears” is to lower the stresses in the Side Box Beam and the Floor Beam Flange developed under the 1g cruise and vertical gust loadings. The characteristics are similar to Concept No.3 with an increase in weight due to addition of elongated “ears”. Additional disadvantage is a required modification of the System Plate located FWD of Frame C43. Elongated "Ears"

Flight Direction

Fig. 7 Concept No.4 (CFRP) and No.5 (steel-CFRP hybrid).

Concept No. 5 (Figure 7) is a hybrid fiber-metal laminate made out of steel sheets and carbon fiber layers. The inner area of the part containing the indentation has the same material properties as Concept No.4, specifically carbon fiber layers oriented at ±45° direction in relation to the flight direction. The elongated "ears" contain steel sheets and carbon fibers oriented in the longitudinal direction to improve 1g and vertical gust load carrying capability. Due to the presence of steel sheets this design weighs significantly more and is susceptible to fatigue damage, but it is damage tolerant. Similarly to Concept No.4 both the Pressure Membrane and the System Plate need to be modified for this design to be installed.

544

N. Cenic, B. Zapf, and T. Haberle Table 1 Weight saving opportunities (+ positive effect, - negative effect).

Design Concepts Current - Rubber Connection Strap Concept No.1 – Integral Aluminum Concept No.2 – Slotted Integral Aluminum Concept No.3 – CFRP Concept No.4 – CFRP with ears Concept No.5 – Hybrid with ears

Weight [g] 343.9 166.2 125.8 148.2 435.3 781.0

Weight Savings + +++ ++ --

3 Test Setup The measurements were performed after the NEF2 section of the A320 full scale fatigue test [1] had reached the required 180000 simulated flight cycles [FC]. As it was not possible due to time constrains to test the straps under a fully simulated flight conditions, only the combined fatigue load cases causing maximum fatigue damage to the critical locations were chosen for the test campaign. Table 2 contains the description of the Combined Load Cases (List ID 1-7) used to test the connection straps. The Combined Load Cases were constructed using the Unitary Fatigue Load Cases. The process is described by Eqn. 1. Table 3 provides an overview of the multiplication matrix for constructing the Combined Load Cases and Table 4 contains the description of the Unitary Fatigue Load Cases. 3

Combined _ Loadcase = Fact _ dp × Loadcase _ dp + ∑ Fact _ i × Loadcase _ i (1) i =0

Table 2 NEF2 Loading Program – Description of Flight and Ground Load Cases. NEF2 Loading Program - Combined Fatigue Load Case Description List ID Combined Loadcase Short Description 1 1000000000000020 Pressure Difference (Δp) 0.564 Bar 2 2278000000000000 Δp + 1g Cruise 3 2278000000610000 Δp + 1g + Coordinated Turn 4 2278006100000000 Δp + 1g + Lateral Gust 5 2278710000000000 Δp + 1g + Vertical Gust 6 2152000000000000 Touch Down MLG 7 2152340000000000 Touch Down MLG + MLG Spring Back 3.5 ft/s

Pressure Difference Fact dp Loadcase dp 1.00 10000000 1.00 10000000 1.00 10000000 1.00 10000000 1.00 10000000 0.00 10000000 0.00 10000000

New Connection Strap Concepts for A320 Wheel Well Area

545

Table 3 Description of NEF2 Loading Program and Unitary Fatigue Load Case components. NEF2 Loading Program List ID Combined Loadcase 1 1000000000000020 2 2278000000000000 3 2278000000610000 4 2278006100000000 5 2278710000000000 6 2152000000000000 7 2152340000000000

1G Increment 1 Increment 2 Increment 3 Fact 0 Loadcase 0 Fact 1 Loadcase 1 Fact 2 Loadcase 2 Fact 3 Loadcase 3 1.00 1.00 1.00 1.00 1.00 1.00

22780000 22780000 22780000 22780000 21520000 21520000

1.402 1.013 0.817

22781010

1.00

21521340

22784010

22782010

Table 4 Description of Unitary Fatigue Load Cases.

ID 1 2 3 4 5 6 7

NEF2 Loading Program - Unitary Fatigue Load Case Description Unitary Fatigue LC Short Description 10000000 Pressure Difference (Δp) 0.564 Bar 22780000 1g Cruise Medium Range Mission 22781010 10 ft/s Vertical Gust Medium Range Mission 22782010 10 ft/s Lateral Gust Medium Range Mission 22784010 Coordinated Turn Medium Range Mission 21520000 Touch Down Main Landing Gear 21521340 Touch Down MLG Spring Back 3.5 ft/s

The Combined Load Cases were applied at loading steps of 0% - 50% - 100% 0% to observe if the structure was deforming linearly. Strain gauges were used to measure the strain levels in the test components as well as in the surrounding structure itself. They included uni-directional strain gauges and strain gauge rosettes. An example of a strain gauge installation is given in Figure 8. Uni-directional strain gauges were marked with a 5 digit number and JPA/JLA notation, where J stands for uni-directional strain gauge, P for pressurized area, L for non-pressurized area and A for Aluminum (or C in case of CFRP). Rosettes were also marked with a 5 digit number and ABC PA / ABC LA, where P,L and A have the same meaning as in the uni-directional strain gauges but the letters ABC mark the three channels found on the rosette. Where it was possible the strain gauges were placed on both sides of the component to estimate the level of out-of-plane bending. For the reference point in evaluating the effect of different strap concepts on the load distribution into the Side Box Beam flange, a strain gauge rosette shown in Figure 9 attached from both sides to the System Plate was used. The relative movement of the System Plate to the Pressure Membrane in the area where the connecting strap is located was monitored using three Displacement Transducers mounted on the System Plate with the measuring point being on the Pressure Membrane (Figure 10). The Displacement Transducers were able to measure the spatial movement in all 3 directions (X,Y,Z) and they were calibrated for displacements of ±5.00 mm.

546

N. Cenic, B. Zapf, and T. Haberle

Fig. 8 Example of strain gauges installed on test specimen for Concept No.2.

Fig. 9 Strain Gauge Rosette 66401ABCPA / 66402ABCLA used for evaluating load distribution to the Side Box Beam flange.

New Connection Strap Concepts for A320 Wheel Well Area

547

Fig. 10 Displacement Transducers installed in NEF2 section.

Verification of the relative displacement under flight and ground loads was also performed using ARAMIS 3D deformation measurement system. The goal of using ARAMIS was twofold. First reason was to gain an insight into the structural deformation under applied loads in visual terms [Figure 11]. Second was to use the ARAMIS data as a means to certify a local Finite Element Model that is to be used for further structural optimization.

Fig. 11 ARAMIS measurement taken during the Pressure Difference Load Case.

548

N. Cenic, B. Zapf, and T. Haberle

4 Evaluation of Concepts The associated increase or decrease in stress levels was observed for each strap concept as can be seen in Figure 12 (data for Concepts 4 and 5 was not available at the time of writing).

Fig. 12 Effect of different strap concepts on the load distribution to the Side Box Beam flange as measured by strain gauge 66402JLA (Figure 9).

The prediction of the crack growth rates (Figure 13) in the Side Box Beam was estimated using Forman's formulation [2] where the stress intensity function (Eqn. 2) was based on the edge crack function from Rooke and Cartwright [3]. The stress intensity function was additionally corrected by α which is a function of crack length 'a' that was obtained by inserting an artificial damage and monitoring the crack propagation rate during the full scale fatigue test. The factor C1 in Eqn. 2 represents a fatigue stress spectrum multiplication factor for adjusting the difference in fatigue stress levels caused by different connection strap concepts as depicted in Figure 12.

ΔK = α ⋅ C1 ⋅ β ⋅ Δσ ⋅ πa

(2)

New Connection Strap Concepts for A320 Wheel Well Area

549

Fig. 13 Predicted crack growth rate in the Side Box Beam with different connection strap concepts installed.

5 Conclusions and Outlook The preliminary analysis outlined in this paper has shown that a redesign of the current rubber connection straps can benefit the airworthiness of A320 in the MLG Bay Area. Concepts No.1 and No.3 seem to be viable solutions. However to effectively judge their effect on the overall airworthiness the fatigue and static strength evaluations of the straps themselves still need to be performed. The challenge now lying ahead of us is to efficiently analyze the information collected during this test campaign and use it towards implementing the most optimal design in the MLG Bay Area for the up-coming A320 variants.

References [1] Rößler, N., Peters, C., Tusch, O., Hilfer, G., Hermann, C.: Concept of the New A320 Fatigue Test. In: Proceedings of the 25th ICAF Symposium, Rotterdam, Netherlands [2] Deutsche Aerospace Airbus, Crack Propagation and Residual Strength - Theoretical Manual, Deutsche Aerospace Airbus GmbH, Germany (1994) [3] Rooke, D.P., Cartwright, D.J.: Compendium of Stress Intensity Factors, H.M. Stationary Office, London, Great Britain (1975)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Full-Scale Static and Fatigue Testing of Composite Fuselage Section Rushabh Kothari Bombardier Aerospace

Abstract. The influence of efficiency in new design concepts and green imperatives are on the horizon for next generation aircrafts. Bombardier Aerospace is exploring greener aircraft and has been evaluating sandwich composite structure for use in a fuselage. Two thin sheets of a high performance material, the facesheets, are adhesively bonded to each side of a thick but considerably lighter core. The benefits of using this configuration layup are the high stiffness and strength to weight ratios, high thermal insulation, energy absorption and integrated manufacturing with reduced part count. As a part of this green aircraft initiative Bombardier Aerospace has been actively engaged in fatigue and static testing of full-scale sandwich fuselage sections as a part of a composite fuselage project. The ground testing consisted of various aspects of the damage tolerance and resistance study of the fuselage section as per FAR-25 guidelines. The damage resistance was measured using drop impactor and damage tolerance testing has been carried out through pressure cycle loading. The damage resistance of the sandwich structure was evaluated using multiple impactor sizes. The various parameters like impactor size, energy imparted, visual and barely visible damage effects, as well as hidden damage levels were studied. The damage tolerance (fatigue and static testing) of the fuselage consisted of various test stages where additional parameters such as moisture ingress, freeze thaw and repairs etc. were evaluated. New NDI techniques were developed and evaluated during this multiphase project to support current ground testing and future in-service inspection methodology development. The current paper outlines impact testing on sandwich construction, NDI results of impact sites using new generation of NDI instruments and DADT results of fullscale testing. Testing is currently on-going with further studies like advanced repair technologies, moisture ingress etc. on sandwich structure and continued improvements to existing design from the test results.

1 Introduction The composite fuselage full-scale testing was conducted as a part of composites research and development demonstration project. The full-scale barrel test article was build having similar geometry as an existing business aircraft model. The constant section barrel was approximately 16’ long and 8.5’ diameter. The *

Oral presentation.

552

R. Kothari

structure was mainly sandwich composite structure where highly stiff CFRP facesheets are separated by honeycomb core. The sandwich structure has high strength-to-weight ratio, excellent bending stiffness, streamlined manufacturing process and economical tooling making them very attractive for the next generation composite fuselage structure. The initial business case studies in sandwich construction were performed during ACT programs where fuselage panel representing Boeing aircraft was studies in various configurations [1]. The full-scale test plan was developed to study the durability and damage tolerant (DADT) response of the structure. The current paper outlines the details of the full-scale DADT testing.

2 Composite Barrel Design The composite barrel was manufactured as a part of collaborative project between Bell Helicopters Textron Canada Limited (BHTCL), Bombardier Aerospace (BA), National Research Council of Canada (NRC) and Composites Atlantic Limited (CAL). The structure of the fuselage mainly consists of sandwich structure. Figure 1 shows manufactured composite barrel test article. The fuselage has some monolithic reinforced areas to facilitate secondary attachment for systems, antennas, etc. The majority of the design consists of sandwich structure with honeycomb core material.

Fig. 1 Composite Barrel Structure.

Full-Scale Static and Fatigue Testing of Composite Fuselage Section

553

Barrel Manufacturing The barrel geometry is part of the constant section fuselage of a current BA business aircraft portfolio. The nominal dimensions are – 16 ft length and 8.5 ft diameter. The manufacturing of the barrel was achieved by 3 steps process. 1. 2. 3.

Manufacturing of the forward constant section fuselage, approx. 8’ Manufacturing of the aft constant section fuselage, approx. 8’ Circumferential joining of the fuselage with co-bonded out-of-autoclave bonding process.

The various novel technologies like fully integrated co-cured sandwich composite fuselage, secondary bonding process etc. were evaluated and demonstrated during the project. Barrel Construction The barrel was manufactured using hand lay-up prepreg tape from Cytec Industries. The tape has quasi-isotropic modulus of about 55 GPa. The sandwich core is a medium density honeycomb core from Euro-Composites. The OML (Outer mold line) and IML (inner mold line) of the sandwich construction were kept close to quasi lay-up to avoid thermal residual distortion. The nominal facesheet thicknesses of OML and IML are about 1mm (0.040”).

3 Barrel Impact Test As part of the DADT (Durability and Damage Tolerance) testing of the fuselage, the barrel was pre-impacted before fatigue testing. The various impacts parameter were evaluated and studies during this process. The impact on sandwich structure is a complex task due to the presence of a sandwich core. It is challenging to predict the response of the structure under various impact parameters. The impact on the fuselage is carried out in accordance with current damage scenario requirements. A typical fuselage structure will endure various manufacturing and in-service damages during its lifetime. The types of damages include tool drop, hail impact, tire debris on runway etc. Each of these damage types have certain energy associated with it. The impacts on the fuselage at various zones included following parameters: − − − −

Steel drop weight impacts Impactor head sizes from 12.7 mm (0.5”) – 89 mm (3.5”) Impact energies of 5 J – 80 J Sandwich and monolithic construction

554

R. Kothari

The typical impact zones with multiple impact sites on the fuselage are shown in Figure 2.

Fig. 2 Typical Impact zones on the fuselage.

Impact Test Results The impact tests were followed by NDI inspection to detect and study imparted damage sizes. The various NDI techniques like Thermography, Shearography and UT were used to evaluate damage areas and types. The following information after every impact event were obtained: − − − − − −

Velocity at the point of impact Impact force Skin damage area Core damage area Damage type (core crush, skin-to-core disbond etc.) Presence of delamination

Figure 3 shows typical damage types in sandwich construction. These areas were measured after each impact for further analysis.

Full-Scale Static and Fatigue Testing of Composite Fuselage Section

Extent of Core damage

555

Visible/surface damage/dent

Extent of skin damage

Fig. 3 Typical damage types for sandwich construction.

Figure 4 gives typical trend of the damage areas vs. impact energies for the sandwich zone of the fuselage.

Damage Area

Skin Damage Core Damage

0.00

10.00

20.00

30.00

40.00

50.00

60.00

Impact Energy (KE)

Fig. 4 Damage area vs. impact energy trend on sandwich structure.

70.00

556

R. Kothari

It is observed, from Figure 4, that the sandwich core usually has much larger damage area than the skin at the same impact energy. This is one of the main concerns in any sandwich structure and one need to design sandwich with acceptable damage resistance threshold to avoid frequent repairs on the aircraft. Figure 5 shows trend of impact energy vs. dent depth on sandwich structure. Dent depths are usually measured in 1/1000” or mm. The 1 mm (0.040”) dent depth is typically identified as BVID (Barely visible impact damage). BVID is an important visual inspection parameter. The aircraft operators are usually able to notice and visually detect BVID dent depth during the inspection process. If a structure has an internal damage with very low dent depth on the surface (less than BVID dent depth), then it may go unnoticed during normal walk-around type inspection and it is expected for the damaged structure to handle ultimate load [2]. Due to this requirement, it is imperative to characterize sandwich and monolithic structures with respect to their dent depths at various energy levels.

Dent Depth [1/1000 in]

0.5" Impactor

1.25" Impactor

3.25" Impactor 2.5" Impactor

0.00

10.00

20.00

30.00

40.00

50.00

60.00

70.00

Impact Energy (KE) [J]

Fig. 5 Dent depth vs. impact energy trend on sandwich structure.

Figure 5 shows typical trend of the dent depth for various impactor head diameter. It can be seen that as impactor diameter increases, hence blunt impact, dent depth does not increase as rapidly. This poses design challenge for sandwich structure. It can be seen from Figure 4 that for higher energy levels the core damage is substantial. Due to lower dent depth of blunt impacts at those energies, the damage will go unnoticed during the normal aircraft operation. This needs to be well understood in the initial design phase of the fuselage so that all the loads requirements are met with damaged structure.

Full-Scale Static and Fatigue Testing of Composite Fuselage Section

557

4 Full-Scale DADT Barrel Testing Test Program The composite fuselage barrel test program was designed to conduct fatigue as well as static testing on the same specimen. Unlike metallic structure, it is possible for composite structure to conduct both fatigue and static testing on the same specimen without any adverse effect. The comprehensive test plan was developed, including above mentioned impact testing. The testing occurs in phases where each phase (or life) consists of fatigue, limit load and ultimate load tests.

DADT 1000 Cycles

Pre-test NDI

NDI of all damage sites

DADT 10 000 Cycles

NDI of all damage sites

Static test (Limit and Ultimate)

NDI of all damage sites

DADT 5 000 Cycles

NDI of all damage sites

NDI of all damage sites

DADT 20 000 Cycles

Results Analysis

Fig. 6 DADT static and fatigue life of the fuselage barrel.

Figure 6 represents typical one life of full-scale fuselage testing. The damage sites are monitored at the specified intervals. The damage growth in sandwich skin as well as core is monitored during the fatigue and static life. Test Results – Strain Gauges There were as many as total of 100 strain gauges present at various locations to monitor barrel’s health during the test. These strain results are also used to calibrate and refine FEA model. Figure 7 represents strain variations at various locations on the fuselage barrel during fatigue cycling. It can be seen that most of the locations do not show large strain variations (less than 5%). The larger strain variations at certain locations are due either damage growth or presence of higher noise at low strains (less than 400 με).

558

R. Kothari

% Change in Strain from 1-20k flights-1st Life 15.0% 12.5% 10.0% 7.5%

% Strain Change

5.0% 2.5% 0.0% -2.5% -5.0% -7.5% -10.0% -12.5% -15.0%

E0 0 R 1 01 4 R A 01 5 R A 01 6 R C 00 9 R B 01 9 R A 03 4 R B 01 2 R A 03 3 R C 01 7 R B 02 0 R A 02 2 R C 02 9 R B 02 1 R0 A 23 R C 02 8 R B 03 6 R A 03 1 R C 03 9 R B 03 5 R A 03 7 R C 04 0 R B 07 0 R A 04 8 R C 05 2B E0 54 E0 60 E0 6 R 5 04 6 R A 05 0 R C 05 3 R B 07 1A

-17.5%

SG

Fig. 7 Strain variation at various locations during fatigue testing.

The fatigue cycling revealed some damage growth areas in the structure. The damage progression was captured and it was observed that damage progression rate decreased over the time. Figure 8 shows the example of a strain gauge capturing such location. The fatigue testing was followed by static limit and ultimate tests to demonstrate ultimate load bearing capacity of the structure with induced damages. The damage size as well as progression is considered within acceptable limits once structure passes specified ultimate load at the end of the fatigue cycling. The composite barrel was demonstrated to pass specified ultimate static load case, thus representing acceptable damage sizes and growth rate. Test Results – Damage Sites The impact damages introduced during the pre-test impact event were monitored throughout the test phase. The damage progressions in sandwich as well as monolithic constructions were studies under fatigue loads. Usually, CFRP are not prone to typical fatigue loads unless loaded out-of-plane. The newer generation honeycomb sandwich core used for current design has much better fatigue properties as well. The Figure 9 shows typical damage progression under fatigue and static load of a sandwich damage site. It should be noted that due accumulation of tolerances in damage detection using current NDI methods, the slight reduction of damage size is reported at certain steps, which should be interpreted as no-damage-growth as damage zone can not reduce over time unless repaired.

Full-Scale Static and Fatigue Testing of Composite Fuselage Section

Fig. 8 An example strain gauge showing damage progression.

Fig. 9 Damage progression trend in sandwich for one of the damage sites.

559

560

R. Kothari

Overall, it was observed that the damage progression was within acceptable limits for sandwich as well as monolithic zones.

5 Conclusion The composite sandwich fuselage barrel was manufactured and DADT tests were performed. The initial impact results, one life fatigue and static test results of sandwich structure was studied in detail for further investigation of cost effective and reduced weight fuselage structure. It is observed that such a structure is a viable solution of future aircraft platforms, which can support green aircraft initiative at Bombardier. The reduced weight, less manufacturing cost and reduced carbon footprint is achievable with this newer generation design.

Acknowledgement The author would like to thank everyone at BA, BHTCL, NRC and CAL for their support in composite barrel test program.

References [1] Polland, D.R., Finn, S.R., Griess, K.H., Hafenrichter, J.L., Hanson, C.T., Ilcewicz, L.B., Metschan, S.L., Scholz, D.B., Smith, P.J.: Global cost and weight evaluation of fuselage side panel design concepts, NASA contractor report 4730 (1997) [2] Federal Aviation Administration (FAA), Advisory Circular AC 20-107B, Composite Aircraft Structure (2009)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Durability and Damage Tolerance Evaluation of VaRTM Composite Wing Structure

Yuichiro Aoki, Yoshiyasu Hirano, Sunao Sugimoto, Yutaka Iwahori, Yosuke Nagao, and Takeshi Ohnuki Japan Aerospace Exploration Agency, 6-13-1 Osawa Mitaka, Tokyo 181-0015 Japan

Abstract. Durability and damage tolerance of subcomponent and full-scale wing box structure fabricated by VaRTM are evaluated. Fatigue spectrum with load enhancement factor was applied to the test articles for 1 DSO of 40,000 flights. The Mini-TWIST fatigue spectrum is used for both tests. Then, impact damages are given to the skin stiffened by co-cured stringer and typical skin part by dropweight to create the delamination. After that, impact damage growth is evaluated during 1 DSO fatigue spectrum and optimal inspection interval is examined. Finally, residual strength of structures is verified by ultimate load test with 150% design limit load. Applied strain level for Subcomponents are intentionally higher than original one in order to evaluate the structural performance in more critical condition. Non-destructive inspection is carried out by 3D ultrasonic scan system with multiple-array sensors to evaluate delamination growth. In Subcomponent test, stringer run-out shows local out-of-plane deformation and that causes disbonding of stringer termination. The disbonding area gradually increases during 1 DSO fatigue test. However, the structure did not show any degradation of structural performance. The damage tolerance tests verify that impact-induced delaminations have not grown throughout the 1 DSO. In the final ultimate load test, the load bearing-capabilities of present VaRTM wing structure have been verified and the structure could survive for 4 seconds without any detrimental deformation and damage growth.

1 Introduction In recent years, manufacturing cost as well as structural weight becomes more important issues for developing a new aircraft. Although composite materials have been successfully introduced to primary structures of both military and civil aircraft, the raw material price for carbon fibre prepreg is still much higher than conventional aerospace-grade aluminium [1]. Manufacturing costs are also higher than conventional metal structure. Therefore, many fabrication technologies have been studied to focus on the affordable composite airframe application. Since the late 1980’s, many fabrication demonstration programs were undertaken to investigate the application of textile reinforced composites as a cost-effective method of *

Oral presentation.

562

Y. Aoki et al.

producing damage tolerant primary aircraft structures. In the NASA- ACT Program [2-3], Boeing fabricated a wing structure with the through-the-thickness stitching and RFI (Resin Film Infusion) process under contract to NASA. The program target of 25% weight reduction and 60% part-count reduction compared with conventional aluminium structure was accomplished. Now, the technology is partly evolved into the blended-wing-body (BWB) damage tolerant fuselage design concept within NASA research project [4-5]. On the other hand, TANGO [6] to be followed by ALCAS [7-8] and MAAXIMUS [9] are the most active European aviation research projects that have been running since the year 2000s, with the targets of higher reduction in weight and cost reduction in comparison to current aircraft structures. The latest project MAAXIMUS also aims at achieving the fast development and right-first time validation of a highly-optimised composite airframe by using a coordinated effort between virtual structure development and state-of-art composite technology. In Japan, there are some integrated aviation research projects on low-cost composite fablication technology. The ISTR (Innovative STRucture) project started in 1999 was a 5-year reserach project funded by NEDO (New Energy and Industrial Technology Development Organization) [10-11]. In the ISTR project, large scale wing structures were fabricated through one-shot process, where pre-cured skin are co-bonded with stringers and ribs fabricated by RTM (Resin Transfer Molding). The project has accomplished the development of innovative composite wing structure with durability and damage tolerance and satisfied 27% lighter weight and 54% less in part-count compared to conventional metal structure [12-13]. Furthermore, recently, JAXA (Japan Aerospace Exploration Agency) focuses on the VaRTM (Vacuum-assisted Resin Transfer Moulding) fabrication technology as one of the most promising candidates to achieve cost reduction of composite airframes. The VaRTM gives the opportunity to increase part size with easier mould design with lower-cost. Part counts reduction is also expected by an integration of structure with one-shot fabrication process. A research project on development of full-scale VaRTM wing box has been conducted by Aviation Program Group and Advanced Composite Group in JAXA since 2003 (Figure 1). In the project, fabrication and manufacturing processes have been established and static performance of the structure has been verified since 2008 by test and numerical analysis in reference to the specific building block approach as shown in Figure 2 [14-16]. The developed VaRTM wing structure offers the potential benefit of 25% fabrication cost reduction compared to the conventional prepreg structure [17]. In addition, the NDI (Non-Destructive Inspection) method for the verification of VaRTM structure has been established and results are reported [16]. However, there still remain some issues to be clarified from the aspects of operation and maintenance. In this paper, fatigue tests of Subcomponent and full-scale wing box structure fabricated by VaRTM are conducted in order to verify durability and damage tolerance of the structures. The first phase is fatigue tests to verify durability of the structure without any artificial damages. The second phase is evaluation of damage tolerant capability by using the structure with visible and barely visible impact damages to estimate optimal inspection intervals. Finally, residual strength of structures with impact damages after fatigue tests is verified by ultimate load test.

Durability and Damage Tolerance Evaluation of VaRTM Composite Wing Structure

563

Fig. 1 Overview of JAXA VaRTM wing box development project.

○：Finished ×：Not finished

○ ○ ○ ○ ○

○ ○ ○ ○ ○

× × N/A N/A N/A

Fig. 2 Project status as of 2008.

2 Test Articles Full-scale wing box A wing box demonstrator was designed to represent a main wing of 30-seat business jet [18]. This demonstrator focuses on reducing cost by combining parts into an integrated wing structure. Size of this demonstrator is 6.0 m span and chord length is 1.4 m at the root. The structure consists of integral spars, upper and lower covers stiffened by blade-type stringers and ribs, where each part is separately fabricated by one-shot VaRTM process. Both covers have curvature and optimized thickness distributions. The lower cover has cut-outs and is integrated with both

564

Y. Aoki et al.

spars to reduce the number of mechanical fasteners needed for assembly. Ribs are integrated by secondary bonding and both covers are assembled by mechanical fasteners at both spars. The cover panel was fabricated by unidirectional carbon fabric (T800SC, supplied by Toray Industries Inc.) and stringers are fabricated by bi-axial Non Crimp Fabric using IMS5131-24K carbon fibre (supplied by SAERTEX GmbH & Co.), and modified low viscosity epoxy resin system, XNR6809/XNH6809 (produced by Nagase Chemtex, Co. LTD.), is used for infusion process. Schematic overview of the wing box structure is illustrated in Figure 3. Subcomponent panel The test panel represents a portion of the lower cover of wing box structure. The panel size is 2.1m long x 0.9m width that has four stringers and a cut-out. Two stringers are terminated at its centre, so-called stringer run-out. Evaluation area is equivalent to the 1 bay between ribs, length of 700 mm x width of 900 mm in which cut-out and stringer run-out are critical regions. In order to grip the panel with hydraulic actuators, steel fixtures are attached to both ends of the panel by mechanical fasteners with adhesive agent.

1.4 m

Upper cover

6.0 m

1.0 m

Spar assembly

Rib assembly

Lower cover

Fig. 3 Overview of the wing box structure.

3 Test Plan General condition Present test consist of strain survey, 2 DSO fatigue and residual strength tests. The same test was conducted for Subcomponent and full-scale structure. Figure 4

Durability and Damage Tolerance Evaluation of VaRTM Composite Wing Structure

565

shows test procedures and evaluation items. First, all the test equipment, devices and hydraulic systems were checked at a low load level prior to actual testing. Then, the static load was applied up to the maximum and minimum load levels that test article will be subjected during fatigue spectrum, and evaluates the structure behaviour compared to numerical model. After strain survey, fatigue spectrum with load enhancement factor was applied to the test article for 1 DSO (Design Service Objective) to verify durability of the structure, where stringer runout is the most critical area in present test. Then, 100% DLL (Design Limit Load) was applied to evaluate residual strength of the structure. Next, BVIDs (Barely Visible Impact Damages) and VIDs (Visible Impact damages) were given by drop-weight with 25mm diameter hemispherical and sharp edge steel tips. The total mass of the drop-weight impactor was 15 kg. Impact energies were ranged from 22.6J to 169.5J shown as Table 1. Impact locations and directions are shown in Figure 5. After that, every impact point was inspected by ultrasonic scanning system to evaluate delamination size and shape. Then, static load was applied to check the behaviour and structural integrity of the test article. Then, another 1 DSO fatigue spectrum was applied to the test article to verify damage tolerance of the structures. The purpose of this test is to demonstrate that BVIDs would not increase to a detrimental size during 1 DSO and VIDs would not increase to a detrimental size within two inspection intervals. NDI was carried out every 10% DSO during the test to check the impact damage growth. Finally, residual strength of the structures after fatigue tests with impact damages was verified by ultimate load (150% DLL) in up bending condition. Consideration of the structural integrity and inspection interval are mainly based on AC20-107B [19].

Fig. 4 Test procedures and evaluation items.

566

Y. Aoki et al.

Upper cover Lower cover Subcomponent

Full-scale wing box

Table 1 Impact test conditions. ID

Location

Energy

Impactor shape

Impact 1

Skin/stringer

62.1 J (550 in-lbs)

Hemispherical

Category BVID

Impact 2

Stringer run-out

62.1 J (550 in-lbs)

Hemispherical

BVID

Impact 3

Skin/stringer

62.1 J (550 in-lbs)

Sharp edge

VID

Impact 4

Stringer run-out

62.1 J (550 in-lbs)

Sharp edge

VID

Impact 5

Skin

39.5 J (350 in-lbs)

Hemispherical

BVID

Impact 6

Skin

39.5 J (350 in-lbs)

Sharp edge

VID

Impact 7

Skin/stringer

135.6 J (1200 in-lbs)

Sharp edge

VID

Impact 8

Skin/stringer

135.6 J (1200 in-lbs)

Hemispherical

BVID

Impact 9

Stringer run-out

169.5 J (1500 in-lbs)

Sharp edge

VID

Impact 10

Stringer run-out

169.5 J (1500 in-lbs)

Hemispherical

BVID

Impact 1

Skin/stringer

135.6 J (1200 in-lbs)

Hemispherical

BVID

Impact 2

Skin/stringer

135.6 J (1200 in-lbs)

Sharp edge

VID

Impact 3

Skin panel

67.8 J (600 in-lbs)

Hemispherical

BVID

Impact4

Stringer web

22.6 J (200 in-lbs)

Hemispherical

VID

Fatigue spectrum and load conditions Present durability and damage tolerance evaluations were performed by the fatigue spectrum, Mini-TWIST [20-21], which represents typical load sequences for flight simulation on transport aircraft wing. The Mini-TWIST consists of 10 distinct flight types (Flight type A to J). Load levels in each flight have been normalized by load factor during cruise conditions. In the Mini-TWIST, specific 4000 flights set is defined as “one block”. The maximum and minimum load levels in “one block” are +2.6G (= +104%DLL in present case) and -0.6G respectively. These peak loads is included in the most critical Flight type A, each load occurs once in one block (0.025% probability). In present tests, one design life was defined as equal to 40,000 flights, which is the design service objective for present aircraft. Furthermore, load level Load Enhancement Factor (LEF) method [22-23] is used in order to economically demonstrate one operational fatigue lifetime. Fatigue spectrum load level is increased by LEF of 1.18. Since there is not sufficient statistical fatigue data for present material to achieve the desired reliability level, the conservative LEF value was chosen to reduce the test duration in reference to Ref [22-23]. However, the present LEF value must be reasonable to demonstrate durability and damage tolerance of composite structure fabricated by VaRTM process because the fatigue scatter (including damage initiation and growth) of present VaRTM composite material tends to be larger than that of conventional autoclaved composite material. Therefore, the peak load level in present fatigue spectrum is 123% DLL at +2.6G.

Durability and Damage Tolerance Evaluation of VaRTM Composite Wing Structure

567

(a) Full-scale wing box: all impacts were given from the outside IML

Impact 1 135.6 J (1,200 in-lbs)

OML

Impact 4 22.6 J (200 in-lbs) Impact 3 67.8 J (600 in-lbs)

Impact 2 135.6 J (1,200 in-lbs)

Impact directions

Impactor shape

(b) Subcomponent panel and impact test for “Impact 3” Fig. 5 Impact locations and directions.

Test systems Equipment for the full-scale test, MTS servohydraulic actuators driven by FlexTest 200 digital controller and client PC running AeroPro Control & Data Acquisition software. The average frequency was 0.25 Hz. Total of 374 strain gage data are collected by HBM MGC-plus and TDS-630 (Tokyo Sokki Kenkyujo) with sampling rate of 10 Hz. The other equipment were SAMOS 32 AE system (Physical Acoustic Corporation) to evaluate the impact damage growth as shown in Figure 6-(a) and fiberscope (Olympus IPLEX-LX) used for inspection inside the structure. For the Subcomponent tests, load was applied to the panel by MTSModel 311.41 hydraulic load frame (2500kN capacity) outfitted with high force Model 647.200 hydraulic grips. The system employs FlexTest 40 digital controller and a PC running Multi-Purpose TestWare software. The load frequency was 1.0 Hz. A multi-channel data logger (Kyowa UCAM 500B) was used as data acquisition devices to obtain load, displacement and 80ch. strains. Figure 6 shows test setup for full-scale test and subcomponent panel. A three-dimensional

568

Y. Aoki et al.

ultrasonic inspection system (Toshiba Matrixeye64) was used to evaluate impactinduced delamination in both tests. Impact areas were inspected by this scanning system with multiple 5 MHz linear-array transducers.

(a) Full-scale wing box

(b) Subcomponent panel

Fig. 6 Test setup.

4 Results and Discussions Full-scale wing box test All tests for full-scale structure were conducted at ambient conditions. No environmental factors were applied to the test articles to account for moisture and/or temperature effects. The strain distributions obtained by strain survey were verified by analytical model that demonstrates the design capability of VaRTM composite materials. Any fatigue damages didn’t occur during 1 DSO fatigue spectrum with LEF of 1.18, where maximum strain levels at peak fatigue load were +2,900μand -2,800μ at the stringer run-out tip of lower and upper covers respectively, +2,400μaround the cut-out of lower panel and other area was within ±1,500μ. These strain levels were no greater than 60 percent of the ultimate capability of present VaRTM composite materials. Therefore, any fatigue failure didn’t appear and the initial structural stiffness remains at the 100%DLL verification after 1 DSO fatigue test. Since the margin of safety for ultimate capability was sufficient, the integrity structure was not significantly affected. Figure 7 shows external view and delamination shape taken by ultra sonic scanning system for “Impact 9” and “Impact 10” that were given by the impact test at the most critical impact condition. For “Impact 9” given by the sharp edge impactor, fibre breakage occurred in the impacted surface and quite large delamination was created inside the panel, and the impactor penetrated through the thickness. Thus, delamination envelop with cracks in back surface can be clearly shown from the ultrasonic inspection result. Despite these damage conditions, any delaminations didn’t grow during another 1 DSO fatigue spectrum. ”Impact” 2, 4, 9, 10 were given to the vicinity of stringer run-out where the skin panel shows local

Durability and Damage Tolerance Evaluation of VaRTM Composite Wing Structure

569

out-of-plane deformation, but the any impact damage size didn’t increase under repeated load condition. Neither fatigue damage nor any propagation of impact damages occurred during the present full-scale fatigue tests. Finally, damage tolerant residual strength capability of present wing box was demonstrated by the ultimate load test after 2DSO fatigue spectrum. The structure could survive for 4 seconds without any detrimental deformation and damage growth, and didn’t show any degradation of stiffness. Figure 8 shows the strain behaviors at major parts.

(a) Impact 9

(b) Impact 10 Fig. 7 External view and NDI result after impact test.

Load (% DLL) 150

100 Cut-out Str. run-out (Lower) Str. run-out (Upper) Skin (Lower) Stringer web (Lower) Skin (Upper) Stringer web (Upper)

50

-4000

-2000

0

2000 Strain (

4000

μStrain)

Fig. 8 Strain histories during ultimate load test.

570

Y. Aoki et al.

Subcomponent test Present test panel represents a part of lower cover of the full-scale wing structure, where tensile load is dominant. For subcomponent test, the average strain level was intentionally increased in order to clarify the failure mode of VaRTM wing structure. More severe load condition was given to the panel than full-scale structure. Applied strain level was 65% higher than full-scale structure. The predicted strain levels at peak fatigue load were +5,000μ at the stringer run-out tip, +4,000μ around the cut-out and other area was about +2,400μ. However, one of the stringer run-out disbanded at the 83% of peak fatigue load during the initial strain survey. Actual local strain at the stringer run-out tip area was +7,000μ which is 69% higher than predicted result at the same load level. That unexpected high strain was caused by concentrated load due to poor quality of the run-out region and local out-of-plane displacement of stringer tip. Since present structure is made by one-shot VaRTM fabrication, quality of this section tends to be scattered. It notes that further and more detail considerations should be needed for numerical modelling of these complex sections. Therefore, 1 DSO fatigue test was started by the panel with initial disbonding of stringer run-out. The disbonding area gradually increases during 1 DSO fatigue spectrum (Figure 9). However, the structure did not show any degradation of structural performance and has residual strength. The damage tolerance tests verify that impact damages have not grown throughout the 1 DSO fatigue spectrum because tensile load is dominant which does not affect the impact-induced delaminations. Although the increasing rate tends to be smaller, disbanding area of stringer run-out shows slow growth behavior during the test. In the final ultimate test, the load bearing-capability of present subcomponent was verified and the panel successfully survived for 4 seconds without any stiffness degradations and impact damages growth.

Fig. 9 Slow growth of stringer run-out disbanding.

Durability and Damage Tolerance Evaluation of VaRTM Composite Wing Structure

571

5 Conclusions Durability and damage tolerance evaluation of VaRTM wing structure were successfully verified for full-scale and subcomponent structures. Neither fatigue damage nor any impact damage propagations occurred during throughout the tests. Although special attention on disbonding of stringer run-out should be required, residual strength capability of present structure was demonstrated without any degradation and detrimental deformation.

References [1] Kruckenberg, T., Paton, R.: Resin Transfer Moulding for Aerospace Structures. Kluwer Academic Publishers, Great Britain (1998) [2] Poe Jr., C.C., Dexter, H.B., Raju, I.S.: A Review of the NASA Textile Composites Research. In: Proceedings of the 38th AIAA/ASME/ASCE/AHS/ASC Structures. Structural Dynamics & Materials Conference, Paper No. 97-13 (1997) [3] Dexter, H.B.: Development of Textile Reinforced Composites for Aircraft Structures. In: Proceedings of the 4th International Symposium for Textile Composites, Kyoto, Japan, October 12-14 (1998) [4] Mukhopadhyay, V.: Blended-wing-body (BWB) fuselage structural design for weight reduction. In: Proceedings of the 46th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics & Materials Conference, Paper No. AIAA 2005-2349 (2005) [5] Yovanof, N.P., Velicki, A., Li, V.: Advanced Structural Stability Analysis of a Noncircular, BWB-Shaped Vehicle. In: Proceedings of the 50th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics & Materials Conference, Paper No. AIAA 2009-2452 (2009) [6] Vincendon, M.: TANGO: Low cost light weight structure. Air & Space Europe 3(3-4), 122–125 (2001) [7] Bommer, H.: Alcas – Advanced Low Cost Aircraft Structures General Overview. In: Proceedings Of Sampe Europe 31st International Technical Conference & Forum (SEICO 2010), pp. 23–28 (2010) [8] De Luzy, A., Vilain, T.: General Overview of ALCAS WP3 – Business jet wing project, Architecture Concepts and Design Aspects. In: Proceedings of SAMPE EUROPE 31st International Technical Conference & Forum (SEICO 2010), pp. 29–34 (2010) [9] MAAXIMUS Project, More Affordable Aircraft through eXtended, Integrated and Mature nUmerical Sizing (2009), http://www.maaximus.eu [10] Yahata, A., Kikukawa, H.: R & D Program for Innovative Civil Aircraft Structures. In: Proceedings of the 14th International Conference on Composite Materials, Paper No.1983 (2003) [11] Toi, Y., Harada, A., Kamiya, T., Inoue, T., Amaoka, K., Kikukawa, H.: Development of Affordable Composite Wing Structure. Advanced Composite Materials 12(4), 321–330 (2003) [12] Aoki, Y., Ishikawa, T., Takeda, S., Hayakawa, Y., Harada, A., Kikukawa, H.: Fatigue test of lightweight composite wing structure. International Journal of Fatigue 28(10), 1109–1115 (2006)

572

Y. Aoki et al.

[13] Takeda, S., Aoki, Y., Ishikawa, T., Takeda, N., Kikukawa, H.: Structural health monitoring of composite wing structure during durability test. Composite Structures 79(1), 133–139 (2007) [14] Nagao, Y., Iwahori, Y., Hirano, Y., Aoki, Y.: Low Cost Composite Wing Structure Manufacturing Technology Development Program in JAXA. In: Proceedings of the 16th International Conference on Composite Materials (ICCM-16), Paper No. MoAM1-05pl (2007) [15] Review of Aeronautical Fatigue Investigations in Japan during the period July 2007 to May 2009. In: Terada, H., Takeda, N. (eds.) Presented at 31st Conference of the International Committee on Aeronautical Fatigue (ICAF), Rotterdam, The Netherlands, May 25-26, pp. 21–23 (2009) [16] Aoki, Y., Sugimoto, S., Hirano, Y., Nagao, Y.: Non-Destructive Inspection Technologies for VARTM Composite Structure. SAMPE Journal 46(1), 22–27 (2010) [17] Nagao, Y., Iwahori, Y., Aoki, Y., Hirano, Y., Kuratamni, Y.: Low Cost Composite Manufacturing Technology Development Program for Wing Structure by JAXA. In: Proceedings of 14th US-Japan Conference on Composite Materials, Dayton, Ohio, September 20-22 (2010) [18] Hirano, Y., Aoki, Y., Iwahori, Y., Sugimoto, S., Nagao, Y.: Development and Mechanical Properties Verification of VaRTM Composite Full-Scale wing Demonstrator. In: Proceedings of SAMPE 2008 Long Beach, May 18-22 Paper No. 269 (2008) [19] Federal Aviation Administration: AC 20-107B – Composite aircraft structure (September 2009) [20] de Jonge, J.B., Schütz, D., Lowak, H., Schijve, J.: A standardised load sequence for flight simulation tests on transport aircraft wing structures, NLR TR 73029 U (1973) [21] Lowak, H., de Jonge, J.B., Franz, J., Schütz, D.: MINITWIST - A shortened version of TWIST, NLR MP 79018 U (1979) [22] Whitehead, R.S., Kan, H.P., Cordero, R., Sather, E.S.: Certification Testing Methodology for Composite Structure, Volume II — Methodology Development, NADC87042-60, DOT/FAA/CT-86/39 (1986) [23] Lameris, J.: The use of load enhancement factors in the certification of composite aircraft structures, NLR TP 90068U (1990)

26th ICAF Symposium – Montreal, 1-3 June 2011 Development of Load Spectrum for Full Scale Fatigue Test of a Trainer Aircraft Andrzej Leski, Piotr Reymer, and Marcin Kurdelski Air Force Institute of Technology, Warsaw, Poland

Abstract. PZL-130 „Orlik” is a turbo-propeller engine trainer airplane, entirely designed and build in Poland. This plane is used in Polish Air Force mainly for primary training of polish military pilots. Current modernization of this airplane (which mainly consists of the engine and wings conversion) is enhanced by major change in the maintenance system. Maintenance system change program (SEWST) is carried through by EADS PZLOkęcie (manufacturer) and ITWL (Air Force Institute of Technology). One of the major program task is a Full Scale Fatigue Test of the airframe which will be conducted by VZLU in Czech Republic. This article will present the methodology which was used to prepare a characteristic load spectrum for this test. Following sections describe the whole process from statistical analysis of data recorded in the onboard flight recorders through strain gauge based flight load measurements to development of a fatigue load spectrum. As a result a characteristic flight load spectrum block was obtained representing equivalent 200 flight hours. This spectrum will be used in a Full Scale Fatigue Test which is designed to estimate fatigue life and critical points in the structure. After completion of all the above described steps the obtained spectrum was both statistically consistent and representative for characteristic flight program in Polish Air Force. This guaranteed that the fatigue test based on this load spectrum will estimate actual fatigue life for these aircrafts operated in Polish Air Force and determine critical points in which structural damage may occur during operation.

1 Introduction The PZL-130 "Orlik" trainer aircraft was designed in Poland in the end of the twentieth century. In 1994 it was introduced to the Polish Air Force (TC-I version). It is a single-engine, two-seated aircraft used for preliminary pilot training and display flying. These aircrafts are still operated and the users are very satisfied with it's performance. The main disadvantages of the TC-I version are necessary frequent overhauls (every 1000 flight hours) and frequent periodic maintenance performed by user. Such maintenance system was a result of suspension of the former research program which goal was to determine actual fatigue life under Polish Air Force operation characteristic flight profile. No Full Scale Fatigue Test was ever carried through. The fatigue life was estimated using analytical methods. The experience gathered throughout over 20 years of operation showed, that the current operation method was not optimal and the need to perform overhaul every

574

A. Leski, P. Reymer, and M. Kurdelski

1000 flight hours results in unavailability of the fleet due to repairs. The TCII version was developed in the 1990's. The most important changes were: - Pratt & Whitney engine (700BHP), - 5 blade propeller, - new wing geometry, - new fin geometry. The manufacturer has provided the opportunity to upgrade the airplanes from TC-I to TC-II version. Two airplanes were upgraded to TC-II version in the early twenty-first century. Positive experiences – mainly a significant improvement in flight performance - resulted in modernization order for another 14 airplanes. In the mean time PZL Warszawa-Okęcie, the manufacturer of the aircraft, became a part of the EADS. Along with the airplane modernization the Polish Air Force ordered development of a new modern maintenance system for the PZL-130 Orlik TC-II. The main requirement for the new system, were: - avoidance of overhaul every 1000 flight hours, - confirmation (and most preferably excess) of the previously estimated total fatigue life of 6000 flight hours, - development of the structure integrity program suitable for this aircraft. Realization of these objectives is the subject of SEWST research program performed by EADS PZL-Okęcie and ITWL.

2 Sewst Program Definition The whole SEWST program covers many different topics. The most important part is the confirmation of the total fatigue life without the need of overhaul. The main objective of the whole program, as it comes to the aircraft structure, is the Full Scale Fatigue Test. As stated above aircrafts intended for the Polish Air Force will be the modernized version of the aircraft previously operated as TC-I. During the modernization aircrafts are fitted with a completely new wing and the fin's geometry is slightly changed. It was therefore decided that the fatigue test will be performed for an equivalent structure. For the Full Scale Fatigue Test purpose EADS PZL-Okęcie has performed a partial modernization (from TC-I to TC-II covering the structural changes only) of an aircraft already withdrawn from operation (No. 015). ITWL is a research institute subordinate to the Polish Ministry of Defense and executes all kinds of aviation investigations for all types of Polish Armed Forces. One of the regular ITWL tasks is monitoring of flight load on all kinds of aircrafts. No major changes are going to be introduced in the training course for TC-II version. Hence data gathered from operation of TC-I version of the training aircraft, which has been recorded throughout many years, were the basis to determine the severity of flight load spectrum characteristic for Polish Air Force and furthermore to determine the test flights schedule. The test flights were performed by military pilots according to a precisely scheduled program which included the most common,

Development of Load Spectrum for Full Scale Fatigue Test of a Trainer Aircraft

575

statistically determined, types of sorties. Basing on data recorded during test flights a characteristic flight profile and the test load sequence were determined. The contractor of the Full Scale fatigue Test is the VZLU Praha in Czech Republic. After finalization of the test a Teardown Inspection is scheduled.

3 Flight Program Analysis All the aircrafts operated in Polish Air Force were equipped with digital flight recorders. The main objective of aircraft instrumentation was to analyze the training process. Unfortunately no usage monitoring system, based on the gathered data, was initially developed. Thanks to ITWL flight data from over 40000 flight hours were gathered and could be used for preliminary flight load spectrum determination and analysis [1]. These spectra are represented by nz vs time characteristic. The first problem that had to be solved during the recorded data analysis was to determine time of takeoff and landing. The recorder was not able to show whether the airplane was on ground or in the air. Solution to this issue was developed by analyzing all the available flight parameters, e.q. velocity, height, engine RPM etc. This enabled separation of ground and airborne states. The main parameter taken into consideration in analysis was the Nz factor (g factor). A detailed statistic of Nz exceedances was determined. Since PZL-130 "Orlik" aircrafts are used for pilot training most of the flights are sorties from training program. The number of the sortie that is going to be performed during each flight is given to the recorded before each flight during debriefing. This additional information was important to determine the number of performed types of sorties per statistical hour of flight. During the analysis of recorded data it appeared that many flights (about 30%) have the sortie type number entered incorrectly. In majority those flight were performed not during the regular flight school program, including display flying which are significantly more severe than sorties performed during training. Example results of analysis described above are presented in the table 1 below. Table 1 Percentage of distinguished flight phases [1].

Basic maneuvers 21%

Intermediate maneuvers 13%

Loiter

Spins

Route flights

Other

8%

4%

23%

31%

The mean number of landings, per hour of flight, was determined:

nl = 2.31

landings fth

(1) More touch and go landings were recorded than those ending with full stop. Among the flight phases mentioned above spins and stalling cause severe loads to the structure. Those maneuvers are not distinguished in the Nz factor analysis. Hence in order to determine how often are they performed per statistical hour of

576

A. Leski, P. Reymer, and M. Kurdelski

flight the training program sorties definition along with sorties statistics were used. Mean number of spins per hour of flight was estimated to be equal 0.16.

4 Flight Test Program Instrumentation In order to determine the actual loads acting upon aircrafts structure during flight and on-ground manoeuvres a series of carefully designed test flights were performed by trained military pilots. One of the PZL-130 "Orlik" TC-II trainer aircrafts operated by Polish Air Force was designated for flight tests purpose (No.37). The design and installation of measurement system was carried out by AFIT with close cooperation with EADS PZL-Okęcie. Overall thirteen measurement sections were chosen in which 86 measurement points were installed (due to required redundancy). The total number of measured loads was equal to 27. Three sections were determined in each wing where bending moment, and shear force were measured. In addition torque in wing was measured in section 1. In horizontal stabilizer as well as in fin bending moment along X axis and shear force were measured in only one section each. Moreover there were two fuselage sections (Frame 1 and Frame 9) where bending moments along X and Y axis were measured (in addition torque in Section 9). Two last sections were located on the main gear, where bending moments along X and Y axis were monitored.

Fig. 1 Location of measurement sections [2].

Strain gauge configurations used are shown on Fig. 2. Due to access problems mainly Poisson's half bridges were used with full temperature compensation. For strain measurement in the main spar due to bending separate channels were designated for the upper and lower flange.

Development of Load Spectrum for Full Scale Fatigue Test of a Trainer Aircraft

577

a)

b)

c)

Fig. 2 Strain gauge configurations for different loads (a) bending moment/tension b) torque c) shear force ) [3].

a)

b)

Fig. 3 Aircraft instrumentation. (a) strain gauges (b) recorder modules [2].

578

A. Leski, P. Reymer, and M. Kurdelski

For majority of the measurement sections main and backup strain gauges were installed. High level of redundancy was applied to maximize the reliability of the measurements. Measurements and recording were performed with KAM-500 recorder. During majority of flights sampling frequency of 25 Hz was used. For recording spins and stalling the sampling frequency was set to 400 Hz for strain gauges located in the empennage region. In addition the KAM-500 recorder was gathering signals from accelerometer and flight parameters, like velocity or pitch/yaw/roll angles, from the onboard flight recorder. A laser rangefinder was attached to the left wings lower surface in order to measure descent speed. Signal from this device was also recorded in KAM-500. Exactly the same strain measurement array will be installed by VZLU on the Full Scale Fatigue Test specimen (No.15) according to detailed technical documentation [2]. Calibration After instrumentation and preliminary flight, performed in order to eliminate possible hysteresis in the strain gauges output signals, calibration was carried out. The idea of this process is to exert known loads upon the aircraft's structure and to simultaneously record the response strain in measurement points [4]. Loads were applied by means of jacks and belts tensioned with belt stretchers and monitored with dynamometers. To distribute the loads evenly and make sure to introduce them in the ribs surfaces as well as to prevent possible dent and damage to the structure a number of specially designed clamps were manufactured for the calibration purpose only. Figure 4 shows how loads were applied to the wings and fin. Recorded signals from right wing are presented on the figure 5.

Fig. 4 Clamps and load application during calibration (wing and fin) [5].

Development of Load Spectrum for Full Scale Fatigue Test of a Trainer Aircraft

579

Fig. 5 Calibration of the right wing gauges. Systems response to upward bending moment caused by force F4 [5].

Data recorded during calibration were carefully analyzed in order to verify that the strain gauges work correctly and react to the applied loads in the predicted manner. Changes to the measurement system could be introduced only at this stage, since after any modification the calibration process should be repeated what would be impossible after forwarding the airplane to the Air Force. Experimental flights Flight loads were recorded in three phases [6]. In phases I and II the aircraft was fully instrumented as shown on Fig. 1. For phase III, which is meant to last till the aircraft is withdrawn from operation, only 8 most crucial strain gauges were left (two in each wing, one in left part of the horizontal stabilizer and fin and two in the tail boom). The phases are described below: Phase I – experimental flights performed according to detailed program, Phase II – routine flights performed by the user (autonomous, full instrumentation), Phase III – continuous recording (autonomous, reduced instrumentation). Test flights with full instrumentation were performed according to carefully developed flight program (Phase I). During definition of the program most care was taken to include all the characteristic elements that occur in the regular operation. Sixteen flights were planned and performed. Among scheduled flights we can distinguish: performance flights, mixed sorties flights and two specially designed for spin and stalling loads monitoring. The flights were carried out throughout summer in 2010 by a professional Air Force instructor pilots. In order to determine the following flight stages a special marker signal was introduced which could be triggered by the pilot. This signal was recognized by the KAM-500 recorder and enabled to determine the beginning and end of each planned maneuver. This signal had greatly helped in interpretation of obtained results.

580

A. Leski, P. Reymer, and M. Kurdelski

Phase II was recorded in automatic mode. The recording started and ended autonomously without pilots interference. The preliminary SEWST program schedule assumed long term recording in Phase II. Unfortunately from reasons beyond the Air Force Institute of technology this phase has been strongly reduced. Four flights were recorded (1 flight route and 3 display flights). Phase III is focused on constant signal recording from a group of selected strain gauges during regular operation. A substantial part of the measuring system was removed before phase III started. The results obtained in Phase III were not used for Full Scale Fatigue Test load spectrum determination. The carried out research proved the constructors prediction, that significant vibration occurs just before stalling in the empennage area (Fig.6).

Fig. 6 Flatter in empennage preceding stalling. Presented signal corresponds to the fin's bending moment.

Since these vibrations are characterized by relative high frequency, they cannot be implemented in the standard block of frequency 0,5 Hz. Therefore a separate block was created in which all the jacks, except ones in the tail section, will be stopped at determined neutral position while fin and horizontal stabilizer will be loaded with higher frequency. This will enable to determine fatigue damage that may occur in the rear section due to recorded vibration.

5 Development of Load Sequence Regression equations During the test flights signals from all the installed strain gauges were recorded. After completition of flights in Phase I and II a preliminary signal selection was

Development of Load Spectrum for Full Scale Fatigue Test of a Trainer Aircraft

581

done taking into account their performance. The main cryterion of selection was the stability and repeatability. Linear regression equations determining flight loads by means of registered strains were developed on the basis of data recorded during calibration [7]. The equations were determined using advanced linear regression module in STATISTICA (GRM). In each case the load applied during calibration was chosen as the dependant variable and the corresponding strain signals as the independedt ones. In the first approach signals from all the strain gauges present in the analyzed section were taken into consideration and with use of the best set method and mean square root method the signal with the best correlation was chosen. Due to the chosen calibration methodology only one strain gauge was chosen for determinining each load. However equations for the remaining strain signals corresponding to the analyzed load were determined as well for comparison purposes. Since high level of redundancy was one of the main assumptions it was possible to choose for each load the strain signal with the highest level of correlation with high level of excitation at the same time. Secondly the linear regresion equations coefficents were determined. The obtained value of loads were compared with the recorded ones using statistical methods available in the software. This was an additional source of information to determine whether chosen signals describe the loads correctly within te whole range. Below some of the results used in equation verification are presented. Figure 8 shows how the determined linear regression equation (represented with the trend line) corresponds to the measured values of load (depicted with markers along the line).

Fig. 7 Expected versus observed values of MxSL1 bending moment [7].

582

A. Leski, P. Reymer, and M. Kurdelski

Figure 9 shows a histogram of raw residuals which depicts how the obtained values differ from the measured ones. Since most of the samples are concentrated near the zero value the obtained results seem to be representative and the achieved difference is within reasoneable range.

Fig. 8 Hiatogram of raw residuals [7].

Data filtering The raw data obtained from flight tests needed to be verified for any data loss or strain gauge malfunctions. Since a high level of redundancy was applied it was possible to compare obtained load values from different gauges in order to check whether they are similar and correspond to manufacturers predictions for given values of Nz. After assuring that signals from the chosen set of parameters is reliable throughout the whole research further post processing was undertaken. The goal was to determine characteristic flight load spectrum for approximately 200 Simulated Flight Hours. The duration of load block during the fatigue test was also determined within reasonably boundaries. Hence data reduction leading to achieving about 150 load lines per hour of flight had to be done. Firstly the peak and valleys values were determined using numerical algorithm. Although 27 loads were monitored not all had to be taken into consideration for extremes definition, since generally shear forces extremes will correspond to the bending moments in common sections. This allowed to achieve further data reduction.

Development of Load Spectrum for Full Scale Fatigue Test of a Trainer Aircraft

583

Secondly filtering of amplitude of 10% was performed. Since higher magnitude cycles introduce relatively more fatigue damage to the structure, the low amplitude cycles could (and had to due to block restrictions) be neglected.

6 Summary As a result of the described process a load history from each flight was obtained. Since the final load spectrum corresponds to 200 simulated flight hours the load block had to be constructed by repeating the recorded flights by a determined number of times (resulting from the statistical analysis of the historical data) [8]. Secondly the landing loads and ground maneuver loads were input into the block to simulate the landings between flights (as mentioned above 2,31 landings per hour of flight). The buffeting loads were defined separately due to high frequency. The final load block will have to be additionally tested before the actual Full Scale Fatigue Test. During the preliminary test the strain measurement array installed on the test specimen will also be verified. Since it was installed according to the detailed technical documentation [2] the strain signals recorded during the fatigue test will be used to verify the exerted loads. Finally the loads within the block were ordered in such way, that it is expected to determine fatigue markers within cracks during Teardown Inspection. It is believed, and was already discussed [9], that due to such markers it can be possible to determine when the crack initiated and how fast was it propagating. The whole SEWST program, after a year, is still in it's preliminary stage. The presented research concerning development of the load spectrum is just the beginning of the whole program. Further results will be presented in time, followed by discussion of the results.

References [1] PZL-130 Orlik Flight school program analysis for determiantion of flight spectrum severity, Report No. 134/31/2010, Air Force Institute of Technology, Warsaw, Poland [2] Technical documentation of the strain gauge instalaltion on PZL-130 Orlik TC-II trainer aircraft number 037, Report No. 89/31/2010, Air Force Institute of Technology, Warsaw, Poland [3] Kottkamp, E., Wilhelm, H., Kohl, D.: Strain gauge measurements on aircraft, NATO Advisory Group for Aerospace Research and Development. AGARDograph 7(160) (April 1976) [4] Skopionski, T.H., Aiken Jr., W.S., Huston, W.B.: Calibration of strain-gage installations in aircraft structures for the measurement of flight loads, NACA report nr 1178 [5] Calibration of the strain gauges installed on PZL-130TC-II Orlik trainer aircraft number 037, Report No. 127/31/2010, Air Force Institute of Technology, Warsaw, Poland [6] Development of the flight load measurement procedure for the PZL-130 Orlik TC-II trainer aircraft, Report No. 57/31/2010, Air Force Institute of Technology, Warsaw, Poland

584

A. Leski, P. Reymer, and M. Kurdelski

[7] Development of the approximation equations for determining inner loads within PZL130 "Orlik" TCII aircraft structure on the basis of recorded strain, Report No. 164/31/2010, Air Force Institute of technology, Warsaw, Poland [8] Leski, A.: An Algorithm of Selecting a Representative Load Sequence for a Trainer. In: 2nd International Conference on Engineering Optimization. Lisbon Portugal. CD, pp.1-8 (September 6-9, 2010), http://www.engopt2010.org/ [9] Molent, L., Barter, S.A., White, P., Dixon, B.: Damage tolerance demonstration testing for the Australian F/A-18. International Journal of Fatigue 31, 1031–1038 (2009) [10] Air Force MIL-STD-1530C Aircraft Structural Integrity Program (ASIP) [11] Anderson, I.A., Parker, R.G.: Full Scale Fatigue Test of the Pilatus PC9/A trainer aircraft. In: 20th ICAF Symposium, Bellevue, Washington, USA, July 14-16 (1999) [12] Klimaszewski, S., Leski, A., Zurek, J.: The Role of AFIT in the Polish Aging Military Aircraft Programs. In: Proceedings of 7th Joint FAA/DoD/NASA Conference on Aging Aircraft, New Orleans, September 9-11 (2003) [13] Leski, A., Klimaszewski, S., Kurdelski, M.: The Assessment of Fatigue-Life Resources of the PZL-130 Orlik’s Structure. In: Sixth DSTO International Conference of Health & Usage Monitoring, Melbourne, Australia, March 6-9 (2009)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

ATR Life Extension Project Maurizio Cajani, Roberto Ciotola, Jérémy David, and Jaco Salvi ATR, Blagnac, France

Abstract. Designed in the mid 80’s, ATR 42 and ATR 72 regional turboprops were conceived under Damage Tolerance design approach for a Design Service Goal (DSG) of 70 000 flights or equivalent 25 years. Since the oldest aircraft in the fleet are approaching the DSG, under a joined impulse of the ATR Operations and Commercial Directions, activities have been launched to provide ATR operators with an Extended Service Goal (ESG) of 105 000 flights for both ATR 42 and 72 models. Feedback from the in service world fleet and the results of major fatigue inspections that do not show major fatigue damage are encouraging. The new ESG would provide more flexibility for fleet management as well as an increase in residual value of ATR aircraft.

1 Overview of ATR and Its Fleet Corporate presentation The ATR programme started in 1981, with the establishment of a joint venture between the Italian Aeritalia (today Alenia Aeronautica, part of Finmeccanica) and the French Aérospatiale (today included in EADS), who merged their two separate, but similar, regional aircraft designs into a single project.

The first prototype of the ATR 42 flew in 1984 with certification following the year after. The ATR 72 model followed in 1989. *

Oral presentation.

586

M. Cajani et al.

Fig. 1 MSN 001 on its maiden flight.

An important milestone for the history of ATR was passed in 1995, with the launch into service of the -500 series, resulting from a concern for constant development. These aircraft feature Pratt and Whitney PW 127 engines, the new Hamilton Sundstrand 6 blades 568F propellers and a composite tail, besides new interior design and improved noise control technologies, providing a level of passenger comfort comparable to those of jetliners. ATR is currently working on the launch of its state-of-art new generation aircraft, the -600 series, planned to enter service by mid 2011.

Fig. 2 Evolution of ATR models since the birth of the program.

Aircraft characteristics The ATR family is built around the design of a high-wing, twin turboprop aircraft. The ATR 72 models are derived from the ATR 42 with a 4.5 m stretched fuselage, adding 8 frame bays to the central section 15 of the fuselage. The two aircraft feature a high degree of commonality: the same cross section, simple systems and cockpit for cross-crew qualification. The fuselage shows a traditional metallic structure. Composite materials are used for the vertical and horizontal stabilizers as well as for all control surfaces, fairings and other secondary structure. On ATR 72 models, also the outer wing

ATR Life Extension Project

587

boxes are made of composite material (CFRP monolithic structure), that were at that time the largest composite elements to be used on a civil aircraft. Fuselage and tail are produced by Alenia Aeronautica in Pomigliano d’Arco, Italy. The wing is made in Bordeaux, France, by EADS Sogerma for Airbus. Final assembly, flight-testing, certification and delivery are the responsibility of ATR in Toulouse, France. ATR numbers The following figures refer to January 2011. 915 aircraft delivered, of which 847 currently in service (356 ATR 42 and 491 ATR 72) with 156 operators in 90 countries. Table 1 Current figures of ATR fleet.

total FH

ATR 42 59 717 (MSN 62) 58 613 (MSN 39) 10 336 433

ATR 72 42 169 (MSN 297) 60 939 (MSN 126) 8 294 239

18 630 673

total FC

11 517 263

9 548 314

21 065 577

fleet leader, FH fleet leader, FC

whole fleet

The oldest aircraft, ATR 42-300 MSN 3, is flying since April 1985.

2 Goals for Atr Life Extension Currently, all ATR models have a Design Service Goal of 70 000 flights. This number is valid for a mixed utilisation of 93% short range flights and the remaining 7% of long range flights. These missions were defined as follows during aircraft design: Table 2 Definition of flight missions.

short range long range

duration (min) 40 130

max altitude (ft) 15 000 25 000

With the life extension program, ATR wants to increase the Limit of Validity (LOV) of its aircraft and validate an Extended Service Goal. The final intention is to reach 105 000 flights, that is, to increase the life by 50%. Before reaching such an ambitious result, a first step in the process would validate an ESG of 90 000 fligths. Further extension will depend on the behaviour of the aircraft along this first increment of life. At first, ATR planned to begin with the Life Extension of the ATR 72, which leader has cumulated the highest number of cycles. Due to change in aircraft operator and mission profile, it now appears that the ATR 42 will be the first to reach 70 000 cycles: referring to the table above, ATR 72 MSN 126 was recently

588

M. Cajani et al.

reconfigured as cargo freighter and flies less than 1 000 flights a year, while ATR 42 MSN 39 is operated at a rate of 3 000 flights/year. Which leaves about 4 years time before it cumulates 70 000 FC.

3 Atr Life Extension Programme As required by regulations, the original DSG was validated at certification time with a Full Scale Fatigue Test simulating the aircraft operation for twice the DSG, that is, 140 000 flight cycles. Due to early fatigue damage, the centre wing box was replaced and the cycles count had to be restarted. As a consequence, all other parts have accumulated more cycles (about 185 000).

Fig. 3 ATR 42 – Full Scale Fatigue Test setup.

Extending the life of the aircraft would normally require resuming the Full Scale Fatigue Test (FSFT), in order to accumulate a total of 2 times the ESG. Although conceptually easier, this option is not economically viable, since the costs for resuming the FSFT would drastically counterbalance the benefits of the Life Extension. Besides, tear down inspections were performed at the end of the original tests and the structures were dismantled. Therefore, ATR intends to perform this exercise mainly by analysis. Possible additional testing on small to medium scale (coupons, elements, details) might be necessary to evaluate the ageing of the composite primary structures. A few activity areas were identified in order to develop the Life Extension Program: ANALYSES AND VERIFICATION

• •

analysis of “late damages” found during the Full Scale Fatigue Test (FSFT) analysis of the tear down inspection results on the FSFT article

ATR Life Extension Project

• • •

589

additional tear down inspections on the FSFT article verification of the current maintenance inspection program with up to date calculation tools Widespread Fatigue Damage (WFD) analysis

ON THE FIELD

• • •

in-service experience feedback from entire fleet end of life inspections on selected fleet leaders real aircraft utilisation versus expected one at design time – fleet survey campaign

ADDITIONAL ACTIVITIES

• • •

Life Limited Components review of repair data: SRAS and SRM impact of Life Extension on composite primary structural components

The above points will be described and discussed in the following paragraphs.

4 Late Damage Fuselage As stated earlier, the FSFT for the ATR 72 was programmed for 140 000 simulated flights but actually run for over 185 000 cycles. This allowed evidencing damage that would normally not have been detected. Although it was not necessary to account for these cases in the normal maintenance plan, such damages were recorded and analyzed, in case of aircraft utilisation beyond DSG. This means that corrective actions would apply only after 70 000 flights. To qualify as “late damage”, the 3 following criteria must be fulfilled: • • •

the damage occurred around or after 140 000 simulated flights the damage must not be addressed by an already existing inspection the estimated crack growth rate is low

The non respect of either of the above clauses would qualify the damage as “normal” or “not late”, requiring corrective actions to the normal maintenance plan. The concept of slow crack propagation may be better explained. The past experience of Alenia shows that cracks in typical aeronautical metallic materials grow at a rate of 10-5 to 10-2 millimetres per flight. In other words, it is reasonable to expect growth rates of 0,01 mm every 1 000 flights at the beginning of the crack propagation and of 10 mm every 1 000 flights at the end, when the element is close to failure. Alenia estimated the propagation rate for “late damages” to be of the order of 10-4 mm/flight (0,1 mm every 1 000 flights). As for the fuselage of the ATR 72, tear down and late damages exercises produced 6 new Significant Structural Items and modified 5 others, either by reducing the inspection intervals or by extending the inspected areas.

590

M. Cajani et al.

Again, these repercussions are effective only after 70 000 flight cycles. A similar analysis for the ATR 42 is on going. Sections 11, 13 and 16 are expected to be in line with ATR 72 due to their similarity, whereas some differences could appear on central section 15. Wings Also on the wing of the ATR 72, some damages were found late into the test. More precisely, the expertise has shown that they initiated towards the end of the FSFT. It was found that the cracks initiated in structural zones which are loaded primarily in compression. Crack initiations are therefore attributed to local tensile stress fields. As soon as the cracks reach the compressive layer around the initiation areas, propagation stops. This was verified for all the late damages. No point was thus added to the current list of inspections due to these late damages from the 1st tear down.

5 Tear Down Inspections Overview The tear down inspection consists of a destructive check of the test article by disassembling components and by cutting away the eventual damages and critical zones, in order to perform a final close inspection and to characterize and establish failure causes. The tear down inspection is much more detailed than those periodically performed during the FSFT: the need to quickly resume the test no longer exists and dismantling operations are possible. Even very small damages can be detected, or damages in areas normally difficult to access. Each structural anomaly is analyzed to determine its cause and to define the corrective actions required to demonstrate that the strength, rigidity, damage tolerance and durability design requirements are met. Modifications to the structure or to the maintenance plan can be required. Additional tear down In the frame of the life Extension Program (LEP) of the ATR 72-200, ATR decided to perform an additional tear down on the original test article, in order to verify that no points were overlooked during the 1st tear down on the wing of the test article. Feedback and experience collected during the 25 years that ATR aircraft have been flying were used to guide the inspection. Other criteria include: high local stresses, low static margins, lower calculation confidence due to complex geometry or loading conditions, load redistribution due to damages in adjacent parts… No additional crack was evidenced during this 2nd tear down.

ATR Life Extension Project

591

Tear down for a retired ATR aircraft ATR is evaluating the option of using an out-of-service aircraft having cumulated significant fatigue cycles (at least ¾ of the DSG) in order to perform a tear down on specific structure and acquire useful data to further adjust the requirements to operate up to the ESG. This operation would provide useful data concerning the possible differences between the FSFT simulated flights and real life exploitation. Particular attention will be granted to the structural details that were found damaged during service life. The final decision will depend on estimated costs vs. benefits balance and the availability of a suitable aircraft.

6 Verification of the Current Inspection Programs Since the beginning of the ATR program, back in 1981, our knowledge of fatigue and damage propagation phenomena has somewhat improved, and the associated calculation tools have seen a dramatic evolution. A verification of the results of the calculations performed at certification time with the PSF tool, using the current calculation software called SAFE® was conducted. This new tool can indeed represent more precisely the different geometrical and loading configurations seen on the aircraft. This exercise has been done considering also those structural details that have proven to be critical during the in-service experience, together with the modifications that were introduced since the aircraft certification. Both fatigue and damage propagation verifications were done.

Fig. 4 One of the analyzed structural details (fuel pump cut-out).

592

M. Cajani et al.

Fatigue results: For the ATR 42, SAFE® gives fatigue lives higher or equivalent than the original ones, except for 2 SSI. In the case of ATR 72, the new lives are more often lower than the original ones. The calculated fatigue lives are used to set the threshold of the maintenance inspections, expressed in flight cycles. In order to inspect each SSI at least once in the aircraft lifetime, it was decided that the maximum value of the threshold would be 36 000 FC. Since the new calculated lives are still very high (> 5 times 36 000 FC = 180 000 FC), these have no impact on the current inspection program. Damage Tolerance results: Calculated crack propagation lives are used to establish the inspection intervals. 3 SSI on ATR 42 and 11 on ATR 72 will have the inspection intervals reduced. In addition, a new SSI has been created for both models. This is the consequence of the production process, for which an opening in the wing panel was systematically reworked, leading to the loss of the shot-peening benefits. All these changes in the maintenance plan will be effective only for aircraft flying beyond DSG.

7 Widespread Fatigue Damage Evaluation Overview Widespread fatigue damage (WFD) in a structure is characterised by the simultaneous presence of cracks at multiple structural details that are of sufficient size and density whereby the structure will no longer meet its damage–tolerance requirement. WFD is one of the topics addressed by the regulatory process known as “ageing programme”. It is a phenomenon typical of older structures and is therefore crucial for the Life Extension of ATR aircraft. Two types of WFD are susceptible to generate multiple fatigue cracks: • •

Multiple Site Damage (MSD), the simultaneous presence of fatigue cracks in the same structural element. Multiple Element Damage (MED), the simultaneous initiation and growth of multiple cracks in multiple load path components. This event zeroes the load re-distribution after one component failure; since each component can fail at the same time, the single structural component becomes a single load path element and the fail safe capability is definitively lost.

Analysis method The Airworthiness Assurance Working Group (AAWG) has developed material to address WFD. The following methodology is based upon such material.

ATR Life Extension Project

593

Since WFD is due to the simultaneous initiation and growth of multiple cracks, it is likely to occur when, at the same time: 1. 2. 3.

the geometric features of the structural detail are repetitive, the stress field is homogeneous in the component, the fatigue life of the structural component is close to the DSG

Fig. 5 WFD prone structural details.

The following entities are defined: • • •

WFD average behaviour - NWFD: point in time when 50% of the fleet is expected to reach WFD for a particular detail. Inspection Start Point, ISP: point in time when special inspections of the fleet are initiated due to a specific probability of developing WFD. Structure Modification Point, SMP: point reduced from the WFD average behaviour (i.e., lower bound), so that operation up to that point provides equivalent protection to that of a two-lifetime fatigue test. No airplane may be operated beyond the SMP without modification or part replacement.

To establish NWFD, a deterministic fatigue approach is used, applying a factor 0,9m to the unfactored fatigue life. m is the S-N curve slope of the material. ISP and SMP are derived from NWFD applying scatter factors: • •

ISP = NWFD / 5 SMP = NWFD / 3

ISP

SMP

k1 = 5 1st flight

k2 = 3 specific inspections

Fig. 6 Definition of ISP and SMP.

N WFD

594

M. Cajani et al.

Results The analyses performed show that the structure of ATR is expected to be very little prone to WFD. Fuselage Two items were identified as potentially prone to MSD: the skin under the piano hinge of the cargo door and the rear pressure bulkhead dome. Crack growth analyses were performed, considering as critical length the growth between two contiguous holes. Two dedicated inspection tasks have been issued. In addition, to improve the local design of the cargo door, where WFD was found during the test, a modification was defined which installs a new hinge with two rows of fasteners instead of one. The only items potentially prone to MED are the cargo door latch fittings. Following this result a dedicated inspection task, based on the single load path crack growth analysis, has been issued, considering as critical crack length the failure of the first latch. In addition, a modification was defined, consisting in changing the material of the forward and aft latches, from aluminium to steel.

Fig. 7. The rear pressure bulkhead and its repetitive structural details.

Wings The following locations on the wings were identified as possibly prone to WFD (MSD): • • •

stiffener stops at rib 4 junction of the lower skin panel with the ribs 2 and 12 underwing box attachment fastener lines

ATR Life Extension Project

595

For all three of them, the calculations have shown that for ATR 72-200 and 72210, the Inspection Start Point is higher than 105 000 cycles. This means that no additional inspections are necessary on the wing structure of these aircraft before they reach this number of flights. The tear down inspections performed after completion of the FSFT confirm these results, since no damages were found at the concerned locations.

8 Data from Aircfaft Operating Life Continuous feedback from operators ATR can currently rely on an important in-service experience, collected on specific databases. As prescribed by the regulation, main damage findings and visit reporting are followed up, reviewed and discussed together with the airworthiness authorities, to whom ATR also presents the major findings reported during the 36 000 FC heavy maintenance check (about half the aircraft DSG). Results so far are encouraging: more than 120 aircraft are over 36 000 FC and no major structural damage have been detected. Inspection of fleet leaders One other step towards the certification of the ESG is the Heavy Maintenance Check (HMC) planned on the fleet leaders of the different aircraft models when they get close to DSG. The exact nature of the check is still to be defined, but it will closely resemble the 36 000 FC HMC. Each model will benefit of the experience gathered on the previously inspected ones, so that ground time and total cost can be kept at a minimum. Only structural details which are different or requiring further verification will be considered. To correctly extract the most useful information from the HMC, it is of the uttermost importance to have a clear picture of the aircraft utilisation during the entire operational life. Cruise altitude and flight length are among the most influential parameters, since they greatly influence fatigue and crack propagation. Suitable coefficients for each structural item will be used to calculate the expected behaviour of the inspected aircraft, as compared to the ideal aircraft defined by design. Many aircraft change operator a few times during their life. Some times, operational records are missing or incomplete. This exercise might therefore be a tough challenge. The importance of the mission profiles ATR aircraft were designed based on a mix of two specific mission profiles, a short range mission (40 minutes with a cruise altitude of 15 000 ft) and a long range mission (130 minutes at 25 000 ft). As explained above, the actual utilisation of the aircraft greatly influences the fatigue life and the crack propagation. An example will help understanding the importance of this aspect.

596

M. Cajani et al.

Considering the fuselage, most SSI are sized by the pressurization loads. Each flight corresponds to one cycle of pressurization. In case of a short range mission, at the cruise altitude of 15 000 ft, the corresponding differential pressure acting on the cabin structure is 5 psi = 0,345 bar. For a long range mission, at 25 000 ft, ΔP = 6 psi. Loads (and stresses) are increased by a factor of 1,2, that is, +20%. The propagation speed of a crack can be expressed by the Paris formula:

da = C ⋅ ΔK m dN

(1)

where ΔK is proportional to the amplitude of the load cycle and m is a parameter of the material. For aluminium, which is the typical material of the ATR fuselage, m = 4. Since 1,24 = 2,07, we can see that with stresses increased by 20%, the cracks propagate more than two times faster. Non pressurized areas are not showing such an increase in propagation speed. Actual fleet utilisation vs. design figures ATR was aware that some operators are flying their airplanes in a different way that the one considered for design. To have a clearer picture, ATR promoted a fleet survey campaign, asking operators to provide a series of operating parameters, such as the already mentioned flight duration and cruise altitude, together with block time, characteristic weights and take-off power. flight duration

flight altitude

35

20

30

15

20

%

%

25

10

15

5

10 5

0

0 6

24

42

60

78 minutes

96

114

132

90

120

150

180

210

240

Flight level

Fig. 8 Average utilisation of ATR aircraft: duration and altitude.

Feedback was less enthusiastic than desired. Nevertheless, it has clearly appeared that the average utilisation is rather that of a 70 minutes flight at a cruise altitude of 18 000 ft. However, these results need still to be analysed to obtain a more accurate view of various mission profiles. ATR has informed the airlines that both maintenance plan and Limit of Validity of the aircraft should be adapted to match their actual way of operating the machines. Upon request, ATR can prepare a customized Maintenance Planning Document (MPD). For these aircraft, adaptations on the maintenance program beyond DSG need to be accounted for.

ATR Life Extension Project

597

9 Other Aspects of the Life Extension Program Life limited components Life Limited Components are those structural elements which are designed following a safe life philosophy. Safe life means that fatigue cracking is not probable within the life of the considered structure. At the end of this life, the part must be discarded and replaced. This design concept has been the first adopted to control fatigue phenomena, and has since been replaced by more secure approaches: fail safe, in the late 50’s and damage tolerance in the late 70’s. Nevertheless, it is still used for parts that have very small critical crack lengths or cannot be effectively inspected. The basic principle of Damage Tolerance, the ability to detect damages before they become critical, cannot be used. Examples of safe life structures are the landing gears, as is the case on ATR models. Typically, the history of landing gears does not closely follow that of the aircraft: parts are frequently swapped, replaced, removed for overhaul or maintenance… It is mandatory to separately follow each component, with regards of the number of cycles they perform and on which version of the aircraft. In the frame of the Life Extension, the Time Limits section of the Maintenance Review Board Report (MRBR) will be updated to allow operation beyond DSG. Further investigations may be necessary for the parts with a current life greater than 70 000 cycles: while they would never require replacement with the current LOV, this might no longer be the case for the extended life. Besides, some MSG-3 analyses may depend on the new LOV. A full assessment of them will be performed for any possible repercussions on the maintenance programme. Repairs and modifications analysis In the case of structural modifications or repairs designed by ATR, as mandated by regulations, fatigue and Damage Tolerance calculations and considerations are made in order to grant the continued airworthiness of the airplane. The LOV of 70 000 flights is retained for such evaluations. The following documents will have to be reviewed in the frame of the Life Extension to demonstrate the continued airworthiness up to the new LOV: • • • •

SRM – Structural Repair Manual, which describes all the “standard” repairs SB – Service Bulletins, for modifications and corrective actions on aircraft in service Structural modifications embodied in production (concessions included) SRAS – Structural Repair Approval Sheets, issued for repairs not covered by the SRM

598

M. Cajani et al.

Additional repairs may have been embodied on each aircraft that are not known to ATR: the assessment of those repairs is envisaged to be performed during the Repair Survey tied to Aging Aircraft Structure Programme. Some elements might require more analyses, the addition of supplementary inspections or, less probably, modification. Composites Special considerations must be done for the primary structural elements made in composite materials. These are the vertical and horizontal stabilizers and all the control surfaces on both ATR 42 and ATR 72 models; and, most importantly, the outer wing boxes on the ATR 72. Since their first introduction, back in the 50’s, much work has been done to understand the evolution in time of composites. Experience and testing show that the degradation of the mechanical properties is rather tied to exposure to environmental agents than to load cycles. Absorption of humidity is the main enemy.

Fig. 9 Use of composite materials on the ATR 72.

To correctly estimate such degradation in practical time lapses, composite structures to be tested are aged artificially, by exposure to high humidity and high temperature. Typically, 70 °C at 85% of humidity. Higher temperatures would affect the chemical composition of the materials, giving false results. The mass increase due to moisture absorption is the key indicator of a successful process. To obtain the certification, ATR has performed a series of tests that showed the continued airworthiness of the composite parts. An outer wing box was artificially aged, mounted on a metallic fixture simulating the centre wing box and set up with all those items that are in direct contact with the carbon parts on the finished aircraft. Intentional damages were caused and recorded, to simulate accidental damages likely to happen during production, maintenance or service, for example the dropping of maintenance tools. Both Barely Visible Impact Damages (BVID) and Visible Impact Damages (VID) were considered; the first type for the first

ATR Life Extension Project

599

70 000 test cycles, the second type for an additional number of cycles corresponding to a maintenance interval. The test demonstrated the good behaviour of the structure from the points of view of fatigue and damage tolerance, showing that possible damages do not propagate. What about the Life Extension? Fatigue in composite materials is generally not a concern. Current knowledge shows that the static requirements, duly verified by testing, largely cover the fatigue aspects. This is due to the particular shape of the Wohler (or S-N) curves for these materials. They are very high, meaning that high stress levels, typically higher than static Limit Loads, are required to induce fatigue damage. And they are rather flat, meaning that the total number of load cycles is little influent. One case in which composites are sensitive to repeated loads, is when there are out of plane stresses. For example, stringer run-outs or areas where the number of stacked up plies changes. Composite-metal hybrid assemblies shall be also verified to insure their continued reliability. As for the damage tolerance, the no-growth demonstration under repeated loads of any accidental damage should in principle be confirmed beyond the DSG and up to the ESG, although the chances of a discontinuity in this specific capability of the structure do not seem to be very high. The above considerations are encouraging and we are confident that the composite structures of the ATR will not suffer from the Life Extension. Some testing activities may help to confirm some aspects. One option under evaluation is that ATR acquires an outer wing box from an old and much flown ATR 72, from which test coupons could be extracted. An additional phase of artificial ageing could be performed, up to complete saturation of the material, especially for the thickest parts, yielding to new environmental knock-down factors (EKDF) to be compared with the values obtained at the time of certification. Certification approach Before an ATR is allowed to fly beyond DSG, the ESG must be approved by the aviation authorities. ATR intends to file a request for (non-significant) “major change” with the European Aviation Safety Agency (EASA). Before the official application, a dialog phase with the Agency would allow to fine tune the documents and the evidences to be presented, based on the approach proposed by ATR on one side and the expectations and requirements of the Agency on the other side. The certification authorities are already informed about this on-going project through the biyearly structure airworthiness review meetings. Current hard point in the process is the definition of the suitable means of compliance for the extension of the composite primary parts. The practical implementation of the Life Extension will be proposed as a retrofit modification, to be embodied through a Service Bulletin.

600

M. Cajani et al.

The filing of the Life Extension for the ATR 42 is planned before 2014, which is the year when the first ATR42 should reach 70 000 flight cycles. Consequences With the first step of the Life Extension, ATR aircraft will have a longer life, up to 90 000 flights. 20 000 more flights than before. That is, operators can exploit their ATR almost 30% longer, without having to buy new airplanes. The ATR has proven so far to have a globally healthy and robust structure and we expect this first step of the Life Extension to require minimum extra work for operators. A few structural items, that didn’t need to be inspected up to DSG, will have to be looked at. Some other tasks will be done more frequently, or in a different way: on a larger area, with a different, more sensitive inspection method or through different access panels. Consequently, Life extension up to 90000 flight cycles will be possible with a minimum increase on the maintenance costs. To achieve the entire Life Extension project up to 105 000 flights, all efforts will be done in the next future to prevent from the need of reinforcing some structural elements. The currently early stage of the project does not allow making accurate predictions. It seems reasonable, however, to state that such modifications will be minor in nature, with no high impact on the structural configuration. For example, cold working at few fastener holes or installation of some reinforcing straps. To be practically acceptable, it must be possible to embody these adjustments through retrofit solutions. The feedback from the first phase of the extension, and the inspections at the end of it, will tell us if completing the Life Extension as per the original objective remains economically viable.

10 Conclusions The Life Extension program for ATR is on its way. The priority goes to the ATR 42 since it will be the first to reach the current life limit. Application for certification will be substantiated mostly through analysis and calculations. Some additional testing might be necessary, possibly on composite structures. Exchanges between ATR and the Aviation Authorities will define the exact requirements.

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Benefits of Using a Risk Process in ASIP – The CF-18 Experience Yves Beauvais L-3 Communications MAS (Canada) Inc., Mirabel, Canada

Abstract. This paper will present a Risk Management Process used by the Canadian Forces. This process, in combination with the Aircraft Structural Integrity Program (ASIP), allows fleet managers to address structural issues and to maintain the fleet to an acceptable level of safety while providing extra flexibility to the fleet manager for improved efficiency in the use of maintenance resources and to meet operational requirements. This paper will also provide an example to highlight the benefits of applying such an approach within an ASIP program, based on the CF-18 experience.

1 Introduction The operation of an ageing aircraft fleet presents many challenges for fleet managers. It is especially true in periods such as today where resources are limited and there is increasing pressure to use them ever more efficiently. And while the resources become limited, ageing aircraft develop new, more complex and more frequent in-service issues. Another aspect must be considered in the case of military aircraft. For fleet managers of a military fleet, operational requirements need to be addressed. In many cases, the risks related to airworthiness and/or structural issues must be weighed against the risk of not performing a certain mission, and it may be necessary to accept the airworthiness risk and go ahead with the mission. However, such decision must be based on a rational and structured process that will ensure that any risk that is accepted can be mitigated or will not result in a significant airworthiness impact. Fleet managers need additional tools, other than classical deterministic analytical approaches, to deal with those issues while ensuring that Flight Safety and Operational Readiness are maintained. Such tools must give the managers flexibility to prioritize the problems and to allocate resources to those that will have a significant impact on flight safety while minimizing the efforts spent on less significant items. It must also address complex issues for which data is not available or sufficient to analyse the problem in a more classical manner. A risk-management approach presents significant advantages in such cases. The Canadian Forces (CF) airworthiness regulations recognize the use of such approaches and provide a detailed Risk Management Process [1]. In the case of the CF-18 fleet, this process is used in parallel with the Aircraft Structural Integrity Program (ASIP) to maintain the fleet to an acceptable level of safety while *

Oral presentation.

602

Y. Beauvais

providing extra flexibility to the fleet manager for improved efficiency in the use of maintenance resources.

2 Risk Management Process The Aircraft Structural Integrity Program (ASIP) is the cornerstone of the Canadian Forces Airworthiness approach for the majority of the Canadian military fleets. One of the main objectives of the CF-18 ASIP program is to ensure the safe operation of the aircraft through its operational life. A rigorous approach is put in place for the design, testing/certification and monitoring of the fleet to ensure all potential issues affecting airworthiness (from a structural perspective) are identified and addressed appropriately to resolve them before they can affect safety. However, as the fleet aircraft become older, more and more structural problems begin to surface. Also, given the complexity of modern aircraft (especially fighters), some of these problems are extremely difficult to analyse accurately and require that conservative assumptions be made or that large safety factors be applied to account for uncertainties. Overall, this results in more maintenance requirement to address these issues in a context where resources are limited, which affects fleet availability. One must also consider that in the context of a military fleet, the risk of not performing a mission can outweigh the risks associated with a structural failure of a certain component. To address this reality, the Canadian Forces have developed and put in place a rigorous process aimed at identifying, assessing and mitigating/accepting Airworthiness risks. This process is based on the following key principles: – Balance among operations, costs and aviation safety (airworthiness). – Accept no unnecessary risk/ accept necessary risk. – Accept risk at appropriate management level. – Assess risk continuously. The last two points are important to note. This process is a continuous one and it is not used in isolation. Once a risk item is assessed, it must be reviewed and accepted at a level of authority that is commensurate with the risk level, and representing all stakeholders in the organisation (technical and operational). Also, much like the case of the ASIP approach, a risk item that has been accepted will be reassessed whenever new information is made available. The Risk Management Process consists of five major steps: Hazard Identification, Risk Assessment, Risk Control, Risk Acceptance and Risk Tracking. The first stage in Hazard Identification involves determination of the applicable Hazard Conditions by identifying combinations of hazards that may reasonably be expected to result in an undesirable event. These generally consist of an Initiator and Associated Hazards. The Initiator of a Hazard Condition is the trigger which starts the chain of events leading to the undesirable event (referred to as root cause). Associated hazards are the hazards or conditions that must exist at the same time as the initiator to reasonably expect an undesirable event to occur. The combination may involve one or more of the following: Part failure or malfunction; Adverse environmental condition;

Benefits of Using a Risk Process in ASIP – The CF-18 Experience

603

Design, maintenance, manufacturing or operating error; Incorrect procedure.

All these factors combine to produce an undesirable event, and this constitutes the Hazard Scenario (or scenarios). It is important that all potential scenarios be identified by ensuring that all contributing or related factors are considered. Once this is done, the Hazard Effect(s) is (are) identified. These are generally expressed in terms of the effect on occupant/crew, aeronautical product and/or the environment. This process is illustrated in Figure 1 below.

Part failure

Loss of function Environmental conditions

Initiator

+

Associated Hazards

Effect on occupants, aeronautical product or environment

Undesirable Event

Consequence

Hazard Condition Hazard Scenario Hazard Effect

Fig. 1 Hazard Identification Process.

The next step consists of assessing the severity associated with the Hazard identified previously. This must be done in a rigorous manner, using all available information/tools, such as structural analyses, finite element models, certification test results, etc. The characteristics and categorisation of the component must be considered. For instance, in the case of a Fracture Critical (FC) part (as defined for the CF-18 fleet), a failure at the main load interface of the part could by definition result in the collapse of the structure. Such an item would be classified as a Catastrophic hazard severity level. However, a failure at a different location on the main part, but not on the main load path could be assessed as being of lower criticality (Hazardous or even Major in some cases). On Durability Critical (DC) parts, a failure should not result directly to structural collapse or loss of the aircraft, but the load redistribution could affect other DC or FC parts. Also, the cost of repair is expected to be significant. Such a part would generally be classified as Major. However, if the part is a single load path of a larger DC component (e.g. an aileron hinge), failure could result in the loss of the component and could be assigned a severity level of Hazardous, depending on the evaluation made of the impact of such

604

Y. Beauvais

a loss on aircraft stability and safety. It is important that all impacts be considered, and that input from various specialities (Stress, aerodynamics, flight science, systems, etc.) be included in the decision making process, as required. This assessment must therefore account for factors such as: • Part’s criticality • Load path redundancy • Potential for crack arrest features to stop/slow crack progression • Potential for failure to be undetected and cause collateral damage • Etc. The Hazard severity can then be assigned based on the criteria defined in Table 1 below. Table 1 Hazard Severity Definitions.

Hazard Severity

Definition

Description

Category

Catastrophic

A

All hazard conditions which would prevent continued safe flight and landing. Could result in death of the flight crew normally with loss of the aircraft

B

Hazard conditions that would reasonably be expected to result in a large reduction in safety margins or functional capabilities, including higher crew workload or physical distress such that crew may not be relied upon to perform tasks accurately or completely. Could result in death or major injury to aircraft occupants or major damage to an aircraft system. Could result in death or major injury to ground personnel or the general public.

C

Hazard conditions that would reasonably be expected to result in a moderate reduction in safety margins or functional capabilities, including a moderate increase in crew workload or physical distress impairing crew efficiency. Possible physical distress, including injuries to Occupants or minor damage to an aircraft system.

Minor

D

Hazard conditions that would not significantly reduce aircraft safety, but would reasonably be expected to result in a slight reduction in safety margins or a slight increase in crew workload.

Negligible

E

No effect on safety. Negligible effect on safety margins.

Hazardous

Major

Once the Hazard Severity is assessed, the probability of occurrence of that Hazard is determined at a specific point in time (e.g. now, in 200 flight hours, in 2 years, etc.). This determination will be done using all the tools and information available, which in some cases can be quite limited. This is why the Risk Management Process includes provisions for the use of a qualitative evaluation of the probability of occurrence. The use of such qualitative criteria must be considered with caution. Whenever available, quantitative assessments are preferred. Such evaluation can be based on a statistical

Benefits of Using a Risk Process in ASIP – The CF-18 Experience

605

analysis of in-service occurrences, on certification test data by making assumptions regarding the distribution of failure time and accounting for safety factors in line with the applicable certification regulations, etc. Typically, for the CF-18 fleet, the distribution is assumed to follow a Log-Normal distribution, and the Bullen methodology [2] is used to derive the standard-deviation. From that, the CPOF and the failure rate can be derived. Other tools exist to account for factors that can affect the distribution: • •

Dynamic loading (caused by buffet) that can impact the scatter in results (thus affecting the standard deviation and the failure rate), Usage tracking methodology. The usage of the CF-18 is tracked accurately on the fleet for certain major loadings (wing root bending moment, Nz, etc.). For locations that do not respond well to these tracked parameters, additional scatter can be expected when evaluating the SLL, thus affecting the results in terms of distribution, CPOF and failure rate.

The Hazard probability level is then selected, based on failure rate per hour, in accordance with the information presented in Table 2 below. Table 2 Hazard Probability Criteria.

Description

Frequent

Probable

Remote

Extremely Remote

Extremely improbable

Level

Hazard Probability Thresholds (Per Flight Hour)

Qualitative Definition

Likely to occur frequently

1

Greater than 1 x 10-3

2

Less than 1 x 10-3

Expected to occur one or more times

3

Less than 1 x 10-4

Unlikely, but possible to occur

Life of individual Aeronautical product Expected to occur frequently during the operational life of an individual aircraft Expected to occur one or more times during the operational life of an individual aircraft Unlikely, but possible to occur during the operational life of an individual aircraft

Life of Entire Fleet

Occurs continuously to the entire fleet

Likely to occur several times per year to the entire fleet

May occur one or more times per year to the entire fleet May occur one or more times during the entire operational life of the entire fleet

4

Less than 1 x 10-5

Not expected to occur

Not expected to occur during the operational life of an individual aircraft

5

Less than 1 x 10-7

So unlikely, it may be assumed that it will never occur

So unlikely, it may be assumed that it will never occur during the entire operational life of all aircraft of the type

606

Y. Beauvais

Finally, the Hazard is assigned a Risk Index based on its severity and its probability as presented in Table 3 below. At that point, the Risk Assessment is completed, and a course of action must be selected and implemented based on the risk index. For risks that are within the Acceptable Level of Safety, as shown in Table 3, no action is required and the item is simply monitored to verify that no new information becomes available that could affect the initial rating. Otherwise, a mitigating action must be selected and implemented before the point in time at which the risk becomes unacceptable. Table 3 Hazard Risk Index HAZARD SEVERITY

A

B

Catastrophic

Hazardous

Frequent

A1 Extremely High

Probable

CATEGORY C

D

E

Major

Minor

Negligible

B1 Extremely High

C1 Medium

D1 Low

E1

A2 Extremely High

B2 High

C2 Low

D2

E2

A3 High

B3 Medium

C3

D3

E3

A4 Medium

B4

C4

D4

E4

A5

B5

C5

D5

E5

PROBABILITY 1 L

2

E V 3

Remote

E L

4

Extremely Remote

5

Extremely Improbable

Depending on the Risk Index, acceptance of the risk will be done at a higher echelon in the chain of command of the fleet, both on the Technical Airworthiness side and on the Operational side. Acceptance will be granted based on the level of airworthiness risk compared to operational requirements, and also based on the proposed Risk Control measures (mitigation of risk through inspections, flight limitations, etc.). A risk item is then monitored until a definitive correction action is implemented to remove the airworthiness risk. Logistic Risk Assessment The process described in the previous section accounts for Airworthiness aspects. However, even in cases where the Airworthiness risk is deemed acceptable, there could be further considerations that would affect the operations of the fleet. In such cases, if damage is often found in-service at a certain location, the cost and the down-time associated with repairs could become prohibitive even if the overall impact on flight safety is within Acceptable Level of Safety.

Benefits of Using a Risk Process in ASIP – The CF-18 Experience

607

To address such situations, a Logistic Risk Assessment process was put in place for the CF-18 fleet. This process follows a similar approach to that of the Airworthiness Risk described previously. However, in this case the parameters being considered are related to the cost and schedule impact of an occurrence, whereas the probability is expressed in terms of cumulative probability of failure (CPOF) at a point in time as opposed to failure rate per flight hours. This Logistic Risk approach is used to determine if an “Acceptable” airworthiness risk can also be managed efficiently from a logistic point of view. In the case of an item which is not acceptable from an airworthiness point of view, the Logistic Risk approach is aimed at comparing the various options available (inspection, repair on condition, preventive modification, etc.) to select the one that is the most effective in terms of cost and downtime. Benefits of the Risk Management Process on CF-18 Fleet Management The Risk Management Process is a powerful tool for the CF-18 fleet managers. It provides them with the flexibility to address structural issues while maintaining fleet operations to the required level. The benefits for the fleet managers include: • Providing for a more thorough assessment of the risks associated with structural failures: o Prioritize further than FC, DC, at the critical area level, o Assess risk at both the critical area and assembly level, accounting for structural redundancy, o Recognize potential effect of failure of less critical structure, e.g. secondary, on systems that can lead to higher risk than originally anticipated. • Prioritising actions to more significant issues. o Better utilisation of resources. For example, more efforts can be expanded on analysing and developing corrective measures (structural modifications) for critical issues, while less significant items are inspected and repaired on condition, if required. • Providing flexibility to maintain operations while structural issues are analysed and solution are developed. o Provide rational process for monitoring issues and continuing operations during that period. o Issues are identified, monitoring/mitigating actions are put in place (when required) and residual risk assessed and accepted at proper level. • Providing tool for selection of most appropriate maintenance action for a particular issue.

608

Y. Beauvais

3 Example: CF-18 Vertical Stabiliser Stub-Frames One particular case where using a Risk Management approach provided significant benefits to the CF-18 Fleet managers is the area of the Vertical Stabiliser Stub-Frames. The six (6) Vertical Tail Stubs are located in the AFT fuselage and serve as the junction from the Vertical Tail to the Fuselage. The Stubs protrude from the Quarter Frames at stations Y598, Y590, Y580, Y574 and Y566, and the Bulkhead at Y557. Refer to Figure 2 for the location of the Vertical Tail Stubs. Note that there are two main configurations for the Canadian fleet at that location; in the first one the various flanges are thinner than in the later models. For the purpose of this article, only the thin configuration is considered.

Fig. 2 Location of Vertical Tail Stubs.

Cracking occurrence on the Vertical Tail Stubs is a well-known problem and has been noted on OEM and IFOSTP tests as well as in the Navy, CF fleet, etc. The most frequent cracking occurs under the flange holes and is referred to as “fretting cracks”, as it was originally believed that fretting was the main cause of cracking. Although the occurrence of the phenomenon is well documented, the mechanism and the variables favouring the development of these cracks are now thought to be attributed to many factors (such as the effect of contact stress, clamp up force, surface finish, shimming, dynamics, etc.). Given the complexity of the

Benefits of Using a Risk Process in ASIP – The CF-18 Experience

609

failure mechanism, analytical prediction of the crack initiation is difficult. As such, no reliable methodology or analytical tool exists to quantify and predict “fretting crack” initiation using analytical models/software used for classical fatigue crack initiation prediction. However, a significant number of base and depot level inspections have been performed on the stubs specifically for fretting crack indications. The results of these inspections, although limited in accuracy and level of detectability, currently provide the best means of estimating an average “time to failure”, with the use of statistical models. Other modes of failure on the Stubs have occurred on Fatigue Tests, as shown in Figure 3. These modes of failure were also considered, although their failure rate is expected to be lower than that of the “fretting cracks”.

Fig. 3 Vertical Tail Stubs Failure Modes.

Loading Mechanisms The main loading affecting the area comes from the Vertical Tail root bending moment and torque. Also, as mentioned earlier, there is significant dynamic loading in the area, caused by the interaction of aerodynamic vortices generated at the wing leading edge extension at high angle of attack. This dynamic loading contributes to a majority of the total fatigue damage, and tends to result in a significantly larger scatter in fatigue life, while making testing and analysis much more complex.

610

Y. Beauvais

Evaluation of Safe Life Limits and Risk Assessment The Safe Life Limit (SLL) at the stub flange holes is determined mostly through a statistical analysis of test and in-service results. As mentioned above, the dynamic nature of the loading environment makes it very difficult to derive a complete and representative loads sequence. However, since this problem is relatively widespread, and occurs relatively early in the fleet life, there are numerous in-service occurrences that can be used to derive a representative evaluation of the Safe Life Limit. At this point in time, the in-service defects consist of cracking that is still small enough to be in a stable crack growth regime. The in-service data can then be processed, using statistical tools to get a good estimation of the mean life and the standard deviation. Fatigue test data from various sources also provide fractographic evidence used to calibrate a crack growth model. This model was in turn used to estimate the life at which point cracking becomes unstable. Using that data in the statistical model, along with the safety factors mandated by the applicable airworthiness regulations, provides the estimation of the SLL. A safe life limit (SLL) was calculated for every individual failure type, configuration, and Stub. The analysis has shown that some locations/failure modes are full life and therefore do not require further effort. However, many locations are not full life and will require maintenance action in the near future. The modifications are in development and are planned for the next major downtime. It is required to assess whether airworthiness risk is acceptable until then, since any requirement to implement a modification for this problem outside of planned downtime would have a tremendous operational impact in terms of aircraft availability. The Risk Management Process was therefore used to see if the problem could be managed while maintenance actions are developed and implemented, while maintaining an acceptable level of safety. The hazard severity for individual Stub failure was considered to be Major, given the redundancy of the attachment and the part criticality of the Stub frames. A failure of one stub will overload other stubs and will reduce the integrity of the adjacent stubs in transferring the V-Tail loads to the aft-fuselage. As the six V-Tail Stubs are sharing the V-Tail loads together, and as the failure of more than one Stub will lead directly to V-Tail departure (for most permutations), risk levels associated with combination of Stubs failures were examined. The hazard severity associated with the V-Tail departure was considered Hazardous. The Cumulative Probability of Failure (CPOF) for each item is calculated using a Log-Normal Distribution as documented in Ref. [3]. When combining CPOF of all potential failures for a stub some simple statistical concepts were applied, depending on whether the various critical points were in series or in parallel. In a series configuration, a failure of any component results in failure for the entire system. The reliability function of the system can be expressed as: R(s) = R(1)*R(2)*R(3)*…R(n)

(1)

Benefits of Using a Risk Process in ASIP – The CF-18 Experience

611

Where R(s) is the reliability function of the system R(i) is the reliability function of unit i ( i = 1, 2,…n) n is the number of units in the system And knowing that R(t) = 1-CPOF(t). We obtain CPOF(s) = 1-(1-CPOF(1))*(1-CPOF(2))*(1-CPOF(3))*…(1-CPOF(n))

(2)

On the other hand, when combining parallel units, the CPOF of a group of units is simply the multiplication of all individual CPOF. Then, as presented in Ref. [3] the failure rate is calculated using the following equation: Z(t) = f(t)/(1-CPOF(t))

(3)

Where f(t) is the probability density function. When combining the failure rates of units forming a mechanical system, the equation depends on the system being of series or parallel units. The failure rate of a combination of events for which any event could cause a failure of the system is simply the addition of all individual failure rates (Ref. [4]). Z(s) = Z(1)+Z(2)+Z(3)+…Z(n)

(4)

For the failure rate of a parallel system it is easier to treat the problem using the reversed failure rate. The reversed failure rate of a parallel system is simply the addition of all individual reversed failure rates: ρ(s) = ρ (1)+ ρ (2)+ ρ (3)+… ρ (n)

(5)

Where ρ(i) = f(i)/(1-R(i)), is the reversed failure rate of unit i, i=1...n Finally we can obtain the failure rate using the following equation (Ref. [3]): Z(t) = CPOF(t)* ρ (t)/R(t)

(6)

These combinations were used to evaluate the failure rate and CPOF for each stub individually. To assess the failure rate for multiple stub failures leading to a Vertical Tail departure, all the permutations of stub combined failures were evaluated. For instance, the failure rate of two specific Stubs fracturing at the same time can be evaluated, considering the following: • • • •

F1 and F2 are the total CPOF per side of the given Stubs at 6300 ETH, Z1 and Z2 are the total failure rates per side of the given Stubs at 6300 ETH, Z1∩2 is the combined failure rate of two simultaneous events at 6300 ETH. Where 6300 ETH is the target life in Equivalent Test Hours (ETH) for the CF-18 aircraft retirement.

612

Y. Beauvais

The definition of failure rate, Z is:

Z=

d (CPOF ) PDF = dt 1 − CPOF 1 − CPOF

(7)

Therefore, 1

,

1

(8)

Now, 1

2

Therefore, (9) Applying the chain rule for the derivative in the numerator, (10) Substituting in equation 8 and recognizing that there are two (2) Vertical Tails per A/C, 2

(11)

Every permutation is considered and the failure rate and CPOF were obtained to form a global risk assessment for the Vertical Stubs. A/W risk levels at the next major maintenance downtime as well as at fleet retirement were evaluated for both groups of failures: the non-fretting group (fillet, hole and kick) and the fretting group. The overall conclusion of this Risk Assessment exercise was that with an inspection program put in place to monitor the “fretting cracks”, the overall risk level at the Vertical Stabiliser Stub Frames is within Acceptable Level of Safety until the next planned major maintenance downtime. At this point in time, maintenance actions will be implemented to correct the situation until fleet retirement.

4 Conclusion This paper presented the Risk Management Process used by the Canadian Forces, in support of the CF-18 fleet Aircraft Structural Integrity Program (ASIP). This process is aimed at providing the CF-18 fleet managers with a rational and

Benefits of Using a Risk Process in ASIP – The CF-18 Experience

613

rigorous tool for the identification, evaluation, acceptance, mitigation and control of potential risk issues. Such a tool is an important asset in situations where flexibility is needed to maintain fleet operation while these potential issues are reviewed and corrective maintenance actions are developed. The case of the CF-18 Vertical Stabiliser Stub frames is a good example showing how this process can be used to ensure the continued safe operation of the fleet while a structural modification concept is developed.

References [1] DG01.003, Airworthiness Risk Management Process, Version G, Department of National Defence, Canada (2010) [2] Bullen, N.I.: A Note on Test Factors, Ministry of Aviation Aeronautical Research Council Reports and Memoranda, R. & M. No. 3166 (September 1956) [3] SES DI 3118, Statistical Methodologies Reference Guide, Rev. Basic, L-3 MAS (March 28, 2007) [4] Finkelstein, M.: Failure Rate Modelling for Reliability and Risk. Springer, Heidelberg (2008)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Integrated Probabilistic Analysis of Damage Tolerance and Risk for Airframe Structural Locations W. Hu and R.F. Torregosa Air Vehicles Division, DSTO, 506 Lorimer St, Fishermans Bend, Australia 3207 [email protected]

Abstract. A new method for the integrated probabilistic analysis of damage tolerance and risk of failure that considers the effects of material property scatter and the load history effect on crack growth rates has been developed. The underlying damage tolerance analyses were conducted using a plasticity-induced crack closure model, with key inputs generated using the Monte Carlo method. The statistical behaviours of the following variables were modelled: the initial crack size, the peak stress of the load spectrum, the fracture toughness, the parameters defining the fatigue crack growth rate and the threshold stress intensity range. Each random variable was defined as either (i) a random variable with a normal, log-normal or Weibull distribution, or (ii) as a deterministic variable with a single specified value. Risk analyses were conducted on two structural elements to demonstrate the use of the new method. The results showed a lower probability of failure compared to those obtained using the master crack growth curve method. It is envisaged that by capturing the material property scatter and the history effects the proposed approach would provide an improved probabilistic risk assessment for aircraft structures.

1 Introduction As the number of ageing aircraft worldwide grows, the probabilistic risk assessment of aircraft structural elements is becoming increasingly important, as a complementary tool to damage tolerance analysis, for the safe operation and management of aircraft fleets. The main objectives of these risk analyses are to determine the risk levels of the aircraft in a fleet, assess the impact of inspection intervals on safety and cost of operation and evaluate the relative cost of inspection plus repair versus modification or retirement [1]. This is done by predicting the probability of failure of structural elements by accounting for the stochastic characteristics of their initial quality state, material properties, geometry, loads, inspection and repair. One of the key inputs to probabilistic risk analysis is the distribution of crack lengths at any given flight hour. The risk analysis method specified in MIL-STD1530C requires that the probability of failure be evaluated for each individual population of details that are governed by a single damage tolerance analysis, and the distribution of crack lengths at different flight hours is determined according to *

Oral presentation.

616

Y. Beauvais

the distribution of the initial crack lengths and the crack growth curve obtained from that damage tolerance analysis [2]. The crack growth curve may need to be expanded by extrapolation to cover the small cracks arising from the distribution of the initial cracks. While this approach simplifies the analysis, it also implies that each initial crack will grow according to a master crack growth curve, represented by the solid curve in Figure 1 (a), to obtain the crack lengths at any specified time, hence the crack length distribution at that time. This approach, referred to as the master CG curve approach in this paper, therefore, does not account for the stochastic nature of fatigue crack growth. Using this approach, any two cracks of the same initial crack size will have the same crack lengths at any specified time, given identical loading. Experimentally, it is well known that even under carefully controlled laboratory conditions, noticeable scatter exists in crack growth rates, i.e., cracks of the same initial size can have significantly different crack growth lives, as indicated by the dashed curves in Figure 1 (a). This scatter is caused by the variability in the manufacturing process, heat treatment, surface finish of the material, and the variation in applied load, which can only be described with a statistical model. Figure 2, extracted from [3], clearly illustrates this variability. More examples may be found in [4]. Another implication of the master CG curve approach is that the history effect on crack growth is neglected. All the crack growth curves, emanating from initial cracks of different sizes, will be assumed to be geometrically similar, e.g., in Figure 1 (b) the crack growth curve from an initial crack size of a2 would overlap with the master curve if translated horizontally to point A. Experimentally it has been shown that the crack growth curve for the initial crack a2 is more likely lie somewhere between the two dashed curves, due to the different load histories (plastic events) experienced by cracks a1 and a2 . Retardation models, based on either the plastic zone ahead of the crack tips or the residual plastic deformation behind the crack tip, can reasonably capture these characteristics, but they are all history-dependent.

a1

Master curve Crack length, a

Crack length a

Master curve

CG rate scatter

a2

a1

A History effect

time

time

(a)

(b)

Fig. 1 Crack growth variability caused by material property scatter and history effect.

Benefits of Using a Risk Process in Asip – The CF-18 Experience

617

Fig. 2 Crack growth data for 38 replicate tests (Figure 22 of [1]).

In this paper, we propose a new method to couple the probabilistic analyses of damage tolerance and risk, by taking into consideration the history effect and the scatter in crack growth rates. A random variable method [5] is applied to the underlying plasticity-induced crack closure model to generate more realistic crack length distributions at specified times, and these distributions are then used for the evaluation of the probability of failure. These distributions are considered to be more realistic because they can at least qualitatively represent the observations shown in Figure 2, while using the master CG curve approach this is not feasible.

2 Probabilistic Crack Growth Model There are generally two approaches to modelling fatigue crack growth stochastically. These are the random process approach and the random variable approach [5]. In this study, we have used the random variable approach to represent the stochastic behaviour of fatigue crack growth. A deterministic differential equation was used for the crack growth rate, but the parameters defining the crack growth rate were considered to be random variables. This particular model was previously studied in the context of damage tolerance analysis [6], and the salient points are detailed below. We also mention the modifications to the algorithms in order to interface with the risk analysis computer code. The plasticity-induced crack closure model developed by Newman in [7] was used as the deterministic crack growth rate equation,

da / dN = C (ΔK eff ) m G / H

(1)

where C is the crack growth rate coefficient and m is the crack growth rate exponent. Note that both C and m are material constants. ΔK eff is the effective stress intensity range. G is a function of the threshold stress-intensity range and the effective stress-intensity range, i.e.,

618

Y. Beauvais

G = 1 − (ΔK 0 / ΔK eff ) p

(2)

where p is a material constant, and ΔK 0 is the threshold stress intensity range. If the effective stress intensity range is below the threshold stress intensity range, no crack growth takes place; hence it is a parameter to be determined experimentally, and depends on the material and the stress-ratio. If the effective stress intensity range approaches the threshold, the function G diminishes, thus reducing the crack growth rate to zero, simulating the threshold phenomenon. Conversely, H is a function of the maximum stress-intensity factor and the cyclic fracture toughness, defined as,

H = 1− (K max / C5 )q .

(3)

Here, q is a material constant, K max is the maximum applied stress intensity factor, and C5 is the cyclic fracture toughness of the material. Under normal fatigue loading, K max is much smaller than C5 so that H is approximately one, but when K max approaches C5 , the function H approaches zero, thus causing the crack growth rate to approach infinity, which simulates the process of fast fracture. In this paper, six variables that affect crack growth were modelled probabilistically. These variables were: the equivalent initial flaw size (EIFS), the peak stress of the spectrum, the crack growth rate exponent ( m ), the crack growth rate coefficient ( C ), the fracture toughness of the material and the threshold stress intensity range. Equivalent initial flaw size

The vital role of EIFS variation played on fatigue crack growth has long been recognized. In studies, such as [8, 9], all the uncertainties in fatigue crack growth rates were attributed to variations in EIFS values. In this study, for the purpose of demonstration, normal and lognormal distributions are used for EIFS data. Realistic EIFS distributions were developed in [10] based on teardown data. Maximum load of the spectrum

Normally, damage tolerance analyses assume a deterministic load spectrum, with a clearly defined maximum load and spectrum. In reality, load spectra are derived from flight parameters or from measured strains, both of which are prone to instrument noise, measurement inaccuracies and instrumentation failure. As a result, the magnitude and number of load cycles in a spectrum should be considered to be random variables. To simplify the analysis, the random variation of the spectrum was encapsulated in the maximum load of the spectrum. The effect of load variation on fatigue crack growth rates is signified by Eqn. (1) through ΔK eff .

Benefits of Using a Risk Process in Asip – The CF-18 Experience

619

Crack growth rate coefficient C and exponent m

Analyses of experimental data show that the parameters C and m in Eqn. (1) show a degree of dependence [11, 12]. An independent variable, ξ , was therefore introduced to model this dependence, which was of the form log10 C = ξ (B + P log10 m )

(4)

where B and P are regression parameters that correlates the dependency of log10 C with log10 m . Fracture toughness KIC

Similarly, an independent variable ζ was introduced to model the dependence of the fracture toughness data on m , log10 K IC = ζ ( A + Dm )

(5)

where A and D are regression parameters that correlates the dependency of log10 K IC with m . The regression parameters B, P, A and D are therefore additional parameters required in the in-house computer code CGAP [13] for carrying out a probabilistic assessment. Note that the crack growth rate coefficient C can be made independent of m by setting P = 0 , and similarly, the critical stress intensity factor K IC can also be made independent of m by setting D = 0 . It has been demonstrated in reference [11] that treating correlated variables as independent random variables can be detrimental to the result of probabilistic fatigue life assessments. It is, therefore, crucial that any correlation between variables be simulated accurately. The above probabilistic crack growth model was implemented in CGAP using the Monte Carlo method. Vectors of random variables { ci , S p , ξ , m, ζ , ΔK th } were generated repeatedly, and the subsequent deterministic problem was solved. In order to allow the exploration of different distributions for different variables, each random variable was allowed to assume one of four distributions: normal, lognormal, Weibull and a distribution defined by a cumulative distribution function given in tabular format. For example, instead of fitting the EIFS data to a known distribution, the data derived in [10] was used directly. In this case, the random variable was generated by considering that for a given cumulative distribution function F (x) and a random variable u uniformly distributed over (0,1), then F ( x) = u , or x = F −1 (u ) . In addition, each variable could be made deterministic, to allow easy verification of the computer code. In order to interface with the risk analysis computer code, which requires the probabilistic distribution of crack length at a given flight hour, CGAP was modified to output crack lengths, relative frequency and cumulative distribution for specified flight hours.

620

Y. Beauvais

3 Probabilistic Risk Analysis In this paper, risk specifically refers to the probability of failure in the form of unstable fracture of an aircraft structural element with existing cracks subjected to spectrum loading. Denoting the distribution of crack size at the beginning of kth flight as f (a ) , and the probability distribution function for the peak stresses during the kth flight as g (s ) , then the probability that unstable cracking will occur considering all possible crack sizes is, e.g., [14] ∞

PoF =

∫ 0

⎛ s RS (a ) ⎞ f (a)⎜⎜1 − g (s )ds ⎟⎟da , ⎜ ⎟ 0 ⎝ ⎠

∫

(6)

where S RS (a) is the residual strength of crack a corresponding to the fracture toughness of the material, i.e., ⎛ K IC S RS (a ) = min ⎜ S ys , ⎜ β (a ) πa ⎝

⎞ ⎟. ⎟ ⎠

(7)

Here s ys and K IC are the yield strength and the fracture toughness of the material, respectively, and β (a) is the geometry correction factor in the stress intensity solution. The probability of failure defined in Eqn. (6) has been used for risk analysis of a C-130 structural element, using a fixed crack growth curve [10]. In that study, the distribution of the equivalent initial flaw size (EIFS) was developed from observed crack data. In the current paper, the same EIFS distribution, in the form of tabular cumulative probability, will be used.

4 Results and Discussion To illustrate the application of the integrated analysis of probabilistic analysis of crack growth and risk of failure, we consider two structural elements, both of rectangular shape, with dimensions of width 0.4 m, height 0.6 m and thickness 0.01 m. The first element is similar to a central crack tension (CCT) specimen and is representative of cracks emanating from slots or oval holes contained in structural components. The second element is a plate with a central through-thickness hole of diameter 0.007 m. The material is aluminium alloy 7075-T6, which is widely used in transport and surveillance aircraft. This material has a Young’s modulus of 71 GPa, a yield strength of 468 MPa, an ultimate strength of 538 MPa and a fracture toughness of 50 MPa√m. Based on available test data the mean values of the parameters defining the crack growth rate equation (1) are assumed to be C = 4.5 ×10−10 , m = 3, C5 = 50, and p = q = 2 . The load spectrum used in

Benefits of Using a Risk Process in Asip – The CF-18 Experience

621

this study was representative of those applied on transport aircraft at a lower wing surface panel. To reduce the number of random variables, the maximum stress was fixed and the threshold stress range was deterministic and only the variability of the EIFS, coefficient C , exponent m , and K IC were considered. We further restricted the distribution of C and m to normal distributions, through ξ and ζ in Eqn. (4) and (5) respectively. Central Crack Specimen

The mean value of the initial crack size was assumed to be a=8 mm and the applied maximum stress was 120 MPa. Using these mean values and the above mentioned crack growth rate parameters, a single crack growth curve was generated, as shown in Figure 3 (a). The exceedance frequency of the load spectrum is plotted in Figure 3 (b), where the peak stress was extrapolated beyond 120 MPa. The analytical solution for the geometry correction factor β = 1 /(cos(πa / 2W a / t )) was used to derive the residual strength curve. Based on these data, the probability of failure was then computed using Eqn. (6). This is shown by the thin green curve Figure 4 (b). Probabilistic crack growth analysis was then conducted using the Monte Carlo method. The EIFS distribution was assumed to be normal, with a mean of 0.008 and a standard deviation of 0.0005. The Monte Carlo simulation was run 20,000 times, to obtain the crack length distributions at 20 nominated flight hours (controlled through the number of cycles). Figure 4 (a) plots the cumulative frequencies of crack lengths at three time instances. These distributions were then used in Eqn. (6) to calculate the probability of failure, which is shown by the thick black curve in Figure 4 (b).

0.06

0

10

ai=8 mm C=4.5x10 m=3 C5=50 C6=2 C3=C4 =0

0.05 0.04 0.03

Exceedance Probability

Crack Length, m

-10

0.02 0.01

10

-2

10

-4

10

-6

10

-8

10

-10

10

-12

-14

0.00 0

20000

40000

Flight Hours

(a)

60000

80000

10

0

50 100 150 200 250 300 350 400 450

Peak Stress, MPa

(b)

Fig. 3 (a) The master crack growth curve for the CCT specimen; and (b) the peak stress exceedance probability for the load spectrum.

622

Y. Beauvais

0.8 Flight Hours Current Method 42 8453 21132 Master CG Curve 8453 21132

0.6 0.4 0.2

Probability of Failure

Cumulative Distribution,ai

10 1.0

0.0 0.004 0.006 0.008 0.010 0.012 0.014 0.016

0

-2

10

-4

10

-6

10

-8

10 10

-10

10

-12

10

-14

10

-16

CCT specimen Master curve method Current method

0

Crack Length, m

40000

80000

120000

Flight Hours

(a)

(b)

Fig. 4 (a) Cumulative distribution of crack lengths at different flight hours and (b) the corresponding risk of failure curve for a CCT specimen.

Plate with a Central Hole

For the second example, we consider a plate with a central hole which has a single crack emanating its edge. The mean value of the initial crack size was taken to be 1 mm and the applied maximum stress was 120 MPa. These values, together with the crack growth rate parameters given above, were used to generate the single crack growth curve shown in Figure 5 (a), and this curve is then used in Eqn. (6) to compute the probability of failure. The thin green curve in Figure 6 (b) represents the probability of failure. The probabilistic crack growth analysis was performed by assuming a lognormally distributed EIFS, with a mean a -4.83 and a standard deviation of 0.02. The mean is chosen so that the corresponding gross crack length, measured from the centre of the hole, is 0.008 m, giving a net crack length of 0.001 m. A total of 20,000 Monte Carlo simulations were run to obtain the cumulative distribution of crack lengths at different flight hours. Figure 6 (a) shows three examples of these distributions. Figure 5 (b) plots the geometry correction factors extracted from [15] and the corresponding residual strength curve computed for the geometry and the material. 0.16

ai=0.001 m R=0.007 m -10 C=4.5x10 m=3 C5=50 C6=2 C3=C4=0

0.10 0.08 0.06

3.0

0.04 0.02 0.00 0

40000

80000

Flight Hours

(a)

120000

500 2.5 400 2.0 300

1.5

200

1.0

100

0.5

0

0.0 0.00

0.05

0.10

0.15

Residual Strength SRS

0.12

Geometry Correction Factor β

Crack Length, m

0.14

0.20

Crack Length (from the centre of the hole), m

(b)

Fig. 5 (a) The master crack growth curve, and (b) the geometry correction factor and the residual strengths for a plate with a central hole.

Benefits of Using a Risk Process in Asip – The CF-18 Experience

0.8 Flight Hours Current Method 42 8453 21132 Master CG Curve 8453 21132

0.6 0.4 0.2 0.0 0.006

0.008

0.010

0.012

Crack Length, m

(a)

0.014

Probability of Failure

Cumulative Distribution, a

10 1.0

623

0

-2

10

-4

10

-6

10

-8

10 10

-10

10

-12

10

-14

10

-16

Master curve method Current method

0

50000

100000

150000

200000

Flight Hours

(b)

Fig. 6 (a) Cumulative distribution of crack length at different flight hours, and (b) the corresponding risk of failure curve for a plate with a central hole.

Discussion

As shown in Figure 4(b), the master CG curve approach produced higher probability of failure for the CCT specimen, even when the mean values of the crack growth parameters were used. If more conservative values are used, the probability of failure will be even higher. The method developed in this paper, by considering the scatter in crack growth rates and the history effect of the load, generated crack length distributions for different flight hours which are believed to be more representative of the real situation. This resulted in a lower probability of failure, as shown by Figure 4(b). It should be noted that the difference in the probability of failure is more pronounced when the life gets longer, as the difference in crack length distributions from the master CG curve approach and the current method is amplified. The results plotted in Figure 6 (b) shows the same trend for the second case. Noted also that while the above observation was made based on the two cases studied here, this may be a general trend. As shown in Figure 4 (a) and Figure 6 (a), using the master CG curve method resulted in crack length distributions that have larger mean values, which leads to higher probabilities of failure. A comparison between the green curves, with and without symbols, in Figure 6 (a) shows that the smaller initial cracks grow much faster than for the master CG curve approach than for the current method. Therefore, the lower probability of failure observed for the two cases may not be coincidental. This may help to alleviate the concern that the probability of failure obtained from the master CG curve approach is overly conservative [16].

5 Conclusion A new method was developed to integrate the probability analysis of damage tolerance and risk of failure. Instead of using a single deterministic crack growth

624

Y. Beauvais

curve, the crack growth was modelled statistically using a random variable approach. The Monte Carlo method was used to simulate the stochastic nature of crack growth as represented by the random variables including the equivalent initial flaw size, the maximum load in the spectrum, the crack growth rate coefficient and exponent, the fracture toughness and the threshold stress intensity range. The results for the two cases studied showed lower probabilities of failure compared to those obtained using the master crack growth curve method. This is believed to be due to a more accurate representation of the distributions of crack length at different flight hours. Further work is planned to apply the current method to practical problems.

Acknowledgement The authors wish to gratefully acknowledge Dr Bruce Crawford of DSTO for carefully reviewing the paper and significantly improving its readability.

References 1. Gallagher, J.P., Badish, C.A., Malas, J.C.: In: The 11th International Conference on Fracture, Turin, Italy (2005) 2. Hovey, P.W., Berens, A.P., Loomis, J.S.: p. 96 (1998) 3. Virkler, D.A., Hillberry, B.M., Geoel, P.K.: Journal of Engineering Materials and Technology 101(2), 148–153 (1979) 4. Ghonem, H., Dore, S.: Engineering Fracture Mechanics 27(1), 1–125 (1987) 5. Maymon, G.: Engineering Fracture Mechanics 53(6), 911–916 (1996) 6. Hu, W., Tong, Y.C., Walker, K.F., Mongru, D., Amaratunga, R., Jackson, P.: In. DSTO-RR-0321, DSTO (2006) 7. Newman Jr., J.C.: In: Chang, J.B., Hudson, C.M. (eds.) Methods and Models for Predicting Fatigue Crack Growth under Random Loading, ASTM STP, vol. 748, pp. 53–84. ASTM (1981) 8. Luo, J., Wowen, P.: Acta Materialia 51(12), 3537–3550 (2003) 9. Fawaz, S.: In. AFRL-VA-WP-TR-2000-3024 (2000) 10. Torregosa, R., Hu, W.: In: AIAC14 Fourteenth Australian International Aerospace Congress: Melbourne, Australia (2011) 11. Tong, Y.C.: Probabilistic fatigue life analysis methods for aerospace vehicles. The University of Sydney (2006) 12. Bigerelle, M., Iost, A.: International Journal of Fatigue 21(4), 299–307 (1999) 13. Hu, W., Walker, K.F.: In: The International Conference on Structural Integrity and Failure, Sydney, Australia (2006) 14. Berens, A.P., Hovey, P.W., Skinn, D.A.: WL-TR-91-3066 (1991) 15. Newman Jr., J.C.: NASA TM-104159, NASA (1992) 16. Tuegel, E.J.C.: Private communication. USAF AFMC AFRL/RBSM (2010)

26th ICAF Symposium – Montreal, 1-3 June 2011 Aircraft Joints and Corrosion Control Ung Hing Tiong1,2 and Graham Clark1,2 1

School of Aerospace, Mechanical and Manufacturing Engineering, RMIT University, Melbourne, VIC 3083, Australia 2 Defence Materials and Technology Centre, 24 Wakefield Street, Hawthorn, VIC 3122, Australia

Abstract. Corrosion damage in aircraft structure, if undetected and/or left untreated, can undermine safety. Currently corrosion prevention and management in many civil and military fleets still relies strongly on the use of traditional ‘find and fix’ maintenance practices, although this has been refined by the increasing use of Corrosion Prevention and Control Plans (CPCP) which provide a framework for targeted inspections and treatment to help with corrosion management. Teardowns of high-life service aircraft and parts can also be valuable tools to help identify corrosion-prone areas and relative severity of the corrosion. This paper describes research which supports the development of improved prognostic capability for corrosion, by investigating one particular factor which appears to play a significant role in the development of corrosion. The focus of this research is to better understand and predict the deterioration and breakdown of protective paint coatings at aircraft joints, primarily due to the influence of mechanical displacement. The impact of in-service mechanical loading on coating degradation has so far received little attention, despite clear evidence that coating tend to fail first at specific site such as sheet ends and fastener heads. Potential service/performance implications of the joint displacement on the protection of ageing aircraft are discussed. More importantly, it is argued that appropriate corrective actions are required immediately after the paint cracking detected, even if active corrosion is not fully evident.

1 Introduction The potential impact of corrosion at joints attracted a great deal of attention after the Aloha Airlines accident [1] in 1986, highlighting the detrimental effect of environmental degradation and multi-site cracking on joint integrity. Visual inspection for such damage is often impossible without a costly process of stripping the paint, removing rivets and opening the joints. A review by Furuta et al. [2] noted that joint specimens subjected to a corrosive environment during cyclic loading exhibited fatigue lives 30-50% shorter than those tested in an ambient environment. Hence, to ensure the durability of structural joints, high quality paint coatings and sealants are normally applied. Unfortunately, of course, paint coatings also mask the initial stages of corrosion One useful corrosion management measure is regular inspection of the surface of paint coatings for irregularities such as blisters, flakes, chips and lumps. When

626

U.H. Tiong and G. Clark

there is coating damage and evidence of significant active corrosion, it is usually managed by a grind out process, which identifies the extent of the damage, and can itself serve as a repair. Unfortunately, even when surface paint film cracking is evident there may be a significant delay before that protective coating is repaired. Clark [3], in discussing prediction of the effects of corrosion on aircraft structure, noted that the prediction of overall service life of a corroded part is critically sensitive to the life of protective coatings, see Figure 1. In other words, the onset of structural damage only occurs when the paint coatings become ineffective, ensuring that a well-maintained durable coating is the most effective means of remaining damage-free. Since the life of paint coating is critical, the development of prognostic tools for the service life of coatings, under realistic service conditions, is an important part of an overall corrosion program. Such tools require an understanding of the various parameters which will influence the coating degradation processes and rate. The objective of this research is in line with Australian work [4] which led to the USAF [5] changing from the ‘find and fix’ approach towards a ‘predict and manage’ philosophy, with the aim of reducing maintenance cost and increasing aircraft availability. Such a new corrosion management philosophy requires prediction of the influence of corrosion damage on the life and residual strength of the corroded parts, so that the most appropriate maintenance strategies can be developed.

Fig. 1 The notional failure progression of a corroded component by fatigue crack growth originating from corrosion [3].

Aircraft Joints and Corrosion Control

627

2 Background A typical exterior military aircraft paint scheme consists of three layers, namely a polyurethane topcoat, inhibited epoxy primer and (particularly in legacy aircraft) a chromium conversion/anodized coating; the internal paint scheme is usually primer-only with extensive use of sealant in some areas such as butt joint gaps and joint ends. Current aircraft coating scheme have been subjected to extensive refinement, and if correctly applied, perform well. Ideally they will last for 6-8 years in service. However, occasionally, paint coatings will fail prematurely, for unexpected reasons and with expensive consequences. The failure or degradation of these paint coatings under in-service environment could result in loss of their originally designated mechanical and physical properties [6-7], leading to localised shrinkage of the coating surface that can result in micro-crack formation. Cracking in the paint can lead to the moisture penetration into the interfaces underneath the paint layers and without rapid attention, this can initiate corrosion damage which could affect the structure later in life. Some common physical changes in polyurethane topcoat include gloss loss, discolouration, tackiness, and weight loss. Blistering [8] may be evident. In contrast, the effect of mechanical environment on coating failure, typically at stress concentrations associated with joints, does not appear to have received a great deal of attention so far. It is inevitable that the movements of a structural joint under service loads would cause the coating system to distort, elongate or bend to some extent, and between various components of a joint exhibiting substantial movement; the strain experienced by coatings would be expected to be substantial. The displacements in these joint locations can play an important role in determining the structural integrity of the coating systems.

3 Aircraft Survey Part of a current research activity includes an initial aircraft survey of a military aircraft to identify the coating type used, application procedure, the service history and environment, and physical evidence of coating failure. In brief, the topcoat examined consists of low volatile organic compound (VOC) polyurethane which possesses excellent ultra violet (UV) resistance and long term durability. The topcoat is estimated to have a minimum life of 2 years. When applied to suitably epoxy primer, it should provide excellent resistance to aircraft hydraulic fluids. The primer used contains strontium chromate for long term corrosion inhibition. Once applied, full cure will normally take up to 7 days. The preliminary survey revealed that the majority of paint coating related failures, in a relatively young Royal Australian Air Force (RAAF) high performance aircraft fleet, can be attributed to three primary causes, viz: UV radiation The coating degradation due to UV is a prime concern for polyurethane topcoat because this layer is constantly exposed to a relatively severe outdoor

628

U.H. Tiong and G. Clark

environment. The presence of UV and oxygen encourages the initiation of photooxidation process in the polyurethane topcoat [9]. This can lead to mechanical and physical changes in the topcoat such as discolouration, loss of surface gloss and blistering, see Figure 2.

Fig. 2 Evidence of degraded topcoat, upper wing skin, around exposed fastener and leading edge (Picture courtesy of BAE Systems Australia).

Particulate damage Abrasion results from scraping, scuffing, and erosion due to moving particulates such as sand or slurries. The coating is worn down by continued abrasion and wear-through may occur. Mechanical damage Various sources of mechanical damage exist; damage may be caused by improper handling of the painted structures especially during maintenance rework and shipping. Impact from dropped tools, stones or other types of mechanical damage may chip and break the paint film. The paint cracking was particularly evident at mechanically-fastened joints. More specifically, prime sites for coating breakdown were exposed fastener and joint ends, see Figure 3 which shows tearing of paint near a wing root fairing. Also noted there were fine ring-type cracks around fastener heads and there was a clear correlation between the most common paint damage/cracking and sites

Aircraft Joints and Corrosion Control

629

where mechanical displacement is likely, i.e. locations where there will be cyclic displacement under service loading. This paper focuses on the extent to which such displacement, coupled with progressive degradation due to UV radiation exposure and exposure to a varying temperature environment, could contribute to paint cracking. The extent of cracked coating bring into question whether the fastener area is protected. Paint cracking will assist moisture penetration underneath the paint and this can subsequently initiate corrosion. Adopting a conservative view point, it is arguable that any paint defects should be repaired immediately upon detection in order to minimise the risk of corrosion later in service.

Fig. 3 Coating damage near the wing root and cracking around fastener heads (Picture courtesy of BAE Systems Australia).

4 Microscopy Analysis In a separate study, scanning electron microscope (SEM) and energy dispersive x-ray spectroscopy (EDS) analyses were conducted on a retired P-3C wing panel, allowing a close examination on the paint condition at the fastener areas. The panel is made of 7075-T6 aluminium alloy and consists of internal and external splices fastened together via a column of countersunk Eddie bolts, see Figure 4. The panel has been used by Defence Science and Technology Organisation (DSTO) for a fatigue test. Samples for metallography were taken from panel 8. The paint around fastener head exhibited ring-type cracking prior to sampling, see

630

U.H. Tiong and G. Clark

Figure 5. Sections were cut in regions where paint contained cracks and were examined in the SEM. Four samples were prepared in this exercise. The paint coating contains titanium pigments which possess excellent corrosion resistance. Surface treatment appeared to be silicone and chromate-based, whilst Eddie bolts were cadmium plated to provide desired wear and corrosion resistance properties. SEM-EDS results also revealed excellent paint surface finish away from fastener head regions; but exhibited substantial paint defects, in the form of cracking and dislocation of layers (delamination); all of which were observed around fastener areas, see Figure 6. Numerous short blunt cracks approximately normal to the paint surface had developed both at the external surface and interfacial layer. The extent of damage was substantially greater than initially assessable from examination of the paint surface state alone. The cracks appeared to be consistent with movement associated with the fastener tilting or the stress concentration associated with joints as substantial displacement is expected at these locations. This displacement induces contraction and elongation of the coating layer and it is likely to influence the progress of such damage. The examination supported a correlation between paint failure/cracking and mechanical displacement. The presence of chlorine near one of the blunt crack tips, see Figures 6 and 7, suggested chloride attack beneath the paint film, although no obvious corrosion damage was noted in the metal substrate.

Fig. 4 Schematic of P-3C lower wing panel cut-out.

Aircraft Joints and Corrosion Control

631

Fig. 5 Paint surface condition prior to machining and sampling.

EDS spot

Fig. 6 Scanning electro micrograph, showing paint defects and delamination underneath the paint film at fastener areas.

632

U.H. Tiong and G. Clark 300

Ti 250

Counts

200

Mo Cl Cd

150

100

Al

Cr

Si

Fe

50

O 0 0

1

2

3

4

5

6

7

8

9

10

Energy (KeV)

Fig. 7 Residual chlorine detected at blunt crack tip.

5 Service Implications of Paint Cracking Tiong and Clark [10], examining a generic lap joint, estimated coating strain of the order of 14% at the joint ends. Their analysis, using published experimental data, demonstrated an applied strain of 14% should have no adverse effect on the mechanical integrity of the fresh polyurethane topcoat. In contrast, aged polyurethane topcoat, i.e. after 6 weeks of accelerated weathering exposure, was shown not be able to tolerate this 14% service strain. As mentioned earlier, elimination (repair) of any paint cracking (such as the one observed at fastener areas) immediately upon detection would provide a better option than waiting for a convenient time for coating repair. Such immediate action would minimise the opportunity for moisture entry and future corrosion. Such rapid repair could be done by applying some form of corrosion protection such as the use of non-penetrating corrosion inhibiting compounds (CICs) [11]. CICs discourage moisture penetration until a more permanent treatment such as paint touch-up or sealant can be used. Another, and perhaps more preferable course, since it would not introduce lubricating CICs to a joint area, would be rapid local repair of any paint cracking using primer and/or sealant containing corrosion inhibitors, which can effectively restore the protective scheme. The effectiveness of each of these methods would be highly dependent on whether or not moisture had penetrated and started the corrosion process; while applying repairs to cases where the cracks had been present for some time might allow corrosion to remain, a consistent program of rapid repair would of course eventually minimise this risk. Rapid repairs would be most effective when applied to the structures which contain no damage or damage categorised as level 1 corrosion. If higher levels of

Aircraft Joints and Corrosion Control

633

corrosion require repair, a penetrating CICs might in fact assist by preventing further degradation in the period before schedule repair.

6 Conclusions In summary, this paper highlights the potentially large influence of mechanical displacement on paint degradation, and argues that to minimise the risk of longterm corrosion development in joint areas, appropriate corrective actions are required immediately after the paint cracking is detected, even if active corrosion is not fully evident. Significant cost savings could be realised in the long run by avoiding expensive and time-consuming corrosion-induced maintenance repairs, particularly where there is a need to extend fleet life beyond the originally intended service life.

Acknowledgements The authors are grateful to the Defence Materials Technology Centre (DMTC Ltd) for their financial support. Grateful acknowledgements are also due to Prof. Graeme George, Prof. Milan Brandt, Dr. Anthony Trueman and Dr. James Waldie for their technical support and useful discussions.

References [1] Airlines, A.: Flight 243, Boeing 737-200, N73711, near Maui Hawaii, April 28, National Transport Safety Board, Washington, D.C (1988) [2] Furuta, S., Terada, H., Sashikuma, H.: Fatigue strength of fuselage joint structures under ambient and corrosive environment. In: ICAF 1997: Fatigue in New and Aging Aircraft, EMAS, pp. 231–249 (1997) [3] Clark, G.: Corrosion and the management of Structural Integrity. In: Rudd, J.L. (ed.) ICAF 1999 Structural Integrity for the Next Millennium. EMAS, Warley (1999) [4] Cole, G.K., Clark, G., Sharp, P.K.: The implications of corrosion with respect to structural integrity, DSTO-RR-0102, Defence Science and Technology Organisation, Melbourne (1997) [5] Kinzie, R., Cooke, G.: Corrosion in USAF aging aircraft fleets. In: RTO AVT Workshop on Fatigue in the Presence of Corrosion, Corfu, Greece (1998) [6] Skaja, A., Fernando, D., Croll, S.: Mechanical property changes and degradation during accelerated weathering of polyester-urethane coatings. Journal of Coatings Technology and Research 3, 41–51 (2006) [7] Tangestanian, P., Papini, M., Spelt, J.K.: Starch media blast cleaning of artificially aged paint filrms. Wear 248, 128–139 (2001) [8] Yang, X.F., Tallman, D.E., Bierwagen, G.P., Croll, S.G., Rohlik, S.: Blistering and degradation of polyurethane coatings under different accelerated weathering test. Polymer Degradation and Stability 77, 103–109 (2002)

634

U.H. Tiong and G. Clark

[9] Ranby, B., Rabek, J.F.: Photodegradation, photo-oxidation and photostabilisation of polymer, principles and application. Wiley-Interscience, New York (1975) [10] Tiong, U.H., Clark, G.: The impact of mechanical strain environment on aircraft protective coatings and corrosion protection. Journal of Aircraft, Article in press (2011) [11] Jaya, A., Tiong, U.H., Clark, G.: The interaction between corrosion management and structural integrity of aging aircraft. Fatigue & Fracture of Engineering Materials & Structures (2011) Article in press

26th ICAF Symposium – Montreal, 1-3 June 2011 A Case Study of Nose Landing Gear Failure Caused by Fatigue V.Y. Guertsman Transportation Safety Board, Engineering Laboratory, 1901 Research Road, Ottawa, ON, K1A 1K8, Canada

Abstract. The nose landing gear of a commercial turboprop airplane failed during landing. A post-occurrence examination found a fracture in the trunnion arm of the oleo strut housing. The landing gear had been installed on the aircraft after an overhaul. The laboratory investigation has established that the eventual ductile overload fracture was precipitated by a pre-existing fatigue crack. The fatigue fracture surface had two distinct zones different in macroscopic and microscopic appearances as well as chemical composition. Analysis of the results has shown that the older region of the fatigue crack pre-dated the overhaul, while the fresher fatigue zone grew during the post-overhaul operation of the aircraft.

1 Introduction A commercial turboprop airplane experienced no problems before landing. During the landing roll, part of the nose landing gear structure collapsed resulting in a brief excursion off the runway before the airplane was brought back onto the runway. There were no injuries, but the aircraft sustained some damage. The nose landing gear was installed on the aircraft 20 months prior to the incident. The gear had been in the overhauled condition, and accumulated about 1900 take-off/landing cycles since the installation. A post-occurrence inspection found a fracture in the trunnion arm of the oleo strut housing and this part was sent for laboratory examination.

2 Laboratory Examination Results A general view of the fracture surface is shown in Figure 1. The main direction of the crack propagation was from the inner surface of the strut housing trunnion arm outwards. A fingernail-shaped fatigue crack was followed by ductile overload fracture. An interesting feature is a bright rim surrounding the dark fatigue region. This shiny band differs from the dull and fibrous appearance of the surrounding overload fracture surface. At higher magnification, features characteristic of fatigue, such as crack propagation marks, could be distinguished in this zone (Figure 2). Hereafter, this region of the fatigue crack is called the fresh fatigue zone, while the larger dark area is called the old fatigue crack.

636

V.Y. Guertsman

After cleaning ultrasonically in acetone and ethanol the fracture surface was examined in a scanning electron microscope (SEM) equipped with an energy dispersive x-ray spectroscopy (EDS) system for chemical analysis. The old and fresh areas of the fatigue crack could be clearly distinguished in the SEM. The old areas were heavily oxidized and in many places were covered with deposits (Figure 3a). A typical EDS spectrum from this region, showing significant carbon, oxygen, silicon and sulphur peaks, is displayed in Figure 3b. Fatigue striations are visible in some places of the old area of the fatigue crack (Figure 4a). These striations are not very sharp, due to rubbing and oxidation, unlike the striations in the fresh zone of the crack (see below). EDS spectrum from this region shows primarily aluminum and oxygen (see Figure 4b). The fresh areas of the fatigue crack showed clear fatigue striations (Figure 5) and, unlike the old fatigue areas described above, they were free from deposits. The closely spaces striations are similar to those characteristic of high-cycle fatigue in wrought aluminum alloys [1]. EDS spectra from the fresh fatigue regions indicated insignificant oxidation and showed mainly the elements that were consistent with an AA2014-type aluminum alloy of the strut housing (Figure 6). An example of the transition region from the old fatigue crack to the fresh fatigue zone is displayed in Figure 7. Figure 8 shows a representative SEM micrograph of the transition from the fresh fatigue region to the overload zone (typical dimpled fracture surface). As mentioned above, the fatigue crack originated on the bore surface of the trunnion arm. The region of the bore surface close to the fatigue crack is shown in Figure 9, where one of the ratchet marks is viewed almost edge-on. A secondary crack was filled with some substance. EDS analysis indicated that the crack filling was a compound rich in carbon, oxygen, silicon and sulphur (Figure 10), similar to the deposits on the old fatigue fracture surface (see Figure 4). The uneven backscattered electron contrast on the SEM image of the bore surface (see Figure 9) suggests compositional differences, namely, different degree of oxidation due to uneven wear of anodized layer on aluminum.

A Case Study of Nose Landiing Gear Failure Caused by Fatigue

6337

a

b 1 Fig. 1 Close-up p photographs of the two mating fracture surfaces.

3 Discussion Visual inspection, opticall microscopy and SEM have given strong evidence thhat the eventual failure of thee strut housing in a single-cycle ductile overstress modde was preceded by fatigue fracture. The fatigue crack consisted of two distinctlly y suggesting that it grew in two stages separated bby different zones, strongly

638

V.Y. Guertsmaan

significant time and/or ev vent. The periphery of the fatigue crack (the fresh fatiguue zone) contained clear fatigue f striations with insignificant oxidation and nno extraneous deposits. The rest of the fatigue crack surface (the old fatigue zonee) was severely altered by oxidation, rubbing and deposits. The fatigue crack haad a evident from the observed ratchet marks. The multiple multiple initiation sites, as fatigue crack origins sugg gest that it was not a single particular defect responsible for the fatigue crack initiaation [2]. The inner surface of the strut housing had som me imperfections that could have served as crack nucleation sites. Since the borre c origin was worn and damaged, it is unclear whhat surface near the fatigue crack caused the initial crack to nucleate in the first place. The main question waas p in the strut housing before the overhaul oor whether that crack had pre-existed developed during the subsequent service. The results of this investigation pointeed to the first scenario. t deposits on the old fatigue fracture surface and thhe The compositions of the filling in the nearby seccondary crack (see EDS spectra in Figures 4 and 100) suggest that this was a hardened bushing locking adhesive (there was a steeel unnion arm bore). It is unlikely that this substance coulld bushing fitted into the tru have penetrated into the narrow cracks in completely hardened solidified statte,

Fig. 2 Opticaal micrograph of the region outlined in Figure 1.

A Case Study of Nose Landing Gear Failure Caused by Fatigue

639

(a)

(b) Fig. 3 Representative SEM micrograph (a) of the old area of fatigue fracture near the bore and the corresponding EDS spectrum (b).

640

V.Y. Guertsman

(a)

(b) Fig. 4 Fatigue striations in the old area of fatigue fracture (a) and the corresponding EDS spectrum (b).

A Case Study of Nose Landiing Gear Failure Caused by Fatigue

6441

Fig. 5 Examples of fatigue striations in the fresh area of fatigue crack photographed at different magnifications.

642

V.Y. Guertsmaan

Fig. 6 EDS E spectrum from the fresh fatigue zone.

Fig. 7 Transition from f the old area to the fresh area of the fatigue crack.

A Case Study of Nose Landiing Gear Failure Caused by Fatigue

Fig. 8 Transition frrom the fresh fatigue zone to the ductile overload zone.

Fig. 9 Bore surfacce near the fatigue crack showing a secondary crack.

6443

644

V.Y. Guertsman

Fig. 10 EDS spectrum from the substance in the secondary crack (see Fig. 9).

which would be the case if the cracks developed after the bushing was locked in place. Therefore, it is reasonable to suggest that the bushing locking compound penetrated into the pre-existing cracks when the bushing was press fit into the housing. Traces of elements from solutions used during overhaul were also found in the deposits on the old fatigue fracture surface.

4 Summary The nose landing gear failure was caused by the fracture of the oleo strut housing trunnion link arm. The eventual overload fracture of the strut housing was precipitated by a pre-existing fatigue crack. The fatigue crack initiated from multiple origins on the strut housing arm inner surface. No specific surface defects or other stress raisers could be established at the fatigue crack initiation location. The fatigue crack had likely propagated mostly during the pre-overhaul service, and progressed further during the post-overhaul service of the aircraft.

References [1] Metals Handbook, Fractography, 9th edn., vol. 12, p. 426. ASM International, Metals Park (1987) [2] Sachs, N.W.: J. Failure Analysis and Prevention 5(2), 11 (2005)

26th ICAF Symposium – Montreal, 1-3 June 2011 Test Method for Determining the Effect of Chromate Primers on Fatigue Crack Growth Yongwon Lee1, Sarah E. Galyon Dorman1, and Matthew J. Hammond2 1

Center for Aircraft Structural Life Extension, United States Air Force Academy 2 formerly of the Center for Aircraft Structural Life Extension, United States Air Force Academy

Abstract. The Center for Aircraft Structural Life Extension (CAStLE) at the United States Air Force (USAF) Academy has undertaken work to produce a standardized test method for determining the effect of a corrosion inhibitor containing coating chromate on small scale fatigue damage. CAStLE is specifically focusing their research on the effect of chromate primers on the pit-to-crack transition and crack growth under the damage tolerant flaw size. Quantifying the chromate effect at the small damage scale is also necessary to provide a baseline to compare the efficacy of new and safer coatings on fatigue life. To correctly account for these factors, bare and chromate primered test specimens of AA7075-T651 were produced with a center hole and a corrosion pit at the edge of the hole, and subsequently fatigued using representative environments and loading schemes until a crack of 1 mm was produced. Another unknown with chromate coatings is whether enough chromate to reduce the fatigue crack growth rate can leach out of the primer in the time it takes to grow from a pit to a 1 mm fatigue crack. The current results suggest that more work needs to be completed on the K-solution for the sample and to quantify the ability of chromate to leach into a solution from a primer coating.

1 Introduction Aircraft structural integrity issues are often caused by corrosion. Atmospheric conditions as benign as humid air can cause large changes in the fatigue life of cracks in aircraft alloys [1,2]. In the presence of atmospheric environments corrosion damage such as pit formation can occur. These pits can act as stress raisers which initiate fatigue cracks. Within damage tolerant design, it is assumed that the majority of fatigue life exists in the short crack region of fatigue. However, most existing experimental procedures and data on corrosion fatigue focus on the formation of the pit or long-crack fatigue propagation. It is possible that as the United States Air Force (USAF) and other military organizations move away from chromate-containing primers and coatings to more environmentally friendly options, there may be an unexpected loss in fatigue life if chromate coatings provide protection from corrosion fatigue below the damage tolerant flaw size. To ensure that there is not an unexpected loss in fatigue life with the implementation of new coatings, a new laboratory technique needs to be developed that better encompasses typical aircraft structure (geometry), loading and damage. The Center for Aircraft

646

Y. Lee, S.E. Galyon Dorman, and M.J. Hammond

Structural Life Extension (CAStLE) at the USAF Academy has undertaken research to develop and validate a standardized test method and specimen that document the ability of a corrosion protection system to affect fatigue damage below the damage tolerant flaw size.

2 Test Protocol Figure 1 shows the specimen developed by CAStLE with a centrally located hole. As in-service cracks are typically found emanating from holes in structure, it is thought that this geometry better incorporates a real world aspect of aircraft design better than an edge- or center-cracked panel). For validation of the specimen geometry, samples were made from legacy aircraft aluminum alloy and temper (AA) 7075-T651. For each specimen the flaw to initiate the fatigue crack was a small controlled corrosion pit (~ 300 μm) inserted at the corner of the bore hole [3]. For all tests completed during the sample validation, an environmental chamber was used to control the test environment. The sample was loaded into a computer controlled servohydralic test frame and tested using a direct current potential drop (dcPD) system. For each test a fatigue crack was propagated from the pit until the crack reached approximately 1 mm in length, and then the test was terminated and the sample pulled to failure by overload. Figure 2 shows the environmental chamber used for testing. Three different environments were used during the validation of the test method: dry nitrogen (N2), humid N2, and immersion in salt water (0.06 M NaCl concentration). These environments were selected because they encompass standard laboratory corrosion fatigue test environments and are aircraft environment relevant.

All dimensions in mm

cfinal

afinal Fatigue Crack

Hole

Cross-section Detail

Fig. 1 Proposed specimen design for standardized test protocol. Specimens can be of any metallic aircraft material. All dimensions are in millimeters. For validation, AA7075-T651 samples were used.

Fig. 2 Drawing of environmental test cell used for test protocol validation. Dimensions of test cell are not standardized and should simply be appropriate for the sample being used.

Test Method for Determining the Effect of Chromate Primers

647

Uncoated specimens were used to produce the AA7075-T651 baseline data for comparison with corrosion inhibitor data. For the baseline data the matrix in Table 1 was used. The matrix included common frequencies (f) to detect any frequency effects with corrosion fatigue. As the test method is designed to test corrosion inhibiting primers, when a sample is referred to as “Bare” the term means no form of primer or coating is applied to the sample. The matrix was repeated, as noted by the “Bare/Primered” under “Sample” in Table 1, to validate the method with coated samples. In this testing the AA7075-T651 samples were chromate conversion coated followed by a chromated primer. Table 1 Test matrix for baseline AA 7075-T651 data with standardized test method (Bare). To validate the specimen design for corrosion inhibiting primers, samples were chromate conversion coated and primered (Coated) then tested using the same matrix.

Sample

Environment

Loading

Bare/Primered Bare/Primered Bare/Primered Bare/Primered Bare/Primered Bare/Primered Bare/Primered Bare/Primered Bare/Primered

Humid N2 Dry N2 0.06 M NaCl Humid N2 Dry N2 0.06 M NaCl Humid N2 Dry N2 0.06 M NaCl

Constant Constant Constant Constant Constant Constant Constant Constant Constant

Frequency (Hz) 20 20 20 1 1 1 0.1 0.1 0.1

∆KIntial (MPa√m) 3 3 3 3 3 3 3 3 3

R

Replicates

0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1

2 2 2 2 2 2 2 2 2

3 Aircraft Loading Analysis One of the methodological goals is to incorporate real aircraft loading and to determine how the presence of a chromated primer changes corrosion fatigue propagation. To determine how the crack growth behavior with aircraft loading is different from constant amplitude, a transport aircraft wing spectrum (miniTWIST) was used. Mini-TWIST only contains the peaks and valleys of load pairs; it does not inherently include loading frequency and rates, which have been shown to be important in corrosion fatigue tests particularly with the use of corrosion inhibitors [1]. Strain-time data from a transport plane were examined to determine the proper loading frequency and stress ratio (R) ratio to couple with mini-TWIST. A “cycle” was defined by the simple range counting method in the ASTM standard for cycle counting, with two half-cycles making one cycle [4]. The frequency was calculated as the inverse of cycle period, and the stress ratio was defined as the cycle’s minimum strain divided by the cycle’s maximum strain. The strain was measured near the wing root of a transport airplane. Because of the known structural response in this area, data was recorded at 32 Hz with a lowpass filter at 8 Hz. Although the discreet data sampling also discretized the cycle period and calculated frequency, the overall average frequency value near 3.5 Hz is clearly visible in Figure 3. All strain data from ground operations, including

648

Y. Lee, S.E. Galyon Dorman, and M.J. Hammond

taxiing, takeoff and landing were removed, and cycles with strain amplitudes less than 0.0001 (100 µstrain) were removed as well. This data trimming was done after cycle counting, so that the cycle periods were not affected. The measured stress ratio was very high, ranging from 0.9 to 0.99.

Normalized Exceedance, Per Flight Segment

20% Climb, Decent and Manuver Crusing 15%

10%

5%

0% 0

2

4

6

8

10

Frequency (Hz)

Fig. 3 Normalized frequency exceedance plot based on six flights. Table 3 Test matrix for aircraft loading spectrum for bare and primered AA7075-T651 samples. Areas labeled TBD will be determined by analyzing real aircraft flight data. Sample

Environment

Loading

Bare Bare Bare Primered Primered Primered

Humid N2 Dry N2 0.06 M NaCl Humid N2 Dry N2 0.06 M NaCl

Mini-TWIST Mini-TWIST Mini-TWIST Mini-TWIST Mini-TWIST Mini-TWIST

Frequency (Hz) TBD TBD TBD TBD TBD TBD

∆KIntial (MPa√m) TBD TBD TBD TBD TBD TBD

R

Replicates

TBD TBD TBD TBD TBD TBD

2 2 2 2 2 2

4 Results Figure 4 shows the results from the first three bare baseline tests (symbols) completed using the standardized test method compared to long crack fatigue crack growth (lines) from published literature [1,2]. Results from the humid N2 experiments overlap published long fatigue crack data [1,2]. The 0.6 M NaCl and dry N2 experiments showed slower growth rates when compared to literature [1,2]. The discrepancy could be due to multiple factors, including but not limited to: mixed crack orientation (meaning the crack grows in both the short-transverse and transverse-short direction) small crack effects, crack closure, difference in loading frequency and experimental scatter associated with fatigue testing. Table 4 shows how the aspect ratios (a/c) developed from pit to final crack and the cycles to detect crack propagation for each test. Cycles to detect crack propagation is defined as the number of cycles required to produce a detectible increase in the dcPD signal: 0.2 microvolts.

Test Method for Determining the Effect of Chromate Primers

649

Figure 5 shows the crack growth rate data for all samples tested to date in 0.06 M NaCl. The bare and chromate primer coated samples tested at 20 Hz have basically the same crack growth rate and the presence of the primer appears not to have affected the fatigue crack propagation. This result is expected as chromate is known to be ineffective at higher frequencies in low concentrations [5]. For the samples tested at 0.1 Hz the sample coated with the chromate containing primer has faster fatigue crack growth rate than the bare sample. This was an unexpected result; other published work [5] has shown, at lower frequencies, chromate is expected to have a greater effect in reducing fatigue crack growth rates. Table 5 shows a possible explanation for the unexpected crack growth rate. It should be noted that the pit aspect ratio (a/c) for the 0.1 Hz chromate primered sample was 2.50, while the ideal starting pit aspect ratio is 1. It is likely that this aspect ratio being so far from 1 makes the comparison between the bare and primered samples at 0.1 Hz invalid. Table 5 notes the cycles to detect crack propagation for all of the samples. For both the 20 Hz pair and the 0.1 Hz pair it appears that the presence of the chromated primer increases the cycles to detect propagation for each test. It is possible that an effect of chromate is showing in this crack initiation to very small crack growth range.

Fig. 4 Measured crack growth rate curves for compared to published data [1,2]. Table 4 Crack development and cycles to detect propagation for baseline tests. Environment Dry N Humid N 0.06 M NaCl

Pit Radius Initial Pit Final Crack Cycles to Detect (mm) Aspect Ratio Aspect Ratio Crack Propagation 0.17 1.62 1.26 30,000 0.22 0.92 1.18 35,000 0.19 1.08 1.29 28,000

650

Y. Lee, S.E. Galyon Dorman, and M.J. Hammond

Fig. 5 Comparison of AA 7075-T651 samples tested at 0.1 Hz and 20 Hz with bare and chromate primered specimens. Table 5 Comparison of pit size, aspect ratio and cycles to detect propagation for bare and primered AA 7075-T651 samples tested at 0.1 Hz and 20 Hz in 0.06 M NaCl.

Sample

Environment

Frequency (Hz)

Bare Primered Bare Primered

0.06 M NaCl 0.06 M NaCl 0.06 M NaCl 0.06 M NaCl

20 20 0.1 0.1

Pit Radius Initial Pit Aspect (mm) Ratio 0.29 1.26 0.24 0.86 0.29 0.67 0.27 2.50

Cycles to Detect Propagation 45,000 85,000 10,000 51,000

Stress Intensity The stress intensity (K)-solution for the specimen was taken from the internal Ksolution database built into the crack growth modeling software, AFGROW [6]. The pit was assumed to be a crack with zero height (b). In all reported fatigue data, AFGROW’s two-point advanced solution was used. A few crack geometries were modeled using the Virtual Crack Closure Technique (VCCT) method and compared to the AFGROW solution for validation [6,7]. The modeled specimen geometry is listed in Table 6. As noted, the VCCT model results are consistently higher than the AFGROW solution. Ignoring the singularities near the c and a-tips, the difference in the two stress intensity solutions, (KVCCT-KAFGROW)/KVCCT, varied from 10% to 22%. In all cases, the difference was smallest near the c-tip (10~18%) and highest near the a-tip (19~22%). An obvious mathematical relationship between the two solutions was not observed and extensive analysis was not performed. The current thought is that the difference is due to the fact that the specimen geometry diameter of the hole versus width (d/W) is outside of the current AFGROW parameter limits. To obtain a useable stress intensity solution across all crack geometries observed in testing, the beta values were pulled from AFGROW based on the initial and final crack sizes for each specimen, and then uniformly increased by 20% [6]. This results in K values that are up to 10% too high, and up to 2% too low. A simple beta

Test Method for Determining the Effect of Chromate Primers

651

correction table can be constructed using the VCCT results, and this refinement can be used in the future instead of a single 20% increase. However, more VCCT modeling work may be necessary to produce an accurate beta correction table. Table 6 VCCT model geometries a (mm) c (mm) a/c (KVCCTKAFGROW)/KVCCT (%)

0.3 0.3 1

0.9 0.9 1

0.9 0.75 1.2

0.9 0.643 1.4

0.9 0.563 1.6

1.5 1.5 1

1.5 1.25 1.2

1.5 1.071 1.4

1.5 0.938 1.6

21.1

15.8

16.2

16.6

17.0

17.8

14.6

13.5

13.2

Evolution of Crack Growth The AFGROW predicted aspect ratio evolution is compared with experimental observations in Figure 6. The aspect ratio evolution as observed on two markerbanded specimens (square and diamond points) was compared to AFGROW predicted crack evolution (solid lines). For specimen Spec2, AFGROW predicts a rapid drop in aspect ratio (a/c) compared to a delayed drop in a/c observed by marker bands. In the AFGROW model, the aspect ratio asymptotically approaches a single value after approximately 1 mm of growth, regardless of the initial aspect ratio. Experimental observations (and AFGROW) also suggest that the aspect ratio reaches approximately 1.3 as the cracks evolve, regardless of the original pit size. Overall, these results suggest more analysis of the effect of the crack shape development on the K-solution needs to be completed. AFGROW predicted crack evolution vs. observed aspect ratios

2.0

AFGROW Parameters: Classic Model Harter-T 7075-T6511 constant amplitude stress= 54 MPa R=0.1

a/c

1.5

a/c 0.67 a/c 1.50

1.0

Experimental Spec1 Spec2 Spec1-AFGROW

0.5

Spec2-AFGROW

0.0

0.5

1.0 crack length, a, (mm)

1.5

2.0

Fig. 6 Measured and modeled crack aspect ratios.

652

Y. Lee, S.E. Galyon Dorman, and M.J. Hammond

5 Chromate Leaching One of the other goals of this work is to determine how much chromate must be present in solution to have an effect on fatigue crack growth rates (FCGR). There is not much research published on the leaching of chromate from aircraft primers, however some amount of work has been completed on chromate conversion coatings [8,9]. One paper reviewed did analyze the amount of chromate leached from a primer [10]. However, none of these papers have taken into account an appropriate surface area of primer to volume of solution ratio (SA/V) to determine what concentration of chromate would leach from a primer into solution. Based on teardown and analysis results and three-dimensional (3D) modeling of aircraft completed by CAStLE, several areas in an aircraft known to collect fluid were examined and their SA/V ratios were calculated. An area in the wing of some transport aircraft known to collect fluid was determined to have an SA/V of 0.022 mm-1, meaning that there was approximately twice as much surface area of primer as volume of solution. Another area in service aircraft widely known to collect liquid is the lap-joints. The same SA/V calculations were completed on the several lap-joint areas and the SA/V ratios varied between 1.72 mm-1 and 3.79 mm-1, meaning the surface area of primer exposed compared to the amount of volume the space can hold was extremely large. For the lap-joint SA/V results the amount of liquid required to produce the required ratio varied between 0.26 mL and 0.61 mL of solution for the specimen in Figure 1; far too little to be considered a full immersion test. This suggests that to accurately mimic the primer/solution conditions in a lap-joint a thin salt film rather than full immersion would be highly applicable [1]. Based on the fatigue results with the chromate primered samples, further analysis of published data on the effect of chromate on fatigue cracks was performed [5]. In that publication, two concentrations (0.03 M and 0.5 M) of chromate salt (Na2CrO4) were added to the bulk 0.6 M (3.5%) NaCl during a series of fatigue crack growth tests. Figures 7-9 show the effect of chromate in the two different concentrations compared to the fatigue crack growth rate in pure NaCl at each of the three ∆K values. Table 7 gives the frequencies over which there was a chromate effect for each ∆K and concentration [5]. From Table 7 it should be noted that the frequency over which chromate lowers the fatigue crack growth rate increases with lower ∆K. That suggests that at ∆K = 3 where the short crack growth testing starts, the critical frequency should be higher than the 0.2 Hz for a ∆K = 6 if the chromate concentration exceeds 0.03 M. However if the concentration is lower, then it is possible that much lower frequencies will be need to observe the effect of chromate leached from the primer. Additionally, the addition of chromate is less potent at lower ∆K, when compared to ultra high vacuum (UHV) and 30-40% Relative Humidity in Air (RH) (both noted by arrows on the side of each chart). The effect of chromate at a ∆K = 12 is near that of UHV; at ∆K = 9 the chromate effect is reduced for ∆K = 9 to dry air for the higher concentration of chromate and to 30-40% RH for the lower concentration. For ∆K = 6 the effect is even less dramatic with the higher concentration of chromate bringing the effect on FCGR to approximately 30-40% RH and the lower

Test Method for Determining the Effect of Chromate Primers

653

concentration only dropping the growth rate to that of high RH. This raises the concern that at a ∆K = 3 the chromate effect on fatigue crack growth rate will be very small with leached concentrations; in the range of the difference between full immersion and high humidity, but not near UHV.

1.0E-03

? K=6 MPavm Chromate in Bulk Solution Effect on AA7075-T651 Fatigue Crack Growth Rates

1.00E-02

AA 7075-T651 ? K=6 MPavm R=0.1

? K= 9 MPavm Chromate in Bulk Solution Effect on AA7075-T651 Fatigue Crack Growth Rates 0.6 M NaCl

AA7075-T651 ?K=9 MPavm R=0.1

0.6 M NaCl + 0.03 M Na2CrO4 0.6 M NaCl + 0.5 M Na2CrO4 30 - 40% RH Air

) 1.00E-03 le ycc / m m ( N d /a d1.00E-04

) e l ycc / m 1.0E-04 m ( N d /a d 0.6 M NaCl 0.6M NaCl + 0.03 M Na2CrO4 0.6 M NaCl + 0.5 M Na2CrO4 30 - 40% RH Air

1.00E-05

1.0E-05 0.01

0.1

1 Frequency (Hz)

0.001

10

Fig. 7 The effect of chromate on the FCGR of AA7075-T651 in 0.6 M NaCl at ∆K=6 MPa√m [5]. Broken line denotes 30-40% RH.

0.01

0.01

0.1 1 Frequency (Hz)

10

Fig. 8 The effect of chromate on the FCGR of AA7075-T651 in 0.6 M NaCl at ∆K=9 MPa√m [5]. Broken line denotes 30-40% RH.

? K= 12 MPavm Chromate in Bulk Solution Effect on AA7075-T651 Fatigue Crack Growth Rates 0.6 M NaCl

AA 7075-T651 ? K=12 MPavm R=0.1

0.6 M NaCl + 0.03 M Na2CrO4 0.6 M NaCl + 0.5 M Na2CrO4 30 - 40% RH Air

) e cly c/ m 0.001 m ( N /d a d

0.0001 0.005

0.05

0.5 Frequency (Hz)

5

50

Fig. 9 The effect of chromate on the fatigue crack growth rate of AA 7075-T651 in 0.6 M NaCl at ∆K=12 MPa√m [5]. The broken line denotes 30-40% RH. Table 7 Comparison of where the highest fatigue crack growth rates occur at each ∆K value and where the chromate effect is noted for each concentration [5]. ∆K (MPa√m)

High FCGR Range

6 9 12

0.5 Hz-10.4 Hz 0.1 Hz-1 Hz 0.04 Hz-0.1 Hz

Effective f Range 0.03 M 0.2 Hz -0.8 Hz 0.1 Hz-0.3 Hz 0.01 Hz -1 Hz

Effective f Range 0.5 M 0.2 Hz -60 Hz 0.01 Hz -80 Hz 0.1 Hz -7 Hz

654

Y. Lee, S.E. Galyon Dorman, and M.J. Hammond

6 Conclusions Repeated tests appear to indicate that crack nucleation time from a small pit at a hole is lengthened by the presence of a chromate coating. However, the effect of chromate containing primer on corrosion fatigue crack growth for very small cracks is not well characterized. Based on the analysis of the K-solution and crack shape development, more work needs to be completed on developing a better solution and fully understanding the crack shape development. The analysis of the chromate leaching suggests that a thin salt film may be more applicable than full emersion testing for lap joint applications. Leaching experiments to determine the amount of chromate leaching from the primer into solution need to be completed to better determine the frequency and loading ranges appropriate for the effect of chromate.

References [1] Warner, J.S., Kim, S., Gangloff, R.P.: International Journal of Fatigue 31, 1952–1965 (2009) [2] Ciccone, M.P.: The Effect of Corrosion Product Formation on Fatigue Crack Closure of AA7075-T6511 and AA7055-T7451. MS Thesis, University of Virginia, Charlottesville, VA (2005) [3] Burns, J.B.: The Effect of Initiation Feature and Environment on Fatigue Crack Formation and Early Propagation in Al-Zn-Mg-Cu. PhD Dissertation, University of Virginia, Charlottesville, VA (2010) [4] ASTM Standard E1049-85, Standard Practices for Cycles Counting in Fatigue Analysis. In: ASTM International, West Conshohocken, PA (2005) [5] Gasem, Z., Gangloff, R.P.: Rate-Limiting Processes in Environmental Fatigue Crack Propagation in 7000-series Aluminum Alloys, Chemisty and Electrochemistry of Corrosion and Stress Corrosion Cracking. In: Jones, R.H. (ed.), TMS-AIME, Warrendale, PA, pp. 501–521 (2001) [6] www.AFGROW.net version 5.01.05.16 (2010) [7] Krueger, R.: Contractor Report NASA/CR-2002-211628 The Virtual Crack Closure Technique: History, Approach and Applications. NASA Langley Research Center, Hampton, VA (2002) [8] Xia, L., Akiyama, E., Frankel, G., McCreery, R.: Journal of the Electrochem. Soc. 147, 2556–2562 (2000) [9] Laget, V., Jeffcoate, C.S., Isaacs, H.S., Buchheit, R.G.: Journal of the Electrochem. Soc. 150, B425–B432 (2003) [10] Scholes, F.H., Furman, S.A., Hughes, A.E., Nikpour, T., Wright, N., Curtis, P.R., Macrae, C.M., Intem, S., Hill, A.J.: Prog. In Organic Coat. 56, 23–32 (2006)

26th ICAF Symposium – Montreal, 1-3 June 2011 Life Extension: Fatigue Lifetime Updating of the French Xingu Fleet P. Madelpech, D. Théret, M. Fressinet, J. Despujols, and B. André DGA Aeronautical Systems

Abstract. Embraer 121 Xingu is a reduced wing span version of the EMB110 "Bandeirante" with its own smaller fuselage. French Forces bought about 50 of the 105 produced aircraft. It mainly helped for the training of the transport aircraft pilots. Primary designed for passenger transport, the French use was slightly different and less severe due to lighter take-off weight and fewer pressurization cycles. This leads EMBRAER, in 1994, to give a fatigue life extension for the fleet, based on a comparison of load spectrum. The study concluded on an extension from 15 000 flight hours to 45 000. Today, the mean use has evolved and the extended potential is no longer applicable to the concerned aircraft. Therefore, the fleet needs an update of its fatigue potential, in an adaptable way taking into account that a change of the mean use could still be possible. As Embraer did not give a satisfactory answer in terms of costs or delivery time, the study was performed by a technical laboratory of the French MoD. The method consists in dividing the standard use of the aircraft in several missions and to determine by available documentation the associated solicitations. A flight test campaign enables to determine the relationship between the flight parameters and the local stress levels. Then the cumulative stress spectrum associated with the defined missions is deduced. Traditional fatigue damaging formulas based on Miner cumulative laws were finally used to compute the fatigue lifetime and appropriate safety coefficients were applied.

1 Introduction General presentation of the aircraft Embraer 121 Xingu is a twin-turboprop (PT6A-28) corporate transport airplane which can carry 9 people (including crew members), 170 kg of luggage and 5670 kg MTOW. Its range is about 2300 km at 26 000 feet maximum. The Embraer 121 Xingu is a 14.45 m wing span, whose conception is largely inpired from the EMB110 "Bandeirante". In fact, they have the same wing structure with 1.6 m shorter wing span. The Xingu has its own smaller fuselage (7 passengers maximum against 26). Less pay-load and reduced wing span lead to significant decrease of the bending moment on the wings. The aircraft was certified according to FAR – Part 23 – Amdt 23-16, with a “safe life” philosophy. No specific ground full-scale fatigue and static test were performed on the Xingu and the potential of 15 000 flight hours was obtain using:

656

-

P. Madelpech et al.

the full scale fatigue test of the EMB – 110 for the wings component tests on the carry-through spars dedicated fuselage fatigue test component test for the landing gear, common to the EMB-110.

Fig. 1 French Navy Xingu.

Fig. 2 Xingu lateral sizes and axis definition.

Primary life extension French Forces bought about 50 of the 105 produced aircraft. It is used by French Air Force and French Navy and mainly helped for the training of the transport aircraft pilots. Primary designed for passenger transport, so for full weight, high altitude and high speed, the Xingu has a slightly different and less severe use in France due to lighter take-off weight and fewer pressurization cycles. The applied loads in terms of wing bending are therefore below the ones considered to certify this aircraft.

Life Extension: Fatigue Lifetime Updating of the French Xingu Fleet

657

Analysing it [1] in 1994, EMBRAER, gave a fatigue life extension for the fleet, based on a comparison of original load spectrum used for potential determination to the one corresponding to the mean current use. The study concluded on an extension from 15 000 to 45 000 flight hours. The study basically compares the cumulative fatigue damaging associated to both loading spectra, for different points of the structure. Need for an update of the fatigue potential Today, the mean use has evolved due to the reduction of the need for training transport pilots and the mutualisation of the instruction between the French Forces. As a result, the Navy only performs basic training with “Xingu” and the number of transport flights has notably increased in its mean use. As those flights are pressurized and undergo heavier payloads than the standard ones, the result is a global increase of the load spectrum severity. As a consequence, the potential of 45000 flight hours is no longer applicable to the concerned aircraft. If the initial potential of 15 000 flight hours is restored, such a decrease made the remaining fatigue lifetime insufficient to reach the operational objective of a retirement of the fleet in 2020. Otherwise, the spectrum is still less severe than the original one. The fleet needs consequently an update of its fatigue potential, taking into account the change of usage. As far as possible, it should be completed in an adaptable way considering that a change of the mean use could still be possible in the future.

2 Overall Strategy The overall strategy is largely based on the one used by Embraer in its life extension of 1994 [1]. In particular, the choice of the critical points, the simplification of the considered sources of load and the damaging references in terms of test results can only be chosen by the aircraft manufacturer. Critical points Six critical points are here considered considering the structural point which encountered a fatigue crack during either a full scale fatigue test or a dedicated component test. The first one is the front spar to fuselage lower attach fitting (Figure 3), and will be referenced in the following as “point 1”, at the most loaded rivet hole. The second one (point 2 on Figure 2) is the engine mount attachment on the front spar of the wing. The third one (point 3) is on the front spar carry-through, which is a beam of 7075-T6 machined out of a forged block. The damage appears in the lower flange, at the edge of a hole for riveting the piece to the lower fuselage skin. The fourth one (point 4) is on the rear spar carry-through in the lower flange, at a rivet hole for fuselage skin attachment. The fifth one (point 5) is the main landing gear and the sixth (point 6) is the fuselage, as the whole.

658

P. Madelpech et al.

Point 3 Point 4

Point 1 Fig. 3 Critical points locations (from [1]).

Point 2 Fig. 4 Critical point 2 location.

Mean use determination

Several mission profiles have been defined to cover the use of the aircraft. Each one is composed of segments, each having a type (climbing, cruise or descent), a duration, a speed and an altitude. In addition, a take-off weight is determined and a mean weight is computed on the segments. The missions (cf. Figure 5) are defined by the end-user according to the statistics on performed flight. To update the data, a special attention has been paid to the take-off weight, which was too much rounded in the past study, though being a very important parameter for the load computation. Similar segments of mission in terms of weight, altitude, speed have been gathered to reduce the number of computation cases.

Life Extension: Fatigue Lifetime Updating of the French Xingu Fleet

659

Fig. 5 Example of a mission profile.

Loading spectra

As the critical points 1 to 4 are all in the lower part of the wing spars or spars carry through, the hypothesis was made by Embraer that only load cases inducing wing bending have to be considered, all the others (yawing, rolling, lateral gusts) being neglected. As a result, the vertical acceleration nz is a key factor for the intensity of the loading spectrum. Ground loads are also considered. For the landing gear, a comparative analysis of loading is performed simply by weight comparison to obtain a new fatigue potential. For the fuselage, as the flight loads are less severe than the original one, the analysis only consists in the comparison of the number of pressurization cycles between the usages. The loading spectra on the wings are therefore composed by those sources of solicitation: -

Manoeuvres: statistics on reached nz for training military aircraft are issued from [2]. Vertical gusts: statistics on encountered speeds of vertical gusts are provided by [3]. Ground loads: landing, braking, turns and engine tests are considered through the mean use. Ground – Air – Ground cycle is also considered.

Combined with the definition of the mean use and therefore the time spent in 1000 flight hours at given altitude and speed, it enables to have a global loading spectrum in terms of bending moment at the different locations of the wing corresponding to the critical points.

660

P. Madelpech et al.

Local stress computation

As explained, the damage calculation and therefore the need to compute the local load spectra are only necessary for the points 1 to 4. The local stress is supposed to be proportional to the stress in the lower cap of the nearest spars. For point 1 and 3, the bending moment at the wing to fuselage junction (y = 0.98 m) and the stress in the front spar cap are considered. For point 4, the same moment and the stress in the rear spar cap are used. For point 2, the bending moment at the engine mount (y = 2.55 m) and the front spar cap are taken into account. The linear relationships are not explicitly given in the reference and have been obtained through a test campaign described below. Damage computation and safety factors

Traditional fatigue damage formulas based on Miner cumulative laws are used to assess the fatigue damage associated to the predetermined loading spectra. For the material data, the Wöhler curves are provided by [4]. The obtained fatigue damages are compared to the ones which made a crack occur at the points during the tests [1] and the fatigue potential is deduced by proportionality. A safety factor is then applied to the theoretical results to determine the potential of the fleet. Taking into account that the justification is purely made by analysis and that the computed stress levels are low (and consequently the scattering important), relatively high safety factors are applied. They represent the coefficient to apply to go from a probability of failure of 50 % to 10 % on the corresponding Wöhler curve and given Table 1. Despites all this, the obtained potentials are high and the global fatigue lifetime of the aircraft is limited by point 3 to 45 000 flight hours in the Embraer study [1] (see Table 1). Table 1 Fatigue potential by point after life extension [1].

Damaging in test

Safety coefficient

Validated potential [1] (fh)

Point 1

0.00616

14.8

223 900

Point 2

0.00996

40.9

286 000

Point 3

0.21014

25.35

45 796

Point 4

0.01094

25.35

1 304 000 000

Life Extension: Fatigue Lifetime Updating of the French Xingu Fleet

661

3 Flight Tests The flight tests aim at: -

-

Giving the missing relationship between the global flight parameters, the corresponding wing bending moment and the local stress at the critical point. Those links are mandatory to pass from the considered use to the local damaging. Comparing the severity of “touch and go” and landing. As a potential is also declined in an authorised number of landings and as this potential is generally reached before the flight hours one, because of the landing training, it could be highly valuable, in term of fatigue life extension not to count the touch and go, as the lift does not disappear. But before that, it must be verified that it is consistent with the stress analysis. This part of the study is not covered by the article.

The campaign was completed at DGA Flight Tests in Istres (France) on EMB-121 n°77 from the Navy from October 2010 to January 2011 for the basic measurement, the validation phase being still on progress. Instrumentation

Only points 1 to 4, the critical ones, are concerned and they are all located in the not pressurized area. To avoid passing wires to the inside of the fuselage, the data acquisition unit, ACRA© KAM-500, was placed at the wing - fuselage junction, under the fuselage. As the available space was reduced only eight measurement slots were available plus one for the GPS time to be synchronized with the parameter data recorded on personal computer inside the cabin, where different parameters were recorded directly from the flight data recorder (FDR): speed, altitude, position angles, load factors, flaps and landing gear positions or engines regimes. Most of them were only exploited to verify the stability of the flight at a measurement point. The 8 strain gauges locations (cf. Figure 6) were chosen as close as possible of the critical points. The gauges have a resistance of 350 Ω and are powered by +/5V. They were adapted to the metal on which they were bonded. As, at the maximum service altitude, the outside temperature was lower than the range of adapted ones, a half bridge was constituted with one identical gauge, in the vicinity, bonded on silicone and a sheet of the same metal to be not mechanically affected, but undergoing the same thermal dilatation (cf. Figure 7). The complement of the half bridge is made inside the data recorder.

662

P. Madelpech et al.

J4 J2 J1

J3

J5-J8

J7

J6

Fig. 6 Overview on gauges locations in the aircraft structure.

More in detail, J1 and J7 were symmetrically located at the vicinity of the wing - fuselage junction, but beyond the attachment fittings (y = 1.5 m), as the load transfer through the 11 fixations would be hard to compute. J2 is located another 20 cm beyond along the wing span. Those gauges serve to estimate the bending moment at point 1. J3 is located at the same lateral abscissa as J1 but on the rear spar cap to complete the global bending moment assessment. J4 is located beyond the engine attachment and is used to estimate the bending moment at the engine attachment. For this particular case the gauges were bonded on upper and lower cap to measure directly the bending stress.

Life Extension: Fatigue Lifetime Updating of the French Xingu Fleet

663

Second gauge on silicone

Strain gauge

Fig. 7 Gauge J7.

J5 J8

Fig. 8 Instrumentation at point 3.

A special attention was paid to critical point 3; two gauges (J5 and J8) were installed at the same lateral station but at different heights on the beam profile (Figure 8) to have redundant stress data and thus to be able to detect measurement errors. Moreover, this aims at obtaining information on local bending of the beam. J6 is located like J5, as close as possible of the location of point 4 on rear carry through spar.

664

P. Madelpech et al.

Finite element models

As mentioned, it was not possible to measure the stress at the exact locations of the critical points. At best, the gauge was located in the vicinity of the hole at the edge of which the fatigue damage occurred. To obtain the relationship, local finite element models were made, on the one hand, to correct the measures of the stress concentration due to holes proximity and thus to have the nominal stress and, on the other hand, to compute the local stress concentration factor Kt at the critical point. Those data are necessary to use corresponding material data on fatigue initiation. On the example of Figure 9 corresponding to critical point 3, the stress concentration coefficient Kt and the stress at point 3, function of the gauge measurements, were determined:

σ po int 3 = Ktσ nom = 2.9 (0.945σ J 5 + 0.075 (σ J 5 − σ J 8 ))

(1)

J8

J5

Fig. 9 Local FE model.

Calibration – error of measurement

Calibration of the measurements acquisition chain is an important first step to validate the data. The theoretical precision of the measure associated to electrical assembly is about 3 MPa. Three additional causes of errors are investigated:

Life Extension: Fatigue Lifetime Updating of the French Xingu Fleet

-

-

-

665

Firstly, the instrumentation was made on an on-the-ground aircraft and therefore a stressed structure. The strain measurement “0” is not the unloaded structure. Secondly, there are always small differences in the resistances of the gauges and also additional resistances due to the wire of the installation, which disturb the strain measures. Finally, there can be a drift of the electrical measurement due to the heating of the electronic acquisition module. Contrary of the other phenomena, this is not a static error and therefore it is more difficult to handle because it results in an evolution of the reference of the measures. It can reappear at each change of temperature, so at each change of altitude. Moreover, the variations are slow and can escape notice if the controls are realised to rapidly.

Calibration typically consists in applying separately the different sources of loading in relatively low but calibrated amplitudes and to verify the coherence of the measured strains. In our case, due to the impossibility of applying the aerodynamical lift and due to the lack of reference, the calibration has only consisted in a data acquisition during the filling up of the fuel tanks. It was not without detrimental consequences for the following. In fact, a slow electrical drift has not been detected during the calibration. Consequently, a part of the data from the first flight tests was not valid In order to quantify the problem a simple flight test was made consisting in 2 hours of on the ground measurement, take off and climbing at FL 190, steady flight at 170 kt during half an hour with turns at high load factors nz, then decent to 5000 feet, 30 minutes of steady flight in the same conditions and return with another 30 minutes of recording on the ground. Parts of the results are drawn in Figure 10 and Figure 11.

On the ground measurement 4

2

0 Stress

J1 Mpa J2 Mpa J4 Mpa

-2

J5 Mpa -4

-6

-8 10:33:36

11:02:24

11:31:12

12:00:00

12:28:48

12:57:36

13:26:24

time

Fig. 10 On the ground measurement: drift of the data.

666

P. Madelpech et al.

In flight measurement (FL 190 - 170 kt) 50 45 40 35 J1

stress

30

J2

25

J4

20

J5

15 10 5 0 13:40:48

13:48:00

13:55:12

14:02:24

14:09:36

14:16:48

time

Fig. 11 In steady flight measurement: drift of the data.

After that, it was decided: -

to start the recording of the data one hour before each flight to wait 15 minutes at each change of altitude causing significant ouside temperature variation before beginning the load factors variations.

Even after that, the problem was not solved and special attention must be paid to results validity. Those problems were relatively heavy to handle and led to a huge waste of time. The data recorder ACRA© KAM-500 respects the norm [5] but the requirement level of the norm are very low for a small aircraft with limited endurance (4 hours of stabilization before the measures). As a conclusion, the best would have been to put the recorder inside the cabin at a controled temperature. Otherwise, the results are remarkably not noisy in flight, in not turbulent conditions. In presence of severe gusts, no correlation could have been done between the gauges and the flight parametres and those aerologic conditions have therefore been avoided. Another point is that no measurement have been made at negative load factors because when trying to, the fuel was not properly injected to the engine anymore and so did stop. At nz positive but inferior at 1, recording problems occur for flight parameters on the PC inside the cabin. It was partially solved by using a flash disk instead of the classical hard disk. In flight testing

The flight testing consists in performing calibrated manoeuvres and thus calibrated vertical load factors nz, at different altitudes, speeds and configurations (flaps and landing gears). The aim is to cover the mean use as much as possible to limit the number of assumptions.

Life Extension: Fatigue Lifetime Updating of the French Xingu Fleet

667

Steady flight h = FL190 - nz = 1 - Vi = 170 kt 25,00 20,00 J2 J3

Stress

15,00

J4 J6 J7

10,00 5,00 0,00 5360

J5

5380

5400

5420

5440

5460

-5,00 Total weight

Fig. 12 Influence of the total weight.

The first influent parameter is the total weight of the aircraft, which should be lifted by the wings: a proportional relationship is expected. Then, the influence of the speed was also investigated.

Steady flight - h = 1000' - nz = 1 25,00

20,00

J1

Stress

15,00

J4 J6 10,00

J7

5,00

0,00 100,0

120,0

140,0

160,0

180,0

200,0

220,0

Vi (kt)

Fig. 13 Influence of the dedicated speed.

As noticed Figure 12 and Figure 13, the variation of stress at the gauges location is very low and there is no clear tendency in the evolution. Results

To minimise the errors on the results, two phases of analysis are proceeded. First the values at “1g” are measured depending on the other parameters. It enables to detect the drift problem. It serves as reference over the different flights. Correction

668

P. Madelpech et al.

due to the weight and remaining fuel volume are applied. In a second time, a variation of stress is associated to a variation of nz from 1 to its value and others parameters on consecutive measures considering that no significant drift can occur during this period inferior to two minutes. The measures were made at four different altitudes corresponding to the mean use of the aircraft, 1000 ft, 3000 ft, 5000 ft and FL190. The indicated airspeeds were also chosen according to the values needed for the fatigue computations. The maximum range of nz respecting the flight envelope of the aircraft and the stall speed was investigated. For each nz greater than one, a stabilisation period of 15 seconds was aimed. As it was difficult to perform it due to the required piloting skills and, above all, the poor manoeuvrability of the aircraft, shorter periods were finally accepted, but only after validation, if a good correlation between the stress and the load factor nz was observed around the measure time.

4 Fatigue Damage Computation The final step of the program is to compute the fatigue potential associated to the actual mean use. Having previously determined the solicitation spectra associated to manoeuvres, gusts and ground loading and having obtained the relationship between those solicitations and the local stress at the critical points, it is possible to compute the fatigue damaging for 1000 flight hours. Cumulative spectra on load factors nz by segment of profile are converted in cumulative spectra of bending moment and then in a cumulative spectrum of local stress at each critical point. To have cumulative spectra of load cycles from σmin and σmax, the hypothesis was made that the association can be made between the same occurrences numbers. A 21 points sampling of the cumulative spectrum was used to compute the fatigue damaging. The ground-air-ground cycle was determine by the minimum of the reached stress during the ground evolution (supposed uniform between the flights) and the σmax having the occurrences number of the desired flights number. As the ground evolutions are of low amplitude, other influence was neglected on the fatigue damage of the structure. Despite it and as the main landing gears are under the wings, the ground-air-ground cycle has low impact on fatigue, except for point 2. Having performed a computation by mission, it is possible to compare the severity. As illustrated Figure 14 on load factors due to gusts by missions, it appears that all the missions have not the same contribution in the fatigue damage. The interest of having homogenized usage appears obvious. The landing gear potential is updated by analysis of the mean mass for ground evolution. Another difficulty deals with the evolution of the pressurization cycles. Their number increase significantly by stopping to use the aircraft for training and may become a sizing parameter. This phenomenon is not correctly taken into account by the mean percentage of the flight time spent in transport. It only considers the duration of the flights and not their occurrences.

Life Extension: Fatigue Lifetime Updating of the French Xingu Fleet

669

Exceedence number of Dnz 2,5 2 1,5

Loading factor

1

mission A

0,5

mission B 0 1,E-07

1,E-05

1,E-03

1,E-01

mission C 1,E+01

1,E+03

1,E+05

1,E+07

-0,5

mission D

-1 -1,5 -2 -2,5

Fig. 14 Exceedences number of load factors due to gusts by missions.

5 Conclusion The study enables to obtain the necessary 25 000 hours potential to extend lifetime until 2020. Moreover the flight safety was increased by pointing out the worst missions and the importance of standardization of the aircraft use over the fleet concerning those missions. An action is currently going on to compile the history of the fleet and make forecast on the fatigue lifetime evolution. The risk taken by working on mean usage and probability of event occurrences is acceptable regarding the safety coefficients and the low stress levels reached in normal flight. Nevertheless the comparison with the test results with such different load spectrum remains arguable.

References [1] Fontana, F.T.: Analyse de fatigue de l’avion EMB-121 Xingu pour les missions d’instruction Armée de l’Air et Aéronautique Navale Françaises, 121-FA-26. vol. 1 (1991) [2] MIL-A-008866B, Military Specification Airplane Strength and Rigidity Reliability Requirements, repeated loads and Fatigue, USAF [3] ESDU 69023, average gust frequencies, Subsonic transport aircraft, Amdt D, The Royal Aeronautical Society (1989) [4] MIL-HDBK-5E, Military Standardization Handbook, Department of Defense (1986) [5] MIL-STD-810 F. Methods 501.4 and 502.4

26th ICAF Symposium – Montreal, 1-3 June 2011 Fatigue Life of Cold Expanded Fastener Holes at Short Edge Margins G.M. Vallières and D.L. DuQuesnay Royal Military College of Canada, Kingston, Ontario Canada

Abstract. The fatigue life of cold-expanded fastener holes with short edge margins was studied on aluminum 7075-T6 specimens with straight open holes. The study was done in two parts: experimentally and through finite element analysis. The experiments measured the total fatigue life and crack growth, and the results from the finite element analysis consisted of tangential residual stress profiles. The experiments showed that, at all edge distances, the fatigue life increased with the level of cold expansion. The edge distance, on the other hand, only had a significant effect at the highest level of cold expansion, with fatigue life decreasing at low edge distances. The finite element results were used to make fatigue life predictions that corresponded reasonably well with the experimental results. Strain-life was used for non-cold expanded holes, but crack growth had to be taken into account for coldexpanded holes.

1 Introduction Fastener holes cause stress concentrations in mechanical joints, and hence are inherently prone to fatigue damage. This stress concentration can be further increased if the edge margin, the distance between the edge of the component and the center of the hole (e) normalized over the hole diameter (d), is small. A minimum edge margin between 1.5 and 2.0 is generally specified for aircraft components [1]. However, this distance can be reduced if the hole is out of alignment, or if it is subject to a repair scheme which involves oversizing the hole. Recent experience on Canadian Forces CF-188 Hornet fighter jets has shown a need to repair fastener holes with edge margins below 1.5, and cold expansion was investigated in this context [2]. Cold expansion is a means of reducing fatigue damage occurring at fastener holes with normal or short edge margins by imparting a beneficial residual compressive stress field around the hole. The most common cold expansion method in the aviation industry, the Spit-sleeve method commercialized by Fatigue Technologies Inc., was used in the present study. Cold expansion can be used by manufacturers on new aircraft, and on aircraft in service as a repair scheme or to enhance fatigue life of critical components. Cold expansion is simple, relatively low cost, and does not add weight to components [3]. The fatigue life

672

G.M. Vallieres and D.L. DuQuesnay

improvement factor credited to cold expansion varies, but is generally between two and ten [4]. One of the factors that affect the fatigue life of cold expanded fastener holes is the degree of cold expansion, %CE. It is expressed as a percentage based on the difference between the maximum diameter of the cold expansion tool (which consists of a mandrel and a split-sleeve), Dtool, and the diameter of the starting hole, Dhole, as shown in Equation 1. The accepted optimum degree of cold expansion varies between 4% and 6% [5]. The entire amount of expansion is not retained after the tool has been pulled through the hole, due to material spring back. After the hole has been expanded, it is reamed to its final size.

%CE =

Dtool − Dhole × 100 Dhole

(1)

Another factor that affects the fatigue life is the edge margin. When the edge margin is small, the residual stress due to cold expansion interacts with the stressfree edge of the plate, and the tangential residual stress profile around the hole is no longer symmetric with respect to the center of the hole [6]. Even though cold expansion is commonly used in practice, its fatigue life benefits have not generally been included in fatigue life computations because of the complexity in quantitative predictions and assessment of the residual stresses [7]. Cold expansion of straight holes has been studied extensively; however, there is little experimental data on the effect of edge margin and its combination with different levels of cold expansion. Furthermore, the residual stress profiles obtained from finite element analysis have not been used successfully to make fatigue life predictions.

2 Experimental Results Specimen manufacturing The material used in this study was aluminum 7075-T6, an alloy commonly used in aircraft structural components. It was received as cold rolled sheets with a nominal thickness of 3.175 mm (0.125 inch). Four hourglass specimens were tested in tension to determine the mechanical properties of the material. The yield stress was experimentally found to be 499 MPa, the ultimate stress 568 MPa, and Young’s Modulus 69.0 GPa. The fatigue specimens consisted of a rectangular plate 228.6 mm (9 inches) by 50.8 mm (2 inches) machined from the aluminum sheets. Ten different series of specimens were manufactured from these blank rectangles with various degrees of cold expansion and edge distances, as listed in Table I. All the holes were reamed to 6.30 mm diameter after preparation.

Fatigue Life of Cold Expanded Fastener Holes at Short Edge Margins

673

Fatigue tests The fatigue tests were performed in force control, at constant amplitude sinusoidal loading, on an axial closed-loop servo-hydraulic testing system, with an R-ratio of 0.1 and a maximum net stress of 225 MPa. Table 1 shows the mean and standard deviation of the fatigue life for each series of specimens. The life improvement factor (LIF) shown in the table represents the ratio of the mean life for the series to the mean life for a hole with no cold expansion at the same edge distance. Figure 1 shows the life in cycles versus the edge distance for different levels of cold expansion. Figure 2 shows the life in cycles versus the degree of cold expansion at an edge distance of 2.0. These plots show the results for individual specimens as points, and the mean life as a curve. The experimental results were analyzed by assuming that the fatigue data had a log-normal life distribution. This assumption was verified in the analysis and found to be valid. A two-way analysis of variance (ANOVA) was done to determine the effect of edge distance and level of cold expansion on the life of the specimens. It showed that the edge distance, the level of cold expansion and their interaction all had a significant effect on the fatigue lives of specimens, with the probability of these results occurring randomly on the order of 10-8. A series of one-way ANOVA gave more detailed information. The effect of edge distance was not significant on open holes with cold expansion levels of 3.4% and below, but was significant at 4.7% cold expansion, with a 0.007% probability of these results occurring randomly. The degree of cold expansion, on the other hand, had a significant effect at all investigated edge distances, with a 0.009% probability of these results occurring randomly. Furthermore, a series of F-tests were performed to determine if there was a significant difference in variability between the different hole treatments. The only significant difference was found between 4.7% cold expansion and no cold expansion. The variability in the fatigue life at 4.7% cold expansion was higher than that of holes with no cold expansion. Table 1 Specimen series description and fatigue test results. Series 1 2 3 4 5 6 7 8 9 10

Level of cold expansion (%) 0 4.7 4.7 4.7 3.4 1.7 0 0 3.4 3.4

Edge distance (e/d) 2.0 2.0 1.75 1.5 2.0 2.0 1.75 1.5 1.75 1.5

Number of specimens 3 4 3 3 3 3 3 3 3 3

Mean life (cycles) 12 953 183 477 51 978 43 295 32 239 20 877 12 780 11 936 30 631 27 429

Standard deviation (cycles) 1577 48 727 14 124 5242 6712 1924 1714 1409 5377 1735

LIF 1.0 14.2 4.1 3.6 2.5 1.6 1.0 1.0 2.4 2.3

674

G.M. Vallieres and D.L. DuQuesnay

Fig. 1 Effect of edge distance on life at different levels of cold expansion. ■ Series 1: e/d = 2.0, no CE, ○ Series 2: e/d = 2.0, 4.7% CE, ▲ Series 3: e/d = 1.75, 4.7% CE, Δ Series 4: e/d = 1.5, 4.7% CE, + Series 5: e/d = 2.0, 3.4% CE, □ Series 7: e/d = 1.75, no CE, ● Series 8: e/d = 1.5, no CE, ◊ Series 9: e/d = 1.75, 3.4% CE, x Series 10 : e/d = 1.5, 3.4% CE.

Fig. 2 Effect of cold expansion on life at e/d = 2.0 edge distance.

Fatigue Life of Cold Expanded Fastener Holes at Short Edge Margins

675

Crack growth Crack growth was monitored using a pair of video cameras connected to the testing system. Prior to each test, the system was set to pause after a given number of cycles, and then at regular intervals, and take pictures of each side of the specimens. The pictures were used after the test to extract crack growth data. Due to limitations in the method, crack growth data is not available for all specimens. Figure 3 shows crack growth data for some of the cold expanded specimens.

Fig. 3 Crack growth results for cold expanded holes. ■ 4-2: e/d = 1.5, 4.7% CE, □ 4-3: e/d = 1.5, 4.7% CE, ● 5-1: e/d = 2.0, 3.4% CE, ○ 5-3 : e/d = 2.0, 3.4% CE, ▲ 6-1 : e/d = 2.0, 1.7% CE, Δ 6-3 : e/d = 2.0, 1.7% CE, + 9-1 : e/d = 1.75, 3.4% CE, ◊ 12-2 : e/d = 1.75, 3.4% CE, x 10-2 : e/d = 1.5, 3.4% CE.

The most interesting specimen from a crack growth measurement perspective was Specimen 4-3 (4.7% CE, e/d = 1.5) because more pictures were available at short crack lengths for this specimen. The data reveal that a very short crack was present for a considerable portion of the fatigue life without much growth, and that it started growing faster only after it reached a certain critical length (approximately 0.51 mm). Qualitatively, some of the other crack growth curves seem to show the end of this phenomenon, but none have enough data available to show it quantitatively. This phenomenon is in line with the literature, where it has been stated that cold expansion increases the fatigue life mostly through an increase in crack growth life [5, 8]. The latter stage of crack growth was similar for all specimens. The crack growth was rapid, spanning only a few thousand cycles, which for most specimens, corresponded to a small fraction of their fatigue life.

676

G.M. Vallieres and D.L. DuQuesnay

Fractography Every failed specimen was prepared for fractography and observed using an optical microscope. Most of the cracks started on the short leg of specimens, more so for edge distances e/d = 1.75 and e/d = 1.5. In non cold-expanded holes, the cracks seemed to start at the entry or exit face corners and inside the bore interchangeably. In cold expanded holes, however, cracks always grew primarily toward the entry face, so that they appeared on the cameras to start at the entry face. This is true in every case, except at the lower cold expansion level of 1.7%. This explains the presence of a shear lip at the exit face of most cold expanded specimens, since often the crack never broke through the exit face until final failure.

3 Finite Element Analysis Model description A three-dimensional non-linear finite element analysis of the experimental specimens was conducted using ANSYS. The 7075-T6 aluminum was modeled as a non-linear kinematic hardening material, based on the experimental results obtained from the tensile tests. The stainless steel of the cold expansion tool was modeled as an elastic material. Cold expansion of the hole was modeled using contact elements. It was assumed frictionless because of the lubricated sleeve. The first load applied to all models was cold expansion. The cycling load was then applied in at least two load steps after cold expansion, corresponding to the maximum and minimum loads. The final reaming step was neglected, as it has been deemed to have a negligible influence on the final residual stress profile [9]. Results Tangential residual stresses are in the same direction as the applied stresses at the notch root, hence they affect the fatigue life of the specimens. The tangential residual stresses at the mid-plane of the plate are presented in this section. Figure 4 shows the tangential residual stresses at different edge distances with 4.7% cold expansion, Figure 5 shows the tangential residual stresses at different edge distances with 3.4% cold expansion, and Figure 6 shows the residual stresses at e/d = 2.0 with different levels of cold expansion. The tangential residual stress profiles obtained from the cold expansion model displayed some specific features at all studied edge distances and levels of cold expansion. First, the maximum compressive tangential residual stress occurred on the exit face, near the edge of the hole. The tangential residual stress profile at the mid-plane was also compressive near the hole, and similar to the stress profile on the exit face. The tangential residual stresses on the entry face, however, were much higher and often tensile. These results are in line with those published by Pavier et al., who also used frictionless contact in their model [10].

Fatigue Life of Cold Expanded Fastener Holes at Short Edge Margins

Fig. 4 Residual tangential stress at mid-plane with 4.7% CE.

Fig. 5 Residual tangential stress at mid-plane with 3.4% CE.

677

678

G.M. Vallieres and D.L. DuQuesnay

Fig. 6 Residual tangential stress at mid-plane with e/d = 2.0.

Second, the residual tangential stresses on the short ligament of the plate did not tend to zero at the edge. This has been observed by other authors [6, 9]. Finally, the residual tangential stress profile showed a peak tensile stress at a distance away from the hole, as would be expected from equilibrium within the material, and was described by Chakherlou and Vogwell [11]. A re-yielding zone was visible on the models with 4.7% and 3.4% cold expansion at the mid-plane and exit face at all edge distances. Due to this re-yielding zone, the maximum compressive stress occurred a short distance away from the edge of the hole, as described by Julien et al. [12].

4 Discussion Effect of edge distance As was expected from the literature [6], cold expansion created residual tangential compressive stresses at the edge of the hole, and was therefore beneficial to fatigue life at all edge distances. Edge distance had a significant effect on the fatigue life of holes with 4.7% cold expansion, but not at lower levels of cold expansion. From Figure 1, it is obvious that there is a large drop in fatigue life between e/d = 2.0 and e/d = 1.75, and that the drop between e/d = 1.75 and e/d = 1.5 is much lower. The finite element results, on the other hand, showed differences in the residual stress profile with varying edge distance at all edge distances with 3.4% and 4.7%

Fatigue Life of Cold Expanded Fastener Holes at Short Edge Margins

679

cold expansion. Figure 4 shows that, at the midplane with 4.7% cold expansion, the maximum tensile residual tangential stress and the residual tangential stress at the edge of the plate increased with decreasing edge distance. The most important difference was seen in the stress profile of e/d = 1.5, where the minimum compressive residual stress on both the short and long leg sides of the plate, and the tensile residual stress at the edge of the plate were substantially higher. In Figure 5, representing the mid-plane stress profile of holes with 3.4% cold expansion, it can be seen that the difference between the stress profiles at various edge distances was less important than with 4.7% cold expansion. The fact that the residual tangential stress profiles with 3.4% cold expansion were very similar can explain why varying the edge distance at this level of cold expansion did not have a significant effect on the experimental fatigue life. Similarly, the fact that there was a noticeable difference in the stress profile at 4.7% cold expansion explains the fact the edge distance had a significant effect on the experimental fatigue life. However, the fact that the largest differences occurred between e/d = 1.75 and e/d = 1.5 with FEA, and between e/d = 2.0 and e/d = 1.75 experimentally could be explained by the fact that the residual stress profiles illustrated are two-dimensional representation of what is really a three-dimensional phenomenon. Effect of level of cold expansion The level of cold expansion had a significant effect on fatigue life at all investigated edge distances. As shown in Figure 2, at e/d = 2.0, the largest difference in fatigue life occurred between 4.7% and 3.4% cold expansion. The difference in fatigue life between 0%, 1.7% and 3.4% cold expansion was much smaller. There was also a significant difference in the variability of the results between 0% and 4.7% cold expansion. The finite element analysis results for residual tangential stress at mid-plane with e/d = 2.0 are shown in Figure 6. It can be seen from this figure that the magnitude of the compressive stress at the edge of the hole, the magnitude of the maximum compressive stress, the maximum tensile stress and the stress at the edge of the plate all increased substantially with increasing level of cold expansion. Most of the variation in tangential residual stress profile occurred between 3.4% CE and 1.7% CE and between 1.7% CE and 0% CE, which would appear as a horizontal line in Figure 6. From these results, it is clear that the magnitude of the compressive stress at the edge of the hole and the minimum compressive residual stress are the critical factors in determining fatigue life. Fatigue life decreased for a decreasing compressive stress magnitude at e/d = 2.0, even if the residual tensile stresses away from the hole also decreased. However, the fact that the largest differences occurred between 0% and 3.4% cold expansion with FEA, and between 3.4% and 4.7% cold expansion experimentally could be explained by the fact that the residual stress profiles illustrated are a two-dimensional representation of what is really a three-dimensional phenomena.

680

G.M. Vallieres and D.L. DuQuesnay

Significant interaction effects were also found between the edge distance and the level of cold expansion. This can be appreciated qualitatively by the fact that fatigue life improved steadily with increasing cold expansion level and increasing edge distance, but there was a jump in fatigue life at the highest studied edge distance and level of cold expansion (e/d = 2.0 and 4.7% CE). This makes sense since the full benefit of a higher level of cold expansion can only be felt by the component if the edge distance is sufficiently large that the free edge does not disrupt the residual stress pattern extensively. A shear lip was present to some extent on almost all cold expanded specimens, but was not present on non-cold expanded specimens. This shows that the residual stresses on the exit face of cold expanded specimens were high enough that the crack had difficulty breaking through this surface. This phenomenon occurred even at the lowest cold expansion level of 1.7%, showing that there was a lasting effect on the material around the hole even at such a low cold expansion level. It also shows that the residual stresses on the entry and exit face were different at all cold expansion levels, the exit face being more compressive, as was predicted from finite element analysis. Finally, the fact that fatigue life variability at 4.7% cold expansion was significantly higher than variability at 0% cold expansion was also expected [13]. There are more steps to the process, therefore more chances of increasing variability, in a cold expanded hole than in a non-cold expanded hole. It is at the highest degree of cold expansion, where the benefits should be fully realized on the fatigue life, that this increased variability is more evident. Fatigue life predictions The finite element analysis results can be used to make fatigue life predictions, which can then be compared to the experimental results. The total fatigue life of a notched component can be calculated using Equation 2, where Nf is the total life, Ni is the initiation life and is found using a strain-life approach, and Np is the propagation life and is found using a fracture mechanics approach. Based on the literature and visible crack growth data, it appears that most of the fatigue life of non-cold expanded specimens was spent in the initiation phase. Therefore, an approach based on strain-life was used to calculate the total fatigue life for these specimens. For the cold expanded specimens, both initiation and propagation were taken into account. The approach for these specimens combined strain-life and fracture mechanics.

N f = Ni + N p

(2)

Predictions for non-cold expanded holes. The local stress and strain at the notch root was determined using the finite element results. A number of different models were tried, and the best results were obtained using the stress and strain values at a node that was at the edge of the hole, on the short leg side, 25% of the thickness down from the entry face. The stresses and strains were taken in the tangential direction.

Fatigue Life of Cold Expanded Fastener Holes at Short Edge Margins

681

Table 2 shows the stresses and strains at this node for the load step at the maximum cyclic load and the load step at the minimum cyclic load. The strain-life relation with Morrow’s mean stress correction (Equation 3) [14] was then used to determine the number of reversals to failure, which represents twice the number of cycles to failure, Nf. The stain life material parameters were taken from Dowling [15]. They are the fatigue strength coefficient σf′ = 1466 MPa, the fatigue strength exponent b = -0.143, the fatigue ductility coefficient εf′ = 0.262, and the fatigue ductility exponent c = -0.619. Young’s modulus E = 71 GPa is from the cyclic stress-strain curve.

Δε σ ′f − σ m 2N f = 2 E

(

)

b

(

+ ε ′f 2 N f

)

c

(3)

Table 2 Non-cold expanded hole specimens strain-life predictions. Series 1 7 8

σmax (MPa) 530.3 532.4 525.2

σmin (MPa) -11.5 -17.7 -23.2

εmax

εmin

σm

Δε/2

Nf

0.009235 0.009456 0.009563

0.001585 0.001692 0.001785

265.2 266.2 262.6

0.003825 0.003882 0.003889

29 000 26 750 26 700

Experimental results 12 953 12 780 11 936

Predictions for cold expanded holes. For cold-expanded holes, both the initiation and the propagation life were taken into account. Some of the literature [5, 8] and some of the crack-growth data (Specimen 4-3) suggested that cold expansion increases propagation life more than it increases initiation life. Therefore, an assumption was made that cold expanded holes develop very short cracks at the same time as the equivalent non-cold expanded hole would, and that the increase in life due to cold expansion is solely due to an increase in the propagation life. Based on this assumption, the initiation life predicted previously for a non-cold expanded hole with a given edge distance was assumed to represent the initiation life of a cold expanded hole with the same edge distance. A finite element analysis that included cyclic stresses was then performed on the cold expanded holes. The maximum and minimum tangential stress profiles in the short ligament of the specimen were extracted from this finite element analysis. A hypothesis was made that a stress profile representative of the crack growth for the entire propagation life could be extracted somewhere between 25% of the thickness down from the entry face and the mid-plane of the specimen. The exact location of this stress profile is one of the parameters that could be adjusted to obtain a fatigue life similar to that measured experimentally. The second adjustable parameter was the initial crack length, which was selected as 3 μm. Because the finite element model did not contain enough elements in the thickness to directly determine the point that would be representative of the crack growth life, it was assumed that the stresses between 25% of the thickness and the mid-plane varied linearly. Using the available data and linear interpolation, it was determined that for specimens with edge distance e/d = 2.0, the stress profile at

682

G.M. Vallieres and D.L. DuQuesnay

33% of the thickness down from the entry face was most representative of the propagation life. Unfortunately, this did not work for specimens with edge distance e/d = 1.75 and e/d = 1.5. Different representative stress profiles had to be determined for these cases. In the case of e/d = 1.75, it was determined that the representative stress profile was 29% of the thickness down from the entry face. In the case of e/d = 1.5, it was determined that the representative stress profile was 28% of the thickness down from the entry face. It is not surprising that stress profiles at different points are representative of different specimen geometries. The idea that crack growth can be represented by a single stress profile is a simplification. In reality the crack grows both down through the material thickness, as well as away from the hole. A more detailed crack growth model that would take into account two-dimensional crack growth, and be more detailed in the material thickness, would likely yield results applicable for all specimen geometries without the need to adjust the location of the stress profile. The maximum and minimum stress profiles were linearized, and the average maximum and minimum stress in each segment were used to calculate the local stress range Δσ. When the stress was negative, a value of zero was used. The stress intensity factor range ΔK could then be calculated using Equation 4 [15]. In this equation, F is the geometry factor and S is the remotely applied stress. In the present case, however, Δσ was used instead of ΔS. The local stress takes into account specimen geometric effects. Therefore, F only had to account for the crack geometry. Assuming that the crack was a quarter-circular corner crack, the literature shows that, in a plate subjected to tension, the value of F is between 1 and 1.5 [16]. For simplicity, it was assumed that F = 1 in subsequent calculations. Therefore, Equation 5 was used to determine the stress intensity factor range. ΔK = F ΔS π a

ΔK = Δσ πa

(4) (5)

Once ΔK was obtained, Forman’s Equation [15] was used to determine the crack growth at every cycle. Equation 6 shows the form of Forman’s Equation that was used. In this equation, Δa is the crack growth for one cycle, R is the stress ratio, and C2 = 5.29 x 10-6 (mm/cycle)/(MPa m1/2(m2-1)), m2 = 3.21, and Kc =78.7 MPa m1/2, are material constants from Dowling [15].

Δa =

C2 ( ΔK )

m2

(1 − R ) Kc − ΔK

(6)

A computer program was written to start with a 3 μm crack length, identify the applicable Δσ, calculate the corresponding ΔK, then the corresponding Δa, update a and add a cycle to N. This process was repeated until the crack length reached the size of the short ligament, or the fracture toughness of the material, KIC = 29

Fatigue Life of Cold Expanded Fastener Holes at Short Edge Margins

683

MPa m1/2 [15], was reached. Then, the number of cycles was added to the initiation life determined earlier to obtain the total fatigue life. This process worked very well for predicting the fatigue life of cold expanded hole specimens. The results are listed in Table III with the experimental results for comparison. The results are very good, considering the simplicity of the process used to calculate crack growth life. The fact that such a simple model can be used to predict the fatigue life of cold expanded specimens is compelling evidence that the life improvement due to cold expansion is due to crack growth retardation. The data from the crack growth computer program for each specimen, especially long life ones, showed that the cracks grow extremely slowly when close to the hole, and that most of the fatigue life was spent at these very short crack lengths. Figure 7 shows the computed crack growth for a Series 4 specimen, as well as experimentally measured crack growth for Specimen 4-3. Both plots show a slow crack growth plateau at approximately 0.5 mm, and then rapid crack growth at the very end of the specimen life. However, the plateau is flatter and lasts approximately 10 000 cycles in the experimental case, while it is steeper and lasts only about 5 000 cycles in the numerical case. Table 3 Fatigue results for cold expanded specimens.

Series

e/d

2 3 4 5 6 9 10

2.0 1.75 1.5 2.0 2.0 1.75 1.5

CE level (%) 4.7 4.7 4.7 3.4 1.7 3.4 3.4

FEA life (cycles) 119 827 46 790 40 526 41 791 29 581 27 958 28 126

Experimental life (cycles) 183 477 51 978 43 295 32 239 20 877 30 631 27 429

Fig. 7 Series 4 computed and observed crack growth.

5 Conclusion This study investigated the fatigue life of aluminum alloy 7075-T6 specimens with cold expanded holes. The edge distance and the level of cold expansion were varied

684

G.M. Vallieres and D.L. DuQuesnay

in a series of experiments and finite element analysis simulations. It was determined that, at all edge distances, the fatigue life increased with increased cold expansion. The edge distance, on the other hand, only had a significant effect at the highest level of cold expansion, with fatigue life decreasing at low edge distances. The present work showed that finite element models can determine qualitatively the combined effects of edge distance and residual stresses due to cold expansion. With calibration, such models can even be used to predict the fatigue life of cold expanded holes. This study also allowed for a better understanding of the combined effects of short edge distances and cold expansion on the fatigue life of aerospace structures. The finite element model and the understanding of fastener hole behaviour can be used to make better decisions relating to the design and repair of fastener holes in aircraft structural components.

References [1] Niu, M.C.Y.: Airframe Structural Design, 2nd edn. Conmilit Press Ltd., Honk Kong (1999) [2] DuQuesnay, D.L.: Hole cold work and interference fit installation and interference fit bushing effects on the life improvement factor for aluminum plates. Mechanical Engineering Report 080701, Royal Military College of Canada, Kingston, Ontario (2008) [3] Forgues, S.A., Bernard, M., Bui-Quoc, T.: Computer Methods and Experimental Measurements for Surface Treatment Effects. In: Aliabadi, M.H., Brebbia, C.A. (eds.) Proceedings of the First International Conference, pp. 61–70. Computational Mechanics Publications, Southampton (1993) [4] Ozdemir, A.T., Edwards, L.: J. Strain Anal. Eng. Des. 31(6), 413 (1996) [5] Burlat, M., Julien, D., Levesque, M., Bui-Quoc, T., Bernard, M.: Eng. Fract. Mech. 75(8), 2042 (2008) [6] Ayatollahi, M.R., Arian, M.: Comput. Mater. Sci. 45(4), 1134 (2009) [7] Seng, H.Y., Ying, Z.G.: Using hole cold working as a fatigue improvement modification. In: Airforce Technology Seminar (2007) [8] Leon, A.: Int. J. Fatigue 20(1), 1 (1998) [9] Kang, J., Johnson, S.W., Clark, D.A.: J. Eng. Mater. Technol. 124(2), 140 (2002) [10] Pavier, M.J., Poussard, C.G.C., Smith, D.J.: J. Strain Anal. Eng. Des. 32(4), 287 (1997) [11] Chakherlou, T.N., Vogwell, J.: Eng. Fail. Anal. 10(1), 13 (2003) [12] Julien, D., Bernard, M., Bui-Quoc, T., Larouche, S.: Fatigue of Aeronautical Structures as an Engineering Challenge. In: ICAF 2003, pp. 663–688. EMAS Publishing (2004) [13] de Matos, P.F.P., McEvily, A.J., Moreira, P.M.G.P., de Castro, P.M.S.T.: Int. J. Fatigue 29(3), 575 (2007) [14] Bannantine, J.A., Comer, J.J., Handrock, J.L.: Fundamentals of Metal Fatigue Analysis. Prentice Hall, Upper Saddle River (1990) [15] Dowling, N.E.: Mechanical Behavior of Materials, Engineering Methods for Deformation, Fracture and Fatigue, 3rd edn. Prentice Hall, Upper Saddle River (2007) [16] Newman Jr., J.C., Raju, I.S.: In: Lewis, J.C., Sines, G. (eds.) Fracture Mechanics: Fourteenth Symposium, vol. 1, pp. 238–265. American Society for Testing and Materials (1983)

26th ICAF Symposium – Montreal, 1-3 June 2011 Environmentally Assisted Cracking in Advanced Aerospace Aluminums E.M. Arnold1, J.J. Schubbe2, P.J. Moran3, and R. Bayles4 1

Midshipman, United States Naval Academy, Class of 2011 2 Asst. Professor, United States Naval Academy, Mechanical Engineering Department 3 Professor, United States Naval Academy, Mechanical Engineering Department 4 U.S. Naval Research Laboratory, Washington D.C.

Abstract. Aerospace alloys, often aluminums, are frequently exposed to corrosive environments resulting from naval service. These environments may produce significant changes in crack growth characteristics in these materials. An experiment was designed to characterize the effects of environment on crack growth rate and fracture mechanism for existing cracks in aluminum 7050-T7451 plate material. This data will be comparatively analyzed against aluminum 7075T7631, an alloy with known susceptibility to corrosion, in order to determine the relative susceptibility of 7050-T7451, generally considered a superior aluminum alloy in terms of strength and corrosion resistance. The resulting data and subsequent analysis can in turn be used in more accurate determination of aircraft component service life in common corrosive environments experienced by aircraft in naval service.

1 Introduction Aircraft designers are constantly seeking materials which will provide the optimal performance to the aircraft which they design. Minimizing weight without sacrificing strength is one of the primary goals in selecting the best material for aviation applications. Resisting the effects of the environment is another important consideration. In the design of new military aircraft, titanium alloys are a primary structural choice due to their high strength to weight ratio and high resistance to corrosion. Historically, aluminums are frequently used in aerospace applications also due to their favorable strength to weight ratio and additionally, their relative low cost to that of alloys such as titanium. In order to reduce weight and costs in manufacturing, designers selected aluminum alloys to include 7050 and 7085 series plate and forgings for use in some recent aircraft variants. Many aircraft components are also constructed out of plate products rather than forged castings in an attempt to reduce the residual stresses generated in deep forgings at corners and to reduce stress gradients due to forming processes. In several newer aircraft designs, aluminum 7050-T7451 was selected as one of the primary structural alloy forms. This alloy, in thick plate form, considered a

686

E.M. Arnold et al.

superior aluminum alloy, was selected prior to full characterization of fatigue cracking in all orientations which could lead to potential life prediction problems in parts manufactured from this plate product. In a study performed by J. Schubbe, it was discovered that at load levels above a certain threshold, 7050-T7451 thick plate has the potential for crack splitting parallel to the load direction when loaded in the LS direction [1, 2]. This poses a particular problem in components formed of plate products such as aircraft bulkheads, which have the potential to crack in directions which are neither expected nor easily evaluated, and which in turn hold the potential for reduced service life and unexpected failure. Inspection of such ‘buried’ parts is also problematic. This study was designed to determine the effects of environment upon cracking within the 7050-T7451 alloy. Specimens were exposed to 3.5% NaCl solution, which is the approximate concentration of seawater, as well as PENAIR M-5571, which is a common emulsifying cleaner used on naval aircraft. These environmental solutions are two of the most common experienced by aircraft in maritime environments and in the naval service. Aluminum 7075-T7631 has historically proven to be an aluminum alloy susceptible to corrosion cracking, and as such, specimens were tested in the same manner and environments as those of 7050-T7451 [3]. This allowed for direct comparison of the response of each alloy to the reagents used in this experiment, and thereby determine the relative susceptibility of 7050-T7451 to environmentally assisted cracking.

2 Environmental Procedure Twelve compact tension specimens were cut for this experiment from rolled plate aluminum products. The grain structure for the 16.5 cm 7050-T7451 plate is shown in Figure 1. All specimen dimensions were in accordance with ASTM standards and shown in Figure 2 [4, 5]. Of these twelve specimens, six were cut from aluminum 7050-T7451 in the longitudinal short-transverse or LS orientation, and the remaining six specimens were cut from 7075-T7631 in the longitudinal transverse or LT orientation (due to availability of material and documented comparison values). These specimens were fatigue pre-cracked using a MTS 810 electric servo-hydraulic 22 kip test stand. An external MTS Flex SE Controller was used in conjunction with MTS PC interface software and Cyctest software in order to control the pre-cracking and maintain constant ∆K and a load ratio of R=0.1 for each specimen. A MTS clip gage was used to measure crack opening displacement throughout fatigue pre-cracking. The fatigue load was such that a tension-tension stress range of approximately 8.79 MPa√m was applied at 10 Hz. Fatigue pre-cracking was stopped once a crack had been grown to an approximate length of 3.5mm from the notch root as calculated using the crack closure compliance method. After fatigue pre-cracking, the specimens were statically pin loaded using the load jig shown in Figure 3.

Environmentally Assisted Cracking in Advanced Aerospace Aluminums

Fig. 1 Typical grain structure and orientation in 7050-T7451.

Fig. 2 ASTM Standard E399-08 compact tension specimen.

687

688

E.M. Arnold et aal.

Fig. 3 Compact tension speccimen with load jig installed to provide constant crack openinng displacement.

The crack closure comp mpliance method was used to calculate the load required tto reach an initial loading off 80% of KIC. Due to the variability of the manual loadinng jig, the value below KIC was w used to prevent overloading of the specimen. It waas found that this loading was w sufficient to generate stable growth and achieve aan experimental value for K comparison. The crack opening displacement (COD D) that would result from this loading was also calculated using crack closurre compliance equations of ASTM E399. With the MTS clip gage, the COD waas used in conjunction with the load jig from Figure 3 in order to apply the proper loading on the specimens [6]. Once the above calculated load is applied, the cracck begins to propagate in aiir as a result of the load. Periodic measurements of thhe crack length were taken using a Starett optical traveling telemicroscope with a n-thousandth of an inch. These measured crack lengthhs digital readout to the ten were plotted versus time. The fixed crack opening displacement maintained by thhe load jig allowed calculatiion of the decreasing load (and thus K) on the specimeen as the crack increased in length. After a period of approximately 330 hours, cracck d to be negligible. This arrest signified that thhe growth slowed to a rate determined experimental threshold (o or KIC ) for the specimen had effectively been reached [77]. After reaching KIC , on ne of the two corrosive agents were periodically drippeed on the notch root and craack on each face of the specimen. Two agents commonlly found in naval aviation en nvironments, 3.5% NaCl solution and PENAIR M-5571, an emulsifying aircraft clleaner at full strength, were used in this experiment. Thhe specimens were tested in n triplicates with three specimens of each alloy beinng exposed to each of the driip solutions. The test specimen matrix for this experimennt is shown in Table 1.

Environmentally Assisted Cracking in Advanced Aerospace Aluminums

689

Table 1 Specimen test matrix with initial load applied to the specimen load line by pin loading. Specimen # AS-01 AS-02 AS-03 BS-01 BS-02 BS-03 AW-01 AW-02 AW-03 BW-01 BW-02 BW-03

Alloy 7050-T7451 7050-T7451 7050-T7451 7075-T7631 7075-T7631 7075-T7631 7050-T7451 7050-T7451 7050-T7451 7075-T7631 7075-T7631 7075-T7631

Reagent 3.5% NaCl 3.5% NaCl 3.5% NaCl 3.5% NaCl 3.5% NaCl 3.5% NaCl PENAIR M-5571 PENAIR M-5571 PENAIR M-5571 PENAIR M-5571 PENAIR M-5571 PENAIR M-5571

Initial Loading (N) 8010.35 8105.66 8034.70 8453.79 8552.47 8587.73 8205.47 7994.18 7983.42 8629.14 8379.48 8593.62

When applying the environment to the specimen, one drop of solution was applied to the notch root and crack area on the front face of each specimen using a needle-less syringe. The solution was drawn to the crack tip by capillary action before excess solution was wiped from the face of the specimen in order to reduce the potential for pitting or other corrosion on the surface of the specimen which would obscure the crack and make optical measurement difficult. After allowing the droplet of solution to remain on the specimen for a period of one to two minutes, the surface was wiped clean and a drop was applied to the other face. The same procedure was followed on the other side of each specimen, and the crack length was measured on both sides. These values were averaged together assuming that the crack tip is linear through the thickness of the specimen and growing in a self-similar manner. Measurements showed no significant evidence to the contrary. Specimens were exposed to their respective environments on a daily basis for a period of approximately 1400 hours.

3 Results Each of the twelve specimens experienced detectable crack growth after loading to 80% of the documented value for KIC. Crack growth, at constant crack opening displacement, resulted in decreased specimen load at the load line, as well as a drop in the calculated ∆K. After a period of approximately 330 hours, this crack growth arrested at an experimental threshold ∆K. As the reagents were applied to the specimens, detectable crack propagation resumed as a result of exposure to these environments causing subsequent drops in load and ∆K. The ∆K values for each specimen at each of the significant events within the experiment are contained within Table 2.

690

E.M. Arnold et al.

Table 2 Fracture toughness of each test specimen at significant times over the course of the experiment. Specimen #

Documented KIC*

80% KIC*

Kthreshold*

KSCC*,†

AS-01

35.05310733

28.0424859

27.998

27.949

AS-02

35.05310733

28.0424859

28.004

27.961

AS-03

35.05310733

28.0424859

28.022

27.993

BS-01

36.26183517

29.0094681

29.089

29.067

BS-02

36.26183517

29.0094681

28.922

28.853

BS-03

36.26183517

29.0094681

28.949

28.901

AW-01

35.05310733

28.0424859

28.011

27.979

AW-02

35.05310733

28.0424859

28.011

27.973

AW-03

35.05310733

28.0424859

28.035

27.986

BW-01

36.26183517

29.0094681

28.893

28.855

BW-02

36.26183517

29.0094681

29.087

29.064

BW-03

36.26183517

29.0094681

28.941

28.901

* All K values are in units of MPa√m. † These values for KSCC are those calculated from each of the specimens at a time approximately 1400 hours after initial loading.

The total crack length of each specimen was optically measured over a period of approximately 1400 hours. In that time detectable cracking as a result of exposure to environment occurred in both the 7050-T7451 and 7075-T7631 specimens in both 3.5% NaCl and PENAIR M-5571. The average crack length in each triplicate is plotted in Figure 4 showing the crack propagation over the course of the experiment. The K values for each specimen resulting from the observed crack length are plotted in Figure 5.

Environmentally Assisted Cracking in Advanced Aerospace Aluminums

691

(a)

Crack Length (mm)

4.25 4.2 4.15 4.1 4.05 4 0

500

1000

1500

Time (hrs) 7050 NaCl

7050 PENAIR (b)

Crack Length (mm)

3.8 3.75 3.7 3.65 3.6 3.55 0

500

1000

1500

Time (hrs) 7075 NaCl

7075 PENAIR

Fig. 4 (a) Crack length over time including initial loading and pop-in crack growth, continued crack growth to arrest at experimental threshold KIC, and crack growth as a result of environmental exposure over the period between 330 and 1400 hours for (a) 7050-T7451 and (b) 7075-T7631.

692

E.M. Arnold et al.

(a) 28.05 28.04

∆K (MPa√m)

28.03 28.02 28.01 28 27.99 27.98 27.97 27.96 0

500

1000

1500

Time (hrs) 7050 NaCl

7050 PENAIR

(b) 29.02 29.01

∆K (MPa√m)

29 28.99 28.98 28.97 28.96 28.95 28.94 28.93 0

500 7075 NaCl

Time (hrs) 1000

1500

7075 PENAIR

Fig. 5 ∆K values plotted over time showing the experimental threshold prior to 330 hours in (a) 7050-T7451 and (b) 7075-T7631.

Environmentally Assisted Cracking in Advanced Aerospace Aluminums

693

The K values listed in Table 2 and plotted in Figure 5 were calculated using the crack closure compliance method, which, in the case of this experiment provides an approximation of the drop in K as calculated from the measurement of the primary crack growth. However, this method cannot account for any branching or splitting in the crack. Such anomalies would increase the effective crack length, decreasing the resulting calculated applied load and value of K. Post-test metallographic imaging showed that splitting and branching was occurring and therefore the resulting threshold values may be significantly lower than plotted. Representative photographs of typical cracking observed in each of the test alloys are shown in Figure 6. Test groups of the same alloy exhibited very similar behavior in both adverse environments, so Figure 6 compares the two alloys rather than attempting a comparison between the environments. Both test alloys experienced splitting and secondary cracking under exposure to both reagents, but this splitting was significantly greater in 7050-T7451 than in 7075-T7631. Secondary cracks for this orientation in 7050-T7451 were typically four to five times the length of their counterparts in 7075-T7631 alloy. Representative photographs of this behavior are also included in Figure 6.

4 Conclusions A viable method for examining the effects of adverse environments on existing cracks was employed to examine the susceptibility of 7050-T7451 thick plate to environmentally assisted or enhanced cracking. It was shown that detectable growth was achieved and was compared to a known susceptible alloy 7075T7631. The specimens of 7075-T7631 experienced detectable crack growth as a result of exposure to both 3.5% NaCl and PENAIR M-5571. This confirms the susceptibility of this alloy to environmentally assisted cracking and confirms its usefulness as a benchmark against which to compare the behavior of 7050-T7451. The latter alloy also experienced detectable crack growth in response to the application of reagents to the crack tip. As a result, 7050-T7451 is also susceptible to corrosion cracking in this orientation. As such, its use as a structural alloy should be carefully monitored, and further studies should be conducted to further characterize the fracture behavior of this alloy. Additionally, the 7050-T7451 exhibited the tendency for cracks to split perpendicular to the primary crack growth direction or branch in an unpredictable manner. This behavior was first exhibited during previous studies characterizing fatigue fracture of the material. The presence of this behavior under environmental cracking conditions calls for considerable evaluation of the material as a structural aerospace alloy. Such behavior poses a major threat in that cracks of significant size may be propagating through currently in-service aircraft components in directions or locations which may be much more difficult to find and evaluate.

694

E.M. Arnold et aal.

(a)

(b) Fig. 6 Representative photog graphs of cracking in each alloy. (a) 5x magnification of typiccal crack in 7050-T7451. (b) 5x magnification of a typical crack in 7075-T7631.

Environmentally Assisted Crracking in Advanced Aerospace Aluminums

6995

(a)

(b) Fig. 7 (c) 100x magnificatio on of splitting and secondary cracking in 7050-7451. (d) 1000x magnification of splitting and secondary cracking in 7075-T7631.

696

E.M. Arnold et al.

5 Future Studies Further study in this experiment will include further metallographic analysis of the specimens in order to better determine fracture mechanisms and the environmental effects on grain structure. Scanning electron microscopy will also be used to determine element concentrations at the crack tip and grain boundaries. Additional specimens of aluminum 7050-T7451 in the SL orientation are currently being exposed to reagents and evaluated for growth in this weak plane orientation. Previous study has shown the tendency of cracks in 7050 in the LS orientation to split into the SL direction. Performing a similar experiment on specimens in the SL orientation will provide additional data for determination of splitting thresholds.

Acknowledgments [1] Center for Corrosion Science and Engineering, United States Naval Research Laboratory, Washington, D.C.

References [1] [2] [3] [4] [5] [6] [7]

Schubbe, J.: Engineering Failure Analysis 16, 340–349 (2009) Schubbe, J.: Engineering Fracture Mechanics 76, 1037–1048 (2009) Hollingsworth, E., Hunsicker, B.: Metals Handbook, 9th edn., vol. 13 (1987) ASTM Standard E399-08, p. 19–20 ASTM Standard E647-08, p. 11–12 Antolovich, S., Antolovich, B.: ASM Handbook, vol. 19, pp. 371–392 Gangloff, R.: Environmentally Assisted Cracking of Metals (1994)

26th ICAF Symposium – Montreal, 1-3 June 2011 An Overview of Fretting Aspects Relating to Aero-Engine Dovetail Attachment Raghu V. Prakash1, K. Anandavel1,2, and P. Balasubramani1 1

Indian Institute of Technology Madras, Chennai, India 2 Infotech Enterprises Limited, Bangalore, India

Abstract. An aero-engine fan and compressor rotor design involves dovetail interface between the blade and the disc, which is critical, as early crack initiation occurs at the interface due to fretting. This paper reviews the different methodologies adopted to understand the fretting behaviour at the interface: experimental, analytical and numerical methods. The effects of geometry and material on fretting behaviour are discussed. The results of a 3-Dimensional finite element analysis of a typical dovetail interface with skew, cone and twist angle are presented. Areas that require attention are highlighted at the concluding section of this paper.

1 Introduction An aero-engine fan and compressor rotor design involves dovetail interface between the blade and the disc. Dovetail attachment is a critical element in the rotor, as early crack initiation leading to premature failure of the interface edge occurs, due to fretting damage at the interface under combined low cycle fatigue (LCF) and high cycle fatigue (HCF) loadings. In general, fretting damage is due to small amplitude oscillations between two contacting bodies, that results in: a) early crack initiation (termed as fretting fatigue) or b) wear of material at the contact interface (fretting wear); the presence of normal and tangential loads with or without bulk load leads to fretting. Today, fretting mechanisms are understood to include damage due to fatigue, wear and corrosion. Early designs did not take into account the fretting aspects; field failures at the interface due to fretting are reported in literature (fretting failure at riveted joints, spline in shafts, dovetails in engines). Fretting fatigue failures are observed in mechanical systems and assemblies, such as dovetail, flanges, pins or fasteners bearing on hole, spring washers, leaf spring and coil springs, keys in keyways of shaft, spline assembly, gears, wheel bosses on shafts, collars and bushes, etc. used in aircraft, power plant, automotive, locomotive, nuclear, orthopaedic and other engineering applications. A systems approach to design that considers the fretting aspects of a blade-disc attachment is emphasized by aero-engine community, to prevent premature failure of rotor and ensure structural integrity of rotor – so that the requirements of a lighter engine, with better performance without any compromise on the safety can be achieved.

698

R.V. Prakash, K. Anandavel, and P. Balasubramani

Fretting research at dovetail interface can be broadly categorized as the ones dealing with the analytical, experimental and numerical simulation of contact at the interface. Inputs, such as, fretting characteristics of mating materials, are used in the development and application of analytical, semi-analytical and numerical tools to assess the macroscopic fretting aspects; this apart, life prediction models, incorporating fretting fatigue, fretting wear are developed to predict safe operating lives of critical interfaces. Fretting control through surface preparations and tribological methods are studied to enhance component life, which makes the field as complex and multi-disciplinary by nature. The objective of this paper is to bring out the state-of-art developments in these various research areas, present results of a 3-dimensional modelling taking into account the geometric features such as skew, cone in blade-disc mounting and look at future challenges. The study suggests that there are several areas that require attention and the same is highlighted in the concluding section of this paper.

2 Origin and Background of Fretting It is exactly completion of a century, since the problem of fretting was reported first and a scientific investigation was started. The first publication related to fretting failure was reported by Eden et al. [1]. The publication reported failure of fatigue specimens at the grips of a testing machine, in which, the presence of iron oxide on the surface of the specimen noticed. Systematic investigation on fretting fatigue started when Tomlinson [2] designed a fretting fatigue machine consisting of contact annuli with small amplitude rotational oscillations. This work concentrated on damage and corrosion associated with fretting. Warlow-Davis [3] examined the influence of fretting corrosion on fatigue life of a specimen which was subjected to fretting corrosion first that was followed by fatigue testing. McDowell [4] carried out simultaneous fretting fatigue tests and indicated the combined effect of fretting and fatigue was more severe than the summation of independent effects. Reduction in fatigue strength from the study was observed to a factor of 5, compared to 18% reduction in study carried out by Warlow-Davis [3]. Fenner et al [5] observed that many service failures in locomotive engine components were due to fretting; further studies by Fenner and Field [6,7] with a bridge type fretting pad showed that fretting accelerated the process of crack initiation. Intensive study of the fretting process took place during 1950s and 1960s. An important review by Hurricks [8] recapitulating various theories of the nature of fretting was published and a logical continuation of this was presented by Waterhouse [9]. In a subsequent publication, Golego et al [10] summarized the results of experimental and theoretical investigations carried out in the area of fretting in the erstwhile USSR and other countries. Parametric studies such as effect of contact pressure, relative slip, environmental effect and stress amplitude were the main areas of research until late 1980s. A standard test method was proposed by Attia and Waterhouse [11]. It may be said that through these studies, the understanding on fretting fatigue developed significantly, and the importance of the fretting damage phenomenon in engineering application become widely understood. Subsequent investigations in

An Overview of Fretting Aspects Relating to Aero-Engine Dovetail Attachment

699

fretting proceeded in several fundamentally related directions such as fretting damage mechanism and process, contact mechanics and fretting variables prediction methodologies, fretting crack initiation mechanism, fretting fatigue life estimation model, fretting wear model and palliatives for controlling or minimizing fretting failure, which are discussed further. Fretting damage mechanism and representation Fretting fatigue and fretting wear are recognized as two main mechanisms of damage, and are studied widely due to the pressing need of industry to prevent fretting fatigue damages. Fretting damages are represented by friction log, velocity accommodation mapping and fretting maps. One of the most important progresses is in the development in fretting maps, which describes the overall behaviour of fretting, such as contact conditions, fretting regime, wear mechanism, crack nucleation and propagation. Fretting map concept was originated by Vingsbo and Soderberg [12] and Vingsbo et al [13]. They isolated three different fretting regimes: stick, mixed stick and slip, and gross slip as shown in fig. 1. Subsequently two more fretting maps such as running condition fretting map [RCFM] and material response fretting maps [MRFM] were proposed by Pellerin [14] and Zhou et al [15]. Representative RCFMs and MRFMs corresponding to characteristics friction loop is shown in fig. 2; these maps provide a design guidance as well indicate the trend of different fretting damages and mechanisms.

Fig. 1 Fretting map [12].

700

R.V. Prakash, K. Anandavel, and P. Balasubramani

Different characteristics fretting loops

Fig. 2 Running condition fretting map and material response fretting map corresponding to different characteristic friction loops [16].

Zhou and Vincent [17] emphasized the importance of mixed fretting regime (MFR), located between stick-slip and gross slip during the fretting fatigue test. MFR was identified as the most critical regime for crack nucleation, crack propagation and failure during fretting wear and fretting fatigue tests. The fretting maps indicate the influence of fretting variables such as slip amplitude, contact loads and contact stresses in determining the fretting damage mechanism and fretting failures. Contact mechanics and prediction of fretting variable The first step in the fretting analysis is the prediction and understanding of contact kinematics and traction at the interface of contacting bodies and estimation of contact stresses. The pioneering work of Hertz [18] formed the basis for subsequent development of analytical solutions in contact mechanics for fretting application. Hertz

An Overview of Fretting Aspects Relating to Aero-Engine Dovetail Attachment

701

proposed an analytical solution for the estimation of normal traction and contact widths of non-conforming contacts, such as, cylinder-on-cylinder and sphere-onsphere, with the assumption of: a) frictionless continuous contact surface, b) strains are small, and c) contacting bodies are homogeneous, isotropic and follow linear elastic theory. The solution to the problem of two cylinders pressed together by a normal force and subsequently loaded by shear force (which is less than friction coefficient times the normal load) was first found by Cattaneo [19] and independently by Mindlin [20]. They proposed analytical solution for normal and shear traction over the contact area and indicated that the stick condition prevails at the central zone of contact and slip situation at the outer regions of contact, as shown in fig. 3. Johnson [21] gave an overview of range of contact problems that are of interest and explained the solution for contact traction, stress and displacement contact fields at the contact surface and sub-surfaces. Nowell and Hills [22] developed the Mindlin solution for contact of cylinders on half space subjected to simultaneous application of shear load and remote bulk stress on the half space specimen. The shear traction over cylindrical contact, for different levels of bulk tension, is shown in fig. 4. Ciaveralla [23] developed an analytical solution for a flat and a rounded edge contact on a half-space and obtained traction variation for various combinations of width and radii. The results define the transition from Hertzian configuration to flat-on-flat contact configuration similar to rigid-flat punch contact on half space, as shown in fig. 5.

Fig. 3 Mindlin condition [17].

Fig. 4 Nowell solution-Hertzian cylindrical contact with tangential load and a) moderate bulk load b) Large bulk load [24 ].

702

R.V. Prakash, K. Anandavel, and P. Balasubramani

Fig. 5 Contact pressure distribution for flat and rounded corners on halfspace for different geometric combination [25].

Barber and Ciavarella [26] reviewed the challenges of contact problem and indicated the potential application of fractal method to characterize contact of rough surfaces and in the general area of thermoelastic/plastic contact. Contact problems are characterized by unilateral inequalities that are described by the physical impossibility of tensile contact tractions and of material interpenetration in normal direction. Additional inequality is introduced in shear direction when frictional laws are taken into account. Stress concentration resulting at the contact zone is another important characteristic of contact problem. Muskhelishvili [27] introduced a complex stress function to solve plane elasticity problems including contact problems. Analytical calculation of the Muskhelishvili potential leads to the determination of stress and displacement fields in contact areas. Hills and Nowell [24] and Truman and Sackfield [28] used the Muskhelishvili’s function to obtain closed form potential function for the standard contact problems. Closed-form potential functions for contact problem with complex geometry involve lengthy procedure and are difficult to solve. Adibnazari and Sharafbabi [29] introduced a new relation that simplifies the determination of Muskhelishvili’s potential function in plane contact problems. Fretting damages at the contact interfaces are studied commonly post-failure through fractographic studies [30]. Prediction of contact tractions and stresses during fretting loading cycle is a challenging task yet to be addressed, though in-situ micro-tomography [31], Synchrotron X-ray micro-tomography [32] and acoustic emission [33] techniques have been recently implemented in fretting fatigue setup, to identify crack initiation and crack growth. The lack of appropriate measurement system makes the prediction of fretting variables at contact interfaces to be totally dependent on either analytical or numerical methods. Analytical solutions developed are applicable for limited cases of contact problems; however, as practical situations involve complex geometries, which are difficult to solve by existing analytical methods, considerable development has taken place in numerical methods and semi-analytical methods. 2-Dimensional finite element method has been used extensively for the prediction of contact traction and stresses over flat fretting test specimens subjected to fretting test by cylindrical pad [34, 35] and for flat with rounded corner pad [35]. The studies indicated

An Overview of Fretting Aspects Relating to Aero-Engine Dovetail Attachment

703

that with sufficient mesh refinement, the stresses at the contact edge can be predicted, by finite element (FE) method. McVeigh et al [36] developed a semi-analytical approach combining analytical and singular integral equations that govern plane strain elastic contact of surfaces. Rajasekaran [37] developed a semi-analytical approach, by considering use of global finite element model for the estimation of contact loads at the interface and local analytical approach for the prediction of traction distribution and contact stresses. The semi-analytical approaches were aimed at reducing the computing time. Murthy et al [38] developed a computationally efficient approach using singular integral equation that governs plane strain elastic contact problem, for the prediction of contact tractions and contact stresses over the interface of twodimensional contact of flat pad with rounded edge on half space. Multi-scale analysis consisting of coupling a semi-analytical contact solver with FE method for structural behaviour is presented by Gallego et al [39] for computation of fretting wear and the same is applied for analyzing the contact at the blade-disc interface. Ambrico and Begley [40] showed the influence of plasticity on stress fields in fretting contacts by a FE model of cylinder and a plate contact. Giannakopoulos and Suresh [41] developed a 3D finite element model for fretting fatigue between a sphere and a planar surface. Leen et al [42] used a 3D FE model for the prediction of fretting variables at contact interfaces of a spline coupling and used the results of the analysis for optimization of contact profile for minimizing the fretting damage. Kim and Mall [35] showed the free edge effect on contact tractions and stresses by 3D finite element analysis on finite width specimen with contact of cylindrical pad and flat with rounded edge pads. The above 3D contact studies emphasize the importance of accurate prediction of fretting variables and contact edge stresses for fretting fatigue analysis. In the recent times, boundary element formulations [43] and combined FEM-BEM methodology [44, 45] have been developed to solve 3D rolling contact problems [46], sliding wear simulation [47], to take the advantage of lesser computational time. Fretting fatigue studies Fretting fatigue is often recognized for reduction in expected component life because of surface degradation due to large cyclic stresses resulting from two bodies that are clamped together and are subjected to relative oscillations. Many researchers have carried out experimental studies for different combination of material, loading conditions, contact configurations, coefficient of friction, surface finish, metallurgical parameters etc. for a detailed understanding of the mechanism of fretting fatigue, to define suitable criteria for fretting fatigue failures and to develop fretting fatigue life estimation methodologies. Some authors have correlated the experimental results with stress, strain and displacement fields in the neighbourhood of contact interface. Experimental parametric studies. Study of Kovalevskii [48] indicates that the reduction in fretting fatigue strength is due to stress concentrations formed by the destruction of total surface layers by the mechanism of LCF. Nowell and Hills [49] have carried out a series of fretting tests on Al-Cu alloy in which contact size is varied, while, all other relevant parameters were held constant. Fretting fatigue

704

R.V. Prakash, K. Anandavel, and P. Balasubramani

life was found to be infinite below a certain critical contact width. Effect of slip amplitude on fretting fatigue behaviour was studied by Husheng et al [50] on several alloy steels; they found that with increasing slip amplitude, the depth of wear scar increases and wear damage becomes more severe. Influence of contact configuration in fretting fatigue testing was studied by Pape and Neu [51] using flat and cylindrical contact. Effect of both contact configurations was concurrently examined under the same applied contact load. The result indicated that the major crack which led to specimen failure always occurred at a cylinder-on-flat contact, while a crack was observed at a flat-on-flat side contact. Adibnazari and Hoeppner [52] illustrated the existence of normal pressure threshold concept (value after which increasing the pressure does not affect life in fretting fatigue). The role of normal pressure and a frictional stresses in fretting fatigue life was the aim of the investigation. At a given applied cyclic stress and below the threshold pressure, fretting fatigue life is a function of normal pressure and change of normal pressure; hence, by the law of friction, fretting fatigue life is a function of frictional stress and change of frictional stress. Above the threshold pressure, fretting failure is a function of normal pressure and frictional stress, not increase of these two variables. In conjunction with these findings, it has been argued that a frictional stress might cause fretting damage to reach its threshold faster. Iyer and Mall [53] studied the effect of contact pressure and stress amplitude on fretting fatigue life through the normal pressure threshold concept. Fretting fatigue loading results in an amplified stress range in the vicinity of contact. The amplification of local stress range during fretting fatigue is derived in part from stress concentration due to normal load, and a significant part of amplification is due to local build up of compressive stresses upon unloading. The amplification is more severe for small cyclic stress amplitudes and thus can be used to explain the remarkable reduction in fatigue life due to fretting near the fatigue strength. The study also shows that the decrease in fretting fatigue life with the increase in contact pressure can be due to increase in local stress amplification, without any regard to interfacial shear stress or slip amplitude. The above studies highlight the importance of contact pressure on fretting fatigue. Nowell and Dini [54] proposed a notch analogy to account for stress gradient effects in fretting fatigue. Nowell et al [55] in the review on recent developments of fretting fatigue summarised the critical characteristics of fretting fatigue are: a) the high stress gradient due to localized stress concentration at the contact, b) loading is likely to be non-proportional near the contact, c) initiated cracks will experience a variable R-ratio as they grow away from contact and d) localized surface damage at the asperity level play a role in accelerating the crack initiation. Mechanism of fretting fatigue. Experimental studies of Endo and Goto [56] and Conner et al [57] suggests that initially the fretting fatigue crack develops and grows at about 45 degrees to the surface and then changes it direction perpendicular to the surface. Venkatesh et al [58] indicated in detail that the mechanism of fretting fatigue involves four stages till failure, as shown in fig. 6 – a) Crack initiation, b) crack propagation under the combined influence of contact and bulk loads, c) crack propagation under the influence of bulk load only and d) component failure.

An Overview of Fretting Aspects Relating to Aero-Engine Dovetail Attachment

705

Fig. 6 Schematic representation of fretting fatigue crack mechanism a) longitudinal cross section view b) lateral cross section view [58].

Criteria for fretting fatigue failure. Peak stresses at the edges of contact result in early nucleation of cracks, while remotely applied loads combined with contact stresses propagate the cracks to failure. The rapid nucleation and steady propagation makes a dual mechanism total life prediction. The dual mechanism total life approach had been suggested by Fellows et al [59], Szolwinski and Farris [60], and Golden and Grant [61] while making several improvements all along the way. Various criteria are considered by Hills et al [62] based either on the condition for crack initiation or crack propagation. As synergistic effects of many parameters are involved, even under laboratory conditions, determination and modelling of crack initiation criteria for fretting fatigue becomes extremely difficult. However, from an engineering point of view, a quantitative evaluation method, based on a few specific parameters, is strongly required to describe the fretting fatigue failure and to estimate the fretting fatigue life for design in industrial applications. This has lead to the development of the following approaches, for crack initiation life. Ruiz parameter. Ruiz et al [63] developed a criterion for studying the fretting fatigue damage at dovetail joints. According to this criterion, the primary surface damage driving factors are the relative slip and contact shear stress at the interface. The fretting damage parameter was introduced as a product of relative slip and contact shear stress. Assuming that crack growth is governed by the maximum tangential stress, a composite fretting damage parameter was defined as the product of frictional work and tangential stress. Composite fretting damage parameter characterizes the severity of fretting damage and probability of crack initiation location. Stress based approaches. Stress based approach typically involves formulation of multi-axial fatigue criteria such as Crossland, Findley, Smith-Watson-Topper (SWT), McDiarmid, etc. One of the appealing features is its relative simplicity for estimation of crack initiation. The limitation is that, it does not incorporate length scale; secondly, prediction of crack location and crack direction in all fretting situations is not possible. Lykins et al (64) evaluated the critical plane SWT parameter and the critical plane maximum shear stress range parameter to predict number of cycles to crack initiation, crack location, and crack orientation angle. The studies suggest that fretting fatigue in a Ti-6Al-4V alloy is governed by the maximum shear stress range on the critical plane.

706

R.V. Prakash, K. Anandavel, and P. Balasubramani

Fracture mechanics based approaches. By incorporating suitable length scale in the analysis of fretting crack initiation and propagation, fracture mechanics approach addresses the principal limitation of stress based approaches. Giannakopoulos et al [65] demonstrated an analogy between the stress state existing at the corner of square contact, and that present in the crack tip, through asymptotic analysis, for quantifying the fretting behaviour. Fracture mechanics methodology was assessed for fretting fatigue by Nicholas et al [66], for a flat specimen fixed at one end, and axial load applied at other end while contact loads are applied on both sides of specimen. The study concluded that the fracture mechanics methodology appears to be more promising for predicting failure or fatigue limit compared to analysis of the stress field. Mutoh and Xu [67] reviewed the fracture mechanics approach for evaluating fretting fatigue life and strength. The authors brought out the limitation of prevailing in fracture mechanics model and proposed a new fracture mechanics approach with a more physical background. Ciavarella and Macina [68] studied the contact of flat punch over half plane under constant normal loads, and oscillating tangential and bulk loads with a crack analogous model for fretting fatigue. Similarity between contact mechanics and fracture mechanics lead to the development of crack analogy method (CAM), which defines the stress intensity factor as fretting fatigue crack initiation parameter. Conner et al [69] presented a fracture mechanics approach to predict crack initiation life, using crack analogy and related concepts. Naboulsi [70] applied the crack analogy for fretting fatigue and indicated its potential in effectively predicting the complex mechanism of fretting fatigue. Attia [71] developed a fracture mechanics model to predict the fretting fatigue strength and service life of structural components. A three-dimensional interface element was developed to express the constitutive laws of interface. The methodology used is based on the threshold and crack propagation analysis until fracture takes place by plastic collapse or by exceedance of material fracture toughness. Fretting wear studies Wear process was broken into particle detachment, third body and particle ejection [72]. Adhesion, abrasion, corrosion and fatigue govern the particle detachment phase of wear. Third body action in fretting is best illustrated using friction logs in which the coefficient of friction is plotted against amplitude and time. Ratsimba et al [73] described a methodology to predict a wear in a complex coupling and validated against results obtained from a reduced scale aero-engine spline coupling subjected to complex cyclic load cases. The methodology uses the modified Archard’s equation to calculate the wear depth from contact pressure and slip distribution using wear coefficients obtained from round against flat fretting tests. Ding et al [74] applied a numerical approach to simulate fretting wear using modified Archard’s equation, for gross sliding and partial slip conditions. Surface wear damage is predicted to have significant effect on the near surface tangential and shear stress for both slip regimes. The implications of these effects were discussed with respect to fretting fatigue predictions, leading to new insight into experimentally observed effects of slip regime on crack initiation. The work establishes a

An Overview of Fretting Aspects Relating to Aero-Engine Dovetail Attachment

707

base for direct incorporation of the effect of slip amplitude on fretting fatigue life prediction. Fretting wear is a surface degradation process, but it has been shown to have an important impact on the near surface stresses which affect the initiation of fretting fatigue cracks. This has lead to fretting fatigue strength estimation considering fretting wear process [75]; it was observed that the fretting fatigue strength decreases in accordance with the increase in fretting wear. This tendency coincided well with fretting failure observations in industrial fields. Further, criticality of fretting wear in the analysis of fretting fatigue is studied in detail by Madge et al [76], using fretting wear results and stress based multi-axial fatigue models. Fretting palliatives for fretting performance improvement Buch [77] investigated the effect of stress level on fatigue and fretting of pin-lugs with and without interference fit, made of aluminium alloy 2024-T3. The investigation revealed that the critical stress level below which the interference fit has a beneficial effect on fatigue life for smaller amplitude. Chivers and Gordelier [78], listed various fretting palliatives such as modifying the geometry and parameters of contact, metallic coatings, non-metallic coating, metallic shims, diffusion coating, residual compressive stresses created by the cold working process and lubricants. Shaffer and Glaeser [79] present some principles for reducing fretting fatigue in engineering applications. Juuma [80] estimated experimentally the torsional fatigue strength of shrink fitted shaft couplings for varying geometry, contact pressure, and load. The experiment indicated the fretting fatigue life increased with an increase in contact pressure, because the slip amplitude decreased to less than 3 micron. Alfredsson [81] studied the fretting-fatigue initiation for a shrink fit pin at rotating bending. Eight assemblies with four different grips made from soft normalized steel were tested at loads well below bending endurance. Shrink fit interface reduced the rotating bending endurance by more than 60% and the fatigue life decreased at increasing interface grip. Kiral [82] studied the effect of clearance and interference fit on the failure mode, failure load and bearing strength of the pin-loaded joints made of laminated composites subjected to traction forces. Improvement in failure load and bearing strength is seen as beneficial effect of interference fit. Chakherlou [83] studied the effect of interference fit on fatigue life of holed plate of mechanical joints both experimentally and numerically. Experimental results show that the interference fitted specimens have improved fatigue life compared to open hole; increasing the interference fit from 1% to 2%, increases the fatigue life. However, further increase in interference fit showed no significant improvement in fatigue life. Shot peening [84,85] and other advanced treatments, that induce deep [>1mm ] residual stresses, such as laser shock peening and low plasticity burnishing [86,87] to produce compressive residual stresses that are beneficial to extend fretting fatigue life. The significant observation of these works suggests that fretting cannot be completely avoided, and one has to live with it through the use of palliative measures.

708

R.V. Prakash, K. Anandavel, and P. Balasubramani

3 Fretting Studies in Aero-Engine Dovetail Interface An aero-engine fan and compressor rotor design involves dovetail interface between blade and disc, and the interface is subjected to combination of low cyclic and high cyclic loading during a mission cycle. One out of six of all in-service ‘mishaps’ attributed to high cycle fatigue (HCF) in USAF engine hardware can be linked to fretting-induced damage [88]. The blade-disc interface is an important element in the rotor assembly, as premature crack initiation [89] occurs, as shown in fig. 7 due to fretting at this interface, which leads to catastrophic rotor failures. Excessive fretting wear at the interface would also result in rubbing of rotating blade with the static engine casing and result in failures as a consequence. In view of this, in recent years, aero-engine dovetail interface is studied from the perspective of fretting fatigue as well as fretting wear to improve the rotor design system considering fretting aspects.

Fig. 7 Cracks in disc dovetail [89].

Features of Bladed-disc Fan Rotor A typical bladed-disc rotor with axial dovetail interface is shown in fig. 8 for a single blade sector, with a description of parts and associated nomenclature. The bladed-disc rotor involves three-dimensional geometric features such as skew angle, nose cone angle, airfoil twist, etc. as shown in fig. 9. Skew angle, in a bladeddisc rotor, is an angular off-set of dovetail slot axis with relation to disc or engine axis. The skew angle is important from the viewpoint of assembly, as it helps in accommodating the maximum number of blades within a given space for the desired airfoil configuration. The cone angle is necessary to maintain a constant mass flow over the flow passage because of increase in pressure over the rotor stage. The blade twists are essential for better aerodynamic performance. These three dimensional geometric features contribute to an uneven mass distribution in the blade with dovetail root and hence results in an offset between the centre of

An Overview of Fretting Aspects Relating to Aero-Engine Dovetail Attachment

709

gravity of the blade and centre of area of dovetail interface where the blade loads are reacted. The offset would create three-dimensional moments at the interface, depending upon the magnitude of offset. Aerodynamic forces act on the blade in the axial and tangential direction. The above geometric and loading features, in addition to vibratory loadings, contribute to three-dimensional nature of loadings at the interface.

Fig. 8 Typical bladed-disc rotor.

Fig. 9 3D geometric features of dovetail and rotor [39].

710

R.V. Prakash, K. Anandavel, and P. Balasubramani

Prediction of fretting variable and contact stresses at interface Accurate assessment of critical fretting variables such as contact traction, contact condition [stick, partial slip, and gross slip] and contact stresses over dovetail interface, is essential to characterize the fretting performance and reliability of rotor Experimental studies. Two-dimensional photo-elastic and holographic technique studies [90, 91] were used to predict the contact stresses at the dovetail interface. These studies were not focusing on fretting aspects, since the strength was important design consideration parameters during earlier days. Ruiz et al [92] employed a high sensitivity moiré technique to measure the normal and tangential displacements of dovetail interface. It was observed that when behaviour changes from stick to slip, the dovetail behaved in an asymmetrical manner. Burguete and Patterson [93] used stress frozen photo-elasticity to model dovetail compressor blade fixings. With known friction coefficient, the method to predict the direction of crack propagation was improved. 2D experimental investigation was carried out by Ruiz et al. [94] to assess the fretting fatigue damage at a dovetail interface on a static test facility with application of pull load on blade dovetail. Representative 2D dovetail geometries were tested by Rajasekaran and Nowell [95] in a static experimental set up which can simulate LCF and HCF loads by hydraulic and shaker loads. Similar experimental studies were also carried out by Szolwinski and Farris [96]. Experimental evaluation of the damage due to fretting, incorporating all the complex loading, will involve spin pit test or full engine test like accelerated mission endurance test. Analytical studies. Ciavarella and Demelio [97] proposed application of 2D analytical solution of flat pad with rounded corners contact on half plane for dovetail analysis, as the geometric configuration of dovetail matches closely with flat pad and rounded corner contact, as indicated in fig. 10. Sinclair and Cormier [98] developed simple 2D physical model for stresses in dovetail attachments. The model was aimed to predict global slip condition and contact stresses during loading and unloading. 2D Numerical studies. The finite element analysis by Alderson et al [99] for optimization of compressor blade–disc attachment used the boundary conditions at the interface, which permitted free motion between disc and blade in the plane of contact surface with zero relative motion normal to the contact surface. Kenny et al [100] carried out the two-dimensional FE analysis on the dovetail joint with approximate pressure application at the interface between the blade and disc. These analyses also indicate the fretting aspects that were not considered in detail in the early design of blade-disc attachment. Boddington et al [101] developed a FE code which predicts the different states of contact in two-dimensional elastic analysis of dovetail interface and had the capability to include the relative motion, friction at the interface. Meguid et al [102] carried out two-dimensional finite element contact analysis of dovetail interface for the overall stress distribution using commercial code ABAQUS. Sinclair et al [103] had demonstrated the conforming contact stresses at the dovetail interface is non-singular (for friction and non-friction

An Overview of Fretting Aspects Relating to Aero-Engine Dovetail Attachment

711

cases) through asymptotic analysis, and finite element analysis of dovetail attachment should produce converging stress by fine mesh at the interface. Rajeev et al [104] applied singular integral approach for load history analysis of dovetail under plane strain elastic conditions. Similarly the semi-analytical approaches [95] and hybrid approaches [105] were also applied for analysis of dovetail interface. However the 2D analysis has the limitation to incorporate 3D geometric features of dovetail interface and airfoil loading of rotor, as applicable for any other applications.

Fig. 10 Representation of dovetail as flat with rounded end on half space [97].

3D Numerical studies. Papanikos et al [106] and Meguid et al [107] have shown that 3D geometric effect plays significant role in increasing contact edge stress at compressor dovetail and turbine fir-tree interfaces using three-dimensional finite element contact analysis, though the mesh refinement level in these analysis appears to be not adequate enough to capture the converged peak stresses. Beisheim et al [108] implemented sub-modelling approach in three-dimensional elastic analysis on a straight dovetail attachment. The sub-modelling approach addresses the issue of contact edge stress convergence at reduced computational time. The significant contact edge stress difference between 3D [106] and 2D [102] analysis results also implies that other macroscopic fretting variables such as contact tractions and slip could also be significantly impacted by 3D geometric and loading features. Three-dimensional finite element studies address the half space assumptions involved in 2D analytical solution, free edge effect associated with 2D studies, apart from its ability to handle the complex three-dimensional loading and

712

R.V. Prakash, K. Anandavel, and P. Balasubramani

geometric features of aero-engine dovetail. Motivated by the advantages of 3D finite element studies and benefits of sub-modelling for achieving results convergence, 3D finite element study [109] is carried out, on straight and skewed dovetail, to understand the influence of skew effect and 3D loading of airfoil on fretting variables. The study showed significant increase (more than 100%) in contact traction and different contact conditions for different combinations of 3D loadings as shown in fig. 11 and fig. 12.

Fig. 11 Contact pressure distribution of a) straight dovetail, b) skewed dovetail.

Fig. 12 Skew effect on slip amplitude.

Fretting studies on Titanium alloys Titanium alloy is conventionally used in both blade and disc, attributed to its high specific strength (High strength and low density) and corrosion resistance, despite its poor tribological characteristic like high friction coefficient. Experimental studies on Titanium alloys are relevant for dovetail interface, as they establish the material response for different fretting conditions. Waterhouse and Dutta [110] showed the effect of fretting and environment on fatigue life of Titanium alloy as shown in fig.13. Fretting effect on reduction of fatigue strength is also reported in [111]. Experimental study on normal load effect [112] on fretting fatigue of Titanium alloy indicates that fretting damage in titanium is very sensitive to the normal pressure. Hamdy and Waterhouse [113] established fatigue curves in presence of fretting for Ti-6Al-4V under fluctuating tension with a mean stress of 247 MPa for a set of temperatures (20, 200,400 and 600°C). Fatigue strength at 107 cycles at temperatures of 200 and 400° C are almost the same and the same drops at 600°C. Cortez et al [114] examined the

An Overview of Fretting Aspects Relating to Aero-Engine Dovetail Attachment

713

phenomenon of fretting fatigue of Ti-6Al-4V when subjected to both high frequency constant amplitude and variable amplitude conditions. The results indicate that high frequency fretting fatigue conditions produced shorter lives than those conducted under low-frequency fretting fatigue conditions. Namjoshi et al [115] investigated the fretting fatigue of Ti-6Al-4V under variable amplitude loading which represent the low-and high-cycle components of turbine engine mission loading.

Fig. 13 Effect of fretting on Titanium alloys at different environment [110].

Fretting fatigue life estimation Szolwinski and Farris [116] examined the impact of the interaction between LCF and HCF load spectra on the near surface stress field associated with the conditions of fretting fatigue, using multi-axial fatigue parameter. Calcaterra and Naboulsi [117] explained the methodology adapted to investigate contact fatigue at a dovetail interface in an aero-gas turbine engine hardware. The methodology uses combination of 3D FEM, 2D singular integral approaches for contact stress prediction and SWT parameter for life estimation. Fretting palliatives in dovetail interface Minimizing the fretting damage at dovetail interface is necessary in dovetail interface as the fretting can’t be eliminated totally. Various palliative measures are observed in the literature and are presented below. Geometric modifications. Sinclair and Cormier [118] introduced precision crowning over the contacting flat on blade in both the directions, to alleviate the fluctuating contact stresses at the contact edges. The effect of in-plane and out of plane crowning to reduce contact edge stress is verified by a two dimensional finite element analysis. Three-dimensional contact analyses were carried out by Beisheim and Sinclair [119] to compare effect of crown and improve the crowning profile [120] for better benefit of contact stress alleviation. Another geometric modification implemented to minimize the fretting damage is the introduction of groove at contact edge blade dovetail [121].

714

R.V. Prakash, K. Anandavel, and P. Balasubramani

Preloading application. Blade dovetail loosely fits in the compressor disc until the rotor spins and the centrifugal forces push the blade dovetail firmly and radially upward against the dovetail slot in the disc. The easy insertion and removal of fan blades from the attachment requires a radial small space between bottom of dovetail root of blade and corresponding slot surface of disc. However, when the engine is not operating, the fan blades are free to slowly rotate or likely to experience windmill due to wind or breezes on the ground at the airport. Preloads are applied [122] on the fan blade with a radial outward force to eliminate wear between blade root and the dovetail slot during wind milling conditions, by introducing a spacer or wedge between blade root bottom and disc slot, to minimize the fretting effect. This aspect has been studied to understand the effect of introducing interference fit at the skewed dovetail interface on the contact stresses and tractions [123]. The study indicates that the peak contact pressure and contact stresses were reduced with the interference fit, as shown in fig.14 and fig 15, for friction coefficient of 0.3 thus offering the potential for minimizing the fretting damage through interference fit.

Fig. 14 Contact pressure distribution for (a) no interference fit (b) interference fit of 15micron (c) interference fit and bulk load-centrifugal load 1050 rad/sec.

Fig. 15 Effect of interference fit on peak contact pressure.

Surface treatments. Grogler et al [124] found that chemical vapour deposition (CVD) of diamond on Ti-6Al-4V is highly effective in reducing fretting fatigue damage. Low wear rates, low coefficient of friction against steel, alumina and diamond were observed after CVD of diamond. Hutson et al [125] carried out an experimental investigation to explore the fretting fatigue behaviour Ti-6Al-4V specimens in contact with varying pad surface conditions and Cu-Ni plasma spray coated specimen. Increase in fretting fatigue strength of 20-25% was observed for specimens tested against Cu-Ni coated pads as compared to non-coated pads.

An Overview of Fretting Aspects Relating to Aero-Engine Dovetail Attachment

715

Discussion and Future directions The studies show the 3D analysis of dovetail interface is essential for accurate prediction of fretting variables which has significant influence on the fretting fatigue crack initiation and propagation life. The fretting fatigue studies are generally limited to constant temperature and constant frictional loading. Temperature difference between the component and thermal gradient with the component would result in relative slip at the dovetail interface. The frictional interface will be subjected to variable friction coefficient during it service life because of changes in the surface conditions. Contact evolution due to fretting wear also would change the contact profiles at the interface as indicated in [126] and it has influence on the fretting fatigue. The 3D loading of interface due to complex geometric feature and rotor operating load also produces moment normal to interfaces, in addition to the radial and axial load. This would result in multimode sliding (combination of translational sliding and torsional sliding) and consequent fretting fatigue damage at the dovetail interface. Independent translational and torsional oscillation fretting studies are observed in the specimen level fretting testing. Multimode fretting also involved in other engineering applications as shown in fig. 16. Based on the review, it can be said that the following are the areas of future fretting research: (a) thermal effects at the contact interface (b) effect of variable friction coefficient at the interface (c) integration of fretting wear effect on fretting fatigue and (d) effect of multimode oscillation for improving the fretting fatigue prediction on dovetail interface.

Fig. 16 Different type of multimode oscillation [127].

716

R.V. Prakash, K. Anandavel, and P. Balasubramani

4 Summary The overview fretting fundamentals, concepts of fretting damage mechanism, fretting fatigue criteria are presented. Application of fretting concepts to aero-engine dovetail interface, severity of fretting, methodologies used for fretting variables prediction at the dovetail interface, fretting fatigue life estimation and fretting performance improvements prevailing in the aero-engine industry also presented. The results of a 3D finite element analysis suggests that the fretting variables are more accurately defined when the geometry is modelled in a 3-dimensional mode; such a modelling and analysis is essential when geometric features such as skew, twist and cone angle are considered. This paper also indicated the emerging areas of fretting research.

References [1] Eden, E.M., Rose, W.N., Cunningham, F.L.: Endurance of metals. In: Proceedings of Institute of Mechanical Engineers, vol. 4, pp. 68–76 (1911) [2] Tomlinson, G.A.: The rusting of steel surfaces in contact. Proceedings of Royal Soc. Ser. A 115, 472–483 (1927) [3] Warlow-Davis, E.J.: Fretting corrosion and fatigue strength; Brief results of preliminary experiments. In: Proceedings of Institute of Mechanical Engineers, vol. 146, pp. 32–39 (1941) [4] McDowell, J.R.: Fretting corrosion. In: STP, vol. 144, American Society for Testing and Matreial, Philadelphia (1953) [5] Fenner, A.J., Wright, K.H.R., Mann, Y.J.: Fretting corrosion and its influence on fatigue failure. In: Proc. Int. conf. Fatigue of Metals, Inst. Mech. E, London, vol. 11 (1956) [6] Fenner, A.J., Field, J.E.: La fatigue dans les conditions de frettment. Rev. Met. 55(57), 475–485 (1958) [7] Fenner, A.J., Field, J.E.: A study of onset of fatigue damage due to fretting. In: Proc. N.E. coast Inst. of Engrs. and Ship Builders, vol. 76, p. 183 (1960) [8] Hurricks, P.L.: Wear 15, 389–409 (1970) [9] Waterhouse, R.B.: Fretting corrosion. Pergamon Press, Oxford (1972) [10] Golego, N.L., Alyabev, A.I., Shevelya, V.V.: Fretting corrosion of metals. Technica Kiev, 284 (1974) (in Russian) [11] Attia, M.H., Waterhouse, R.B.: Standardization of fretting fatigue test methods and equipment. In: Proceedings from a symposium, ASTM-STP, San Antonio, vol. 1159 (1990) [12] Vingsbo, O., Soderberg, S.: Wear 126, 131–147 (1988) [13] Vingsbo, O., Odfalk, M., Shen, N.E.: Wear 138, 153–167 (1990) [14] Pellerin, V.: Thesis, Ecole Central de Lyon, Lyon (1990) [15] Zhou, Z.R., Fayeuille, S., Vincent, L.: Wear 181-183, 531–536 (1995) [16] Vincent, L., Berthier, Y., Dubourg, M.C., Godet, M.: Wear 153, 135–148 (1992) [17] Zhou, Z.R., Vincent, L.: Wear 39, 1068–1073 (1995) [18] Hertz, H.: J. Renie Agnew. Mat. 92, 156–171 (1882); (in German: for an account in English, see Johnson, ch.4.2) (1985) [19] Cattaneo, C.: Rendiconti dell’Aecademia Nazionale dei Lincei 27, 342–348 (1938)

An Overview of Fretting Aspects Relating to Aero-Engine Dovetail Attachment [20] [21] [22] [23] [24] [25] [26] [27]

[28] [29] [30] [31]

[32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52] [53] [54]

717

Mindlin, R.D.: Journal of Appl. Mech. 16, 259–268 (1949) Johnson, K.L.: Contact Mechanics. Cambridge University Press, Cambridge (1985) Nowell, D., Hills, D.A.: Int. Journal of Mech. Sciences 29(5), 355–365 (1987) Ciavarella, M.: Int. Journal Solids and Structures 35(19), 2363–2378 (1998) Hills, D.A., Nowell, D.: Mechanics of fretting fatigue. Kluwer Academic Publishers, Dordrecht (1994) Ciavarella, M., Hills, D.A., Monno, G.: I Mech. E Part C. J. of Mech. Engg. Science 2124(4), 319–328 (2003) Barber, J.R., Ciavarella, M.: Int. J. Solids and Structures 37, 29–43 (2000) Muskhelishvili, N.I.: Some Basic Problems of the Mathematical Theory of Elasticity, Noordhoff, Groningen. (Originally translated into English from the 3rd Russian edition (Izdatel’stvo Akademii Nauk, Moscow-Leningrad, by J.R.M. Radok) (1963) Truman, C.E., Sackfield, A.: Int. J. Appl. Mech. 68, 817–819 (2001) Adibnazari, S., Sharafbafi, F.: Tribology International 41(5), 341–347 (2008) Garcia, D.B., Grandt Jr., A.F.: Engineering Failure Analysis 12(4), 537–548 (2005) Buffiere, J.-Y., Proudhon, H., Ferrrie, E., Ludwing, W., Maire, E., Cloetens, P.: Nuclear Instruments and methods in Physics Research Section B: Beam Interactions with Materials and Atoms 238(1-4), 75–82 (2005) Proudhon, H., Buffiere, J.-Y., Fouvry, S.: Tribology Int. 39, 1106–1113 (2006) Meriaux, J., Boinet, M., Foury, S., Lenain, J.C.: Tribology Int. 43(11), 2166–2174 (2010) McVeigh, P.A., Farris, T.N.: ASME Journal of Tribology 119, 797–801 (1997) Kim, S.H., Mall, S.: Finite Element Analysis and Design 41, 1140–1159 (2005) McVeigh, P.A., Harish, G., Farris, T.N., Szolwinski, M.P.: Int. J. of Fatigue 21, S157–S165 (1999) Rajasekaran, R.: D. Phil. Thesis, University of Oxford (2004) Murthy, H., Harish, G., Farris, T.N.: ASME Journal of Tribology 126, 56–64 (2004) Gallego, L., Fulleringer, B., Deyber, S., Nelias, D.: Tribology Int. 43(4), 708–718 (2010) Ambrico, J.M., Begley, M.R.: Journal of the Mechanics and Physics of the Solids 48(11), 2391–2417 (2000) Giannakopoulos, A.E., Suresh, S.: Acta Materialia 46(1), 177–192 (1998) Leen, S.B., Richardson, I.J., McColl, I.R., Williams, E.J., Hyde, T.R.: Journal of Strain Analysis 36(5), 481–497 (2001) Gun, H.: Computers & Structures 82, 555–566 (2004) Elleithy, W.: Comp. Methods Appl Mech. Engineering 197, 3687–3701 (2008) Rodriguez-Tembleque, T., Abascal, R.: Computers and Structures 88, 924–937 (2010) Rodriguez-Tembleque, T., Abascal, R., Aliabadi, M.H.: Int. Journal of solids and structures 47, 2600–2612 (2010) Sfantos, G.K., Aliabadi, M.H.: Int. Journal of Solids and Struct. 43, 3626–3642 (2006) Kovalevskii, V.V.: Wear 67, 271–285 (1981) Nowell, D., Hills, D.A.: Wear 136, 329–343 (1990) Husheng, G., Haicheng, G., Huijiu, Z.: Wear 148, 15–23 (1991) Pape, J.A., Neu, R.W.: Wear 225-229, 1205–1214 (1999) Adibnazari, S., Hoeppner, D.W.: Wear 160, 33–35 (1993) Iyer, K., Mall, S.: ASME Journal of Engg. Materials and Tech. 123, 85–93 (2001) Nowell, D., Dini, D.: Tribology Int. 36, 71–78 (2003)

718 [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80] [81] [82] [83] [84] [85] [86] [87]

[88]

R.V. Prakash, K. Anandavel, and P. Balasubramani Nowell, D., Dini, D., Hills, D.A.: Engg. Fracture Mechanics 73(2), 207–222 (2006) Endo, K., Goto, H.: Wear 38, 311–324 (1976) Conner, B.P., Huston, A.L., Chambon, L.: Wear 255(1-6), 259–268 (2003) Venkatesh, T.A., Conner, B.P., Lee, C.S., Giannakopoulos, A.E., Lindley, T.C., Suresh, S.: Metallurgical and Materials Transactions A 32A, 1131–1146 (2001) Fellows, L.J., Nowell, D., Hills, D.A.: Fatigue and Fract. Engg. Mater. Struct. 20(1), 61–70 (1997) Szolwinski, M.P., Farris, T.N.: Wear 198, 93–107 (1996) Golden, P.J., Grandt Jr., A.F.: Engg. Fracture Mechanics 71(15), 2229–2243 (2004) Hills, D.A., Nowell, D., O’Connor, J.J.: Wear 125(1-2), 129–146 (1988) Ruiz, C., Boddington, P.H.B., Chen, K.C.: Exp. Mech. 24, 208–217 (1984) Lykins, C.D., Mall, S., Jain, V.K.: Int. Journal of Fatigue 23(8), 703–711 (2001) Giannakopoulos, A.E., Lindley, T.C., Suresh, S.: Acta Materialia 46(9), 2955–2968 (1998) Nicholas, T., Hutson, A., John, R., Olson, S.: Int. Journal of Fatigue 25, 1069–1077 (2003) Mutoh, Y., Xu, J.Q.: Tribology Int. 36, 99–107 (2003) Ciavarella, M., Macina, G.: Int. Journal of Mech. Sciences 45, 449–467 (2003) Conner, B.P., Lindley, T.C., Nicholas, T., Suresh, S.: Int. Journal of Fatigue 26, 511–520 (2004) Naboulsi, S.: Engg. Fracture Mechanics 72, 1610–1623 (2005) Attia, M.H.: ASME Journal of Tribology 127, 685–693 (2005) Berthier, Y., Vincent, L., Godet, M.: Wear 125(1-2), 25–38 (1988) Ratsimba, C.H.H., McColl, I.R., Williams, E.J., Leen, S.B., Soh, H.P.: Wear 257, 1193–1206 (2004) Ding, J., Leen, S.B., McColl, I.R.: Int. Journal of Fatigue 26(5), 521–531 (2004) Hattori, T., Watanabe, T.: Tribology Int. 39(10), 1100–1105 (2006) Madge, J.J., Leen, S.B., Shipway, P.H.: Wear 263(1-6), 542–551 (2007) Buch, A.: Wear 43(1), 9–16 (1977) Chivers, T.C., Gordelier, S.C.: Wear 96(2), 153–175 (1984) Shaffer, S.J., Glaeser, W.A.: Fretting Fatigue, Fatigue and Fracture. ASM Handbook 19, 221 (1996) Juuma, T.: Wear 231, 310–318 (1999) Alfredsson, B.: Int. Journal of Fatigue 31(10), 1559–1570 (2009) Kiral, B.G.: Materials and Design 31(1), 85–93 (2010) Chakherlou, T.N., Mirzajanzadeh, M.: European Journal of MechanicsA/Solids 29(4), 675–682 (2010) Liu, D., Tang, B., Zhu, X., Chen, H., He, J., Celi, J.: Surf. Coat. Technology, 116–119, 2340238 (1999) Benrabah, A., Langlade, C., Vannes, A.B.: Wear 224, 267–273 (1999) Shepard, M.J., Prevey, P.S., Jayaraman, N.: In: 8th National Turbine HCF Conference, Universal Technology Corporation, Dayton, OH (2003) Clauer, A.H., Sokol, D.W., Garcia, D.B., Bernadel, L., Ravindranath, R.: 8th National Turbine HCF Conference, Universal Technology Corporation, Dayton, OH (2003) Thomson, D.: The National high cyclic fatigue program. In: Stange WA, editor. 3rd National Turbine Engine High Cycle fatigue (HCF) Conference, San Antonio (TX): Universal technology corporation, CD-ROM Proceedings (1998)

An Overview of Fretting Aspects Relating to Aero-Engine Dovetail Attachment

719

[89] Xi, N.S., Zhong, P.D., Huang, H.Q., Yan, H., Tao, C.H.: Engineering Failure Analysis 7, 385–392 (2002) [90] Parks, V.J., Sanford, R.J.: Experimental Mechanics 18(9), 328–333 (1978) [91] Durelli, A.J., Dally, J.W., Riley, W.F.: Stress and Strength Studies on turbine blade attachments. In: Proceedings of Society for Experimental Stress Analysis, vol. XVI(1), pp. 171–186 (1958) [92] Ruiz, C., Poon, C.Y.: Measurements 12(1), 45–57 (1993) [93] Burguete, R.L., Patterson, E.A.: Journal of Strain Analysis 32(6), 425–433 (1997) [94] Ruiz, C., Boddington, P.H.B., Chen, K.C.: Experimental Mechanics 24(3), 208–217 (1984) [95] Rajasekaran, R., Nowell, D.: Tribology International 39, 1277–1285 (2005) [96] Szlowinski, M.P., Farris, T.N.: Wear 221, 99–107 (1997) [97] Ciavarella, M., Demelio, G.: Int. J. of Solids and Struct. 38, 1791–1811 (2001) [98] Sinclair, G.B., Cormier, N.G.: ASME J. Engg. Gas Turbine Power 124, 325–341 (2002) [99] Alderson, R.G., Tani, M.A., Tree, D.J., Hills, R.J.: J. of Aircraft 13(2), 994–999 (1976) [100] Kenny, B., Patterson, E., Said, M., Aradhya, K.S.: Strain 27(1), 21–24 (1991) [101] Boddington, P.H.B., Chen, K.C., Ruiz, C.: J. of Comp. and Strut. 2(4), 731–735 (1985) [102] Meguid, S.A., Refaat, M.H., Papanikos, P.: Journal of Mat. Tech. 56, 1539–1547 (1996) [103] Sinclair, G.B., Cormier, N.G., Griffin, J.H., Meda, G.: J. Engg. Gas Turbine Power 124, 182–189 (2002) [104] Rajeev, P.T., Murthy, H., Farris, T.N.: ASME J. Engg. Gas Turbine Power 126, 385–390 (2004) [105] Hartle, M.S., Farris, T.N., Matlik, J.F., Rajeev, P.T.: A hybrid approach to 3D edge of contact stress analysis using singular integral equations. In: Proceedings of 2001 National High Cycle Fatigue Conference, Jacksonville, FL (2001) [106] Papanikos, P., Meguid, S.A., Stjepanovic, Z.: Finite Element in Analysis and Design 29, 668–677 (1998) [107] Meguid, S.A., Kanth, P.A., Chekanski, A.: Finite Elements in Analysis and Design 35, 305–317 (1998) [108] Beisheim, J.R., Sinclair, G.B.: ASME J. Turbo. Mech. 125, 372–379 (2003) [109] Anandavel, K., Prakash, R.V.: Tribology Int. (2010), doi:10.1016/j.triboint.2010.10.014 [110] Waterhouse, R.B., Dutta, M.K.: Wear 25, 171–175 (1973) [111] Harris, W.J.: AGARD Advisory report No. 45 (1972) [112] Goss, G.L., Hoeppner, D.W.: Wear 24, 153–159 (1974) [113] Hamdy, M.M., Waterhouse, R.B.: Wear 71(2), 237–248 (1981) [114] Cortez, R., Mall, S., Calcaterra, J.R.: Int. Journal of Fatigue 21(7), 709–717 (1999) [115] Namjoshi, S.A., Jain, V.K., Mall, S.: ASME J. of Engg. Matls. and Tech. 124, 222–228 (2002) [116] Szolwinski, M.P., Farris, T.N.: Wear 198, 93–107 (1996) [117] Calcaterra, J., Naboulsi, S.: Int. Journal of Fatigue 27(9), 1133–1141 (2005) [118] Sinclair, G.B., Cormier, N.G.: ASME J. Engg. Gas Turbine Power 125, 1033–1041 (2003) [119] Beisheim, J.R., Sinclair, G.B.: ASME J. Turbomachinery 130(2), 021012-1:8 (2008)

720

R.V. Prakash, K. Anandavel, and P. Balasubramani

[120] Beisheim, J.R., Sinclair, G.B.: ASME J. of Engg. for Gas Turbines and Power 132, 064504-1:4 (2010) [121] Gautreau, J.C., Martin, N.F., Rickert, C.A. (2006), Compressor blade with dovetail slotted to reduces stress on the airfoil leading edge, Patent No. US7121803B2 (October 17, 2006) [122] Przytulski, J.C., Hemsworth, M.C. (1992), Apparatus for preloading an airfoil loading in a gas turbine engine, Patent No. US5123813 (June 23, 1992) [123] Anandavel, K., Prakash, R.V., Davis, A.: Effect of preloading on the contact stress distribution of a dovetail interface. In: International conference on Mechanical and Industrial Engineering (ICMIE 2010), Singapore, vol. 69, pp. 360–365 (2010) [124] Grogler, T., Franz, A., Klaffke, D., Rosiwal, S.M., Singer, R.F.: Diamond and related Materias 7, 1342–1347 (1998) [125] Hutson, A.L., Niinomi, M., Nicholas, M., Eylon, D.: Int. Journal of Fatigue 24(12), 1223–1234 (2002) [126] Madge, J.J., Leen, S.B., McColl, I.R., Shipway, P.H.: Wear 262, 1159–1170 (2007) [127] Kenneth, B.G.: In: Fretting tests, Guide to Friction, Wear and Erosion Testing, ASTM International (2007)

26th ICAF Symposium – Montreal, 1-3 June 2011 Fatigue Analysis of the Compressor Blades with V- Notches Lucjan Witek Department of Aircraft and Aero Engines, Rzeszow University of Technology, Powstancow Warszawy Ave. 8, 35-959 Rzeszow, Poland [email protected]

Abstract. This paper presents results of experimental and numerical fatigue analysis of damaged compressor blades, subjected to vibration. The blades used in experimental investigations were preliminary defected to simulate the foreign object damage. The crack propagation process was conducted in resonance condition. During the fatigue investigations, the crack length and amplitude of the blade tip displacement were monitored. The main result of experimental test is the crack growth rate obtained for the blade including V-notch. In the second part of work, the results of the complex stress and fatigue analysis for the helicopter turbo-engine compressor blades were presented. A nonlinear finite element method was utilized to determine the stress state of the blade during the first mode of transverse vibration. In this analysis, the numerical models without defects and also with V-notches were defined. Obtained results were next used as an input data into crack initiation (ε–N) analyses performed for the load time history equivalent to one cycle of the transverse vibration. As a result of ε–N analysis, the number of load cycles to the first fatigue crack appearing in the compressor blade was obtained. Moreover, the influence of the blade vibration amplitude on the number of cycles to the crack initiation was analyzed. The numerical results were compared with the results of an experimental high-cycle fatigue (HCF) tests performed for the first stage compressor blades.

1 Introduction Foreign object damage (FOD) is a prime reason for maintenance and reparation of military jet engines which operate on landing grounds. The damage induced by small hard objects of millimeter size in conjunction with the typical load spectra experienced by airfoils can lead to non-conservative life prediction and unexpected fatigue failures. Damage of the compressor blades of engine is normally caused when a particle is hit by the rotating blade. There is high relative velocity due to the motion of the blade and acceleration of the particle causes high forces and local damage of the blade. Often this damage is at or close to the attack edge of the compressor blade and takes the form of a dent or notch in the leading edge (Fig. 1).

722

L. Witek

a)

b)

c)

Fig. 1 View of the aeroengine compressor blades after foreign object damage (a, b) and artificial defect (V-notch) created on the attack edge of investigated blade (c).

The failure analysis of the compressor blade has received the attention of several investigations. The problem of fatigue fracture of the compressor blade was described by: Lourenco et al. [1], Kermanpur et al. [2], Silveira et al. [3], Poznanska et al. [4]. The stress and failure analysis of the compressor blades were also described in [5-6]. This paper is the continuation of work [6] in which undamaged blades were tested in the fatigue conditions. The main objectives of presented investigations are to determine both the stress state and also the number of fatigue cycles to the first crack appearance in the compressor blades (without defects and also including artificially created V-notch), subjected to vibration. These defects (notches) simulate foreign object damage of the blade.

2 Experimental Investigations The investigated blade was made out of EI-961 steel (0.11C; 11Cr; 1.5Ni, 1.6 W; 0.18V; 0.35Mo; 0.025S; 0.03P) with the following properties (measured in temperature 20OC): Ultimate tensile strength (UTS) 900-1000 MPa, Yield stress 800-900 MPa, Young modulus 200 GPa, Poisson ratio 0.3. Before the test, each blade was mechanically damaged (by hitting of a sharp beam made of hardened steel). During this stroke, the blade was fixed in the dovetail region. As a result of this operation, the V-notch presented in Fig. 1c was created. The apex angle of the notch was 90O whereas the depth about 0.5 mm. In presented fatigue examinations twenty five blades were investigated. The majority of them had a single notch located on the attack edge of the blade. In a few cases, the single notch was created on the trailing edge of blade. Moreover the blades with double and multiple notches located on the attack edge were also investigated. Experimental investigations were performed at Research and Development Laboratory for Aerospace Materials of Rzeszow University of Technology. In fatigue examinations the LDS-V830 electrodynamic vibration system, presented in Fig. 2a were used.

Fatigue Analysis of the Compressor Blades with V- Notches

723

Preliminary damaged blade (containing notch) was horizontally mounted on the movable head of the shaker (Fig. 2b). The investigation was started from searching resonance frequency (for 1st mode of transverse vibration). The fatigue test was just started from this frequency. During the test, both vibration amplitude of the blade tip and the size of the crack were periodically monitored. For control of amplitude, the optical measuring microscope MPB-3 and laser scanning vibrometer POLYTEC PSV-400 were used. To measure the length of the crack a non-destructive fluorescent penetrant inspection was utilized. View of luminous crack in UV light detected during fatigue test is presented in Fig. 2c. Investigations were conducted for different intensity of vibration, which was defined by amplitude of acceleration of the movable vibrator head. Amplitude of acceleration (excitation) was defined in g units, where 1g equals 9.81 m/s2. The fatigue tests were performed for intensity of vibration between 5 and 20 g.

a)

b)

c)

Fig. 2 View of vibration system LDS-V830 and laser scanning vibrometer Polytec PSV400 used in experimental fatigue investigations (a), blade fixed to shaker during fatigue test (b) and view of luminous crack in UV light (c).

Size of this paper is limited and from this reason, the results for only two from twenty five investigated blades will be in detail described. The results presented in Figs. 2c, 3 and 4 are obtained from the tests in which intensity of vibration was defined as 11g (for blades nos. 7 and 8).

724

L. Witek

a)

b) Fig. 3 Shape of crack front in first phase of cracking, blade no. 8 (a), and blade fracture with marked beach marks, created by different number of cycles, blade no. 7.

The fractured blade number 7 is shown in Fig. 3b. The crack was initiated on the concave surface of profile (from notch), about 5 mm above the blade locking piece. The resonant frequency for the blade no. 7 is 780 Hz, and from this frequency the test was started. After 7×105 total number of cycles, an amplitude of blade tip decreased from 2.47 mm to 2.34 mm. During cracking, the bending stiffness of the blade is decreased. This information is very important, because the decrease of amplitude by constant intensity of excitation is often related to the start of a crack initiation process. In this case 1.2 mm long crack was detected. The intention of the test was to maintain the blade tip amplitude at the constant level during the fatigue test. This assumption is very difficult to perform. Under cracking process, the resonant frequency decreases and the excitation frequency also must be changed. From this reason, after 11×105 cycles, the excitation frequency decreased with variable rate. The crack length in function of the total number of cycles (crack growth rate), obtained for the blade no. 7 is presented in Fig. 4. As seen from this figure, as long as the line of a graph is horizontal, the blade has no crack. The time to crack initiation (counted from the beginning of a test) is about 5×105 number of cycles. From this moment a crack growth line is ascending. The blade was ruptured after about 12.5×105 number of cycles (counted from the start of crack initiation process). The critical size of crack (ac) at which the blade was ruptured was 18 mm.

Fatigue Analysis of the Compressor Blades with V- Notches

725

Fig. 4 Crack length in function of total number of cycles for compressor blade (intensity of vibration 11g, blade no. 7).

Fig. 5 Dimensions of crack propagated from the attack edge of blade.

The fatigue test for blade number 7 was conducted in short time and the crack was propagated in non-corrosive environment thus the beach marks presented in Fig. 3b are not well visible. For better determine of the crack front shape in the early stage of fracture, blade no. 8 was vibrated until the crack length (a dimension shown in Fig. 5) achieved about 6 mm. Next, the blade was statically tensioned and ruptured using the testing machine. Shape of crack in preliminary phase of growth is presented in Fig. 3a. As seen from this figure, the crack in the first phase of growth propagates more quickly along the concave surface of the blade profile than on the convex surface. The damage schemes of a large majority of remaining blades are similar to presented in Fig. 3. The crack origin was usually located on the attack edge of blade and the cracks were propagated from the v-notch. In the case of a blade with multiple notches (located on the attack edge) the crack was initiated from the notch positioned about 4 mm above the blade locking piece.

3 Numerical Stress Analysis of the Blade The finite element (FE) models used in this work can be divided into two groups. The first group consists of the models which do not have any defects. In this group, the different numbers of finite elements were used to models definition. For example, the model presented in Fig. 6a consists of 60477 nodes and 13440

726

L. Witek

HEX-20 elements. The HEX-20 isoparametric finite element has quadratic shape functions and gives a good convergence of the numerical solution.

a)

b)

c)

Fig. 6 Numerical model of the non-defected blade (a), magnified view of model with sharp notch (b) and model with finite radius of notch (c).

A few blades with different notch radius belong to the second group of FE models (Figs. 6b and 6c). In the stress analysis, models with the following notch radius: 0 mm, 0.025 mm, 0.05 mm; 0.075 mm and 0.1 mm were considered. The V-notch in FE model was located about 5 mm above the blade locking piece. During analysis, the blade was fixed on the bottom surface of the dovetail. In FE analysis, the Abaqus 6.9 program was used [7]. In the first part of investigations, a free vibration analysis was performed to obtain the resonant frequencies frez and vibration modes for the non-defected blades. These blades are consisted of a different number of finite elements. In further analysis only the first mode of transverse vibration will be considered. The quality of FEM solution depends on many factors. One of the main factor is the number of finite elements used for creation of the model. The main index during the convergence analysis was value of the resonant frequency (1st mode) obtained from calculations. Results of the convergence analysis are presented in Fig. 7. Value of the resonant frequency for the model which consists of about 2500 HEX-20 elements is equal 814 Hz. When the number of elements in numerical model grows, the resonant frequency decreases. Value of frez approaches a limit of about 788 Hz. Thus, in the presented case the satisfactory results can be obtained when the model has more elements than 13000. However, the maximum number of FE must be finite because it is limited by the computational power of a workstation. Values of frez (obtained from experimental investigations) were in the range of 770-790 Hz [6]. The Von Mises stress distribution does not show if the material is tensioned or compressed. Owing to this fact, the maximum principal stress (σ1) distributions were analyzed in this paper. This stress is particularly interesting from the point of view of the fatigue strength because just the tensile stresses the most contribute to the fatigue crack initiation and subsequently to crack propagation.

Fatigue Analysis of the Compressor Blades with V- Notches

727

Resonant frequency [Hz]

815 810 805 800 795 790 785 2000

6000 10000 14000 Number of finite elements

18000

Fig. 7 Resonant frequency in function of the total number of finite element used in the numerical model of the blade without preliminary defect.

Figs. 8a and 8b show that the highest tensile stress area is located on the convex surface of the blade where the profile is connected to the dovetail. When the amplitude of the blade tip displacement equals 1 mm, the σ1 stress achieves 248 MPa. A bit lover value of the tensile stresses (192-211 MPa) is observed in the zone located between 1 mm and 15 mm above the dovetail, on the central part of convex blade surface. In this zone the large number of fatigue cracks in the compressor blades was observed in experiment [6]. Near the neutral axis of the blade cross-section, the tensile stress decreases to 0 MPa. Between maximum stress region and the neutral axis, the high gradient of stresses is observed.

a)

b)

Fig. 8 Maximum principal (σ1) stress distribution for blade during resonance (a) and stress distribution in cross-section of the blade, 5 mm above the dovetail (b). 1st mode of transverse vibrations, amplitude of blade tip 1 mm.

In the next stage of the stress analysis, the blades with different V-notch radius were considered. The zero radius of notch is the worse case, in which the highest stress is observed. Figure 9a shows that the σ1 stress in the notch zone of the blade vibrated at amplitude 1 mm achieves 878 MPa.

728

L. Witek

Max. Princ. Stress in notch tip [Mpa]

900 850 800 750 700 650 600 550 500 450 400 0

0,02

0,04

0,06

0,08

0,1

Notch radius [mm]

a)

b)

Fig. 9 Maximum principal stress distribution in the notch vicinity (a), (1st mode of transverse vibrations, amplitude of blade tip: 1 mm, notch radius r = 0, and the σ1 stress values in function of the notch radius (b).

Relation between the notch radius and the maximum principal stress value is presented in Fig. 9b. As seen from this figure, the maximum tensioned stress depends on the notch radius. For the sharp notch (radius r=0 mm) the maximum σ1 stress in the notch zone achieves 878 MPa. When the notch radius increases, than the stress in the notch tip is lower. For example, if the notch has a radius 0.075 mm, than the stress reduces to about 630 MPa.

4 Fatigue Crack Initiation Analysis Purely static loading is rarely observed in modern engineering components or structures. By far, the majority of structures involve parts subjected to fluctuating or cyclic loads. For this reason, design analysts must address themselves to the implications of the repeated loads, fluctuating loads, and rapidly applied loads. Such loading induces fluctuating or cyclic stress than often result in failure of the structure by fatigue [8]. The compressor blade has a thin profile and in consequence a low bending stiffness. The fluctuation of the compressed air or rotation of the unbalanced shaft induces vibration of the blade. To estimate the number of cycles to occurring of the first fatigue crack in the blade, the program MSC Fatigue 9.0 was used. This program enables to perform the ε-N (crack initiation) analysis for the geometrically complicated numerical models defined by the user [9]. The program Fatigue needs (as the input data) the results obtained from the static stress analysis. In ε-N analysis only the maximum principal stresses are considered. In presented example, the Smith-Watson-Topper (S-W-T) mean stress method was used.

Fatigue Analysis of the Compressor Blades with V- Notches

729

The fatigue calculations were performed for the load time history equivalent to one cycle of the transverse vibration (Fig. 10a). Maximum value of 1 on the vertical axis is associated with the stress state obtained during the blade vibration (for one extreme deflection of blade).

1

Load

0,5 0 0

1

2

3

4

-0,5 -1

Tim e

a)

b)

Fig. 10 Load time history equivalent to one cycle of the transverse vibration (a) and strainlife (ε-N) plot for the EI-961 alloy.

The static (monotonic) properties of the EI-961 alloy were described in chapter 2. In order to perform a crack initiation analysis, the additional material properties used in the fatigue calculations will be needed. Table 1 presents values of the fatigue properties of the EI-961 alloy calculated according to Baumel-Seeger equations. In these calculations the Young modulus were defined as 200 GPa while the UTS as 1090 MPa. The ε-N analysis was also performed for UTS=1000 MPa and 950 MPa. The strain-life curve plotted for the EI-961 alloy is presented in Fig. 10b. Table 1 Values of the fatigue properties of EI-961 steel obtained on the base of BaumelSeeger solution (UTS=1090 MPa, G=200 GPa). Fatigue properties Fatigue strength coefficient [Mpa] Fatigue strength exponent [-] Fatigue ductility exponent [-] Fatigue ductility coefficient [-] Cyclic strain-hardening exponent [-]

Symbol S'f b c ε 'f K'

Formula 1.5 UTS -0.087 -0.58 0.59 α 1.61 UTS

Value 1635 -0.087 -0.58 0.4015 1799

The results of ε-N analysis for the non-defected compressor blade which works in the resonance conditions are presented in Fig. 11. The results are displayed in the exponential form. The practical meaning of these results has the minimum estimated number of cycles. In the zones displayed as red, the first fatigue cracks are expected. For example, when the amplitude of the blade tip during the 1st

730

L. Witek

mode of vibration is 1.5 mm, the estimated number of cycles to the first fatigue crack is 106.13 (1348963 cycles) (Fig 11). The first fatigue crack is understood in this case as the detectable crack, which has a length about 1.5-2 mm [9]. There are two critical fatigue zones in the blade: central part of the convex surface (Fig.11a) and the attack edge of the blade. Location of the critical (red) zones overlaps to the areas, where the cracks were detected during experimental fatigue analysis. The fatigue cracks in non-defected blades were initiated from the central part of the convex surface.

9 Results of the E-N analysis (No. of cycles to the crack initiation = 10 X

Rm= 950 MPa

8

Rm=1000 MPa Rm=1090 MPa

7 6 5 4 3 1

1,5

2

2,5

3

3,5

4

Amplitude of crack tip displacement [mm]

a)

b)

Fig. 11 Number of cycles at which the first fatigue crack can be initiated in the blade without preliminary defects (vibration amplitude 1.5 mm) (a) and number of cycles (showed in exponential form) to the crack initiation in function of the vibration amplitude for diversified UTS (Rm) of blade material. Table 2 Number of cycles to the first fatigue crack for the blade made out of material with different UTS, (blade without preliminary defects). UTS Blade tip amplitude [mm]

1 1.5 2 3 4

950 Mpa Exponent 8.29 6.8 5.59 4.16 3.45

1000 Mpa 1090 MPa Number of cycles to the crack initiation Integer Exponent Integer Exponent Integer 194984460 8.53 338844156 8.97 933254301 6309573 7.04 10964782 7.37 23442288 6.13 1348963 389045 5.79 616595 14454 4.27 18621 4.42 26303 2818 3.52 3311 3.67 4677

The FE fatigue computations were made for the following amplitude of the blade tip displacement: 1, 1.5, 2, 3 and 4 mm. Moreover, the influence of the ultimate tensile strength (UTS, Rm) of the EI-961 alloy on the fatigue life of blade was considered. In this analysis the fatigue properties were defined on the base of

Fatigue Analysis of the Compressor Blades with V- Notches

731

following values of UTS: 950; 1000 and 1090 MPa. Results obtained for different vibration amplitude and also for diversified UTS of the EI-961 alloy were shown in Tab. 2 and Fig. 11 (for the blade without preliminary defects). There are two forms used for description of the numbers of cycles: integer and exponential in this table. For the exponential form the base of power is 10. The results presented in Fig. 11a (106.13 = 1348963 number of cycles) are displayed in Tab. 2 with use of the bold fonts.

Results of E-N analysis

X

(No. of cycles to crack initiation=10 )

5 r=0 r = 0,025 mm r = 0,075 mm

4,5 4 3,5 3 2,5 2 1,5 1 1

1,5

2

2,5

3

3,5

4

Amplitide of crack tip dispacement

a)

b)

Fig. 12 Result of the ε-N analysis for the blade with notch radius r= 0.075 mm (a) and for the notched blade with different notch radius (b).

When the blade has the notch (or mechanical defect as FOD), the estimated number of cycles is much lower than computed for new blade without any defects. For the blade with notch radius of 0.075 mm (vibrated at amplitude 1.5 mm), the estimated number of cycles to crack initiation is 103.97 = 9333. When the notch is sharp (r=0 mm), the crack can initiate after about 2630 cycles (Tab. 3). Table 3 Number of cycles to crack initiation for different notch radius. Notch radius [mm]

Blade tip amplitude [mm] 1 1.5 2 3 4

r=0 Exponent 4.11 3.42 2.92 2.13 1.47

r = 0.025 Integer 12882 2630 832 135 30

r = 0.075

Number of cycles to crack initiation Exponent Integer Exponent 4.29 19498 4.69 3.97 3.62 4169 3.06 1148 3.37 2.25 178 2.49 1.64 44 1.99

Integer 48978 9333 2344 309 98

732

L. Witek

In order to check the quality of the solution obtained from the numerical calculations, the comparison between FE and experimental results was made. As seen from Fig. 13, the experimental results for the non-defected blade have a wide dispersion for the blades tested in the same conditions. The dotted line represents average results of the experimental investigations. The number of cycles to crack initiation computed on the numerical way is about 100% lower than the results of experimental investigations performed for the same conditions. Numerical calcul. Rm=1090 MPa Experimental test, Blade no.1 Experimental test, Blade no.2 Experimental test, Blade no.3 Experimental test, Blade no.4 Experimental test, Blade no.5 Experimental test, Blade no.6 Experimental test, Blade no.7 Experimental test, Blade no.8 Experimental test, Blade no.9 Experimental test, Blade no.10

Results of the E-N analysis

(No. of cycles to the crack initiation = 10X

9

8

7

6

5

4

3 1

1,5

2

2,5

3

3,5

4

Amplitude of crack tip displacement [mm]

Fig. 13 Comparison of the experimental and numerical results for the blade without preliminary defects.

The verification performed for the blades with the notches revealed that the numerical estimation gives an about 3-5 times lower values of the number of cycles to crack initiation than the experimental results. For example, the estimated number of cycles to crack initiation of the notched blade vibrated witch amplitude of 1.5 mm (for notch radius r=0.075 mm) is about 10000 (0.1 × 105 cycles) (Tab. 3). By comparison, in the real (experimentally tested) blade vibrated at the same amplitude, the first crack was usually detected at about 0.3-0.5 × 105 cycles. The numerical fatigue results are conservative but the difference between FE estimation and experimental results (from the science point of view) is big. The next analysis should be performed to explanation of this divergence.

5 Summary In this paper, the results of experimental and numerical fatigue analysis performed for the compressor blade were presented. The blades used in experimental investigations were preliminary defected to simulate the foreign object damage. The crack propagation process was conducted in resonance condition. During the fatigue investigations, the crack length and amplitude of the blade tip displacement were monitored. In the next part of work, the complex stress and fatigue numerical analysis of the compressor blades was performed. In this

Fatigue Analysis of the Compressor Blades with V- Notches

733

analysis the blades both without preliminary defects and also with the notches were considered. In both cases, the influence of the blade vibration amplitude and also UTS of the blade material on the number of cycles to the crack initiation was investigated. In the notched blades, the influence of the notch radius on the fatigue strength of the blade was additionally examined. The finite element results were compared with the results of an experimental vibration tests. Based on presented analysis, the following conclusions can be formulated: 1. The maximum tensile stress in the blade without defects, vibrated with amplitude of 1 mm is about 248 MPa (Fig. 8). When the blade has the notch with radius of 0 mm, the computed stress near the notch tip is about 3 times larger (878 MPa, Fig. 9). Thus, the most unfavorable case is when the blade is damaged by a sharp foreign object during the engine operation and the notch has a radius close to 0.2. 2. The estimated number of cycles to the crack initiation for the blade without the mechanical defects is about 1,4 million (mln), (for the blade vibrated with amplitude of 2 mm). When the amplitude increases to the value of 4 mm, the fatigue durability of the blade decreases to 4677 cycles (Tab. 2). Results of ε-N analysis strongly depends on the amplitude of blade tip displacement. 3. The estimated life of the blade with the notch or FOD is very low (Fig. 12, Tab. 3). For the notched blade vibrated witch amplitude of 1mm (for notch radius r=0mm), the estimated number of cycles to the crack initiation is 12882. It means that the blade vibrated with resonant frequency equaled about 800 Hz, will work only 12 second to the first crack appearance. When the radius of the notch is 0.075 mm, the crack will initiate after about 49000 cycles. 4. Results of ε-N analysis strongly depends on the Ultimate Tensile Strength (UTS, Rm) of the blade material. The estimated number of cycles to the crack initiation (for the following parameters: UTS = 1090 MPa, vibration amplitude of 2 mm, not defected blade) is about 1,4 mln. cycles. When the UTS decreases to 1000 MPa, the initiation process begins at 0.6 mln cycles. Result of ε-N analysis for UTS=950 MPa is about 0.4 mln of cycles. The real UTS of the EI961 alloy depends on its heat treatment. Quality of the heat treatment should be controlled during manufacturing process because of its meaningful influence on the fatigue life. 5. The observed variation of the experimental results is more than 300% (Fig. 13). In the experimental fatigue analysis this dispersion is typical. 6. Based on the obtained results, it seems that the numerical fatigue calculations are conservative (from engineering point of view). The results of numerical calculations for non-defected blade are about 100% lower than the real number of cycles to the crack initiation obtained in experiment. 7. The results divergence could be caused by not accurate fatigue material properties estimation. The experimental tests (or different analytical method to obtain these properties) should be performed before next analysis.

734

L. Witek

8. The blade during manufacturing process was shoot peened. The shoot peening introduces the compressive (residual) stress into the surface layer of the material. Superposition of the actual (tensioned) stress with the residual (compressed) stress causes that the fatigue life of the shoot peened components is larger. In the ε-N analysis, the influence of the residual stresses was not considered. This fact could be the reasons for divergence of the numerical and experimental results. 9. The divergence between numerical and experimental fatigue results for the blades with the notch is very big (300-500%). One of possible reasons for the low fatigue resistance estimation is that during the notch creation process (notch was created by hit of the sharp object in the attack edge of the blade), very large plastic strain is appeared. In consequence of this, the large compressive (residual) stress zone is also formed in the notch vicinity. The influence of the residual stresses on the fatigue life of the blade should be also investigated.

Acknowledgement This work was supported by Polish Ministry of Science and Higher Education (Project No. N N - 504 346736).

References [1] Lourenço, N.J., Graça, M.L.A., Franco, L.A.L., Silva, O.: Fatigue failure of a compressor blade. Engineering Failure Analysis 15, 1150 (2008) [2] Kermanpur, A., Sepehri Ami, H., Ziaei-Rad, S., Nourbakhshnia, N., Mosaddeghfar, M.: Failure analysis of Ti6Al4V gas turbine compressor blades. Engineering Failure Analysis 15, 1052–1064 (2008) [3] Silveira, E., Atxaga, G., Irisarri, A.: Failure analysis of a set of compressor blades. Engineering Failure Analysis 15, 666–674 (2008) [4] Poznanska, A., Sniezek, M., Wierzbinska, M.: Pitting corrosion – main factor generating fracture of the compressor of aeroengine blades under operation. In: Proceedings of IX conference of Turbomachinery, Rzeszow (2003) [5] Witek, L., Wierzbińska, M., Poznańska, A.: Fracture analysis of compressor blade of a helicopter engine. Engineering Failure Analysis 16(5), 1616–1622 (2009) [6] Witek, L.: Experimental crack propagation and failure analysis of the first stage compressor blade subjected to vibration. Engineering Failure Analysis 16(7), 2163–2170 (2009) [7] ABAQUS Users Manual, ver. 6.10, Abaqus Inc. (2011) [8] Kocańda, S., Szala, J.: Fundamentals of fatigue calculations. PWN Warszawa (1997) [9] MSC-Fatigue 9.0, Users Manual-Fatigue theory, Los Angeles (2000)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Fatigue and Residual Strength Performance of Bonded Repairs to Metallic Fuselage John G. Bakuckas Jr.1 and Bud Westerman2

1

FAA William J. Hughes Technical Center, Atlantic City International Airport, NJ 08405, USA 2 Boeing Research & Technology, 9725 East Marginal Way, Seattle, WA 98108, USA

Abstract. Test and analysis were performed in a phased approach to study the fatigue and residual strength performance of adhesively bonded repairs to metallic fuselage under a variety of loading conditions and initial damage scenarios. Both boron/epoxy and aluminum bonded patches were used to repair lap joint scribes and through-the-thickness cracks. Various loading conditions were applied, including fatigue under simulated in-service and elevated loads and thermalmechanical static loading to measure residual strength. During all test phases, damage formation and growth of cracks and disbonds were monitored and recorded using several nondestructive inspection methods. Results revealed that properly designed and installed bonded repairs are durable and effective over long periods of fatigue and exceed typical design service goals of transport category airplanes. In addition, bonded repairs can successfully contain large damage under severe loads in excess of ultimate load requirements.

1 Introduction The application of bonded repairs to aircraft structures has been studied extensively over the past three decades, particularly for military applications [1]. Bonded repairs have been shown to be a viable alternative to metallic fastened repairs. Obvious advantages include the aerodynamic and structural efficiency of bonded repairs and the ability to significantly reduce stress concentrations. Despite these advantages, the use of bonded repairs in commercial applications is limited due to the lack of confidence in bonding. Ensuring bond quality and durability is of major concern. As such, in certification programs involving bonded repairs to a primary structure, inspection credits are not typically provided unless bondline integrity can be substantiated over the life of the part. In this study, the fatigue and residual strength performance of bonded repairs to a metallic fuselage was investigated through testing and analysis. The overall objectives were to characterize the fatigue behavior of bonded repairs under simulated service load (SL) conditions and to determine if the repairs meet strength, deformation, and damage tolerance requirements in residual strength tests. This initial program followed four phases to study different damage *

Oral presentation.

736

J.G. Bakuckas Jr. and B. Westerman

scenarios and corresponding repair configurations using both boron/epoxy (B/Ep) and aluminum patches. An overview of the results obtained from all four test phases are provided in terms of bond repair durability and damage tolerance. Several nondestructive inspection (NDI) techniques were used to detect disbonds and fatigue cracking. Test results revealed that properly designed and installed bonded repairs are durable and effective over long periods of fatigue and exceed typical design service goals of transport category airplanes. In addition, bonded repairs can successfully contain large damage under severe mechanical and thermal load conditions in excess of ultimate load requirements.

2 Experimental Procedure Tests were conducted using the Federal Aviation Administration (FAA) Full-Scale Aircraft Structural Test Evaluation and Research (FASTER) facility as summarized in this section. Further details can be found in references [2-4]. Description of Test Panel The fuselage panel used in this study was removed from the crown of a retired passenger service Boeing 727 airplane having dimensions of 3.175 m by 1.85 m and a radius of 1.88 m as shown in Figure 1. The substructure included six stringers (S) and six frame stations (FS). The skin material was 2024-T3 aluminum with a thickness varying from 1.0 mm between FS-720C and -720D to 2.0 mm between FS-720F and -740. A longitudinal lap joint was located along stringer S-4L and a circumferential butt joint was located along FS-740. The panel was reinforced with aluminum doublers along the four edges. Test Phases Table 1 summarizes the applied load history for each of the four test phases. A description of the test phases is provided in the subsequent sections. Phase 1: Durable Patch Design The purpose of this test phase was to demonstrate and evaluate the durability and fatigue performance of bonded repairs and to correlate results with analytical models. Two damage scenarios were considered, namely, a mid-bay through-the-thickness crack (fatigue presharpened 76.2 mm long) and a lap joint scribe (152.4 mm long by 0.3 mm deep), as shown in Figure 1. Both B/Ep and aluminum patches were used to repair this damage. For mid-bay Crack 1, a B/Ep patch, C1BE, was installed with fibers oriented in the hoop direction to provide a stiff repair. For mid-bay Crack 2, an aluminum repair, C2A, was installed using aluminum that matched the properties of the skin material.

Fatigue and Residual Strength Performance of Bonded Repairs to Metallic Fuselage FS -720B

FS- 720C

FS- 720D

FS-720E

FS- 720F

737

FS-740

Boron/Epoxy Alum Patch S-2L S-3L

S1BE

S2BE

S3A

Scribe 1

Scribe 2

Scribe 3

S-4L

1.85 m

S-5L

C1BE

S-7L

C2A Crack 2

Crack 1

S-6L Hoop Axial

Scribes: 152.4 mm Long by 0.3 mm Deep Notches: 76.2 mm Long

3.175 m

Fig. 1 Panel configuration and Phase 1 repair designations.

The effect of stiffness was studied using the lap joint scribe repairs. Fibers were oriented in the axial direction for Scribe 1 B/Ep repair, S1BE, to yield a lowstiffness, compliant patch in the hoop direction. For Scribe 2 B/Ep repair, S2BE, the fibers were directed in the hoop direction to provide a high-stiffness patch. For Scribe 3 aluminum repair, S3A, the patch material was selected to match the skin. The test sequence for Phase 1, summarized in Table 1, consisted of fatigue and residual strength tests. The loads used for the fatigue test simulated the SL conditions, including cabin pressurization (61.4 kPa) and fuselage vertical bending, and was represented by an equivalent constant-amplitude spectrum. Fatigue loading was applied for 60,000 cycles at a frequency of 0.33 Hz with an R ratio (minimum to maximum load) of 0.1. For the residual strength test, the applied load corresponded to the damage tolerance requirements of Title 14 Code of Federal Regulations (CFR) 25.571 “Maximum value of the normal operating pressure including the expected external aerodynamic pressure during 1g level flight multiplied by 1.15” (70.6-kPa pressure). Phase 2:No-Damage Growth Repair Design In this phase, a no-damage growth repair design and installation was demonstrated. For this, the original mid-bay B/Ep repair (C1BE) was removed and replaced with an aluminum patch repair (C1A) that would prevent crack formation and growth in the skin. The crack tips developed during the Phase 1 tests were removed using a 19.1-mm drill stop hole. All other repair patches remained intact. Figure 2a shows the damage scenarios and the corresponding repairs for Phase 2.

738

J.G. Bakuckas Jr. and B. Westerman Table 1 Test phases and loading history.

Description

Load Type

Maximum Load Pressure (kPa)

Hoop (kN)

Axial (kN)

Frame (kN)

Cyclic, R = 0.1

61.4

42.3

39.6

6.7

Residual Strength, 1.15 SL

Quasistatic

70.6

48.7

36.6

7.7

Fatigue, 1.15 SL, 10,000 cycles

Cyclic, R = 0.1

70.6

48.7

36.6

7.7

Fatigue, 1.33 SL, 10,000 cycles

Cyclic, R = 0.1

81.6

56.3

39.6

8.9

61.4

42.3

39.6

6.7

70.6

48.7

36.6

7.7

81.6

56.3

39.6

8.9

105.4

73.0

39.6

9.3

170.5

117.6

39.6

18.7

70.6

48.7

36.6

7.7

3: Large-Damage Repair Capability

Fatigue, SL conditions, 60,000 cycles

Fatigue, SL, 20,000 cycles

4: Elevated Temperature Capability

2: No-Damage Growth Design

1: Durable Patch Design

Phase

Residual Strength, 1.15SL (104°C and 121°C)

Cyclic, R = 0.1 Quasistatic Quasistatic Quasistatic Quasistatic Quasistatic

Residual Strength, 1.33 SL (104° and 121°C)

Quasistatic

81.6

56.3

39.6

8.9

Residual Strength, 1.5 limit load (121°C)

Quasistatic

105.4

73.0

39.6

9.3

Residual Strength, Failure (121°C)

Quasistatic

141.3

97.5

39.6

15.5

Residual Strength, 1.15 SL Residual Strength, 1.33 SL Residual Strength, 1.5 limit load Residual Strength, 2.8 SL

The test sequence for Phase 2 is summarized in Table 1 and consisted of fatigue tests at two elevated load levels corresponding to (1) the damage tolerance requirements of CFR 25.571 (70.6-kPa pressure) and (2) the pressurized compartment load requirements of CFR 25.365 (81.6-kPa pressure), “The airplane structure must be designed to be able to withstand the pressure differential loads corresponding to the maximum relief valve setting multiplied by a factor of 1.33 ...” For each of these loading conditions, 10,000 fatigue cycles were applied.

Fatigue and Residual Strength Performance of Bonded Repairs to Metallic Fuselage

a. Phase 2 Boron/Epoxy Alum Patch

739

Hoop Axial

C1A Scribes: 152.4 mm long by 0.3 mm deep Notch 1: 120.9 mm Long, 19.1 mm drill stop hole Notch 2: 76.2 mm long

b. Phase 3 and 4

LJC3A

Lap-Joint Notch

Scribe 3: 152.4 mm long by 0.3 mm deep Lap-Joint Notch: 660 mm long Notch 1: 120.9 mm Long, 19.1 mm drill stop hole Notch 2: 76.2 mm long

Fig. 2 Initial damage scenarios and repair designations for Phase 2 through 4.

Phase 3: Large-Damage Repair Capability The purpose of the Phase 3 test was to demonstrate the fatigue and residual strength performance of a large-damage repair design and installation. For this, the original B/Ep lap joint scribe repairs (S1BE and S2BE) were removed. A 660-mm-long through-the-thickness notch was inserted along the lap joint where the original scribes were located. The twobay notch was centered over FS-720D. An aluminum repair patch, LJC3A, was applied in the lap joint region. All other repairs from Phase 2 remained intact. Figure 2b shows the damage scenarios and the corresponding repairs for Phase 3. The test sequence for Phase 3, which is summarized in Table 1, consisted of a fatigue test using the simulated SL conditions for 20,000 cycles. Afterwards, the center frame and tearstrap were severed along the plane of the notch. Three quasistatic residual strength tests were conducted under room temperature conditions to load levels corresponding to (1) the damage tolerance requirements of CFR 25.571 (70.6-kPa pressure), (2) the pressurized compartment load requirements of CFR 25.365 (81.4-kPa pressure), and, (3) the strength and deformation requirements of CFR 25.305 (105.4-kPa pressure), “The structure must be able to support ultimate loads without failure for at least 3 seconds,” where ultimate load is equal to 1.5 times the limit load. A fourth residual strength test was conducted to the maximum pressure capacity (170.5 kPa) of the FASTER fixture.

740

J.G. Bakuckas Jr. and B. Westerman

Phase 4: Elevated Temperature Capability The purpose of the Phase 4 test was to demonstrate the load-carrying capacity of a large-damage repair design and installation under thermal-mechanical loading. For this, all repairs and initial damage were the same as in Phase 3. In Phase 4, elevated temperatures of 104º and 121ºC were applied to degrade the bond properties and strength in repair LJC3A to determine if there was a reduction in load-carrying capacity compared to the four residual strength tests in Phase 3. Heat blankets in combination with vacuum bags were used to apply the temperature profile consisting of linear rampup to, and then held at, 104º and 121ºC. Mechanical loads were applied once steady-state temperature was reached, as summarized in Table 1. Inspection and Measurement Procedures Several NDI methods were used to monitor and record damage formation and growth of cracks and disbonds. Crack growth was monitored continuously using high-magnification visual inspections, particularly for mid-bay patches C1BE and C2A. Detailed inspections were made every 5000 cycles: eddy current was used to inspect for crack growth, and flash thermography and a computer-aided tap tester were used to inspect for disbonds. The full-field strain and displacements of the patches and the area surrounding the structure were recorded using, noncontact, 3-D deformation measuring system, ARAMISTM. In addition, standard strain gages were used to monitor strains throughout the test. Baseline measurements were made at the beginning of the test and then at 5000-cycle intervals.

3 Analytical Procedure Finite element analysis was conducted to predict the strain distributions as detailed in references 2, 4, and 5. The initial design and sizing of the repairs were done using the U.S. Air Force/Boeing-developed tool, Composite Repair of Aircraft Structure (CRAS). The basic analysis methodology in CRAS was developed by Hart-Smith, Duong, Wang, and Yu [1, 6]. Their approach was based on the pioneering work of Rose [1, 7] incorporating several enhanced capabilities. The CRAS program was verified at coupon and limited subcomponent levels using a building-block approach for flat panel applications. In this study, the CRAS program was calibrated and enhanced to more complex fuselage repairs.

4 Results and Discussion Representative results are outlined in the subsequent sections for each phase. Detailed results are provided in reference 4. Phase 1: Durable Patch Design Initial results from Phase 1 were reported in reference 2. In summary, both B/Ep and aluminum bonded repairs were subjected to simulated SL conditions up to

Fatigue and Residual Strength Performance of Bonded Repairs to Metallic Fuselage

741

60,000 cycles (one DSG). Two damage scenarios were considered, namely, a midbay through-the-thickness crack and a lap joint scribe. There were no indications of damage development in the form of crack growth or disbonding for the lap joint scribe line repair patches, S1BE, S2BE, and S3A. In addition, no disbonding occurred in the mid-bay repair patches, C1BE and C2A; however, slow crack growth was measured, as shown in Figure 3. For the B/Ep repair, C1BE, the crack lengths measured visually and using eddy current inspections were in good agreement, Figure 3b. For the aluminum patch, C2A, the crack extension was measured using the internal eddy current only because the crack tip could not be reliably located for visual measurements, Figure 3c. Repair patch C2A was very effective in constraining the crack and keeping the two surfaces in contact. For both repair patches, the observed crack growth was symmetric and colinear, and the measured rate was nearly constant. Note that the crack growth rate measured in the thinner skin gage region under C1BE was approximately 5 times higher than that measured in the thicker skin gage region under C2A.

a. Designation of mid-bay patches and skin thickness Fatigue Loads: Pressure = 61.4 kPa Hoop = 42.3 kN Axial = 39.6 kN Frame = 6.7 kN

UP C1BE, 1.12 mm

c. Crack length under patch C2A

Aft (Visual) Fwd (Visual) Aft (Internal Eddy Current) Fwd (Internal Eddy Current)

60

55

50

45

AFT

6

Crack Extension, Δa (mm)

Half Crack Length,a (mm)

b. Crack length under patch C1BE 65

C2A, 1.83 mm

da = 26.7 × 10−5 mm / cycle dN

40

Aft (Internal Eddy Current) Fwd (Internal Eddy Current)

5 4 3

Δa

2

da = 5.1× 10−5 mm / cycle dN

1 0

0

10000

20000

30000

Cycles

40000

50000

60000

0

10000

20000

30000

40000

50000

60000

Cycles

Fig. 3 Results from Phase 1: (a) location of patches and skin thickness, (b) crack growth under patch C1B, and (c) crack growth under patch C2A.

The mid-bay repair patches were quite effective in reducing the crack growth rate, as shown in Figure 4. The measurements made during fatigue precracking (before the patches were installed) and during fatigue at SL conditions (after the patches were installed) were compared as shown in Figure 4b. In both repairs, the crack growth rates were reduced substantially. It should be noted that the

742

J.G. Bakuckas Jr. and B. Westerman

precracking loads were 75% of the SL used during fatigue (Figure 4a). In addition, the measured rates in the B/Ep patch, C1BE, were higher than the target rate provided by the original CRAS analysis. After further evaluations, enhancements were made to the CRAS program to obtain better correlation with the experiments by accounting for curvature and applied pressure and by using measured thermal strains during the cure process. Results of the test and analysis correlation are shown in Figure 4c. a. Applied loads Pre-Cracking (before Patch Installation) 46.0 kPa

Fatigue (after Patch Installation) 61.4 kPa

Hoop

32.7 kN

42.3 kN

Axial

29.7 kN

39.6 kN

Frame

5.0 kN

6.7 kN

Load Component Pressure

b. Measured rates before and after patch installation Pre-Cracking Fatigue

0.0010 0.0008 0.0006 0.0004

Design Target Rate

C1BE

0.0002

C2A

Half Crack Length, a (mm)

da/dN (mm/cycle)

0.0012

c. Crack length test and analysis correlation 65 Visual Internal Eddy Current External Eddy Current CRAS Analysis

60

55

50

45

40

0.0000 Notch 1

Notch 2

Strain gage

0

10000

20000

30000

40000

50000

60000

Cycles

Fig. 4 Results from Phase 1: (a) applied loads, (b) effect of patches on crack growth rates, and (c) correlation of test and analysis.

Test and analysis results revealed that the installation of the mid-bay repairs resulted in eccentric loading and inward deformation of the mid-bay region, with high tensile strains on the skin surface and compressive strain on the outer patch surface. Representative results are shown in Figure 5 for mid-bay patch, C1BE, measured from several strain gages. Gage S9 is located in the center of the patch on the outer surface and is under compression. Gages S8 and S10 are located on the outer surface of the skin 19 mm from the patch boundary. Gages IS26 and IS27 were at the same location on the inner skin surface. Comparing the back-toback gages, there is a large amount of bending along the patch boundary with the outer surface in tension and inner surface in compression, as shown in Figure 5b. Good agreement was obtained between the finite element analysis and measured strains. The bending strains remained relatively constant throughout the fatigue test, as shown in Figure 5c.

Fatigue and Residual Strength Performance of Bonded Repairs to Metallic Fuselage

743

The full-field hoop strain in the vicinity of C1BE, which was measured using the ARAMIS system, is shown in Figure 6. In the figure, the patch boundary and initial defect (Crack 1) are indicated. The hashed regions are areas where data could not be processed because of interference from strain gage wires. Details of the hoop strain variation in the vicinity of mid-bay patch are shown along three vertical sections. In Figure 6, the distance between sections is 38-mm. As shown, the hoop strain varied most toward the middle of the patch (section 1, located 19 mm from the patch vertical centerline) where the values ranged from approximately 1200 με in the skin outside the patch to approximately -1000 με in the center of the patch. The strain gradient was quite high in a narrow band of approximately ±12.7 mm around the crack (section 1); over this short distance the strain ranged from 0 to -1000 με. Strain also changed rapidly at the patch boundary crossing from the skin into the patch (sections 1-2) at ±38 mm about the centerline. The severity of the strain gradient did reduce substantially in sections outside the patch on the skin (section 3).

a. Strain gage location, C1BE Section AA 19 mm S8 IS26

A

76.2 mm S9 Mid-Bay Notch

IS27

S8 S9 S10 IS26 IS27 Analysis

3000

Hoop Strain (με)

Hoop Strain (μm)

1000 500 0 -500

S9 S10

A

c. Hoop strains during fatigue

b. Measured strains indicate bending 1500

S8

19 mm S10

2000

10000 20000 40000 50000

Strain Survey Loads: Pressure = 46.0 kPa Hoop = 32.7 kN Axial = 29.7 kN Frame = 5.0 kN

1000

0

-1000

-1000 -1500 0

10

20

30

Pressure (kPa)

40

50

S8

S9

S10

Strain Gage Number

Fig. 5 Results from Phase 1: (a) location of strain gage, (b) strains measured along patch boundary and centerline, (c) strains measured during fatigue.

After the fatigue test, the panel was subjected to quasi-static loading to 70.6kPa pressure, a level that simulates the damage tolerance requirements defined in 14 CFR 25.571. The load levels are listed in Table 1 for the residual strength test. All five repair patches were effective in preventing failure of the damaged panel. No disbond or crack growth was observed during inspections made after the residual strength test. In addition, there was no evidence of load redistribution;

2000

Strain Survey Loads: Pressure = 46.0 kPa Hoop = 32.7 kN Axial = 29.7 kN Frame = 5.0 kN

Patch Boundary

S8

1500

50

CL

Crack 1

μm/m

100

J.G. Bakuckas Jr. and B. Westerman

2 0

1000 500 S9

3

1 -50

0 -500

Y-Coordinate (mm)

744

S10 -1000

Strain (με)

-100

-1000

-500

0

25.4 mm

500

1 1000

2

Section

1500

3

Fig. 6 Hoop strain variation in the vicinity of patch C1BE.

strain survey results after residual strength test were similar to the baseline strain survey. Phase 2: No-Damage Growth Repair Design In Phase 2, a no-damage growth repair design was demonstrated. The mid-bay patch, C1BE, was removed and replaced with a no-damage growth repair, C1A. The crack tips were stop-drilled and repaired using an aluminum patch. The panel was subjected to an elevated load of 115% of the SL for 10,000 cycles and then to an elevated load of 133% of the SL for an additional 10,000 cycles. In general, crack growth and disbonding was monitored during the fatigue test for all five patches shown in Figure 2a. There were no global changes in the strain field measured using the ARAMIS system, which would indicate load redistribution due to disbonding or crack growth. The strains measured at gages also remained unchanged. Representative results are shown in Figure 7 for the nogrowth mid-bay patch C1A. The locations of the gages are shown in Figure 7a. Strains measured in gages on the outer and inner skin surfaces are provided in Figure 7b and 7c, respectively. As indicated, the values of strains remained relatively constant and the strain field was symmetric about the repair. The highest tensile strains were measured in the inner skin surface at gages located approximately 6.35 mm from the notch, IS32 and IN2. The magnitude of these notch region strains was slightly higher than strains measured in the outer skin surface at the gages along the patch boundary, S8 and S10. Strain levels were low enough not to cause crack initiation in the notch region and patch boundaries. High-frequency, eddy-current inspections confirmed that no cracks developed in the notch region during Phase 2.

Fatigue and Residual Strength Performance of Bonded Repairs to Metallic Fuselage

745

a. Strain Gage Locations, C1A Section AA 19 mm S8 IS26

A

76.2 mm S9 IS24

S8

19 mm S10

19 mm IS32

S9 S10

IS25 IS27 Crack 1

IS32

A

6.35 mm

b. Strains on outer surface

c. Strains on inner surface 2000

2000

Hoop Strain (με)

Hoop Strain (με)

1000

0

-1000

-2000

Crack 1

Survey Loads: Pressure = 61.4 kPa Hoop = 42.3 kN Long = 39.6 kN Frame = 6.7 kN

After 10K Cycles at 115% After 5K Cycles at 133% After 10K Cycles at 133%

1500

1000

500

0 -3000 S8

S9

S10

Strain Gage

IS26

IS24

IS25

IS27

INS2

IS32

Strain Gage

Fig. 7 Results from Phase 2, C1A: (a) gage locations, (b) strains measured on the outer surface and, (c) strains measured on the inner surface.

Phase 3: Large-Damage Repair Capability Large-damage repair design and installation capability was demonstrated in Phase 3. Lap joint repair patches S1BE and S2BE were removed. A 660-mm, two-bay lap joint through-the-thickness notch was inserted and repaired using an aluminum patch, LJC3A. An additional 20,000 cycles were then applied using simulated SL conditions. During fatigue, crack growth and disbonding were monitored at all four patches. No indications of damage growth occurred from measured strains. Representative results are summarized in Figure 8, which shows hoop strains measured on the outer surface of repair LJC3A using strain gages. As shown, all strains remained relatively constant during the fatigue, indicating no load redistribution, which would signify damage formation. The measured maximum tensile strains were in the skin along the patch boundary at gages S29 and S30. Figure 9 shows measured hoop strains on the inner skin surface in the notch-tip vicinity during the fatigue test in repair LJC3A. As shown, there is a strain variation in the notch-tip region, with strains highest near the notch, as measured using an array of gages with a 25.4 mm pitch. Gages P6-1 through 3 were located in the thin gage section (t = 1.17 mm) and gages P6-10 through 12 were located in the thicker gage section (t = 1.52 mm). Thus, the strains were not symmetric about the notch centerline. Measured strains remained relatively constant during fatigue. The patch was very effective in reducing the notch-tip strain field to nondamaging levels. High-frequency, eddy-current inspections confirmed that no crack growth occurred in the notch-tip region. The magnitude of strain in the notch regions

746

J.G. Bakuckas Jr. and B. Westerman 813 mm 165 mm

S3

S31

S29 12.7 mm

12.7 mm

S4 S24

S25

S26

S30

1500

Strain (με)

1000

S27

S28

S32

S5

5K 10K 15K 20K

720C

720E

720D

500

0

Survey Loads: Pressure = 61.4 kPa Hoop = 42.3 kN Long = 39.6 kN Frame = 6.7 kN

-500

-1000

-1500 S24

S25

S26

S27

S28

S29

S30

S31

S32

Strain gage Fig. 8 Hoop strains measured in region of patch LJC3A during fatigue. 720D

720C

720E

S3

12.7 mm

S4

500

25.4 mm

12 11

3

10K Cycles 15K Cycles

S5

2 P6-1

P6-10

Strain (με)

400

Survey Loads: Pressure = 61.4 kPa Hoop = 42.3 kN Long = 39.6 kN Frame = 6.7 kN

300

200

100

0 P6-1

P6-2

P6-3

P6-10

P6-11

P6-12

Strain Gage Fig. 9 Hoop strains measured in notch-tip region of LJC3A during fatigue.

Fatigue and Residual Strength Performance of Bonded Repairs to Metallic Fuselage

747

(e.g, P6-3 in Figure 9) was much lower than the strains in the skin along the patch boundary (e.g, S29 in Figure 8). Figure 10a summarizes the number of cycles each repair patch was subjected to after the Phase 3 fatigue test and demonstrates the durability of the patches installed. The large repair, LJC3A, was installed in Phase 3 and was subjected to 20,000 cycles. There were no indications of crack growth from the notch-tips or disbonds from NDI and strain measurements. The no-growth repair, C1A, was installed in Phase 2 and was subjected to 45,000 cycles with no indications of disbonds or crack formation from the drill-stop holes. More significant is the fatigue performance of the original scribe-line and mid-bay repairs, S3A and C2A, which were subjected to 110,000 cycles. For the scribe-line repair, there were no indications of damage formation. For the other mid-bay repair, C2A, there was no measurable disbonding; however, limited crack extension occurred. Figure 10b

a. Cycle count for repairs after Phase 3 Phase 3 S3A, 110,000 cycles

LJC3A, 20,000 cycles

C1A, 45,000 cycles

C2A, 110,000 cycles

Hoop Axial

b. Crack Growth measured under C2A

Crack Extension, (Δa) (mm)

12

Aft (Internal Eddy Current) Fwd (Internal Eddy Current)

10

100% SL Conditions: Pressure = 61.4 kPa Hoop = 42.3 kN Axial = 39.6 kN Frame = 6.7 kN

8

100% 133% 115%

100%

6

4

Phase 4

2

Phase 1 Phase 2

Phase 3

0 0

20x103

40x103

60x103

80x103

100x103

120x103

Cycles Fig. 10 (a) Fatigue cycle count summary for each repair after Phase 3 fatigue test and (b) Crack extension measured in skin under patch C2A for all test phases.

748

J.G. Bakuckas Jr. and B. Westerman

shows the crack growth measured in the mid-bay crack under C2A for all test phases. As shown, the crack growth rate increased during Phase 2 compared to the results from Phase 1 due to the elevated applied load levels. The rate then reduced substantially during Phase 3 when subjected to SL conditions. The total amount of crack extension from the notches was under 12 mm with a average rate of less than 6.3 x 10-5 mm/cycle. After the fatigue test, the central tear strap and frame were severed at FS-720D along the plane of the 660-mm center notch under LJC3A. A quasi-static residual strength test was then conducted to load levels corresponding to (1) damage tolerance requirements of 14 CFR 25.571 (1.15 SL), (2) the pressurized compartment load requirements of 14 CFR 25.365 (1.33 SL), and (3) the strength and deformation requirements of 14 CFR 25.305 (1.5 limit load). A fourth residual strength test was conducted to the maximum pressure capacity of the fixture (170.5 kPa). The load levels applied during the residual strength test are summarized in Table 1. In general, there was no measurable disbonding or notchtip crack extension in patch LJC3A or in the other three patches (C1A, C2A, and S3A). Phase 4: Elevated Temperature Capability The purpose of Phase 4 was to demonstrate the load-carrying capacity of a largedamage repair design and installation under thermal-mechanical loading. For this, all repairs and initial damage were the same as in Phase 3. In Phase 4, elevated temperatures of 104° and 121°C were applied to reduce the adhesive bond properties and strength in repair LJC3A and determine if there was a reduction in load-carrying capacity compared to the residual strength tests in Phase 3. The effect of temperature on the local notch-tip strain in the skin under LJC3A is shown in Figure 11. The measured strain at gage P6-3 located the thin gage section (t = 1.17 mm) and gage P6-10 located in the thicker gage section (t = 1.52 mm) are shown in Figure 11b and 11c, respectively. As shown, the value of strain at the notch tip increased with an increase in temperature indicating less load transfer into the patch. Results from the residual strength test at an elevated temperature of 121°C are shown in Figure 12. As shown, gage P6-2 is near the notch-tip of the thin gage section, gage P6-10 is near the notch-tip of the thicker gage section, and gages S31 and 32 on the inner skin in the center of LJC3A along the boundary. The strains measured from these gages indicated a load transfer, most likely due to bond failure, as shown in the figure. As the strains measured in the skin in the middle of the patch reduced (S31 and 32), strain at the notch-tip gages increased substantially.

Fatigue and Residual Strength Performance of Bonded Repairs to Metallic Fuselage

749

a. Strain gage locations 720D

720C

720E

S3

t = 1.52 mm

t = 1.17 mm S4

12.7 mm

12.7 mm

P6-3

P6-10

S5

b. Measured strain, gage P6-3, t = 1.17 mm 1600 1400

1400

121°C

1200 1000 800 600 400

Room Temperature

200

Hoop Strain (με)

Hoop Strain (με)

c. Measured strain, gage P6-10, t = 1.52 mm

1600

121°C

1200 1000

104°C

800 600 400

Room Temperature

200 0

0 0

20

40

60

80

0

100

20

40

60

80

100

Pressure (kPa)

Pressure (kPa)

Fig. 11 Effect of temperature on measured strains: (a) location of gages, (b) notch-tip strain in thin gage section, and c. notch-tip strain in thick gage section. 720D

720C

720E

S3 12.7 mm

S4

6000

Hoop Strain (με)

12.7 mm

38.1 mm

P6-2

5000

S31

12.7 mm

S32

P6-10

S5

4000

Load transfer

3000 P6-2 P6-10 S 31 S 32

2000 1000 0 0

100

200

300

400

500

600

700

Time (seconds) Fig. 12 Strains during residual strength test at 121°F illustrating load transfer due to bond failure.

750

J.G. Bakuckas Jr. and B. Westerman

The panel failed catastrophically as shown in Figure 13a, at an applied pressure of 141.4 kPa, well above ultimate strength requirements. Bond failure occurred in the patch overlap region between the first doubler layer, S1, and the second filler layer, F2, as shown in Figure 13b. In addition, crack turning occurred from the notch-tip resulting in final skin failure path along the outer rivet row of the lap joint in S-4L. The majority of the disbonded surfaces of S1 and F2 layers were coated with residual adhesive, which indicated cohesion failure, as shown in Figure 13c. There were also some small regions where the surfaces were clean (no adhesive) which indicated interfacial or adhesion failure of the bond. a. Catastrophic Panel Failure

b. Failure Path Crack Path

S1

F2

Crack Path

Original 660 mm Thru-Notch Disbond Overlap Notch-Tip

Crack Path Along Rivet Row

Crack Turning

Crack Path Along Rivet Row

c. Matching Surfaces of Disbonded Overlap

Doubler Layer Surface, S1

Filler Layer Surface, F2

Fig. 13 Results from Phase 4: (a) panel failure, (b) failure path, and (c) disbond surfaces of overlap.

5 Summary A four-phase approach was used to study the fatigue and residual strength performance of boron/epoxy (B/Ep) and aluminum bonded repairs to metallic fuselage under a variety of loading conditions and initial damage scenarios. Several nondestructive inspection (NDI) methods were used to monitor and record

Fatigue and Residual Strength Performance of Bonded Repairs to Metallic Fuselage

751

damage formation including crack growth and disbonding. Full-field strain and displacements were measured using photogrammetry to ascertain the load transfer in the vicinity of the repair patches. In Phase 1, the durability of B/Ep and aluminum bonded patches was demonstrated in a fatigue test using simulated SL conditions up to 60,000 cycles. There were no indications of damage development in the form of crack growth or disbonding for the lap joint scribe repairs. For the mid-bay repairs, no disbonding occurred; however, slow crack growth was measured and correlated with analysis. Subsequent residual strength loads representative of damage tolerance requirements in 14 CFR 25.571 revealed that all repairs were effective in containing the damage. A no-damage growth repair design was verified in Phase 2 using an aluminum patch with drill-stopped holes to remove fatigue crack-tips. The panel was subjected to elevated loads of 115% of the SL for 10,000 cycles and then 133% of the SL for an additional 10,000 cycles. During both tests, there were no indications of damage growth under the new repair using all methods of inspection. For Phase 3, a large-damage repair design and installation capability was demonstrated. A long 660-mm, two-bay lap joint through-the-thickness notch was inserted and repaired using an aluminum patch. An additional 20,000 cycles were then applied using the simulated SL conditions with no indications of damage growth. After the fatigue test, the central tear strap and frame were severed. The panel was subjected to residual strength tests to applied pressures of 70.6 kPa (115% SL), 81.6 kPa (133% SL), 105.4 kPa (1.5 limit load), and 170.5 kPa (280% SL, maximum capacity of fixture). Inspections verified that the repairs were intact with no measurable damage growth. The load-carrying capacity of a large-damage repair design under thermalmechanical loading was evaluated in Phase 4. For this, all repairs and initial damage were the same as in Phase 3. Elevated temperatures of 104° and 121°C were applied to reduce the adhesive bond properties and strength in the large repair and determine if there was a reduction in load carrying capacity compared to the residual strength tests in Phase 3. The applied temperature degraded the adhesive properties causing a reduction in load transfer capabilities of the largedamage repair patch. Catastrophic failure occurred at 141.3 kPa at an applied temperature of 121°C which was 17% lower than the maximum applied pressure of 170.5 kPa obtained during the room temperature residual strength test. Results from this study revealed bonded repairs that are properly designed and installed are durable under fatigue and can effectively contain large damage under severe static loads in excess of the ultimate load requirements. Data from this effort will be used to further calibrate and verify predictive models.

Acknowledgements This paper summarizes a team effort consisting of representatives from several organizations, including the FAA, the Boeing Company, and Drexel University. From the FAA, the authors wish to acknowledge Reewanshu Chadha, Yongzhe Tian, and Jeff Panco for their diligent efforts running the tests using the FASTER

752

J.G. Bakuckas Jr. and B. Westerman

facility; Dave Galella and Paul Swindell for providing NDI support; Curtis Davis for providing ARAMIS support; and Ian Won for his co-sponsorship. From Boeing, thanks are due to Ron Turner, Russell Keller, Keith McIver, Carly Schlottman and Kelly Greene for managing the activities of Boeing; to Ching Hsu, Cong Duong, and Trica Carr for their technical contributions in the design and analysis of the repairs; to Mike Evans and Joel Baldwin for installing the repairs; and to Bill Tapia, Morteza Safai, and Kimberly Meredith for their efforts in providing NDI support. Finally, the authors would like to recognize Professors Jonathan Awerbuch and T.M. Tan from the Department of Mechanical Engineering and Mechanics of Drexel University for their input in all aspects of the work reported herein.

References [1] Baker, A.A., Rose, L.R.F., Jones, R. (eds.): Advances in the Bonded Composite Repair of Metallic Aircraft Structure, vol. 1& 2. Elsevier, Amsterdam (2002) [2] Bakuckas, J.G., McIver, K., Hsu, C.: Durability and Damage Tolerance of Bonded Repairs to Metallic Fuselage Structure. In: Proceedings of the 25th ICAF Symposium, Rotterdam, May 27-29 (2009) [3] Chadha, R., Bakuckas, J.G., Won, I., Westerman, E.A., Keller, R., McIver, K., Hsu, C., Awerbuch, J., Tan, T.: Characterization of Adhesive-Bonded Repairs to Fuselage Structure. In: Proceedings of the 2010 Aircraft Airworthiness & Sustainment Conference, Austin, Texas, May 10-13 (2010) [4] Chadha, R., Bakuckas, J.G.: Adhesively Bonded Repairs to Metallic Fuselage Structure: Test 1, Fatigue and Residual Strength Performance. Final Report, DOT/FAA/AR-11/4 (to be published) [5] Duong, C.: Bonded Repair of a Commercial Airframe Curved Fuselage Panel: Design/Analysis, Installation and Validation Test. In: Proceedings of the 12th Joint NASA/FAA/DoD Conference on Aging Aircraft, Kansas City, Missouri, May 4-7 (2009) [6] Duong, C., Wang, J.J., Yu, J.: An Approximate Algorithm Solution for the Elastic Fields in Bonded Patched Sheets. Int. J. Solids and Structures 38, 4685–4699 (2001) [7] Rose, L.R.F.: An Application of the Inclusion Analogy for Bonded Reinforcements. Int. J. Solids and Structures 17, 827–838 (1981)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Life Extension Techniques for Aircraft Structures – Extending Durability and Promoting Damage Tolerance through Bonded Crack Retarders P.E. Irving1, X. Zhang1, J. Doucet1, D. Figueroa-Gordon1, M. Boscolo1 M. Heinimann2, G. Shepherd3, M.E. Fitzpatrick4, and D. Liljedahl4 1

Cranfield University, Cranfield, Bedford, UK 2 Alcoa Technical Center, Alcoa Center, PA, USA 3 Airbus Operations Limited, Filton, Bristol BS34 7AR, UK 4 Materials Engineering, The Open University, Milton Keynes, UK

Abstract. This paper explores the viability of the bonded crack retarder concept as a device for life extension of damage tolerant aircraft structures. Fatigue crack growth behaviour in metallic substrates with bonded straps has been determined. SENT and M(T) test coupons and large scale skin-stringer panels were tested at constant and variable amplitude loads. The strap materials were glass fibre polymer composites, GLARE, AA7085 and Ti-6Al-4V. Comprehensive measurements were made of residual stress fields in coupons and panels. A finite element model to predict retardation effects was developed. Compared to the test result, predicted crack growth life had an error range of -29% to 61%. Mechanisms and failure modes in the bonded strap reinforced structures have been identified. The strap locally reduces substrate stresses and bridges the crack faces, inhibiting crack opening and reducing crack growth rates. In the absence of residual stress, global stiffness ratio accounts for effects of both strap modulus and strap cross section area. In elevated temperature cure adhesives, retardation performance was best in aluminium and GLARE strap materials, which have the closest thermal expansion coefficient to the substrate. Strap materials of high stiffness and dissimilar thermal expansion coefficient such as titanium had poor retardation characteristics.

1 Introduction Requirements for green aircraft can be interpreted as lighter aircraft, and/or aircraft with extended life and less maintenance. Extended lives can be achieved on safe life designs by increasing material fatigue strength and reducing stress concentration factors. Requirements for damage tolerance in extended-life aircraft imply real increases in fatigue crack growth resistance, fracture toughness and *

Oral presentation.

754

P.E. Irving et al.

residual strength. Extensive research over the past 30 years suggests that little improvement in crack growth resistance in monolithic aluminium alloys can be achieved through metallurgical innovation alone [1]. Indeed, if weight reduction is attempted via use of increased strength materials at augmented stress levels, fatigue crack growth rates increase rather than decrease, driven by the increased stress ranges. In turn damage tolerance of the structure will be reduced. Alternative approaches to increased damage tolerant life include use of fibremetal laminates, such as GLARE, or the use of bonded features such as bonded stringers, bonded repairs, and the recently developed bonded crack retarder concepts [2-5]. These techniques have the capability to produce step changes in resistance to fatigue crack growth in metallic structural components. The potential to improve crack growth resistance has been demonstrated in a number of previous papers [2-5]. However, many details of application of bonded retarders and accurate modelling of their effects remain to be established. Previous work shows that cracks propagating in substrates containing bonded straps have growth rates which are not exclusively determined by the stress intensity calculated from substrate geometry and stress. Calculation of growth rates requires data characterising the strap-substrate stiffness ratio and adhesive strength and toughness as well as the position of the crack tip in relation to the strap location. In this research coupon samples of 7085 aluminium alloy containing bonded straps of a wide range of geometry and modulus have been manufactured and tested to characterise fatigue crack growth resistance. The local stress state around the strap and crack has been modelled using finite element analysis, and the residual stress field induced by thermal expansion mismatch between the strap and substrate has been measured using neutron diffraction. Fracture mechanics analysis was performed using finite elements to calculate the effect of the strap on stress intensity factors at a range of crack tip locations. Output from the numerical models was used to interpret the experimental test results and to extend them to predict optimum strap properties and geometry for life extension at minimum weight and cost. The performance of bonded straps was further evaluated by testing 5-stringer integrally stiffened panels with and without bonded straps under a realistic flight load spectrum. Panels with bonded straps demonstrated significant crack growth retardation resulting in increased panel lives by a factor of 2 at stress levels increased by 20% (to account for the additional cross sectional area of the straps). These results permit selection of strap materials and geometry to optimise fatigue crack growth resistance in aluminium alloys. The implications for life extension and damage tolerance capability in aircraft will be discussed in the paper.

2 Coupon Specimens The substrate metal for all the testing work was aluminium alloy 7085-T7651 supplied as 10 mm plate. The specified chemical composition (wt%) is Zn 7.0-8.0, Mg 1.2-1.8, Cu 1.3-2.0, Fe < 0.08, Si < 0.06, Zr 0.08-0.15. Measured mechanical properties from [6] are listed in Table 1.

Life Extension Techniques for Aircraft Structures

755

Table 1 Measured tensile mechanical properties of supplied 7085 plate [6].

Sample orientation Longitudinal Transverse

0.2% proof / MPa 476 462

UTS / MPa 510 503

Elongation / % 7.3 5.0

Coupon tests were performed primarily using single edge notched tension (SENT) samples 140 mm wide, 400 mm long. Straps were bonded to one side of the specimens at the location shown in Fig. 1(a). The initial notch was 17 mm long. Sample length between the grip ends was 270 mm. Four strap materials were tested; these were fibre-metal laminate GLARE-1(3/2), titanium alloy Ti-6Al-4V, glass fibre reinforced polymer (GFRP) Hexcel 913G-E-5-30%, and the substrate material AA 7085. Strap nominal dimensions were 200 x 20 x 2 mm (1.8 mm for the GLARE strap). For selected strap materials a range of other geometries were tested. The strap edge was located 20 mm from the notch tip. Mechanical properties of the strap materials are given in Table 2. For the SENT geometry all tests were performed under constant amplitude loading. Tests were also conducted using the middle cracked tension M(T) geometry. These tests were used to establish the effects of higher substrate stresses and also variable amplitude loading. The M(T) samples were the same overall dimensions as the SENT but with a centre notch and with the strap edge located 13.5 mm from the centre line of the specimen as shown in Fig. 1(b). In addition to tests on specimens with a bonded strap, samples of 7085 were tested to determine the substrate fatigue crack growth characteristics at R ratios of 0.1, 0.3 and 0.6. Aluminium substrates were surface treated following ASTM D 2651-90 prior to applying the primer BR 127 to the side to be reinforced. Retarder straps were cut to size and bonded to the samples using either FM94, a high temperature curing film adhesive or Redux 810, a room temperature curing adhesive. FM 94 was cured in an autoclave following Cytec recommended procedures at 120°C. The residual stress field induced by thermal expansion mismatch between the strap and substrate was measured using the neutron diffraction technique [7]. The REDUX 810 was cured at room temperature for 5 days. Table 2 Adhesive and strap material properties.

Material

FM 94

GLARE-1 (3/2)

GFRP

Ti-6Al-4V

E11 (GPa) E22 (GPa) G12 (GPa)

1.9 1.9 0.62 0.52 / / 1.1

65 48 16

40 10 5 0.28 3.6 21 2

114 114 44 0.34 8.6 8.6 4.5

υ12 α11 (με °C-1) α22 (με °C-1) ρ (g cm-3)

16.3 25.5 2.49

756

P.E. Irving et al.

(a) SENT specimen

(b) M(T) specimen

Fig. 1 Coupon test samples and dimensions (unit: mm).

3 Coupon Test Results Effects of strap/substrate stiffness ratio on fatigue crack growth rates Results reported here refer to the SENT samples bonded with the complete range of strap materials (GLARE, GFRP, Ti-6-4 and AA7085). These straps had a wide range of stiffness values, from 35 GPa for GFRP to 114 GPa for the titanium alloy (Table 2). In terms of material type there were metal (aluminium and titanium), polymer composite (GFRP) and fibre-metal laminate. The other parameter in the expression for the global stiffness ratio (μ), eq. 1, is the cross section area of the strap, which depends on the thickness and width of the strap. In a large number of the tests the strap width was kept constant and the thickness changed to vary the strap cross section area.

μ=

∑ (E

strap

Astrap )

Esubstrate Asubstrate + ∑ ( E strap Astrap )

(1)

Figure 2 shows the effect on growth rate of changing global stiffness ratio μ from 0.02 to 0.16, a factor of 8 change, in comparison with fatigue crack growth rates in the unstrapped sample at the same nominal Δσ value of 18.6 MPa. In these tests the straps were bonded using Redux 810 at room temperature; hence there was no residual stress present.

Life Extension Techniques for Aircraft Structures

757

In comparison with the unstrapped substrate, all of the samples with straps show a reduction in growth rate. The reduction is least for the GFRP strap (μ=0.02) and greatest for the titanium (μ = 0.16). It can also be noted that the retardation effects begin when the crack tip is at least 5 mm from the strap edge, immediately at the start of the test. Hence constraint of crack opening by the strap plays no role in retardation at this point, but can only be a consequence of local substrate stress reduction by the presence of the strap. The extent of retardation gradually reduces as the crack propagates under the strap. Retardation is still present when the crack tip is beyond the strap, but the difference in growth rates between the various strap materials is slowly reducing, becoming zero when the crack tip is 20 mm beyond the strap edge.

Fig. 2 Effect of global stiffness ratio (μ) on fatigue crack growth rates as the crack approaches and tunnels under the strap; (20oC, Δσ = 18.6 MPa).

Residual stresses in substrate due to elevated temperature curing process Results reported here refer to the SENT samples bonded with the complete range of strap materials. Longitudinal residual stress profiles for the strap reinforced SENT specimens are shown in Fig. 3, [8]. The measuring position was 2.5 mm from the strap/substrate bond interface. For the GFRP and GLARE straps the residual stresses are much lower than that of the Ti-6Al-4V alloy. To summarise, measured maximum longitudinal stresses in the substrate are listed in Table 3.

758

P.E. Irving et al.

Fig. 3 Measured longitudinal residual stresses (average values) in a 10 mm thick SENT sample bonded with 20 mm wide, 200 mm long straps . Measured residual stresses are 2.5 mm from the bond interface. Table 3 Measured longitudinal residual stresses for the different strap materials.

Strap material

Ti-6Al-4V GFRP GLARE-1

Strap dimensions length x width x thickness (mm) 200 x 20 x 2 200 x 20 x 2 200 x 20 x 1.8

Maximum residual stress in substrate (MPa) 28 ~ 30 10 ~ 12 5~8

Effects of different strap materials at elevated temperature cure (at same μ) The elevated temperature curing process used for FM94 produces tensile residual stresses in the aluminium substrate as the Coefficient of Thermal Expansion (CTE) of aluminium is greater than that of any of the strap materials, and if there is zero residual stress at the cure temperature of 120°C, when the bond is fully cured, residual stresses will develop under and around the strap as it cools down to room temperature. The magnitude of the residual stress will depend on the mismatch between CTE of substrate and strap, and also on the stiffness mismatch of the substrate and strap and the thickness mismatch. For a strap-substrate system in which the strap covers all of the substrate area the substrate stress arising from thermal mismatch is:

σ res =

ΔT (α s − α r ) Es Er t r E s t s + Er t r

(2)

Life Extension Techniques for Aircraft Structures

759

where, ΔT is the temperature change, αs and αr are CTE, Es and Er are elastic modulus, ts and tr are thicknesses of substrate and reinforcing strap, respectively. In the absence of residual stresses, growth rates in samples with different strap materials but the same global stiffness ratio will be identical. Under elevated temperature curing, residual stresses will exist and differences in growth rate will be related to different residual stresses induced by different strap materials. The smallest residual stresses will have the best retardation effect. An example of this behaviour is shown in Fig 4 where straps of aluminium 7085, GLARE, titanium and unidirectional GFRP are compared. The optimum retardation behaviour is exhibited by the 7085 strap with zero residual stress, next is GLARE with a residual stress field of 5-8 MPa, GFRP and Ti-6-4 straps have much higher growth rates and have greater residual stresses than the field developed by GLARE. Similar growth rates in GFRP and Ti-6-4 reinforced coupons may be due to the fact that the titanium and the GFRP straps did not have identical stiffness ratios.

Fig. 4 Comparison of growth rates in samples with different strap materials having approximately the same value of μ = 0.07 and elevated temperature cure, showing influence of thermal residual stress.

Substrate stress levels of 60 MPa A number of tests were performed on M(T) samples at substrate stresses of 60 MPa to establish the influence of stress level on the crack growth rate behaviour observed. An example of an equivalent comparison for aluminium, titanium and

760

P.E. Irving et al.

GLARE straps is shown in Fig 5, where growth rates are plotted against nominal (assuming no strap present) ΔK instead of crack length, in order to make the comparison under high stress and different sample geometry & crack length with substrate material. It can be seen that the behaviour is rather different to that seen in the SENT samples. Firstly the difference in crack growth rates is least at the start of the tests and steadily increased as the crack tips moved under the straps. This is the reverse of the behaviour observed in the SENT samples at low stresses. This difference may be related to differences in the residual stress field induced by the thermal cycles in the two geometries of sample. Secondly there is little difference in the growth rates in the three strap materials, although all of them show significant retardation compared with growth rates in the base material. Nominal ΔK values are between 10 and 30 MPa√m, instead of 8-12 MPa√m for the stresses of 18.6 MPa used in the SENT samples.

Fig. 5 Log da/dN vs. nominal ΔK for un-strapped and strapped (20 mm width) M(T) samples tested at substrate stress σmax = 60 MPa, R =0.1; μ = 0.07 for titanium, GLARE and aluminium straps; adhesive FM 94 cured 120°C.

The reduced effect of residual stress at larger values of ΔK is believed to arise from two sources. Firstly there is reduced effect of R ratio changes on fatigue crack growth rates in the substrate as ΔK values are increased [9]. Secondly, changes in the effective R ratio produced by a residual stress are expressed by the stress intensity factors resulting from the applied and residual stresses as in eq. (3). As the applied stresses σmin and σmax are increased, the corresponding stress intensity factors, Kmin and Kmax, also increase and the effect of Kres on Reff becomes proportionately less.

Life Extension Techniques for Aircraft Structures

Reff =

K min + K res K max + K res

761

(3)

Variable amplitude load testing A variable amplitude load spectrum representative of a transport aircraft upper wing cover was used to test the relative performance of different straps bonded to M(T) samples (all with μ = 0.07). The spectrum had a maximum substrate tensile stress of 49 MPa. The results are shown in Fig. 6, plotted as crack length vs. cycles where cycles are individual spectrum cycles. The data show that under the upper wing cover spectrum, the strap producing the longest life is GLARE, with a life more than double that of the unstrapped substrate. Next best is aluminium, followed by titanium and by GFRP. Under variable amplitude loading the differences in growth rates between the straps, possibly due to residual stress differences, are increased compared with those observed at 60 MPa constant amplitude loading. This may be because of the reduced ΔK values of the small cycles in the spectrum, which will occupy a range of the ΔK vs. da/dN curve with greater mean stress sensitivity.

Fig. 6 Strap comparison: μ = 0.07, 120oC cure, spectrum loads σmax = 49 MPa.

Performance of strap materials – summary of effects of substrate stress & spectrum A comparison of the performance of the different strap materials expressed as the ratio of strapped life to unstrapped life is shown in Fig. 7. This shows the life

762

P.E. Irving et al.

improvement factor for substrate stresses of 18.5 MPa and 60 MPa of constant amplitude (CA) loading compared with variable amplitude (VA) loading with maximum tensile stress of 49 MPa, all samples with μ = 0.07. Strap bonding was with FM 94 and a 120°C cure; hence residual stresses are present. The greatest difference between the straps was found at CA loading of 18.5 MPa, and least under the variable amplitude loading. GLARE straps were consistently the best, except for the small constant amplitude stress conditions. The life factor for GLARE is 1.8 for variable amplitude loading, 2.7 for the 60 MPa loading and 2.4 for 18.5 MPa.

Fig. 7 Fatigue life increment ratio for different strap materials under constant amplitude (σmax=18.5, 60 MPa, R = 0.1) and variable amplitude (σmax= 49 MPa) loads.

4 Coupon Models Crack growth modelling A two-dimensional finite element model has been developed, in which the substrate and straps are modelled by the plate elements and the adhesive is represented by a combination of the rigid and spring elements. This assembly of elements can mimic the interactions of strap, adhesive and substrate. The effective crack-tip stress intensity factor (SIF or K) of the substrate crack is calculated for each propagating crack length under the influences of the external loads, residual stresses, and strap and adhesive constraints. Adhesive disbond and/or delamination damage area is used to modify the substrate crack stress intensity factor, which is then used to predict the crack growth rates of the substrate plate using the material data developed from the coupon tests of the project.

Life Extension Techniques for Aircraft Structures

763

Predicted crack growth rates and lives for various plate and strap configurations are validated by coupon tests. A total of 13 different coupon samples have been modelled (SENT and M(T) with various strap materials and configurations). Comparing to the test result, predicted crack growth life has an error range of -29% to 61%. The absolute mean error is 20%. Details of calculation and validation can be found in [10-11]. Stress transfer from cracked substrate to bonded straps In order to estimate the stress distribution between the substrate and strap, a 3D model of a quarter of the M(T) sample was built using the commercial software ABAQUS®. As a demonstrator, aluminium alloy 2024-HDT is used for both the substrate and strap. The strap/substrate stiffness ratio is 0.215 calculated by eq. (1). A refinement of the mesh was developed at the crack path in order to get the best approximation of stress intensity at the crack tip. The analysis was linear elastic. A delaminated area was implemented in the model in order to take into account the phenomenon of delamination taking place in the samples. The applied stress was 70 MPa. Among several cases studied, the cases of a 26 mm half crack length in the substrate (crack is in middle of strapped region) and 48 mm half crack length (crack has passed the strapped region) are chosen to illustrate the results of the stress transfer from cracked substrate to bonded crack retarder, Fig. 8. As the straps were bonded on one side of the substrate only, secondary bending was observed in with a maximum deflection of 3 mm at the middle of the sample, Fig. 8(a). Due to the secondary bending effect, the strap top surface is under compression and the bondline surface in tension. The mean stresses in the strap in the loading direction are illustrated in Fig. 8(b). In the no-crack case (1), the mean stress is about 15 MPa at the middle of the strap and decreasing towards zero at the free edge. When the substrate crack propagates under the strap, cases (2) & (3), higher stresses are picked up by the strap with a maximum stress at the delamination tip. Hence, as the crack propagates, the strap picks up more stresses from the cracked substrate. The stress distribution in the substrate was influenced mainly by the presence of the crack tip. In the unstrapped area, the longitudinal stress was 70 MPa, for the free surface and bonded surface, corresponding to the nominal applied stress (not shown). Along the crack line, the stress increases to high levels as the crack tip is approached, Fig. 8 (c). This result arises from the linear elastic model used. In reality crack tip plasticity will occur where local stresses exceed the yield stress. In models for crack growth prediction, the stress intensity factor has been used to represent crack tip stress fields which then of course, are not explicitly calculated. Significant difference in stress between the free surface and the bonded surface was found, indicating the influence of the strap and the secondary bending. It also shows the crack is predicted to grow faster at the free surface than at the bonded surface.

764

P.E. Irving et al.

(a) M(T) sample geometry and modelling

(1)

(2)

Min = - 4.7 MPa Max = 15 MPa

(3) Min = - 5.3 MPa Max = 51 MPa

Min = - 5.6 MPa Max = 55 MPa

Crack line

(b) Mean stress in strap: (1) no crack, (2) 26 mm half crack, and (3) 48 mm half crack ; half strap is shown (top line is the strap free edge)

Crack line

(c) Quarter of substrate: Mean stresses for the 48 mm crack Fig. 8 Stresses in the M(T) samples for different crack lengths.

Life Extension Techniques for Aircraft Structures

765

5 Skin-Stringer Panels Specimen and test setup To further test the principles of bonded crack retardation at a larger scale, a skinstringer panel representing the upper wing skin of a transport aircraft was designed and tested. The panel was made of AA 7085 and bonded retarder straps were made of GLARE and of 7085 alloy. A test panel is shown in Fig 9. It was 1200 mm long and 650 mm wide containing five integrally machined stringers. The stringer pitch was 130 mm, with stringer thickness of 7 mm and a height of 50 mm. The skin thickness at the thinnest areas was 4 mm and the skin-doubler thickness 6 mm. The stringer cross-sectional area to skin bay area ratio (Ast/bt) is 0.838. The initial damage scenario was an initial skin crack length of 22 mm under a broken central stringer. Two strap materials were selected to reinforce the panel. The choice of strap materials and dimensions came from the data produced by the coupon tests and residual stress measurements reported earlier. Optimum strap materials should have coefficients of thermal expansion close or equal to that of the substrate. For this reason GLARE and AA7085 were chosen as strap materials. The most important variable in the design of a strap is the stiffness ratio. For the case of the AA7085 straps, the stiffness will be slightly greater than that of GLARE, and although the widths and lengths of the straps were identical to those used for the wide GLARE straps, the thickness was reduced to 3.25 mm to maintain the stiffness ratio at 0.2. The straps were placed in the bays between the stringers with half bay widths at the panel edges.

Stringers

Straps

Gripping Fig. 9 Picture of panel with wide GLARE straps ready for testing.

The load spectrum was based on service strain measurements on the upper wing skin of a transport aircraft. It consisted originally of 21 different blocks each containing a sequence of up to 300 turning points. These were concatenated in a defined order with a total length of 4,800 blocks. The general form of the stress sequence is shown in Fig. 10 showing that the spectrum consists of load changes

766

P.E. Irving et al.

between a tensile mean of approximately 30 MPa, and a compressive mean of around 40-50 MPa. The spectrum was gated to remove smaller amplitude cycles, thus permitting completion of fatigue tests in a feasible length of time.

Fig. 10 First part of upper wing cover spectrum.

Residual stress measurement in the test panels reinforced with GLARE The residual stress field induced by thermal expansion mismatch between the strap and substrate was measured using neutron diffraction at the ENGIN-X instrument at the UK ISIS facility. Fig. 11 shows the longitudinal residual stresses in the panel skin reinforced with GLARE straps. The measurement was performed in the centre of the skin along the width of the specimen. The measured maximum residual stress was about 10 MPa. This shows a key advantage of the GLARE straps in that the coefficient of thermal expansion is close to that of the aluminium substrate so inducing low residual stress.

RD, skin 2 mm, centre 50 40 30

Stress (MPa)

20 10 EXPL

0 -10

0

100

200

300

400

500

600

700

FEA

-20 -30 -40 -50 Width (mm)

Fig. 11 Residual stresses in the panel substrate reinforced with wide GLARE strap.

Life Extension Techniques for Aircraft Structures

767

Panel Fatigue Test results Figure 12 shows the plot of half crack length vs. cycles for the panel tests showing that initial behaviour of the three tests was similar up to a half crack length of 2030 mm. After this the unstrapped and strapped curves diverged with the unstrapped propagating at a much faster rate eventually failing at a skin crack length of 320 mm, as the skin crack approached the second set of stringers. The tests of the wide GLARE and aluminium strapped panels were terminated where the skin crack was between 150-200 mm, largely for reasons of test time. Examination of the failed unstrapped panel after test showed that the crack had become very asymmetric on the skin and on the stringer where the separate growth of the two sides is recorded. At the first stringer, the crack deviated parallel to the loading direction and no further propagation across the stringer occurred. Crack turning behaviour in 7085 has been observed on a number of occasions before in the course of the coupon test reported in the earlier sections of this report, and its occurrence in the stringer webs under conditions of stringer in plane bending is consistent with the observations made on the coupon samples. In all the strapped tests, the crack on the stringer side approached and then tunnelled under the first strap. In the case of the panel with the wide GLARE straps the straps remained intact and the crack tip subsequently emerged on the other side of the strap, before entering the first pair of stringers. In the case of the wide aluminium strap, the substrate fatigue crack caused initiation of a fatigue crack in the strap itself, which then propagated together with the panel crack for the rest of the test duration. In summary, the panel testing has shown that the wide GLARE straps work well, increasing the crack growth life of the panel compared with the unstrapped one by in excess of a factor of 2. Unlike the other strapped panels the wide GLARE ones remained intact throughout the test. The aluminium strap although it initially reduced crack growth rates to a smaller value than the GLARE straps as it had no thermal residual stresses, initiated a fatigue crack very soon after the crack tip tunnelled under it, and rapidly lost retardation capability. Prediction For the un-strapped panel, the predicted crack growth life is 22% shorter than the test result probably because: a) the load interaction model in the AFGROW may be inadequate to fully model a tension-compression loading history, and a compression dominated spectrum; b) After the first stringer the crack in the stringer turned to propagate parallel to the loading direction. This change of failure mode will slow down the crack growth rate of the skin crack and was not modelled. For the wide Glare strap reinforced panel, the predicted FCG life agrees very well with the test result, Fig. 13. For the wide aluminium strap reinforced panel, predicted FCG life is much longer than the test result as the fatigue failure of the strap (crack initiation and propagation) was not modelled.

768

P.E. Irving et al.

350.0 Panel 1, No Retarder, 237kN Panel 2, Glare Retarder (W), 284kN

Avg. Half Crack Length (mm)

300.0

Panel 3, Al Retarder, 284kN

250.0 200.0

strap

150.0 100.0

strap

50.0 0.0 0

500,000

1,000,000

1,500,000

2,000,000

2,500,000

3,000,000

Cycles

Fig. 12 Half crack length vs. cycles for panels tested under a flight spectrum.

Fig. 13 Predicted and test measured crack growth lives (wide GLARE strap).

6 Final Discussion Bonded straps influence crack growth life in four major ways, which are: (1) by transferring stress locally away from the substrate; this reduces crack growth rates as the crack approaches the strap edge; (2) strap effectively bridging the

Life Extension Techniques for Aircraft Structures

769

crack as the crack tunnels under it, reducing the crack growth rates further; (3) delamination developed between the strap and the substrate stops the strap being fractured by the crack tip stress field but also increases substrate crack growth rates; (4) tensile residual stress fields created in the curing process accelerate the crack growth rates. The overall crack growth rates result from the combined influence of all the above. In the absence of residual stresses, the stiffness ratio that is related to strap cross-section area and elastic modulus determines the extent of retardation caused by the strap. Residual stresses induced by the curing process at elevated temperature act in opposition; they accelerate crack growth rates. Of the strap materials studied, AA7085 and GLARE had the least residual stress and had best crack growth retardation performance. The finite element model is able to predict crack growth rates in many of the configurations studied. Although AA7085 straps were best at substrate stresses up to 20 MPa, at 60 MPa, the stresses in the crack tip region were sufficient to initiate a crack in the strap, which then progressed together with the substrate crack with little subsequent retardation effect. Due to the low elastic modulus in the GFRP strap material, stiffness ratio was kept low at 0.07 for most of the coupons. Therefore the life improvement factor found in this study is not generic for future designs. At higher stiffness ratio of 0.2, the skin-stringer panels with wide GLARE or AA7085 straps have substantial life extension by a factor of 2 at stress levels increased by 20%.

7 Conclusions (1) The major parameter influencing retardation behaviour in the absence of residual stress is the global stiffness ratio, which represents the relative stiffness of both strap and substrate, incorporating modulus and cross section area. (2) Residual stresses arising from elevated temperature curing are detrimental to retardation capability; with elevated temperature cure, strap materials having the closest match to substrate CTE such as GLARE are the best. (3) Aluminium straps are superior to GLARE at low substrate stresses; stresses of 60 MPa and above cause fatigue cracking of the aluminium strap. (4) Under variable amplitude loading GLARE maintains its superiority as a strap material. (5) The following effects have been modelled by finite element method: a) Secondary bending due to one-side bonded strap; b) Thermal residual stresses and redistribution due to crack growth; c) Coupling of thermal and mechanical stresses; d) Influence of disbond growth on lead crack tip stress intensity factors; e) Strap size effect (variable width & thickness for a fix stiffness ratio).

770

P.E. Irving et al.

Acknowledgements Cranfield University Innovative Manufacturing Research Centre (EPSRC) is thanked for sponsoring part of this work. The authors also wish to thank the UK ISIS neutron facility for access to the ENGIN-X instrument. MEF is supported by a grant through The Open University from The Lloyd's Register Educational Trust, an independent charity working to achieve advances in transportation, science, engineering and technology education, training and research worldwide for the benefit of all.

References [1] Holwerda, F.: In: Vermeeren, C. (ed.) Around Glare a new material in context, pp. 115–120. Kluwer, Dordrecht (2002) [2] Heinimann, M., Bucci, R., Kulak, M., Garratt, M.: Structural integrity of advanced aircraft and life extension for current fleets. In: Dalle Donne, C. (ed.) Proceedings of the 23rd ICAF Symposium, Hamburg, vol. I, pp. 197–208 (2005) [3] Zhang, X., Figueroa-Gordon, D., Boscolo, M., Allegri, G., Irving, P.: Durability and damage tolerance of aircraft structures: metals vs. composites. In: Lazzeri, L., Salvetti, A. (eds.) Proceedings of the 24th ICAF Symposium, Naples, vol. I, pp. 188–205 (2007) [4] Zhang, X., Boscolo, M., Figueroa-Gordon, D., Allegri, G., Irving, P.: Eng. Fract. Mech. 76, 114–133 (2009) [5] Plokker, M., Daverschot, D., Beumler, T.: Bridging the gap between theory and operational practice. In: Bos, M. (ed.) Proceedings of the 25th ICAF Symposium, pp. 375–385. Springer, Heidelberg (2009) [6] Chakrabarti, D.I., Lui, J., Sawtell, R.R., Venema, G.B.: New generation high strength high damage tolerance 7085 thick alloy product with low quench sensitivity. In: Nie, J.F., Martin, A.J., Muddle, B.C. (eds.) Materials Forum, vol. 28, pp. 969–973. Inst. Mat. Eng. Australasia Ltd. (2004) [7] Liljedahl, C.D.M., Fitzpatrick, M.E.: Residual stresses in bonded crack retarders. Final report, Open University (2010) (unpublished) [8] Liljedahl, C.D.M., Fitzpatrick, M.E., Edwards, L.: Composite Structures 86, 344 (2008) [9] Bonded reinforcement and crack retarders in integral aluminium aircraft structures. Final Report, Cranfield University (2010) (unpublished) [10] Boscolo, M., Zhang, X.: Eng. Fract. Mech. 77, 883–895 (2010) [11] Boscolo, M., Zhang, X.: Eng. Fract. Mech. 77, 896–907 (2010)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Damage Tolerance of Adhesive Bonded Stiffened Panels: Experimental and Analytical Investigation of the Fatigue Crack Propagation Underneath the Stringers Ivan Meneghin, Gianluca Molinari, Goran Ivetic, and Enrico Troiani University of Bologna, Aerospace Engineering, Forlì, Italy

Abstract. On the basis of well-known literature, an analytical model was developed by the authors to provide reliable predictions on the fatigue propagation of cracks growing through the skin underneath adhesively bonded stringers of aeronautical stiffened panel. In order to exploit the significant damage tolerance proprieties of adhesively bonded stiffened panels it is indeed fundamental to take into account the slow fatigue crack growth under the bonded stringers, currently completely neglected as a consequence of the limits of the employed prediction tools. The correlation between the model predictions and experimental fatigue crack propagation test results of wide stiffened panels permitted to validate the model and to provide a deep insight into the mechanisms involved during the skin crack propagation under the bonded stringers of fatigue loaded panels. The implemented model confirmed its reliability in describing the fatigue skin crack propagation under the bonded stringers, but phenomena that can occur during this propagation phase, as adhesive delamination that can develop around the advancing crack tip and the fatigue failure of the bonded stringer must be taken into account in order to obtain reliable predictions.

1 Introduction Adhesively bonded solutions provide significant benefits concerning the damage tolerance (DT) proprieties of the primary aircraft structures in comparison with the more conventional riveted and monolithic structures, as described by J. Schijve [1] and T. Swift [2]. The stiff adhesive joints permit indeed a high load transfer between the cracked and the intact elements of a built-up bonded structure, while maintaining the differential structural behaviour with fail-safe features provided by the multi-load paths. As a consequence, slow fatigue crack growth together with crack arrest capabilities of large damages can be effectively obtained by adhesive bonded *

Oral presentation.

772

I. Meneghin et al.

structures, driving to significant improvements in aircraft structural reliability, weight and operational maintenance costs [3]. Nevertheless, all the benefits that the adhesive bonded structures can provide in terms of the DT proprieties are not yet fully exploited in order to design more effective aircraft solutions. This is due to the difficulties in predicting the behaviour of cracked bonded structures as a consequence of the complexity of the interacting phenomena involved during the crack propagation [4]. The intense load redistribution that can occur between all the elements of the bonded structures during the crack propagation, due to the stiff adhesive joint, can indeed determine the static or fatigue failure of the overloaded intact elements and the development of delaminations at the adhesive interfaces of the bonded structures; these are interacting phenomena which affect the main crack propagation itself. Furthermore, the effect on the fatigue crack propagation of thermal residual stresses established in the structures when different materials (i.e. with different coefficients of thermal expansion) are hot-bonded together must be taken into account in order to obtain reliable DT predictions [5]. Panels with adhesively bonded stiffeners are widely employed in aircraft areas where, as a consequence of the predominant in-service tensile stress state, the DT is the most important design criterion, such as in the crown of pressurized fuselages and in the wing bellies. The DT approach integrates the slow crack growth (SCG) design and scheduled inspections to maintain airframe ability to carry regulatory loads in presence of partial fatigue, corrosion, and/or discrete-source damage. The expected fatigue growth behaviour of the damage is indeed at the base of the definition of the inspection plan which should ensure to detect the damage before it reaches the critical dimension that could determine the catastrophic failure. Figure 1 shows the general fatigue crack propagation (FCP) behaviours of a skin crack propagating on both sides of a central broken stringer of, respectively, a riveted and bonded stiffened panel (due to the symmetrical behaviour of the two crack tips only a stringer bay is reported in the figure); the development of the critical two-bay crack over broken stringer damage scenario is thus depicted. In accordance with the SCG design, bonded stiffened panels could take significant advantage over the riveted panels by the long FCP period under the stringers (ΔNA’B’ >> ΔNAB in Figure 1), since its significant contribution to the total propagation period (ΔNB’ >> ΔNB) to exploit in order to establish an unburdensome and reliable inspection plan or, equally, to increase the allowable stress level. Nevertheless, up to now, as a consequence of the aforementioned limits in predicting the DT performances of bonded structures, in particular when a crack is propagating through the skin underneath the bonded stiffeners, the SCG design of the bonded stiffened panels is based only on the FCP up to the first stringer edge (ΔNA’ in Figure 1). As a consequence, the long crack propagation period through the skin underneath the bonded stringer (ΔNA’B’ in Figure 1) is currently completely disregarded, as highlighted by H.J. Schmidt [6].

Damage Tolerance of Adhesive Bonded Stiffened Panels

Broken Stringer

Broken Stringer

Intact Stringer

Riveted Semi-crack length, a

σ

Crack

773

B Intact Stringer

Bonded

B’

A’ A

a

a0 Broken Stringer

ΔNA σ

NA

NB

ΔNAB

Load Cycles, N

ΔNB

Riveted

ΔNA’ Bonded

ΔNA’B’

NA’

NB’

ΔNB’

Fig. 1 Fatigue crack propagation through riveted and bonded stiffened panels.

Significant benefits, in terms of structural weights, reliability and maintenance costs, could be achieved by taking into account the FCP period underneath the intact bonded stringers during the design of the adhesive bonded stiffened panels. Therefore, a deep investigation of the phenomena involved during this propagation phase is fundamental, as well as the analysis of the mechanisms that can drive the stringer to failure while the skin crack is still in the stringer covered area. In accordance with the effective crack restrain capability of intact stiffeners and the detrimental crack opening action of the failed ones, the latter is indeed a fundamental aspect that drives the FCP through the stiffened panel. On the basis of well-know literature (C.C. Poe [7][8], T. Swift [9][10], L.R.F. Rose [11], F. Erdogan [12]) an analytical tool was developed by the authors [13] using a Linear Elastic Fracture Mechanics (LEFM) approach to predict the DT performances of adhesive bonded stiffened panels and was validated by means of fatigue crack propagation test results obtained by Airbus on wide stiffened flat panels representative of typical aircraft constructions for pressurized fuselage applications. The prediction tool, named LEAF (Linear Elastic Analysis of Fracture), provided the requested insight into the mechanisms involved during the experimentally observed fatigue skin crack propagation thorough the bonded stiffened panels.

2 Experimental Investigation Adhesive bonded stiffened panels, representative of typical fuselage constructions, were manufactured and tested by Airbus in order to investigate the DT

774

I. Meneghin et al.

performances of built-up bonded solutions for pressurized fuselage applications. The effects of different combinations of skins (different materials and thicknesses of monolithic and laminate sheets), stringers (materials and dimensions) and additional doublers bonded between and/or under the stringers were investigated; the results are reported in [14]. The three most meaningful tested configurations are taken into account in the article to support the present investigation on the FCP underneath the bonded stringer. The specimens under consideration were constituted by a wide flat skin with seven equally-spaced parallel stringers, as sketched in Figure 2, together with the clamping system and the anti-bending device employed during the FCP test. Specimen

Clamping system

Anti-bending device

Intact Stringers Load

Notch 2a0

Load Cut Stringer

Skin

Crack path

Fig. 2 Sketch of the tested specimens, clamping system and anti-bending device.

The skin was a 1224mm wide, 1300mm long, 2524-T3, 1.6mm thick one-side clad sheet. Seven identical “Z-shaped” 7349-T76511 extruded stringers where adhesively bonded to the uncladded side of the skin of each panel with a stringer pitch of 185mm. The three investigated panels differed for the cross-sectional area of the bonded stringers: 89mm2, 151mm2 and 207mm2 stringers were bonded, respectively, in the panel named A089, A151 and A207. The panels were provided with a through-the-thickness machined notch (2a0=50mm long) orthogonal to the stringers. The notch was centred at the middle stringer, which was cut as well. During the FCP test, the cyclic load was applied by a servo-hydraulic machine through the clamping system orthogonally to the notch, promoting the immediate crack nucleation at the notch tips and their subsequent fatigue propagation orthogonally to the stringers.

Damage Tolerance of Adhesive Bonded Stiffened Panels

775

The FCP was driven by a merely tensile stretching of the panel thanks to the anti-bending device which prevented the out-of-plane deflection of the stiffened panel induced by the load-path eccentricity inherent to the one-side bonded stringers. The constant amplitude (CA) fatigue load applied to the three panels, characterized by a load ratio [R] of 0.1, had a maximum magnitude that varied in accordance with the panel gross-section in order to obtain the same nominal stress.

3 Analytical Investigation In order to provide a reliable prediction tool for the FCP performances of adhesively bonded stiffened panels, the C.C. Poe displacement compatibility method has been implemented. This method allows the evaluation of the stress intensity factor (SIF) of cracked stiffened panels with uniformly spaced riveted intact/broken stringers and was extended by T. Swift to take into account the effect of adhesively bonded stiffeners. In spite of the T. Swift work, the displacement compatibility method showed its reliability in predicting the crack driving force when the skin crack tip is outside the stiffener covered areas, whereas it proved to be unsuitable when the skin crack tip is propagating underneath the bonded stiffeners [15]. Figure 3 shows the prediction of the normalized SIF (i.e. the geometry factor [β]) plotted as a function of the semi-crack length [a] for the panel A207, together with the relative experimental data, confirming the previous conclusion. The reason of the discrepancy between the predictions and the recorded data in the stringer covered area (i.e. for b ≤ a ≤ c in Figure 3) is the inadequacy of the displacement compatibility method to capture the real mechanisms which drive the skin crack propagation under the adhesively bonded stiffener.

Geometry factor, β

2.5 2 1.5

Prediction Experimental data

Stringer-covered area

Central Cut Stringer

3

1

Limits of the displacement compatibility method

0.5 0 0

a0

92.5

b 185 c

Semi-crack length, a (mm)

Fig. 3 Geometry factor [β] as a function of the semi-crack length [a] for the A207 panel: predicted curve obtained by the displacement compatibility method in comparison with the recorded experimental data.

776

I. Meneghin et al.

Distance from the crack axis, y (mm)

During this crack propagation phase, the most effective stringer hindering action on the crack propagation is the restraint on the crack opening due to the bonded reinforcement. This mechanism, known as crack-bridging, acts along the crack faces under the stringer, as shown in Figure 4; the predicted skin crack opening displacements (CODs) of a crack approaching the stringer (a=160mm) and a crack extended through the whole bonded stringer covered area (a=199.5mm) are herein plotted for the A207 panel. Figure 4 COD predictions are the result of a crack-bridging model proposed by L.R.F. Rose on the basis of the F. Erdogan work, that was implemented by the authors to analytically describe the skin crack propagation underneath the bonded stringers.

1

a = 199.5 mm

Stringer covered area

0.5

0

a = 160 mm

-0.5

Crack axis

Crack Bridging

-1 138.75

185 b c Distance from the crack origin, x (mm)

231.25

Fig. 4 Predicted CODs of a skin crack approaching the bonded stringer edge (a =160mm) and of a skin crack with the crack tip at the end of the bonded stringer covered area (a =199.5mm). CODs calculated for the A207 panel.

The basic simplification proposed by L.R.F. Rose is that the restraining action exerted by the crack-bridging mechanism can be modelled by the crack-closure effect of a continuous distribution of linear springs acting between the crack faces under the bonded stringers, as sketched in Figure 5. The restrained COD under the stringer (2v(x)) is indeed driven by the local stress field relief due to the crack-bridging stress induced by the springs [σS], as depicted by the Eqn.(1): v( x) = f (σ 0 − σ S ) = f (σ 0 − kEP v(x ))

(1)

where σ0 is the local skin stress due to the redistribution of the remote stress [σ∞] between the skin and stringer. σ0 can be obtained from the one-dimensional theory of bonded joints in accordance with the skin/stringer stiffness ratio.

Damage Tolerance of Adhesive Bonded Stiffened Panels

777

y σ Stringer covered area

v(x)

σS(x)

x

x σS(x)

σ

Fig. 5 Crack-bridging mechanism of the bonded reinforcement modelled by a continuous distribution of springs acting between the crack faces in the stringer covered area.

The linear behaviour of the springs acting between the crack faces is expressed by the last term of Eqn.(1), where k is the spring stiffness constant and Ep is the Young’s modulus of the skin material. The k constant can be determined from a one-dimensional analysis of a singlestrap joint representative of the load transfer from the cracked plate to the bonded stringer, as sketched in Figure 6. δA adhesive

σ∞

d reinforcement

σ∞

σ∞

uP

skin

delamination

k

σ∞ y

Fig. 6 Single-strap joint representative of the load transfer from the cracked plate to the bonded stringer.

In the case of perfectly bonded single-strap joint (i.e. no delamination), the remote stress determines an opening of the joint [uP] only due to the adhesive displacement contribution [δA]. A delamination around the crack provides an additional displacement contribution induced by the deformation of the reinforcement over the delaminated area “2d” wide [εRd]; εR is the linear deformation of the reinforcement. In accordance with Eqn. (2), a delamination reduces the stiffness proprieties of the adhesive single-strap, and thus the magnitude of its spring stiffness constant k. k=

σ∞ EPu P

=

σ∞

E P (δ A + ε R d )

(2)

778

I. Meneghin et al.

The solution of the skin crack propagation underneath the bonded stringer comes from the solution of the function in Eqn.(1). This function is described by means of the dislocation theory, translating the Eqn.(1) into an integral equation in the unknown v(x), with a numerical solution as described by F. Erdogan. Under the hypothesis of the LEFM, the relation between the crackdisplacement [v(x)] and the SIF [K] for the plane-stress condition is expressed as: K (a ) = v( x)

G (1 − υ ) 2π a−x 2

(3)

where G and υ are, respectively, the shear modulus and the Poisson’s ratio of the skin material. Figure 7 plots the normalized SIF (geometry factor [β]) as a function of the semi-crack length [a] predicted by the implemented crack-bridging model for the panel A207 (curves), together with the experimental data (spots).

1.5

Geometry factor, β

Stringer covered area

1

del ↑

exp del d=0mm del d=10mm

0.5

del d=20mm del d=30mm del d=0 - 30mm

del d=0 -30 mm 0 162

b

185 Semi-crack length, a (mm)

c

208

Fig. 7 Geometrical factor predicted for several degrees of delamination extension [d] and experimental data of the A207 panel.

The effect of no delamination and delaminations extended 10mm, 20mm and 30mm on both sides of the advancing crack are considered [d]. Furthermore, the effect of a delamination that likely extends during skin crack growth underneath the stringer is considered (“del d=0 - 30mm”); a linear growth of the delamination with the number of the applied load cycles up to 30mm was assumed. Longer is the crack extension through the stringer covered area and more effective is the crack-bridging mechanism which reduces monotonically the crack driving force, and thus the geometry factor [β]. A delamination that can develop at the adhesive interface between the bonded stringer and the cracked skin around the advancing crack tip can significantly

Damage Tolerance of Adhesive Bonded Stiffened Panels

779

Distance from the crack axis, y (mm)

affect the effectiveness of the crack-bridging mechanism. As shown in Figure 8, where the COD of the skin crack propagated through the whole stringer covered area in A207 panel is plotted for various delamination extensions, the delamination reduces the crack-opening restraint exerted by the bonded stringer.

1

Stringer covered area

0.8 0.6

del ↑

0.4 0.2

Crack axis

0 -0.2 -0.4 -0.6 -0.8 -1 138.75

del 0mm

del 10mm

del 20 mm

185 b c Distance from the crack origin, x (mm)

del 30 mm 231.25

Fig. 8 Predicted COD of a skin crack 199.5mm long (semi-crack length, a) in the panel A207 as a function of the delamination extension under the adhesive bonded stringer.

4 Test - Analytical Result Correlation Fatigue crack propagation underneath the adhesively bonded stiffeners The predicted β for the A207 panel in Figure 7 provides a good agreement with the experimental data, showing the reliability of the crack-bridging model to describe the fatigue crack propagation through the skin underneath bonded stiffeners. Furthermore, the test-analytical results correlation suggests that the FCP under the bonded stiffener is coupled with the simultaneous fatigue growth of a delamination at the adhesive interface around the advancing crack tip. This conclusion is in accordance with the Alderliesten model [3] describing the delamination growth at the adhesive interface of Fiber Metal Laminates (FML) around the through-thethickness crack based on a Paris type relation, i.e. the delamination extension as a function of the applied load cycles (i.e. d=f(N)). Figure 9 for the A151 panel, and Figure 10 for the A089 panel, confirm the aforementioned conclusions. Smaller delaminations in comparison with the A207 panel are expected under the stringer of the A151 (“del d=0 - 15mm”) and A089 (“del d=0 - 10mm”) panels as a consequence of the shorter FCP through the skin underneath the stringer as shown in Figure 11.

780

I. Meneghin et al.

1.5

Geometry factor, β

Stringer covered area

1 exp

del ↑

del d=0mm del d=5mm del d=10mm del d=15mm

0.5

del d=20mm del d=0 - 15mm

del d=0 - 15 mm 0 162

b

185 Semi-crack length, a (mm)

c

208

Fig. 9 Geometrical factor predicted for several degrees of delamination extension [d] and experimental data of the A151 panel.

1.5

Geometry factor, β

Stringer covered area

1 exp

del ↑

del d=0mm del d=5mm del d=10mm

0.5

del d=15mm del d=20mm del d=0 - 10mm

del d=0 - 10 mm 0 162

b

185 Semi-crack length, a (mm)

c

208

Fig. 10 Geometrical factor predicted for several degrees of delamination extension [d] and experimental data of the A089 panel.

Stiffener Failure A sudden increase of the experimental geometry factor while the skin crack is still in the stringer covered area, not captured by the crack-bridging model, is shown in Figure 7, Figure 9 and Figure 10. The failure of the bonded stringer, overloaded when the skin crack tip propagates underneath it, is the driver of the observed phenomenon. In Figure 11 the experimental FCP performances of the three investigated panel are reported: as a consequence of the stringer failure, the crack-bridging mechanism cannot be exerted by the failed stringer which, on the contrary, promotes the FCP by means of traction forces exerted to the crack surfaces, thus increasing the crack propagation rates. As a consequence of the significant effect of the stringer failure on the FCP performances, the development of a reliable stringer failure criterion is

Damage Tolerance of Adhesive Bonded Stiffened Panels

781

fundamental in order to exploit the FCP period under the bonded stringers for the SCG design. For this reason, it is important to get a deep insight into the mechanisms which drive the bonded stringer to failure when a skin crack fatigue propagates through the bonded stiffened panel. A151

A089

A207

200

Semi-crack length, a

Stringer covered area 150

A151 A207

Stringer Failure

A089

100

50

Central Cut Stringer

0 0

10000

20000

30000

40000

Load Cycles, N

Fig. 11 Experimental fatigue crack propagation within the first two stringer bays over the central broken stringer.

The maximum stress acting in the stringer foot during the skin crack fatigue propagation was calculated by the analytical tool developed by the authors [13] and plotted in Figure 12 as a function of the number of applied load cycles [N] for the three investigated panels. The vertical lines, which intercept the predicted curves, represent the stringer edges reached after a different number of load cycles for the three panels. The stringer starts experiencing the most significant stresses when the skin crack enters in the stringer covered area. These stresses increase for thinner stringers.

Max stringer stress (MPa)

1000 900

A089

800 700 UTS 7349 -T76511 YTS 7349 -T76511

600

A151

500

A207

400 300 200 100 0 0

5000

10000

15000

20000

25000

30000

35000

40000

Cycles, N

Fig. 12 Maximum predicted tensile stress acting in the stringer foot as a function of the applied cycle loads during the FCP.

782

I. Meneghin et al.

In accordance with a fatigue failure mechanism, lower are the stresses applied at the stringer foot and longer is the fatigue life of the stringer itself. This behaviour determines the precocious failure of the thin stringer in the A089 panel after a few thousands of load cycles with the skin crack underneath it, and the long FCP period under the intact thick stringer in the A207 panel. Table 1 reports the experimentally recorded FCP periods taken by the skin crack to reach the first stringer edge (“bay” in Table 1) and to propagate through the stringer covered area (“under”) for the three investigated panels. The reported FCP periods are expressed as percentage of the total FCP period taken by the skin crack, for each panel, to pass beyond the stringer. It is interesting that in the case of a stringer with a cross-sectional of 207mm2 75% of the total FCP was spent underneath the stringer. This period is completely neglected in the current sizing of the bonded stiffened panels in accordance with the SCG design criterion. Table 1 FCP periods recorded during the experimental tests taken to reach the first stringer edge (“bay”) and to propagate under the stringer (“under”). FCP periods expressed as percentage of the total period to pass beyond the first stringer.

Panel A089 A151 A207

FCP performances bay [%] under [%] 78.7 21.3 44.5 55.5 24.3 75.7

5 Conclusions The crack-bridging model implemented by the authors provided an effective and reliable tool for the prediction of the Fatigue Crack Propagation (FCP) of cracks growing through the skin underneath the adhesively bonded stringers of aeronautical stiffened panels. As a consequence, the long FCP period under the bonded stringers, completely neglected in the current sizing of the bonded stiffened panels due to the limits of the adopted methods, could be fully exploited in accordance with the Slow Crack Growth design criterion. The employment of the crack-bridging model permits to optimize the Damage Tolerance proprieties of the bonded stiffened panels, providing significant benefits in terms of reliability, maintenance costs and weights.

Acknowledgments Grateful thanks are given to Nikolaus Ohrloff and Marco Pacchione (Airbus) for the experimental data and technical advices.

Damage Tolerance of Adhesive Bonded Stiffened Panels

783

References [1] Schijve, J.: Tech. Report LR-589, Delft University of Technology, Delft, Netherlands (1989) [2] Swift, T.: AGARD-AG-176, H. Liebowitz, Ed. AGARD Advisory Group for Aerospace Research and Development, Neuilly sur Seine, France (1974) [3] Alderliesten, R.C.: Bridging the Gap Between Theory and Operational Practice. In: Bos, M.J. (ed.) Proceedings of the 25th ICAF Symposium, Rotterdam, Netherlands, vol. I, pp. 73–90 (2009) [4] Boscolo, M., Zhang, X.: Eng. Fract. Mech. 77(6), 883 (2010) [5] Meneghin, I., Ivetic, G., Troiani, E.: In: Proceedings of the 8th European Conference on Residual Stresses (ECRS8), Riva del Garda, Italy (2010) [6] Schmidt, H.J.: Structural integrity of advanced aircraft and life extension for current fleets - lessons learned in 50 years after the comet accidents. In: Plantema Lecture of the 23rd ICAF Symposium, Hamburg, Germany (2005) [7] Poe, C.C.: MSc Thesis, Virginia Polytechnic Institute, Blacksburg, Virginia, USA (1969) [8] Poe, C.C.: Tech. Report TM X-71947, NASA, Langley Research Center, Hampton, Virginia, USA (1973) [9] Swift, T.: Fracture Mechanics Design Methodology, AGARD-LS-97, North Atlantic Treaty Organization, London, England (1979) [10] Swift, T.: Trans. ASME, Ser. D 100(1), 10–15 (1978) [11] Rose, L.R.F.: In: Baker, A.A., Jones, R. (eds.) Bonded Repair of Aircraft Structures, pp. 77–106. Martinus Nijhoff Publ. (1988) [12] Joseph, P.F., Erdogan, F.: NASA Contractor Report 178328, NASA Langley Research Center, Hampton, Virginia, USA (1987) [13] Molinari, G., Meneghin, I., Melega, M., Troiani, E.: In: Proceedings of the 2nd Aircraft Structural Design Conference RAeS 2010, London, UK (2010) [14] Meneghin, I., Pacchione, M., Vermeer, P.: Bridging the Gap Between Theory and Operational Practice. In: Bos, M.J. (ed.) Proceedings of the 25th ICAF Symposium, Rotterdam, Netherlands, vol. I, pp. 427–447 (2009) [15] Meneghin, I., Molinari, G., Troiani, E., Pacchione, M.: Submitted to Eng. Fract. Mech. (2010) [16] Kim, J.H., Lee, S.B.: Eng. Fract. Mech. 67, 303 (2000)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Development of a New Fiber Metal Laminate Variant Optimized for Cold Expansion and Riveting of Holes David Backman1, Thomas Sears1, and Eann A. Patterson2 1

National Research Council – Institute for Aerospace Research, Ottawa, Canada 2 Michigan State University, Composite Vehicle Research Center, East Lansing, USA

Abstract. The increased use of fiber metal laminates (FML) in aerospace structural applications led to an investigation into whether current FML layups could be improved to provide a better response during cold expansion and riveting of holes. Static measurements of the residual strain during cold expansion of holes were made using digital image correlation. Various grades of FML as well as 2024-T3 aluminum were investigated with the results leading to the design of a new FML layup. Static strength testing of this new FML variant demonstrated that its elastic modulus compared well to more traditional FML.

1 Introduction The use of fiber metal laminate materials in aerospace applications is increasing as manufacturers investigate and integrate fiber metal laminates (FML) into aircraft structures. Fiber metal laminate materials, especially those based on a combination of aluminum and glass fibers, are being promoted as potential replacements for aluminum based on their lower weight and increased fatigue life. The focus on manufacturing processes such as hole cold expansion and riveting of holes is also more critical in FML due to the inherent tensile residual stresses that are locked into the material during the manufacturing process. Compared to monolithic aluminum alloys, crack nucleation and growth to a detectable level tends to occur more rapidly, but the overall propagation of these cracks to failure is greatly retarded by crack bridging from the glass laminate layers [1,2]. Early crack nucleation in FML is a result of three factors. The first is the tensile residual stresses in the aluminum layers arising during the heating/curing process due to the difference in the coefficient of thermal expansion between the metal and the fiber reinforced thermoset layers. The second factor is that crack bridging only becomes fully effective for cracks on the order of 0.51.5 mm [3], depending on the laminate. The third factor is the difference in stiffness between the aluminum layer (75 GPa) and the glass layer (50 GPa) that results in a higher stress in the aluminum layer. To date, little experimental work has been performed to look at the effect of riveting and cold expansion of holes on the residual strain field in fiber metal laminate materials. The focus of this research was to measure the residual strains *

Oral presentation.

786

D. Backman, T. Sears, and E.A. Patterson

in standard grades of FML during the cold expansion and riveting processes and to use this information to drive the development of a modified variant of FML that would be more suited to these types of manufacturing processes.

2 Methods The experimental methods section details the procedures involved in both the static and fatigue testing of the aluminum and fiber metal laminate coupons, including both the cold expansion and riveting procedures as well as information regarding the digital image correlation method used for strain measurement. An overview of the production methodology for the FML panels is also provided along with residual stress measurements for each panel. Production of FML Coupons The FML panels used for this research were all produced at the National Research Council composite facility. All FML material was manufactured using FM-94 film adhesive embedded with S2 glass fibers (typically referred to as a prepreg), along with 0.3 mm thick 2024-T3 aluminum that had been phosphoric acid anodized (PAA) and coated with BR127 primer to improve adhesion to the FM-94 film adhesive. This combination of aluminum and S2 glass fibers is commercially known as GLARE, and is available in several standard grades, defined by the number and orientation of the glass prepreg layers. For this research the focus was on FML 4 which is comprised of three layers of 2024-T3 aluminum interspersed with three layers of unidirectional glass prepreg. If the aluminum layer are abbreviated as // and the unidirectional glass prepreg is designated by the fiber direction then the laminate composition can be described as [//90/0/90//90/0/90//]. Since the laminate is symmetrical about the central ply of aluminum its composition will be abbreviated by the layup of the glass prepreg layers i.e FML 4 [90/0/90]. The cure cycle used during the autoclave process for all FML sheets with a maximum temperature of 121 ºC and a peak pressure of 517 kPa. Digital Image Correlation Digital image correlation (DIC) is a white light, non-contact optical strain measurement technology that can provide measurements of the full-field strain tensors in the area of interest. The basic concept behind the use of image correlation for strain measurement is that a set of points on an undeformed object can be matched to the same set of points on the deformed object [4-6]. For image correlation to work effectively, it is important that the surface of the object examined has enough contrast to ensure that each subset of points on the object is statistically different from every other subset of points. It is for this reason that a black and white speckle pattern is often applied to the surface of the object being tested. A cross-correlation function is then used to characterize the quality of the match between the reference and deformed images [5]. Once this has been completed over the entire area of interest, the displacement field obtained can be differentiated to determine the full strain tensor [5].

Development of a New Fiber Metal Laminate Variant

787

Static Process Setup Identical test fixtures and DIC stereo camera setups were used to measure surface strains during both the hole cold expansion and the riveting process for both the aluminum and FML coupons (Figure 1). The DIC system was comprised of two digital video cameras (Retiga 1300, Qimaging Inc, Burnaby BC) with resolutions of 1280 x 1024 pixels. For the field of view used in the static test, the effective spatial resolution was 0.052 mm/pixel and the coupon surface was illuminated using two high intensity LED light arrays (Siemens-Nerlite Inc, Nashua NH). The sensitivity of a particular DIC setup is a function of the speckle distribution, the image capture hardware and the correlation algorithm. The sensitivity of the DIC setup used was determined empirically after system calibration by capturing multiple reference images and processing them through the correlation algorithm. The overall strain sensitivity in all cases was better than ±50 με and the out of plane displacement (w) sensitivity was better than ±0.0015 mm.

Fig. 1 Isometric diagram of coupon retention fixture for process testing.

Procedure for Cold Expansion of Holes Hole cold expansion was performed using a manual cold-working tool (HP-10, FTI Inc) that was positioned behind the coupon and supported by a V block. The cold expansion mandrel was pulled through the hole by turning the nut at the back of the manual cold expansion tool. Images were captured at regular intervals until the mandrel had passed completely through the hole at which time the mandrel was removed from the coupon and a final image was taken. Hole cold expansion was performed using a size 8-1-N split sleeve with a nominal starting hole diameter of 6.40 mm and total interference of approximately 4.1%. Incremental deformed images were taken from the mandrel entry side of the coupon as the mandrel was displaced through the coupon thickness. The deformed images were processed using Vic-3D software (Correlated Solutions Inc, Columbia SC) over an area of interest spanning the entire height of the coupon. All images were

788

D. Backman, T. Sears, and E.A. Patterson

processed with a subset size of 17 pixels, a step size of 5 pixels and a strain window of 15 pixels. Riveting Procedure The same test fixture and camera setup was used for the riveted coupons. For the static testing, universal head rivets (MS20470AD-5) with a nominal diameter of 3.96 mm (0.156 inches) made from 2117 aluminum were used. These rivets were positioned in pilot holes having a nominal diameter of 4.09 mm (0.161 inches) which left approximately 0.12 mm of clearance around the rivet after installation. An initial measurement of the rivet protrusion was made before the riveting process commenced. Riveting was performed in two steps, with the first step being to use the manual rivet press to set the rivet in the hole at which point a reference image was taken. Next, the coupon was taken to a hydraulic rivet press were the final riveting was completed. The coupon was brought back to the test frame where another image was taken and the final rivet diameter and protrusion height were measured. The deformed images were processed using Vic-3D software (Correlated Solutions Inc, Columbia SC) with a subset size of 17 pixels, a step size of 9 pixels and a strain window of 15 pixels. For the riveting portion of the fatigue test, coupons were riveted using an instrumented arbor press with a squeeze force of approximately 13.3 kN. Since the coupons were going to be used in fatigue, a special support fixture and loading platen were integrated into the arbor press to help standardize the riveting process and reduce any chance of bending the coupons. Standard 3.96 mm diameter aircraft grade rivets (MS20470AD-5) requiring a clearance hole of 4.09 mm diameter were used and all aluminum and standard FML 4 coupons were tested at a net stress of 263 MPa. On the driven head side of the riveted coupon the same optical system was employed to measure crack length. Fatigue Test Setup Fatigue experiments were conducted in an MTS servohydraulic frame equipped with a 90 kN load cell. Three load levels (8.3, 8.9 and 10 kN) were chosen with a load ratio of R=0.1 and a loading frequency of 10 Hz. These corresponded to net section stresses of 168, 175 and 198 MPa respectively for the coupons with unexpanded holes and approximately 0.8% higher net section stresses for the coupons with cold-expanded holes. Net section stress was defined as the applied load divided by the cross-sectional area (perpendicular to the load) of the specimen at the position of the hole. On the mandrel entry face of the aluminum coupons, optical images were captured using a single high resolution digital video camera (Allied Vision Technologies, Newburyport, MA) with a 1000 x 1000 pixel spatial resolution and a Canon FD zoom lens, coupled to a fiber optic based ringlight (Olympus Highlight 3000, Center Valley, PA) providing a spatial resolution of approximately 15.1 μm/pixel. Surface crack length was measured using ImageTool (University of Texas Health Science Center, San Antonio, TX) image processing software.

Development of a New Fiber Metal Laminate Variant

789

3 Results and Discussion Static Strain Measurements: Cold Expansion and Riveting The baseline strain field from riveting (Figure 2a) and cold expansion (Figure 2b) was determined using 2024-T3 aluminum coupons. The residual strain results after cold expansion (Figure 2b) showed that although the strain field was affected by the location of the split in the sleeve, monolithic 2024-T3 showed relatively concentric residual strains. The residual strains after riveting (Figure 2a) showed similar behaviour, although the strain field closest to the pilot hole was obscured by the manufactured and driven head of the rivet. Equivalent data for FML is shown in Figures 3 and 4 for the coupons with the open rivet and with coldexpanded holes respectively. The FML coupons with holes that had been coldexpanded (Figure 4a) showed much more pronounced strain gradients at 45º degree increments around the hole.

(a)

(b)

Fig. 2 Maximum principal strains (a) after riveting, and (b) after cold expansion of holes in 2024-T3 aluminum coupons.

Fig. 3 Maximum principal strains on driven head side of FML coupon.

790

D. Backman, T. Sears, and E.A. Patterson

In order to further investigate the orthotropic nature of FML and the residual strains after cold expansion, additional coupons were machined with the longitudinal direction oriented at 45° degrees to the long axis of the coupon. The split sleeve direction remained the same (to the right in the image), and Figure 4 shows the dramatic change this causes in the maximum principal strain field. From a structural point of view, rotating the coupons’ axes effectively orients the split in the split sleeve in the direction of minimum elastic modulus and minimum yield strength. The effect of this can be seen clearly in Figure 4b, where the typical butterfly shape that results from the split sleeve is replaced with a more concentric annulus of compressive residual strain. This observation inspired the development of a new layup for FML 4 with a more isotropic behaviour.

(a)

(b)

Fig. 4 Maximum principal strains on entry face of FML coupon (a) specimen with standard orientation and (b) FML specimen cut at 45º orientation.

One of the first steps in the design process for a new FML variant was to determine a method of analytically estimating FML material properties such as the elastic modulus, shear modulus and Poisson’s ratio. The properties of any fiber metal laminate material can be predicted analytically using an approach that combines micromechanic strength of materials methods with classical laminated plate theory. The strength of materials approach is used to model the properties of the prepreg material (resin + glass fiber) while classical laminated plate (CLP) theory can be used to determine the properties of the entire fiber metal laminate based on the number of plies and their orientation. The final proposed FML variant stemmed from a review of classical laminated plate theory, specifically the concept of a quasi-isotropic composite [7,8]. It has been shown in classical laminated plate theory that if three or more identical plies are stacked at equal angles (i.e the angle separating each of the n plies is π/n radians) then the in-plane properties of the laminate are isotropic. For an FML 4 variant with n=3 plies, this would mean an angle of 60º between plies i.e a layup of [-60/0/60].

Development of a New Fiber Metal Laminate Variant

791

Static strength testing confirmed that the elastic modulus in the longitudinal, transverse and 45º direction agreed well with the analytical findings (Figure 5). A comparison between FML 4 (Figure 4a) and the new FML 4 variant (Figure 6a) showed that a significantly more uniform residual strain field existed after hole cold expansion,.

Elastic Modulus (GPa)

60.00 55.00 50.00 45.00 40.00 35.00 30.00 0 deg

15 deg 30 deg 45 deg 60 deg 75 deg 90 deg Material Orientation

FML 4 FML 4 Variant (CALC)

FML 4 Variant FML 4 (CALC)

Fig. 5 Elastic modulus for FML 4 [90/0/90] and FML 4 [-60/0/60] measured from static failure tests (columns) combined with results from elastic modulus calculated using classical laminated plate theory (symbols); note that the dotted lines are for visualization purposes only.

(a)

(b)

Fig. 6 Maximum principal strain for a quasi-isotropic FML 4 variant [-60/0/60] after (a) cold expansion and (b) riveting.

792

D. Backman, T. Sears, and E.A. Patterson

Fatigue Test Results The static strength test results (Figure 5) showed that the in-plane elastic modulus of the new FML 4 variant was independent of material orientation. The resultant strain field after cold expansion (Figure 6a) and riveting (Figure 6b) was more uniform than with other grades of FML, and appeared to more closely resemble that of monolithic aluminum. Based on these results it was expected that the new FML 4 variant would have a fairly constant fatigue life irrespective of material orientation, both in the unexpanded and cold expanded configuration. For the fatigue testing, additional panels of FML 4 [90/0/90] and FML 4 variant [-60/0/60] were produced. The cut plan for the coupons was modified to allow n=2 coupons to be produced in the longitudinal, transverse and 45º degree directions respectively. With the reduced number of coupons per panel, fatigue testing of the new FML 4 variant was focused on only one stress level (198 MPa, R=0.1) and two conditions: unexpanded and cold-expanded open hole. The hole diameter after cold expansion and reaming was measured using a ball gauge and the applied load was reduced in order to ensure that both the unexpanded and cold expanded coupons had a net stress of 198 MPa. Comparisons of the fatigue life for coupons of the FML 4 [90/0/90] and the FML 4 variant with both unexpanded and cold-expanded holes, were based on the number of cycles required for a crack length of 3.7 mm as this was the largest crack size visible in the field of view of the camera. For the coupons with riveted holes, a displacement limit of 0.89 mm was used as the basis for comparing fatigue performance because their fatigue life was extremely long as a consequence of multiple small cracks being propagated. The fatigue testing was designed to determine the off-axis fatigue performance of both FML 4 [90/0/90] and the new FML 4 variant. Although FML 4 [90/0/90] coupons cut in the transverse material orientation had over twice the life of FML 4 variant coupons cut in the same orientation (Figure 7), the FML 4 variant coupons tested in the longitudinal and 45° degree material orientations showed an improvement in fatigue life over FML 4 [90/0/90]. Both grades of FML 4 showed an improvement in fatigue life as a result of cold expansion. For FML 4 variant coupons with rivets, a significant increase in fatigue life was seen when compared to that of a coupon with either an unexpanded or coldexpanded hole (Figure 8). Individual crack growth curves from both grades of FML were analyzed for coupons having holes both with and without cold expansion to determine the number of cycles required to grow a 1 mm crack, as well as the crack propagation rate (da/dN). The error bars presented for each data point were calculated from the standard deviation of the average value of the n=2 samples from each material orientation and are presented as ± one standard deviation. The results for standard FML 4 [90/0/90] are provided in Figure 9 and the results for the new FML 4 variant are provided in Figure 10. Since FML 4 [90/0/90] has a lower modulus in the longitudinal and 45º degree orientations than in the transverse direction, crack growth rates are significantly higher compared to those in transversely oriented coupons; however, cold expansion is still effective at reducing crack growth rates.

Development of a New Fiber Metal Laminate Variant

793

Cold expansion appears to relatively ineffective at retarding short crack growth (crack growth to 1 mm), with a retardation of no greater than 5,000 cycles on average. The same analysis performed for all the FML 4 variant coupons showed more constant crack propagation rates (da/dN) in all material orientations, and a fairly uniform decrease in crack propagation rates (da/dN) after cold expansion of the holes. As with the FML 4 [90/0/90] coupons, the effect of cold expansion on short crack growth was fairly minimal with crack growth being retarded by no more than 6,500 cycles on average. 80,000 FML 4 (Unexpanded)

Cycles to Crack length = 3.7 mm

70,000 FML 4 Variant (Unexpanded)

60,000

FML 4 (Cold Expanded) FML 4 Variant (Cold Expanded)

50,000 40,000 30,000 20,000 10,000 0

Long

45 Deg

Transverse

Material Orientation

Fig. 7 Fatigue results for a nominal net stress of 198 MPa for coupons of FML 4 and FML 4 variant both with unexpanded and cold expanded holes.

Cycles to Coupon Run-out

250,000

200,000

150,000

Unexpanded Hole Cold Expanded Hole Rivet Net Stress = 198 MPa

100,000

50,000

0 FML 4 Variant

Fig. 8 Comparison of fatigue results at a nominal net stress of 198 MPa for FML 4 variant showing cycles to failure of coupons with an unexpanded hole, a cold-expanded hole and a rivet.

250.0

25,000

200.0

20,000 150.0 15,000 100.0 10,000 50.0

5,000 0

da/dN (mm/cycle) x 10

30,000

-5

D. Backman, T. Sears, and E.A. Patterson

Cycle Count

794

0.0 Long

Trans 45 Degree Material Orientation FML 4 OH FML 4 OH

FML 4 CX FML 4 CX

70.0

15,000

60.0 -5

18,000

50.0

12,000

40.0 9,000 30.0 6,000

20.0

3,000

10.0

0

da/dN (mm/cycle) x 10

Cycle Count

Fig. 9 Variation in number of cycles to 1 mm crack length (columns) and crack growth rate, da/dN (symbols) as a function of material orientation for FML 4.

0.0 Long

Trans Material Orientation

FML 4 Variant OH FML 4 Variant OH

45 Degree

FML 4 Variant CX FML 4 Variant CX

Fig. 10 Variation in number of cycles to 1 mm crack length (columns) and crack growth rate, da/dN (symbols) as a function of material orientation for the FML 4 variant.

Development of a New Fiber Metal Laminate Variant

795

The overall conclusions from this data set suggest that cold expansion is effective at reducing crack propagation rates in FML, in all material orientations, but is less effective at retarding short crack growth, at least on the entry face.

4 Conclusion Measurements of the residual strain in standard grades of FML motivated the development of a new FML 4 layup that provides a more isotropic material response, and is thus better suited to cold expansion and riveting of holes. Static tensile testing showed that the FML 4 variant with a quasi-isotropic layup had an elastic modulus similar to that predicted by the closed form solution. Strain measurements after cold expansion and riveting showed that this FML 4 variant had a much more uniform residual strain field after cold expansion than either standard FML 3 or FML 4. Fatigue testing results showed that the fatigue life of the FML 4 variant was relatively constant, independent of material orientation, and that it held significant residual strength and stiffness after complete fatigue crack growth on both coupon faces. Although a quasi-isotropic FML may not be ideal in fatigue rated applications, such as in an airplane fuselage, for static design applications or applications with multiaxial loading and/or unknown loading, this FML 4 [-60/0/60] may prove to be extremely useful.

References [1] Alderliesten, R., Homan, J.: Int J. Fatigue 28(10), 1116 (2006) [2] Alderliesten, R., Hagenbeek, M., Homan, J., Hooijmeijer, P., de Vries, T., Vermeeren, C.: Appl. Compos. Mater. 10, 223 (2003) [3] Homan, J.: Int J. Fatigue 28, 366 (2006) [4] Chu, A., Ranson, W., Sutton, M., Peters, A.: Exp. Mech. 25(3), 232 (1985) [5] Sutton, M., Mcneill, S., Helm, J., Chao, Y.: Top. Appl. Phys. 77, 323 (2000) [6] Sutton, M., Wolters, W., Peters, W., Ranson, W., Mcneill, S.: Image Vision Comput. 1, 133 (1983) [7] Kobayashi, A.: Handbook of Experimental Mechanics, 2nd edn. VCH Publishers, New York (1993) [8] Reddy, J.N.: Mechanics of Laminated Composite Plates. CRC Press, Boca Raton (2007)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Applying the Damage Tolerance Approach to Expanded Bushing and Rivetless Nut Plate Installations Len Reid Fatigue Technology Seattle, WA, USA

Abstract. Existing aviation regulations require that structure demonstrate damages tolerance capabilities and retains residual strength for a period after sustaining a given level of fatigue, corrosion, accidental or discrete source damage. For fatigue critical elements it is customary to establish threshold inspection based on half the life to grow a manufacturing flaw, usually considered a 1.25 mm (0.050 inch) crack at a fastener hole, to a critical size at limit load. The benefit of the residual compressive stresses associated with cold worked holes have been used to establish inspection thresholds based on manufacturing flaws by allowing a smaller initial flaw size of 0.125 mm (0.005 inch). The General Harmonization Working Group Report on the Harmonization Effort for FAR/JAR § 25.571, Damage Tolerance and Fatigue Evaluation of Structure, includes recommendations for determining inspection thresholds where the beneficial effects of split sleeve cold expansion could be used to affect the threshold. This paper proposes a similar case could be made to affect inspection thresholds using high interference fit expanded ForceMate bushings or ForceTec rivetless nut plates. These processes are expanded into the hole with high interference fit and like split sleeve cold expansion; induce beneficial residual stresses to the surrounding structure.

1 Introduction The generally accepted damage tolerance approach in establishing inspection thresholds for fatigue is based on the assumption that most critical areas of the structure or joints contain the maximum probable-sized initial material or manufacturing flaw or discrete damage. The time to grow that “flaw” to a critical crack size, under the expected load conditions, is then factored to determine safe inspection intervals. For fastened joints, holes and cutouts, The United States military specification MIL-A-83444 determined that initial flaw size shall be 1.27 mm (0.050 inch) through the thickness in material less than 1.27 mm (0.050 inch) thick and for other holes a 1.27 mm (0.050 inch) radius corner flaw at one side of the hole. Similar requirements have been generally accepted by civil airworthiness authorities and manufacturers [1]. It is assumed these requirements would also apply to attaching lugs and fittings. *

Oral presentation.

798

L. Reid

MIL-A-83444 includes the beneficial effects of interference fit fasteners and other specific joint design including cold expanded (also known as cold worked) holes to achieve compliance with the flaw growth requirements of the specification. These beneficial effects, demonstrated by test, may allow a smaller initial flaw of 0.127 mm (0.005 inch) radius corner flaw with a cold expanded hole, due to the presence of beneficial residual stresses induced, to determine crack growth life and inspection thresholds. The reduction in stress intensity factor due to the presence of the residual stresses has been well documented and discussed [2], particularly in relation to developing analysis methods to predict crack growth lives from cold expanded holes. Although this allowance has been used for fastener holes it is generally not applied to high interference fit cold expanded (ForceMate) bushings such as those used in attaching lug applications and to re-size damaged and discrepant holes, or the now available cold expanded (ForceTec) rivetless nut plates. The General Structures Harmonization Working Group (GSHWG) Report on Harmonization Effort for FAR/JAR § 25.571, Damage Tolerance and Fatigue Evaluation of Structure [3], contains similar approaches that can be used when considering methods for threshold inspection determination. Appendix 3 of this reference further allows for the establishment of the lower bound threshold for inspection to be based on half the life to grow a crack from an initial corner flaw of radius 1.27 mm (0.050 inch) at a single fastener hole to the critical crack size. In the calculation of threshold values the report acknowledges that any component or full scale fatigue tests can be “calibrated” to account for the effects such as residual stresses as well as crack growth retardation. The cold expansion processes provide these beneficial effects on fatigue life and crack growth retardation due to a combination of the crack closure effect and the reduction in stress intensity factor these residual compressive stresses impart. In the case of rotor craft, for rotating parts such as attaching lugs and clevises in rotor hubs operating under high cycle fatigue conditions, FAR/JAR § 25.571 requires a flaw tolerant safe-life evaluation to show that the structure, with flaws present, can withstand variable magnitude loads without detectable flaw growth for the determined interval. For these evaluations, most manufacturers assume that the hole has a 0.375 mm (0.015 inch) radial corner flaw. In all these cases the crack growth life may be very short depending on the load conditions; which can impose restrictive threshold inspection intervals on rotor components. To mitigate this, the design of such components requires the local thickness of the lug be increased to reduce the local operating stress levels appropriately. This imposes an unacceptable dynamic weight increase in design of helicopter rotors. Installation of ForceMate bushings, which are radially expanded into lugs at a high interference fit, have been shown to provide significant fatigue and flaw tolerance life improvement over conventional shrink fit bushing installation methods used for helicopter rotor assemblies [4]. ForceMate allowed the design life target to be achieved without reduction of design nominal stresses while simultaneously achieving a significant increase in safe life in the presence of induced artificial flaws.

Applying the Damage Tolerance Approach

799

2 Expanded Bushings and Rivetless Nut Plates Aircraft designers and maintainers use the benefits of ForceMate high inference fit expanded bushings and ForceTec rivetless nut plates, which are installed in a similar manner, to improve fatigue life and damage tolerance based on test data. Users of these advanced products are seeking the benefits from the manufacturing processes in establishing the “damage tolerance” requirements used to determine inspection thresholds, similar to those afforded cold expanded holes. An allowance to use a smaller initial flaw size when calculating the fatigue tolerance life associated with these expanded products would also be useful. The following benefits derived from the processes, plus the accompanying test and usage examples, could be used to support this request. Obviously individual cases would have to be supported by tests; however, the extent of testing could be minimized. ForceMate Process Most lug assemblies are bushed to provide a durable wear surface. The ForceMate high interference fit expanded bushing method radially expands an initially clearance fit bushing into a hole or lug assembly at high interference fit. A specially sized bushing, with a proprietary lubricant on the inside surface, is radially expanded into the hole using an expansion mandrel as shown in Figure 1. The process of expanding the bushing will yield it into the hole creating a high interference fit. Depending on the combination of the relative yield strength of the bushing material and the parent material, the installation process may also cold expand the base material and induce beneficial residual compressive stresses as described in one of the following actual examples. In addition to providing a rapid, more consistent and non-damaging bushing installation, the combination of the high interference and possible cold expansion of the parent material will result in an installation that will resist bushing rotation/migration which could lead to fretting damage and crack initiation, have a greatly enhanced fatigue life and significantly improved damage or flaw tolerance over alternate bushing installation methods.

Fig. 1 ForceMate Process.

800

L. Reid

A number of significant studies have been carried out to investigate the fatigue life improvement and crack retardation effects of the ForceMate process relative to fixed and rotary wing aircraft in establishing the flaw tolerant effectiveness in accordance with FAR/JAR requirements. The results from these following tests all showed either crack arrest or significant retardation compared to the traditional shrink fit bushing installations. Commercial Aircraft Wing Nacelle Support Fitting This example compares the fatigue crack growth life of shrink-fit and ForceMate bushing installations in 7050-T7451 aluminum double-ended lugs in both the L-T and S-L grain orientations [5]. The comparisons included cycles to failure at different stress levels, crack growth and environmental testing. The test results in Table 1 for the S-L lugs, shows the results of a crack growth test conducted to compare the total lives of the two bushing configurations. Both lug specimens had EDM induced 1.2 mm corner notches prior to bushing installation. The Shrink fit installed lug failed at a relatively low number of cycles 729,137. The ForceMate specimen was cycled for about 2 million cycles without any further crack growth and then the load was progressively increased and cycled for about 1 million cycles each time. As shown in Figure 2, the maximum applied stress was increased by 80% before failure occurred, and at over 6 times the life of the shrink fit installation. This test showed that the ForceMate bushing installations provided significant fatigue and crack growth life improvement over shrink fit bushing installations in the 7050-T7451 aluminum material and influenced the decision to incorporate ForceMate into the nacelle support fitting design. An additional specimen was tested in a corrosive environment of a flowing NaCl solution. Although the overall fatigue lives of the shrink fit and ForceMate lug ends was less than in ambient air, the fatigue life improvement of the ForceMate bushed end was 17 times the life of the shrink-fit bushed end. Part of the reason for this is that the higher interference fit of the ForceMate installation prevents the corrosive solution migrating down the bore of the hole. The residual compressive stress also reduces the effective stress intensity factor associated with the crack as shown in the next example of a wing attachment lug on a fighter aircraft. Table 1 Summary Details of ForceMate Vs Shrink Fit with Increasing Load.

Bushing Material

Lug Material

Thickness (mm/ inch)

Width (mm/ inch)

SHD (mm/ inch)

Al-Ni-Br (AMS 4640)

7050T7451 (S-L)

8.89 / 0.350

38.6 / 1.52

14.3 / 0.563

Shrink Fit Load (kN/ kip)

Cycles

14.9 / 3.356

729,137

ForceMate Load (kN/kip) 14.9 17.9 22.4 26.8

/ / / /

3.356 4.024 5.030 6.037

Cycles 2,066,938 2,957,750 3,730,906 4,728,330

Applying the Damage Tolerance Approach Test Requirements: Specimen: B7SCA4 Specimen Type: Double End Lug

Legend: Shrink Fit Bush End (max load) Testing at 3.356 kip (14.84kN)

Width: 1.5211 inch (38.636 mm) Hole Dia: 0.5626 inch (14.290 mm) Thickness: 0.3506 inch (8.905 mm) Corner Notch: 0.050 inch (1.27 mm)

ForceMate Bush End (max load)

Load Conditions: Constant Amplitude Fatigue Load: See Legend R-Ratio: +0.1 Frequency: 5 Hz Environment: Ambient Lab Air

Testing at 3.356 kip (14.84 kN) Testing at 4.024 kip (17.90 kN) Testing at 5.030 kip (22.37 kN) Testing at 6.073 kip (27.01 kN)

0.50

12.7

0.45

11.4

0.40

10.2

0.35

8.9

0.30

7.6

0.25

6.4

0.20

5.1

0.15

3.8

0.10

2.5

0.05

1.3

0.00

Crack Length (mm)

Crack Length (inch)

801

0.0 0.0

5.0x10 5 1.0x10 6 1.5x10 6 2.0x10 6 2.5x10 6 3.0x10 6 3.5x10 6 4.0x10 6 4.5x10 6 5.0x10 6

Cycles (N)

Fig. 2 Crack Growth Life Improvement of ForceMate Bushing Compared to Shrink Fit with Progressively Increasing Load.

Demonstrated Damage Tolerance of F-22 Wing-Attach Lugs The tear down inspection of the F-22 full scale fatigue test after 2.5 lifetimes revealed cracks in the lower wing-attach lugs. The lower lugs are the most critical for durability and damage tolerance as they react the wing up-bending loads in tension. Although ForceMate bushings were used in the installation, there were circumstances associated with the test article supporting a conclusion that the cracking was actually an anomaly of the testing. It was also felt that the original ForceMate bushings had not been installed at the optimum expansion for the application, the material used and the expected high in-service loads in the lug. An investigation was conducted that included comparative specimen configuration testing as well as damage tolerance evaluation of the ForceMate installation at the determined optimum applied expansion [6]. A total of 9 test specimens were manufactured from Ti-6Al-4V forgings to simulate the wing-attach lugs. ForceMate bushings, installed to the original specification, along with original configuration shrink fit bushings, were compared to

802

L. Reid

an optimized increased expansion ForceMate bushing. A 1.27mm (0.050 inch) nominal starter notch was located at the position of peak stress in the hole. Testing was carried out at the designated spectrum load to correlate the results obtained from the full scale fatigue test article. The second ForceMate specimen test, with the revised expansion bushing, was eventually terminated without failure at greater than 8.9 lifetimes as shown in the summary Table 2 and results in Figure 3. It was interesting to note from the conclusion of the test and analysis carried out that the increased expansion ForceMate bushings verified the solution to the F22 wing-attach lower lug cracking problem. A design change was implemented on production aircraft and a program to retrofit the fleet with the revised ForceMate bushing was proposed. Implementation of these design changes and proposed retrofits eliminates the need for further inspections of this critical location on the aircraft, and therefore validates use of ForceMate to achieve damage tolerance design of these critical lugs. Table 2 Summary of test Results. Bush Type Shrink Fit ForceMate

Thickness (mm/

Width (mm/

SHD (mm/

Maximum Spectrum Load

inch)

inch)

inch)

(kN/kip)

Ti

44.48 /

6Al -4V

1.75

107.9 / 4.248

Bushing Material

Lug Material

Copper Bronze Copper Beryllium

50.8 / 2.0

Lifetimes to Failure

Comparison Fatigue Life

1.01 1099 / 247

>8.9

1.44 (normal exp)

56.54 / 2.226

>8.9 (optimized exp)

1.4 Net Fit Specimen A-K-SP Net Fit Specimen B-K-SP ForceMate Specimen A-UK-5 ForceMate Specimen B-UK-6 Increased Expansion ForceMate Specimen

Total Crack Length

1.2 1.0 0.8 0.6 0.4 0.2 0.0 0

1

2

3

4

5 Life

6

7

8

9

10

Fig. 3 Comparative ForceMate and Shrink Fit Damage Tolerance Crack Growth Lives (Courtesy Ref 5).

Applying the Damage Tolerance Approach

803

Crack Growth Test of Lugs with Residual Stresses As part of an evaluation by FTI to model lugs using the AFGRO program for life prediction, with the inclusion of residual stress fields, four different configuration lugs were evaluated [7]. These lug configurations included; a baseline lug without bushings, a lug with cold expanded holes and no bushings, a lug with conventional shrink fit bushings and a lug with ForceMate (FmCx) installed bushings. The test hole in each specimen was the same diameter prior to installing the respective bushings. The matrix of the test program is shown in Table 3. Table 3 Test Matrix. Starting Hole Diameter (mm/inch)

Description

As-Reamed Hole (Baseline) [NCx] 30.9/1.219!.001 Cold Expanded Hole

[Cx]

29.7/1.168 to 29.77/1.172

Hole with Shrink Fit Bushing [SF] 30.9/1.219!.001 Hole with FmCx Bushing

[FmCx] 30.9/1.219!.001

Processing

Final Hole Diameter (mm/inch)

None

30.9/1.219!.001

Cx with 38-2 -N

30.9/1.219!.001

SF Bush

25.4/1.000!.001

FmCx Bush

25.4/1.000!.001

Stress MPa/ ksi

Quantity

68.9/10 103/15 68.9/10 103/15 68.9/10 103/15 68.9/10 103/15

1 1 1 1 1 1 1 1

The double ended lugs were manufactured from 7075-T73 aluminum with only one end hole of the lug used for the test hole (as shown in Figure 4). To force crack growth and failure to the test hole the other end lug hole was filled, shielded, and then gripped in the test frame. All specimens had a 45 degree 1.25 mm (0.050 inch) corner notch cut into the hole at 90 degrees to the load line. 2.97 0.500 ±.005 1.485

7.0±.03

CL

∅D See Table 3.0-1

Apply shim over filled hole 2.2 All dimensions are in inches. Not drawn to scale.

Fig. 4 Specimen Configuration.

804

L. Reid

The shrink fit lug specimens had 17-4 stainless steel (H925) bushings installed at approximately 0.05 mm (0.002 inch) interference. The other bushed lugs had ForceMate bushings made from the same material and installed per the FTI specification 9901. The baseline and cold expanded specimens had net fit bushings installed to allow use of the same diameter loading pin. After the bushings were installed they were all reamed to a final bushing inner diameter of 25.4 mm (1.0 inch). The residual stress fields derived by finite element analysis for each of the hole conditions evaluated is shown in Figure 5. The shrink fit bushing residual stresses were all tensile, while the ForceMate bushing had a significant residual compressive stress surrounding it. Testing was carried at constant amplitude with a maximum bearing stress of 68.9 MPa (10 ksi) and 103 MPa (15 ksi) with a stress ratio (R) of +0.05. The results of the fatigue tests are shown in Table 4. Crack growth measurements were taken for all configurations to establish the respective crack growth curves for the shrink fit and ForceMate (FmCx) bushings and are shown in Figure 6. Unfortunately the testing was terminated at 2 million cycles and the 68.9 MPa test ForceMate lug was damaged in test. The termination of the ForceMate test and the accidental damage of the other specimen did not show the true potential of the ForceMate bushed lugs. However, based on the minimum achieved life at the 103 MPa levels, the life improvement was in excess of 40:1 compared to the baseline configuration and 7.5:1 compared to the traditional shrink fit bushing installation. Distance from Edge of Hole (mm) 2.5

5.1

50000

7.6

10.2

12.7

15.2

17.8

20.3

22.9 345

25000

172

0

0

-172

-25000

-50000

Cold expanded hole with final ream Shrink-fit bushing installation ForceMate bushing installation

-75000 0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Distance from Edge of Hole (inch)

Fig. 5 Residual Stress Fields for Each Hole Condition.

-345

-517 0.9

Residual Hoop Stress (MPa)

Residual Hoop Stress (psi)

0.0

Applying the Damage Tolerance Approach

805

Table 4 Fatigue Test Results. Specimen

Hole

Identification

Configuration

Hole

Notch Length

Bearing Stress (MPa/ksi)

Cycles to

Fatigue Life

Failure

Improvement

FS373-NCx-1

Baseline

1.2196

0.0525

FS373-NCx-2

Baseline

1.2188

68.9 / 10 103/15

126,306

Baseline (10 ksi)

0.0541

42,689

FS373-Cx-1 FS373-Cx-2

Cx Hole Cx Hole

1.2190 1.2187

Baseline (15 ksi)

0.0375 0.0384

68.9 / 10 103/15

2,000,000 No Failure 2,000,000 No Failure

15.8 (minimum) 46.9 (minimum)

Diameter (inch) (Corner Flaw) (inch)

FS373-Cx-3

Cx Hole

1.2194

0.0584

103/15

2,000,000 No Failure

46.9 (minimum)

FS373-SF-1

Shrink Fit Bushing

1.2185

0.0505

1,893,407

15.0

FS373-SF-2

Shrink Fit Bushing

1.2187

0.0479

68.9 / 10 103/15

263,889

6.2

FS373-FmCx-1

FmCx Bushing

1.2187

0.0488

Note 2

11.1 (minimum)

FS373-FmCx-2

FmCx Bushing

1.2190

0.0510

68.9 / 10 103/15

2,000,000 No Failure

46.9 (minimum)

Notes:

1. Testing was performed at an R- Ratio of +0.05 and a frequency was 10 Hz. Testing was performed in ambient lab conditions. 2. Specimen FS373-FmCx-1 was damaged at 1,400,000 cycles caused by a computer malfunction in the test control software. The test was stopped at this time. 3. Fatigue life improvement compares the one result to the baseline results [Cycles to failureConfig X / Cycles to failureBaseline ].

0.9

22.9

Legend Testing at 10 ksi Shrink Fit Bushed Hole ForceMate Bushed Hole

0.8 0.7

17.8

Testing at 15 ksi Shrink Fit Bushed Hole ForceMate Bushed Hole

0.6

15.2

2.5

0.0

0.0 2.00x10

1.75x10

1.50x10

1.25x10

1.00x10

7.50x10

5.00x10

2.50x10

6

0.1

6

5.1

6

0.2

6

7.6

6

0.3

5

10.2

5

0.4

5

12.7

0.00

0.5

Crack Length (mm)

Crack Length (inch)

20.3

Cycles (N)

Fig. 6 Crack Growth Plots for Shrink Fit and ForceMate installations at Two Different Stress Levels.

Fighter Aircraft Damage Tolerance Evaluation In another non-published evaluation for a fighter aircraft program a damage tolerance assessment was carried out using a double ended titanium lug configuration incorporating 28.5 mm (1.125 inch) diameter 17-4 stainless steel ForceMate

806

L. Reid

bushings. The initial flaw was a 1.25 mm (0.050 inch) corner notch, simulating a manufacturing defect. Loading was 100% load transfer at 90 kN load. The ForceMate life improvement compared the original shrink fit installation was approximately 20:1 (as shown in Figure 7). Using a conservative inspection threshold determination assuming 1/3 the life to the critical crack length the shrink fit configuration would have to be inspected at around 2,000 cycles (A), whereas the ForceMate threshold inspection would have been at around 35,000 cycles (B); over double the total shrink fit life. This test further reinforced the benefit of the high interference fit expanded ForceMate bushing in extending an initial inspection threshold. Helicopter Flaw-Tolerant Rotor Design FAR and JAR 29.571 currently require that for dynamic rotor assemblies, flaw tolerance capabilities must be demonstrated for compliance. That means for lug sections, such as those commonly used in main and tail rotor assemblies in helicopters, have to demonstrate tolerance form accidental damage or manufacturing/assembly flaws. Additionally, under the high cycle fatigue load conditions, fretting can rapidly lead to fatigue crack initiation under dynamic rotor loading. Since most rotors are bushed to provide a durable wear surface, installation of ForceMate high interference fit bushings can provide significant reduction in fretting and a greatly enhanced fatigue and flaw tolerance life improvement over conventional bushing installation methods.

100% Load Transfer Axial Fatigue Technology Inc. Material: Titanium Maximum Load: 90 kN (20 Kip), R-Ratio: +0.10 Frequency: 20 Hz, Environment: Ambient Lab Air Note: All specimens were corner notched and precracked prior to testing. 10.00

0.394

A

B

8.00

0.315

6.00

0.236

Crack Length (mm)4.00

Crack Length (inch)

Shrink Fit

0.157

2.00

0.079

ForceMate 0.00 1

10

100

1000

10000

100000

0.000 1000000

Cycles

Fig. 7 ForceMate 20:1 Increase in Crack Growth Life Compared to Shrink Fit.

Applying the Damage Tolerance Approach

807

High Cycle Fatigue (HCF) and Damage Tolerance Testing To evaluate the effectiveness of 1.0-inch diameter ForceMate 17.4 PH stainless steel bushings under HCF and simulated damage tolerance, double-ended 7075T73 aluminum lug specimens (100% load transfer) were prepared. Two different tests were conducted at different gross stresses to compare the baseline shrink fit bushings against ForceMate with and without an anti-fretting “BlueCoat” coating applied to the outside diameter of the bushing [8]. The second phase of the test compared the same configurations; however, the lug specimen was pre-flawed with a 0.375 mm (0.015 inch) corner flaw prior to installing both the shrink-fit and ForceMate bushings. Test results in Figure 8 show that all ForceMate specimens ran to run-out (no failure) at 10 million cycles, providing relative fatigue life improvement to the baseline specimen (no flaw) of at least 5:1 and for the damage tolerance specimens at 13:1. Follow-on testing of specimens at higher stress levels show similar trends, however the crack growth life improvement factor decreases as stress levels increase. 68.9/ 10.0

Constant Amplitude Fatigue R-Ratio: +0.1, Freq.: 15 Hz Room Temperature Conditions Specimen Material: 7075 -T73 Bushing Material: 17-4PH SS Specimen Type: Double Ended Lug (w/D=1.75)

9.5 62.0/ 9 . 0 8.5

Gross 8 . 0 Stress (MP a/ksi)7 . 5

As Reamed Hole Shrink Fit Bush FmCx Bush FmCx with BlueCoat Bush

48.2/ 7 . 0

Hole with 0.015 Corner Flaw Shrink Fit Bush FmCx Bush FmCx with BlueCoat Bush

6.5 41.3/ 6 . 0 5.5 10,000

100,000

1 ,000,000

10,000,000

5E7

Cycles

Fig. 8 HCF Damage Tolerance Comparison between Shrink Fit and ForceMate Bushings.

ForceMate in Rotor Design Several recent helicopter rotor designs have been certified with “flaw tolerant” rotors using ForceMate. As part of the testing and validation the generally accepted 0.375mm (0.015 inch) initial corner crack flaw has been used to simulate the manufacturing damage. By way of example, the extensive test program undertaken by Agusta [3], to validate ForceMate, provided a significant data base for lugs in titanium and aluminum alloys. The objectives were to establish appropriate S-N curves and demonstrate the damage or flaw tolerance capability in terms of threshold to propagation of cracks and fatigue scatter. For the test specimens, double ended lugs made from Al 7475-T7351 and titanium Ti 6Al-4V annealed were

808

L. Reid

used and 17-4 PH (H1025) stainless steel bushings were installed. Constant amplitude testing was conducted. The results of these tests at maximum and minimum ForceMate interference were compared to standard bushed lugs. An interesting observation (seen before with cold expanded load transfer joints) was that despite the scatter of results, the absence of fretting was extremely favorable in terms of fatigue strength, so that the fatigue allowable with ForceMate bushings installed was much higher than those found with comparative high interference fit shrink fit bushings. Based on the coupon and component testing, ForceMate was validated to meet their “flaw tolerant” design life objectives. ForceTec Rivetless Nut Plates Similar to the ForceMate bushing installation method the cold expanded ForceTec rivetless nut plate expands a bushing like retainer into the fastener hole (see figure 9) at a high interference fit that will locally yield the material around the hole and induce beneficial residual compressive stresses. As has been shown with the ForceMate bushings, these residual stresses greatly increase the fatigue, durability and damage tolerance life of the nut plate installation.

Fig. 9 Schematic of ForceTec Retainer Installation.

Compared to a traditional riveted nut plate in new construction, ForceTec has reduced stress concentration due to the elimination of the satellite rivet holes and prevents fretting and corrosion by providing a bushing type protection in the fastener hole. ForceTec has also been used in repair scenarios to replace existing riveted nut plate installations where fatigue cracking had been identified. A major rework of the F-16 fighter rear fuselage access panel attaching structure included removal of the riveted nut plates; then cold expansion and plugging of the satellite rivet holes, opening up the existing fastener hole and installation of ForceTec retainers into the holes. In full scale coupon tests of the access panel installation,

Applying the Damage Tolerance Approach

809

with existing cracks remaining in some repaired holes, the crack growth life was extended from around 2500 spectrum flight hours to over 23,000 hours [7]. Similar testing of repaired specimens in support of the rework of the US Navy P-3 Orion fleet has been reported [8]. In these tests, existing repaired riveted nut plate holes with residual cracks left in the ForceTec repaired holes of up to 0.6 mm (0.024 inch) at 8000 spectrum hours exceeded 45,000 hours and were deemed to be terminating repair action when used on the P-3 fleet. In both this example and the F-16, determination of the re-inspection threshold interval with ForceTec, compared to rework with the legacy riveted nutplates was in excess of the residual design service life of the airplane. In both cases they were deemed terminating repair actions based on damage tolerance life enhancement test.

3 Conclusions The significance of high interference fit, in conjunction with the induced residual beneficial residual stresses associated with ForceMate bushing installations and ForceTec rivetless nut plates, has been shown. Based on numerous fatigue and crack growth tests of high load transfer lugs tested with pre-existing 1.27 mm (0.050 inch) corner flaws then fitted with the expanded ForceMate bushings, the fatigue life was significantly increased to a point where virtually all lugs exceed the life expected (or testing was terminated without failure or significant crack growth) when compared to conventionally installed shrink or freeze fit bushings. Other similar HCF based fatigue and damage tolerance tests have confirmed that helicopter rotor designs can now meet the FAA/JAR flaw tolerant requirement with ForceMate bushings installed without increasing the size of the lug. Extending the smaller initial flaw size allowable for cold expanded fastener holes to the high interference fit expanded ForceMate bushings, or cold expanded ForceTec nut plates, as an evaluation method for fatigue tolerance evaluation would seem appropriate. The processes would seem to comply with current regulations that permit manufacturing quality and residual stresses in demonstrated damage tolerance evaluations. Applying the same principle as applied to cold expanded holes could greatly extend the inspection thresholds for most lugs or nut plate installations; thereby saving unnecessary inspection downtime and increasing aircraft availability. Analytical damage tolerance prediction methods supporting the test data are needed as a threshold determination design tool.

References 1. Swift, T.: Damage Tolerance Capability. Fatigue 16(1), 75–94 (1994) 2. Reid, L., et al.: The Evaluation of a Modified Stress Intensity Factor for Cold Expanded Holes. In: Proceedings of the 20th ICAF Conference, Bellevue, WA (1999) 3. General Structures Harmonization Working Group Report, Harmonization Effort for FAR/JAR § 25.571, Damage Tolerance and Fatigue Evaluation of Structure. Ref. L350-03-115

810

L. Reid

4. Mariani, U., Ratti, G., Reid, L.: Attaining Flaw-Tolerant Design Life Objectives in Helicopter Rotor System Lugs. In: Proceedings of the 30th European Rotorcraft Forum, Marseilles. France (2004) 5. Fatigue Technology Technical Report # 163731, Fatigue Life Enhancement Test ForceMate versus Shrink Fit (September 1, 2004) 6. Cayton, M., Bunch, J., et al.: In: Test demonstrated damage tolerance of F-22 wingattach lugs with ForceMate bushings. In: USAF ASIP Conference (2007) 7. Fatigue Technology Technical Report # 96262, Prediction of Crack Growth for Lugtype Specimens with Residual Stress Fields (June 5, 2002) 8. Fatigue Technology Technical Report # 92279 (2001). High Cycle Fatigue and damage Tolerance Testing of Lugs with Shrink Fit and ForceMate Bushings (December 31, 2002) 9. Ransom, J., et al.: F-16 Fighting Falcon Upper Fuselage Skin Fatigue Life Enhancement. Proceedings of USAF ASIP Conference, San Antonio TX (1999) 10. Reid, L.: Service Life Extension of the Lockheed P-3 Aircraft. In: Proceedings of USAF ASIP Conference, Savannah, GA (2003)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Fatigue Life Improvement of Metallic Aerospace Structures via Crenellations M.V. Uz, Y.J. Chen, and N. Huber Helmholtz-Zentrum Geesthacht, Institute of Materials Research, Materials Mechanics, Max-Planck-Strasse 1, D-21052, Geesthacht, Germany

Abstract. In a former study it was shown that systematic thickness variations (crenellations) can improve the fatigue crack growth (FCP) life of the integrally stiffened aluminum panels significantly [1]. The source of this improvement is the stress intensity factor (K) modification due to the geometry change caused by implementation of the crenellations. The crenellation pattern controls the K-modification and therefore indirectly the FCP life improvement that can be achieved. In the current study, the limits of this improvement were investigated by optimizing the crenellation pattern numerically. For the optimization, finite element analysis (FEA) and artificial neural networks (ANN) methodologies were utilized in a coupled way. The life calculations were performed according to the Paris Law considering uniaxial and biaxial loading conditions. In the second part, an additional mechanism contributing to the life extension on the crenellated panels, namely loading history effects, explained with the help of the specially designed FCP tests. It was demonstrated that the fluctuation of the K-factors on crenellated panel leads to significant crack growth retardation, which improves the FCP resistance of the panels further. Finally, the application potential of crenellated panels under the more service-related condition of biaxial loading is discussed in conjunction with a currently running research program.

1 Introduction The load distribution on an aircraft fuselage is far from being uniform. Different loads arising from different sources like the cabin pressure and longitudinal bending operate together to form a rather complex loading scheme. Therefore, the fuselage has not a homogeneous structure but is tailored according to the direction dependent loading conditions at a given section. Normally an aluminum sheet of homogeneous thickness is used to carry the load, where the sheet thickness is chosen according to the load level of the more severely loaded direction. The concept of crenellations was developed to respond to the orientation dependent loading in a more sophisticated way. With the help of systematic thickness variations, the FCP rate of a crack growing in a specific direction is *

Oral presentation.

812

M.V. Uz, Y.J. Chen, and N. Huber

reduced. This preferentially strengthened direction can be aligned with the more severely loaded direction of the specific section of an airframe. In this way, the damage tolerance of the component can be tailored according to the need and the weight of the component can be minimized at the same time. Figure 1 shows schematically a crenellated stiffened panel together with a conventional one. As can be seen, at certain sections, the thickness of the crenellated panel is higher than the reference panel and at other sections it is vice versa. The extra weight coming from the thicker regions are compensated by the reduced thickness in the neighboring areas. For later comparison, the overall inplane panel dimensions and the panel weights were exactly the same for these two panel types. In what follows, the weight of the panel is not increased and merely distributed in a different way by the introduction of crenellations.

Fig. 1 Schematic representation of a) conventional and b) crenellated panel with one stringer bay crack under tensile loading. The simplified crenellation pattern shown in b) is given as an example.

Crenellations retard the fatigue crack growth basically by two mechanisms. The first mechanism is related to the nature of the relation between stress intensity factor range (ΔK) under cyclic loading and the resulting FCP rate. As can be realized from the form of Paris law, this relation is not linear. Actually, da /dN is, for all engineering metals, a very strong function of ΔK, generally taking an exponent between 2 and 4 [2]. Simplified, when ΔK is reduced and increased with the same factor for equal lengths, the fatigue life gain in the “slow” growth region will be higher than the life shortening in the “fast” growth region. This mechanism was assumed to be the primary mechanism that retards the FCP lives of the crenellated panels. Further explanations related to this mechanism can be found in [1].

Fatigue Life Improvement of Metallic Aerospace Structures via Crenellations

813

Fig. 2 Stress intensity factor profiles of a middle crack on stiffened reference and crenellated panel having 2.9 mm equivalent thickness. The illustration in the lower side of the figure shows schematically the thickness profile of the crenellated panel. The point “0” on the X-axis corresponds to the symmetry plane of the panels.

The second retardation mechanism is related to loading history effects. It is well known that an overload applied during a constant amplitude loading scheme can retard the crack growth dramatically (among others [3 - 5]). The enlarged plastic zone, caused by the overload, is responsible for this retardation. Strong negative gradients in K at the “thinning” steps of a crenellated panel generate a similar effect. Figure 2 shows the K-factors for a middle crack on a crenellated panel with five laser beam welded (LBW) stringers under uniaxial loading for the crack lengths between 10 and 150 mm. The K-factors of a reference panel having equal weight are also given for the same crack range. It should be noted that the crenellation pattern yielding this K-factor profile was empirically chosen at the initial stage of this work. As can be seen, crenellations give rise to significant fluctuations on the K profile of a transverse propagating crack. When a crack grows from a thin to a thick section, K shows a decreasing trend. Vice versa, when a crack grows from a thick to a thin section, K steeply increases. As the thin and thick sections systematically repeat in a crenellation pattern, the crack passes several zones with continuously changing K level. Therefore, even under constant amplitude loading, a crack in a crenellated component experiences variable amplitude-like loading conditions.

814

M.V. Uz, Y.J. Chen, and N. Huber 250 FCP life comparison, Stiffened ref. and cren. panels Al2139-T8, RT, Beq = 2.9 mm, 2W = 740 mm, a0/W = 0.1, Smax = 50 MPa, R = 0.1

200

65%

a (mm)

150

100 Reference

50 Crenellated

5.2L-left (stif. ref.) 5.2L-right (stif. ref.) 5.5L-left (stif. cren.) 5.5L-right (stif. cren.)

0 0

50

100

150

200

Cycles (N/1000)

Fig. 3 Comparison of fatigue lives of reference and crenellated stiffened panel tested under constant amplitude loading. The crenellated panel showed 65% longer FCP life than the conventional panel having the same weight [1].

The test results of the Al2139-T8 panels containing such a crenellation pattern clearly demonstrated that it is possible to significantly extend FCP lives of the integrally stiffened panels by implementing crenellations (Figure 3). The resulting FCP life depends on the chosen geometry and boundary conditions. For given boundary conditions (initial crack length and load amplitude) and a given equivalent thickness, an optimal crenellation pattern should exist, which provides the maximum possible FCP life. In this context, the goals of the present work are: (i) determination of the crenellation pattern with the highest efficiency, (ii) investigation of the FCP behavior of the crenellated panels under biaxial loading as a more realistic loading condition for the fuselage structures, and (iii) investigation of loading history effects at the “thinning” steps and their added value to FCP life improvement of crenellated panels.

2 Modelling and Optimization In this part, first the Finite Element (FE) models of crenellated panels were created to obtain the K-factors. Artificial neural networks were employed to generalize the FE results for further optimization of the crenellation pattern with respect to maximum FCP life. Finite Element Model As the basis for the modeling, the investigated crenellation pattern, which has an equivalent thickness of 2.9 mm, was selected (Figure 4). This pattern was implemented onto the inner 400×400 mm region of a 600×600 mm sheet (Figure 5-a).

Fatigue Life Improvement of Metallic Aerospace Structures via Crenellations

815

The square form corresponds to the cruciform specimen geometry that can be tested using the available biaxial testing machine [6] for validation during the later stages of the program. The inner section of the panels is surrounded with a 100 mm wide strip of 4.5 mm thickness. The machine arms are connected to the cruciform specimen at this outer section using bolts. The additional thickness here is intended to reduce the stress level around the bolt connections to avoid an undesired crack initiation.

Fig. 4 The crenellation pattern having 2.9 mm equivalent thickness. This symmetrical pattern constitutes a length of 150 mm. It is originally designed to be machined on the 150 mm wide stringer bays.

The finite element model presented in Figure 5-b uses the modified virtual crack closure technique (MVCCT) method [7] to calculate the K-distribution. The crack propagates along the red line indicated in the figure. The approach has been validated by two other models (half model and quarter model) using a nodal release technique. Both approaches gave equal results for the stress intensity factor as function of crack length. The material of the crenellated panel is aluminum alloy 2139, with a Young’s modulus of 72.4GPa and a Poisson’s ratio of 0.33. Both uniaxial loading case and biaxial loading case are studied. For the uniaxial loading case, the loading of 60750 N is applied in the direction of 2 (the coordinates are shown in Figure 5-b). For the biaxial loading case, the loading in direction of 1 is half of that in the direction of 2. These simulations are carried out with ABAQUS 6.8-1 [8]. The geometry of the studied crenellation pattern is given in Table 1. For the corresponding reference panel, the thickness of the inner panel was 2.90mm.

816

M.V. Uz, Y.J. Chen, and N. Huber

Fig. 5 Schematic representation of the a) crenellated panel; b) MVCCT Finite Element half model. The crenellation pattern in Figure 4 is implemented onto the middle 400 mm section of the crenellated panel. Table 1 Definition of the crenellation geometry.

* In the MVCCT half model, the position x=0.0 is at the left side of the model.

For the biaxial loading case, the K factors calculated by these three models are shown in Figure 6 together with the K factors of the corresponding reference pane. The curves of the quarter model (I), the half model (II) and the MVCCT half model (III), are in excellent agreement, i.e. all these three models are valid for the K computation. The resulting K ratios of crenellated to the reference panel are shown in Figure 7. The comparison of the K distributions between the biaxial loading case and the uniaxial loading case is shown in Figure 8. It can be seen that there is a small difference: the K value of biaxial case is a little smaller than that of the uniaxial case. The reason is that the loading in the direction of 1 has a tendency to close the crack through lateral contraction and thus the K value slightly decreases. The corresponding K ratios of the crenellated panel to the reference panel are shown in Figure 9 for uniaxial and biaxial loading conditions. Here again, only a very small difference of K ratio between the two cases can be found. If we assume that only the K ratio determines the extension in the fatigue life, the optimal crenellation pattern should be the same for uniaxial and biaxial loading. In the following optimization, only the MVCCT half model (III) will be used for the simulations.

Fatigue Life Improvement of Metallic Aerospace Structures via Crenellations

817

Fig. 6 K factors under biaxial loading.

Fig. 7 K ratios under Biaxial loading.

Fig. 8 Comparison of the K factors under uniaxial and biaxial loading.

Fig. 9 Comparison of the K ratios under uniaxial and biaxial loading.

Optimization According to Paris law [9-10], the crack propagation rate, da/dN, can calculated from

da m = c ( ΔK ) , dN

(1)

where a is the crack length, N is the number of cycles, ΔK=Kmax-Kmin is the amplitude of the stress intensity factor at crack length a, and c and m are material constants. Thus the fatigue life for the reference panel and the crenellated panel can be obtained from the integrals

N ref = ∫ dN = ∫

af

a0

1

c ( ΔK ref

)

m

da ,

(2)

818

M.V. Uz, Y.J. Chen, and N. Huber

N = ∫ dN = ∫

af

a0

1 c ( ΔK )

m

da ,

(3)

where a0 is the initial crack length, af is the final crack length and the material constants c and m for Al2139-T are 2.70e-7 and 2.60, respectively. The objective of the optimization is to find the maximum FCP lifetime for an uncracked crenellated panel, or the maximum remaining FCP lifetime for an already cracked crenellated panel with the initial crack length a0. Since the K distribution is dependent on the geometry of the crenellation, the fatigue life also depends on the geometry of the crenellation. With a dimensionless analysis, the objective function can be defined by ξ

=

N ( l ) -N ( a0 )

N_ref ( l ) -N_ref ( a0 )

= h ( a0 l , l1 l , l3 l2 , t3 t2 )

(4)

where ξ is the FCP lifetime ratio of crenellated to reference panel and a0 is the initial crack length. The geometry parameters l1, l2, l3, l, t2, t3 are given in Figure 10. In the following, the values of l, l0, t0 and t1 are fixed to 150.00mm, 5.00mm, 3.90mm and 2.90mm, respectively. The optimization will be focused on the parameters: socket width l1 , width of the thin section l2, half width of the thick section l3, thickness of the thin section t2, and thickness of the thick section t3.

Fig. 10 Geometry parameters of the crenellated panel.

The artificial neural network method (ANN) is employed to study the optimization of the crenellation patterns. ANN is a flexible mathematical structure, which is able to approximate complex nonlinear relationships between multiple input and output data. More details about this method could be found in [11]. A python program has been developed to generate 1180 random crenellation patterns. For each crenellation pattern the K factors and the corresponding FCP lifetime were calculated. These fatigue lifetime data served as database for the training and validation of the artificial neural network.

Fatigue Life Improvement of Metallic Aerospace Structures via Crenellations

819

The ANN C++ code FFNN developed during earlier work (see e.g. [12]) has been used here. In this work, an artificial neural network with 4 layers is created. The first layer (the input layer) has 6 input neurons, representing the dimensionless parameters a0/l, l1/l, l2/l, l3/l, t2/t1 and t3/t1. Both the second and the third layers have 6 hidden neurons. The 4th layer (the output layer) has a single output neuron for predicting the FCP lifetime ratio ξ (see Eq. (4)). The Mean Square Errors (MSE) for training and validation of the neural network are shown in Figure 11. The term “epoch” is used to define one increment of improvement of the ANN simultaneously for all training patterns (batch training). It can be seen from the figure that the artificial neural network approaches a minimum MSE value after 32000 epochs.

Fig. 11 Training.

Fig. 12 Quality of the training.

Figure 12 represents the quality for all patterns (training and validation) after the training. The 45° line (B) reveals a perfect fit indicating the excellent quality of the training. The trained neural network will be used in the following for a continuous interpolation the FCP lifetime in the space of the six dimensionless input parameters. This is an efficient approach for parametric studies (e.g. for plotting of relationships) and for optimization of the crenellation pattern. Optimization results parametrized with l1 and a0 In this part, the parameters l1 and a0 are kept constant. The maximum FCP lifetime ratio ξ is reduced as a function of l3/l2 and t3/t2 only:

ξ=

N ( l ) -N ( a0 )

N_ref ( l ) -N_ref ( a0 )

⎛l t ⎞ = h⎜ 3 , 3 ⎟ . ⎝ l2 t 2 ⎠

(5)

The results predicted by the ANN are presented in Figure 13 - Figure 16. The red point on Figure 13 corresponds to the already tested crenellation geometry (Figure 4). As can be seen from the figures, with increasing l1, FCP lifetime improvement that can be obtained by crenellations decreases. Additionally, with increasing l3/l2, the achievable FCP life extension decreases. Especially, with a higher value of t3/t2, the reduction in achievable FCP life extension with increasing l3/l2 becomes more pronounced. This is

820

M.V. Uz, Y.J. Chen, and N. Huber

a result of the equivalent thickness condition, i.e. for increasing width l3 of the thick section, its thickness t3 has an upper bound. Above this we have no solution for simple geometrical reasons and the fatigue life is limited to that of maximum t3/t2 ratio for a given value of l3/l2.

Fig. 13 l1=10, a0=0. The red point corresponds to the already tested crenellation pattern (Figure 4).

Fig. 14 l1=20, a0=0.

Fig. 15 l1=30, a0=0.

Fig. 16 l1=40, a0=0.

On the other hand, increasing t3/t2 increases the achievable FCP life extension. Furthermore, the maximum FCP lifetime increases with the decreasing of l1 and l3/l2. For a small value of l1 a variation around the optimum is less sensitive with respect to a reduction in the FCP lifetime. In order to achieve maximum FCP lifetime extension, l1 and l3/l2 should be as small as possible, while t3/t2 should be as large as possible. This study shows that the crenellations can extend the FCP life of the panels with the chosen value m=2.60 up to 44% just based on the non-linear relation between FCP rate and ΔK. The position of the red point on Figure 13, which locates the empirically designed crenellation pattern, demonstrates that ANN technique can improve the efficiency of the crenellations very effectively. Even more importantly it was seen that the minimum the FCP life ratio of crenellated

Fatigue Life Improvement of Metallic Aerospace Structures via Crenellations

821

and reference panels is always larger than 1 (Figure 13 – Figure 16). In other words, independent of the chosen parameters l1, l3/l2 and t3/t2, the crenellations always improve the FCP life.

3 K-Reduction Tests Based on the results of the optimization study and the initial test results of the panel containing non-optimized crenellations (Figure 3) it is clear that the nonlinear relationship between FCP rate and the ΔK cannot explain the effect of crenellations on FCP process alone. Obviously the loading history effect plays also an important role.

0.01

da/dN (mm/cycle)

Predicted and experimentally measured fatigue lives (stif. cren. panel) Al2139-T8, RT, B = 2.9 mm, C=2.74e-7, m = 2.60

Prediction

0.001 Test

Prediction Test (5.5L left) Test (5.5L right)

0.0001 0

25

50

75

100

125

150

175

a (mm)

Fig. 17 Comparison of the predicted and the measured da/dN vs. a curves of the stiffened crenellated panel.

By comparison of the experimentally measured and predicted FCP rates on the large panels, the retardation due to loading history can be easily identified. Figure 17 shows the da/dN vs. crack length ( a ) diagram for the crenellated M(T)740 panel containing 5 LBW stringers. This panel has an equivalent thickness of 2.9 mm and the initial crack length was 37 mm (for details see [1]). Right after the beginning of the test, the FCP rate observed on the panel started to deviate significantly below the Paris law prediction. At about 55 mm crack length, predicted and measured FCP rates started to get close to each other again. This trend continued until about 95 mm of crack length. After 95 mm, the Paris law again started to significantly overestimate the FCP rate. Revisiting Figure 2, it can be realized that both regions, where the measured FCP rate was clearly overpredicted by Paris law, correspond to the regions of strongly decreasing K-factor.

822

M.V. Uz, Y.J. Chen, and N. Huber

Under normal service conditions, it is expected that the average crack tip loading level increases with increasing crack length. Therefore, it is not common to make FCP tests under decreasing ΔK. Consequently, there is no extensive research in literature about the effect of decreasing ΔK on FCP, with the exception of fatigue growth threshold measurement tests. These tests are conducted to determine the ΔKth level under which an existing crack doesn’t propagate and thus are always performed using some form of ΔK reduction [13]. Difficulty of these tests is that as ΔK progressively decreases the crack tip may be affected from the plastic zone created by the former cycles. Due to the loading history effect the determined value for ΔKth may depend on the chosen form of K reduction. It was suggested to reduce the K-factor such that the fractional reduction of the plastic zone remains constant throughout the test to avoid such uncertainties [14];

1 ⎛ dry ⎞ ⎜ ⎟ = C' ry ⎜⎝ da ⎟⎠

(6)

Then the K-factor should follow the equation

ΔK = ΔK 0 e C ( a 0 − a ) ,

(7)

where C is equal to C’/2. In other words, the constant C controls the fractional change in monotonic plastic zone size during the test. The ASTM Standard [13] suggests the use of C =-0.08 mm-1 or higher to avoid loading history effects. This means that the size of the monotonic plastic zone should be reduced maximum 16% per millimeter of fatigue crack growth. It should be noted that the offered C value of -0.08 mm-1 is somewhat arbitrary and cannot assure a loading history free FCP process. The literature data on this aspect reveals contradicting results for different materials [15 - 17]. The C value offered by ASTM was shown to be nonconservative for Al2024-T3 [15] and for Al7075-T73 [16]. High strength titanium alloys on the other side, showed no loading history effect even for much higher K gradients [17]. As mentioned above, the general interest on the effect of the decreasing K on fatigue is related to the single value of ΔKth and not the FCP behavior in the Paris regime. For the determination of a reliable ΔKth, tests with constant Kmax seem to be more appropriate [18]. In such tests, instead of reducing Kmax and Kmin together, the ΔK is reduced by increasing Kmin. Consequently, most of the literature data is related to the constant Kmax tests, and the data on the influence of K reduction gradient under constant R ratio is limited. However, during the test of crenellated panels the R ratio remains constant, i.e., both Kmax and Kmin decreases or increases concurrently. Therefore, it was decided to further investigate the effect of decreasing K-factors on FCP rate for the material Al2139-T8 with a series of tests conducted on M(T)200 specimens. The specimen thickness was 2.9 mm. All the specimens had 15 mm long notches introduced by electro discharge machining. By loading with a rather low cyclic load (about one fourth of the starting load level during the test), 2.5 mm fatigue pre-cracks were initiated at both sides of the

Fatigue Life Improvement of Metallic Aerospace Structures via Crenellations

823

notches. In this way an initial crack length of 10 mm (2 a = 20 mm) was achieved. The tests were conducted using a servo-hydraulic test machine of 400 kN maximum capacity. Initial ΔK value was 36 MPa.m1/2 for all panels. K-reduction was applied continuously based on the crack lengths measured by compliance calibration. Concurrently a low magnification optical microscope was used to check the crack lengths. The electronic control system utilized was designed so that it was possible to manipulate crack length by adding an offset to the value, which has been determined by compliance calibration. When the optically measured crack length differed 0.2 mm from the value determined by compliance calibration, the latter was equated to the former by adding the corresponding offset. As M(T) type specimens were used for testing, in some of the cases an unsymmetrical crack growth was observed. When the crack length difference between two sides of the specimen exceeded %5 of the crack length, the test was discarded. Before the K reduction tests are started, a reference test with constant maximum load (Fmax) was performed. The initial ΔK at this test was 15 MPa.m1/2. Then K-reduction tests were conducted with C gradients of -0.02, -0.04, -0.08 and -0.16. The R-ratio was 0.1 for all the tests. Figure 18 shows the da/dN vs. ΔK result for these five tests. As can be seen, even with the very low K gradient of -0.02, the FCP rate was about 30% lower than the reference test. With decreasing C value (increasing gradient), the FCP rate was reduced, which demonstrates that the FCP behavior of the material Al2139-T8 is very sensitive to K reductions. Actually, the K-reduction profile on a crenellated panel is not regular. The constant C takes wide range of values between -0.35 and +0.7 at different regions of the panel (Figure 19). However, considering that even very slight gradients affect the FCP rate on this material significantly, it is obvious that the loading history effects contribute substantially to the crack retardation via crenellations. Consequently, in order to determine the real potential of the crenellations by a numerical study, instead of the Paris law, a model which takes the loading history effects into account should be preferred.

824

M.V. Uz, Y.J. Chen, and N. Huber 0.01

da/dN (mm/cycle)

Al2139-T8, M(T)200 R = 0.1, B = 2.9 mm

0.001

reference C = -0.02 C = -0.04 C = -0.08 C = -1.16

0.0001 10

100 1/2 ΔK (Mpa.m )

Fig. 18 Comparison of the da/dN vs. ΔK curves of the M(T) panels tested under different K-reduction rates. The pink curve shows the reference panel test result, which was tested with constant maximum load. 0.8

K reduction constant C for the 2.9 mm thick crenellation pattern 0.6

-1

C (mm )

0.4

0.2

0 C = -0.02 -0.2

C = -0.16

-0.4 0

25

50

75

100

125

150

a (mm)

Fig. 19 The K-factor reduction constant C for the crenellation pattern with 2.9 mm equivalent thickness.

4 Conclusion and Outlook Coupling finite element analysis and artificial neural networks methodologies an optimum crenellation pattern was developed which takes the advantage of the non-linear relationship (Paris law) between stress intensity factor range and the

Fatigue Life Improvement of Metallic Aerospace Structures via Crenellations

825

fatigue crack propagation rate. It was shown that just based on this retardation mechanism, it is possible to extend the fatigue crack propagation life of the selected panel geometry by 45%. As the former test results revealed a higher value of (65%) life extension in spite of the non-optimized crenellation pattern, it is obvious that loading history effects in areas with negative stress intensity factor gradients play an important role on the fatigue crack propagation behavior of the crenellated panels. In order to investigate the extent of the loading history effects arising from negative stress intensity factor gradients on the material Al2139-T8, a series of tests were performed with small scale specimens. Test results revealed that the fatigue crack propagation behavior of this material is highly sensitive to stress intensity factor gradients and even a gradient constant of as high as -0.08 mm-1 reduces the crack growth rate significantly. Based on this result it became clear that the loading history effects should also be taken into account in the optimization process to utilize the full potential of the crenellation methodology. In this context, instead of the Paris law, a new model which can reflect the loading history effects resulting from gradual plastic zone size reduction should be employed during further optimization. This study is a part of a research program which aims the development of the crenellations concept to improve the damage tolerance of aluminum shell structures. The retardation of the fatigue crack growth is only one of the program targets. The whole concept aims to systematically improve the “tailorability” of damage tolerance of metallic airframes using different methodologies including crenellations, laser surface treatment, friction-stir surface treatment and prestraining before welding. Currently one focus of the program is the potential of crenellations to promote crack turning under biaxial loading conditions. The idea here is that in addition to crack growth retardation, a systematical variation of stress intensity factors can be utilized to drive the crack in a preferential orientation, which can impose a limit on the damage size. The research work is being conducted both numerically and experimentally on large scale panels. The results of the ongoing work will be published in a following paper.

References 1. Uz, M.V., Koçak, M., Lemaitre, F., Ehrstöm, J.-C., Kempa, S., Bron, F.: Int. J. of Fatigue 31, 916–926 (2009) 2. Suresh, S.: Fatigue of Materials, 2nd edn. Cambridge University Press, Cambridge (2004) 3. Skorupa, M.: Fat. and Frac. of Eng. Mat. and St. 21, 987–1006 (1998) 4. Shin, C.S., Hsu, S.H.: Int. J. of Fatigue 15, 181–192 (1993) 5. Paris, P.C., Tada, H., Donald, J.K.: Int. J. of Fatigue 21, 35–46 (1999) 6. Heerens, J., Schödel, M., Schwalbe, K.-H.: Ėng. Frac. Mec. 76, 101–113 (2009) 7. Rybicki, E.F., Kanninen, M.F.: Eng. Frac. Mec. 9(4), 931–938 (1977) 8. ABAQUS, Version 6.8.1, Dassault Systems Simulia Corp. (2008) 9. Paris, P.C., Gomez, M.P., Anderson, W.E.: The Trend in Engineering 13, 9–14 (1961) 10. Paris, P.C., Tada, H., Donald, J.K.: Int. J. of Fatigue 21, 35–46 (1999)

826

M.V. Uz, Y.J. Chen, and N. Huber

11. Kohonen, T.: Neural Networks 1 (1), 3–16 (1988) 12. Huber, N., Nix, W.D., Gao, H.: Proceedings of the Royal Society London A 458, pp. 1593–1620 (2002) 13. ASTM-E647-00, Test Method for Measurement of Fatigue Crack Growth Rates 14. Saxena, A., Hudak, S.J., Donald, J.K., Schmidt, D.W.: J. of Test. & Eva. 6, 167–174 (1978) 15. Saxena, V.K., Malakondaiah, G., Radhakrishnan: Eng. Frac. Mec. 49, 153–157 (1994) 16. Forth, S.C., Newman Jr, J.C., Forman, G.: Int. J. of Fatigue 25, 9–15 (2003) 17. Sheldon, J.W., Bain, K.R., Donald, J.K.: Int. J. of Fatigue 21, 733–741 (1999) 18. Clark, T.R., Herman, W.A., Hertzberg, R.W., Jaccard, R.: Int. J. of Fatigue 19, 177–182 (1997)

26th ICAF Symposium – Montreal, 1-3 June 2011 Fatigue Life Improvement Using In-situ Robotic Processes Zahi Hajjar and Benoit Leblanc L-3 Communications, Military Aviation Services (MAS), Mirabel, Canada

Abstract. Several repairs were developed by L-3 MAS on the CF-18 aircraft (Canadian Forces F/A-18) to remove fatigue damage at localized regions. These repairs generally consist in polishing or blending the component surface to remove cumulated fatigue damage and improve surface finish. Additionally, in some cases, shot-peening is used after surface renewal to improve the fatigue life. Currently, no clear data or methodology exists to allow accounting for the benefit of surface renewal in analytical fatigue life predictions. Additionally, due to variability of the improvement provided by peening, no generic life improvement factors exist for shot-peening. Certification of surface renewal and peening modifications is usually completed using coupon fatigue testing specific to the location studied. This paper summarizes some of the work completed in order to provide tools and data to account for surface renewal without requiring specific coupon programs. Robotic surface improvement technologies developed and summarized in this paper have allowed obtaining more consistent and more predictable surface improvements. Finally, a coupon program has shown the benefit of surface renewal after several levels of fatigue exposure.

1 Introduction In the framework of structural repair and overhaul, engineers need to address fatigue related issues at localized areas such as flange radiuses, web tapers, or kick points. Repairing these highly stressed locations often involve material removal to eliminate damages such as dents, cracks and detrimental surface finishes. After this removal, life extension processes are needed to restore or even increase the original fatigue life of the location. One life extension technique typically used is shot-peening. Such techniques were used quite successfully at several locations on the CF-18 aircraft. However, concerns were raised about the consistency and reliability of conventional manual shot-peening methods, especially on parts of complex geometry and where human access is limited. For these reasons, robotic shot-peening methods were developed at L-3 MAS in the last years. Also, in order to increase productivity and consistency through the use of a single production cell, material removal and crack chasing techniques using robotic technology were also developed more recently.

828

Z. Hajjar and B. Leblanc

Some of the locations where these robotic techniques were used include the wing fold shear tie. Indeed, robotic machining was used in order to remove cracks and re-profile the wing fold aft shear tie lug on the 68% spar shown in Figure 1. Finite element modelling of the blended region had shown that the post-blend stress level was very sensitive to the final geometry. Robotic blending has therefore allowed removing existing cracks while providing the exact final geometry required.

Fig. 1 Aft Wing Fold ShearTie Lug.

Another successful application of the robotic surface treatment was on the Y453 wing carry-through bulkhead. Indeed, a web taper area was known to be a potential fatigue issue and shot-peening was seen as an adequate corrective action. However, this area could not be accessed manually as it was located inside tank # 3. Therefore, shot-peening of this region was completed using the robotic technology developed. In this case the robot was inserted upside down through the access hole. Finally, a third location where this technology was used is at the inboard leading edge flap (ILEF) lugs. Indeed cracks were seen at the base of the lugs in the radiuses as shown in Figure 2.

Fig. 2 Inboard Leading Edge Flap (ILEF) lug cracks.

Fatigue Life Improvement Using In-situ Robotic Processes

829

The modification consisted in blending the cracks and re-profiling the lug radius region. Blending also allowed improving the surface finish which initially was an Ion Vapour Deposited (IVD) aluminium coating for corrosion protection. Researchers [1] have clearly demonstrated in the past ten years the detrimental effect on fatigue of the etching process used prior to the deposition of the IVD coating on low stress concentration structural features. After blending is completed, shot-peening is used in order to improve the fatigue life at this location. Initially, the blending operation was completed using a 5 axis Computer Numerical Control (CNC) machine and the ILEF was then transferred to a robotic shot-peening cell in order to complete the peening process. In order to reduce production time and costs, a robotic cell was developed recently in order to provide machining as well as shot-peening capabilities with a single robot for the ILEF. The production cell is presented in Figure 3. The main components of the cell are a robot and its controller (1 and 4), a mobile peening machine (2), a tool and cutter stations (9 and 10), a personal computer (6) and a holding fixture for the ILEF (7). The computer is the master controller for all the robotic system components as well as an interface tool for the operator. Software was designed in order to allow a technician with no specific background in robotics to operate the system.

Fig. 3 Robotic cell for machining and peening of the F/A-18 ILEF.

830

Z. Hajjar and B. Leblanc

2 Benefits of Surface Renewal on Fatigue Life Surface renewal is defined here as any mechanical process that will remove a certain amount of surface material on a component. Surface renewal may range from a light polish to a deep blend. Surface renewal has a beneficial effect on the fatigue life of a component for three main reasons: 1) Stress concentration change: surface renewal allows reprofiling the geometry in order to decrease the stress concentration or to move the peak stress location away from the critical location. 2) Surface finish improvement: surface renewal allows improving the surface finish, particularly by removing the detrimental effect of preIVD etching or anodizing which can be found on the CF-18. Indeed, the etching process completed prior to the IVD process creates etch pits which act as preferential initiation sites for fatigue cracks (Reference [1]). Anodizing (mainly sulphuric) is also known to create a brittle surface layer which decreases fatigue life. 3) Removal of fatigue damage: surface renewal allows removing the surface material which has been subjected to fatigue damage due to previous in-service loading. The stress concentration change (effect number 1) is usually easily catered for by conventional stress analysis such as handbook stress concentration factors or finite element models. The surface finish improvement (effect number 2) can be isolated by comparing the lives of as machined coupons and pre-IVD etched or anodized coupons. Surface finish improvement can be accounted for in analytical fatigue life predictions by applying a life improvement factor derived from these coupon tests. The effect of the pre-IVD etching has been studied in details by DSTO (Reference [1]). For the removal of fatigue damage (effect number 3), the only analytical method available to account for this effect is Miner’s rule. When assessing the life remaining after incorporation of a modification (called the post-mod life) one must determine the amount of fatigue damage already cumulated in the part and the analyst must be able to assess the effect of the surface renewal on the postblend life. As stated earlier, the only analytical method available is the use of Miner’s rule. An example of applying current methodologies is provided with the geometry shown in Figure 4. The original part geometry is shown by the solid lines. After blending, the new surface is defined by the dashed line that passes through point A. In order to determine the life after blending of this component, the following steps are currently followed.

Fatigue Life Improvement Using In-situ Robotic Processes

831

Fig. 4 Illustration of pre-blend and post-blend geometry.

Two finite element models are built: one with the original part geometry and one with the post-blend geometry. The stress level at point A is determined for the pre-blend and post-blend configurations. Using a stress-life fatigue curve, the lives at these two stress levels are determined and will be called: NA_pre-mod and NA_post-mod. Finally using Miner’s rule one can write:

n pre−mod N A _ pre−mod

+

n post −mod N A _ post − mod

=1

(1)

where npre_mod is the number of cycles or hours spent in the pre-mod condition and npost_mod is the remaining life in the post-blend configuration. Specific coupon programs completed by the CF and L-3 MAS as well as test data found in the literature have shown that surface renewal could provide a significant life extension which cannot be predicted using Miner’s rule. Such a rule results in highly conservative life assessments. Indeed, reference [4] has shown the benefit of surface renewal on Ti-6AL-4V alloy. In this study, specimens have been pre-cycled for different percentages of the baseline CI life (up to 90%). After a removal of a surface layer of 100 μm (0.004”) it was shown that the fatigue life of the specimens was almost restored. Additionally, reference [5] shows a study where specimens are cycled for approximately 60% of the baseline life to failure and then 0.75 mm (0.030”) of material is removed from the surface of the specimens. Subsequent cycling (60% of the baseline total life) is repeated followed by another surface removal. This procedure was repeated 3 times without any specimen failure. Every specimen remains intact after twice the number of cycles to failure with two surface removal operations. This validates that removal of the outer surface does increase the fatigue life. However it should be noted that the removal depth is quite large and that the stress level was kept constant after cross-section reduction due to surface removal. In other words, load applied on the specimen was decreased after each surface renewal operation.

832

Z. Hajjar and B. Leblanc

Similarly, Reference [2] presents a repair method where 128 μm (0.005”) of material depth was removed and was followed by a shot peening process. The total life obtained after a single or multiple reworks at various levels of pre-cycling is shown. It appears that a mid-life rework process can restore life several times. Finally, the Canadian Forces (CF) have completed a coupon program for a modification on the CF-18 Web taper for the Y453 bulkhead (Ref. [3]). In this case, pre-IVD etched coupons of 7050-T7451 aluminum were pre-cycled for 3,900 Flight Hours. Then, some coupons were shot-peened and some others were polished by removing 75 μm to 150 μm (0.003”-0.006”) and then shot-peened. A Life Improvement Factor (LIF) of 18.9 is obtained for shot peening only. However, for the polished and shot peened coupons, a LIF of 27.9 is obtained. Therefore, there is clear evidence that surface renewal can provide a significant life improvement and even provide a full fatigue reset in some cases. However, as of today there is no specific methodology available to the analyst to account for the beneficial effect of the surface renewal process when predicting the remaining fatigue life of a component. A coupon program was therefore planned in order to gather generic data on the benefit of surface renewal on fatigue damage removal (effect number 3).

3 Experimental Program As discussed in the previous section, surface renewal affects fatigue life due to three distinct effects. The objective of this study is to determine the fatigue damage removal (effect # 3) caused by surface renewal. However, one of the challenges we face when doing surface renewal on a coupon is that the three effects are always combined. Therefore, a strategy to isolate the third effect is presented herein and illustrated in Figure 5. First, a baseline test series is tested with the original geometry and the original surface finish. This series will provide the pre-modification life. This will allow determining the percentage of fatigue damage incorporated by pre-cycling in the upcoming series. The next two series called series A and series B are also manufactured with the original geometry and original surface finish representative of the part tested. Series B represents the typical process of a part that has a surface renewal mod after a certain level of fatigue exposure. Therefore, series B will be pre-cycled to a certain fatigue level and then blended. Finally, fatigue cycling is resumed and the post-blend life is measured. Series A will be completed similarly to series B but without any pre-cycling. Therefore, after blending both series will have the same geometry, the same surface finish but one has some cumulated fatigue damage and the other one has never seen any fatigue. By comparing the post-blend fatigue lives of both series, one can evaluate the benefit of surface renewal on providing a certain fatigue reset. The post-blend life of series B will be equal or lower than the life of series A. If the lives of both series are equal, then the surface renewal has removed all fatigue damage and a full fatigue reset has been obtained.

Fatigue Life Improvement Using In-situ Robotic Processes

833

Fig. 5 Strategy to study effect of surface renewal on fatigue reset.

Test Specimens A quick survey of surface renewal modifications developed on the CF-18 fleet has been completed to determine representative notch geometry for the coupon. It was determined that a medium–low net stress concentration factor (Ktn) value of approximately 1.5 would be selected. The notch radius is chosen to provide a lowmedium stress gradient zone. To prevent failure in the grip area, a stress ratio of at least 3 is selected between grip section and peak stress at notch. The final specimen geometry selected is shown in Figure 6. A detailed Finite Element Model was built in order to derive the stress concentration factor associated with the coupon geometry presented in Figure 6. It is seen that the net stress concentration factor is 1.44. All coupons were manufactured from a 7050-T7451 aluminium (AMS 4050) plate. 150

50

Longitudinal

R19 TYP

25

Long Transverse

300 6

Fig. 6 Specimen geometry (all dimensions in mm).

Test Procedure All specimens were pre-IVD etched following the manufacturer’s process specification. Etching time was adjusted in order to obtain the same etch pit depth typically seen on the aircraft structure. Two different loading types were used in order to determine if the loading has an impact on the fatigue removal by surface renewal. Most test series in the

834

Z. Hajjar and B. Leblanc

program are tested under constant amplitude loading with a net section stress from -86 MPa (-12.5 ksi) to 287 MPa (41.7 ksi). Few series were tested under a spectrum typical of the CF-18 fuselage down bending moment. Net section stress levels in the spectrum range from -105 MPa (-15.3 ksi) to 354 MPa (51.4 ksi). Surface renewal was completed on the notch surface on all coupons. However, some coupons have had an additional polishing made on the side surfaces of the coupon. Notch surface and side surfaces are shown in Figure 7. Coupons where renewal is only done on the notch surface represent cases where a part cannot be blended on all surfaces since it is mated to another part. Several renewal depths were used as shown in next sub-section in order to establish the effect of the depth of material removed on the fatigue life recovery.

Notch surface

Side surface (front or back)

Fig. 7 Surfaces where renewal was completed.

Test Matrix Table 1 presents the test matrix for the constant amplitude series on the left. Series C.A.1 provides the life of the baseline geometry and surface finish. Similarly, the table on the right presents the test matrix for the variable amplitude series. Series V.A.2 is the baseline (or pre-mod) test series. Note that each series contains 3 to 7 specimens for the same test condition for a total of 120 specimens for the program. Table 1 Test Matrix for constant amplitude and variable amplitude series. Series

Surface Renewal Depth

Pre-Cycling

C.A. 1 C.A. 2 C.A. 3 C.A. 4 C.A. 5 C.A. 6

None 75 μm (notches) 75 μm (notches) 75 μm (notches) 75 μm (notches) 75 μm (notches) 75 μm (notches) 75 μm (front and Back Surfaces) 0.38 mm (notches) 0.38 mm (notches) 75 μm (front and Back Surfaces) 0.38 mm (notches) 75 μm (front and Back Surfaces) 0.38 mm (notches) 0.76 mm (notches) 1.5 mm (notches) 75 μm (front and back surfaces)

None None 4,000 cycles 4,500 cycles 5,000 cycles 5,500 cycles

C.A. 7 C.A. 8 C.A. 9 C.A. 10 C.A. 11 C.A. 12 C.A. 13

5,500 cycles

Series

Surface Finish

V.A. 2 V.A. 3

Pre-IVD Etched As Machined1

Surface Renewal Depth None None

V.A. 4

Pre-IVD Etched

75 μm (notches)

V.A. 5

Pre-IVD Etched

V.A. 6

Pre-IVD Etched

V.A. 7

Pre-IVD Etched

None None 5,500 cycles 5,500 cycles 6,500 cycles 10,300 cycles

75 μm (notches) 75 μm (front and Back Surfaces) 75 μm (notches) 75 μm (front and Back Surfaces) 0.38 mm (notches) 75 μm (front and Back Surfaces)

PreCycling None None None 7,498 SFH 9,128 SFH 9,128 SFH

Fatigue Life Improvement Using In-situ Robotic Processes

835

Test Results In order to analyze the test results, the expended fatigue life of each series is first defined. First, the pre-cycling level was defined as a percentage of the log average CI life of the baseline series. However, since the data is to be used in a safe-life philosophy context, it is seen that it is more relevant to define the level of precycling as a function of the safe life of the baseline series. This safe life is obtained by dividing the log average CI life of the baseline series by a scatter factor that accounts for the scatter measured in the sample tested and allows defining the life at a cumulative probability of failure (CPOF) of 0.001. This approach will allow us to account for scatter differences seen between coupon testing and actual fleet failure data. In parallel, the remaining life (or post-mod life) of each tested series is determined using the strategy shown in Figure 5. Indeed, the ratio of the CI life of the pre-cycled and blended series (series B) over the CI life of the series blended without pre-cycling (series A) is calculated: Remaining Life Ratio =

Post − Blend Life of Pre − Cycled Series Post − Blend Life of Non Pre − Cycled Series

(2)

This ratio is expected to be between 0 and 1. When a full reset is achieved, the remaining life ratio should be equal to 1.0. One should note that this ratio has been calculated using the safe life values for each series (log average CI life divided by the scatter factor of the same series). Such an approach allows to account for a change in scatter with surface renewal after fatigue exposure. Indeed with such an approach, even if two series (one pre-cycled and one without pre-cycling) have the same log average CI life, the remaining life ratio will be smaller than 1 if the series with pre-cycling exhibits a larger scatter than the series not pre-cycled. Figure 8 presents a graph of the measured remaining life ratio as a function of the expended fatigue life for all test series tested. If no fatigue recovery is achieved after blending, the sum of the expended life and remaining life ratios should be equal to 1.0, per Miner’s rule. This is the solid line shown on the same graph. One can see that all data points are on the right hand side of the Miner’s rule curve showing a certain fatigue life recovery after blending. The two charts of Figure 8 have the same Y axis. However the expended life on the x-axis is presented in two different ways. The chart on the left of Figure 8 presents the pre-cycling on the x-axis as a function the CI life of the baseline series. In other words, this represents the expended fatigue life at the peak stress location shown by point B in Figure 4. One can see that the surface renewal effect is as expected a function of the blend depth. The deeper the blend, the further the point is from the Miner’s rule curve. However, on the chart on the right of Figure 8, the expended fatigue life was calculated at the depth of blend location (point A in

836

Z. Hajjar and B. Leblanc

Figure 4). This was done by using an analytical stress life curve and results from a finite element model in order to determine the stress gradient at the notch location. One can see that by presenting the expended fatigue life at the depth of blend, then all data points line up and a simple relationship can be derived between the remaining life ratio and the expended life. This relationship is valid for any blend depth and is presented as a dashed line in Figure 8.

110%

110% Miner's rule (no reset) C.A. 0.075 mm notches (CA4, CA6) C.A. 0.075 mm notches and surfaces (C.A.7) V.A. C.A. C.A. V.A. C.A.

Remaining Life Ratio (Based on factored CI)

90%

80%

100%

0.075 mm notches and surfaces 0.38 mm notches 0.38 mm notches, 0.075 mm surfaces 0.38 mm notches, 0.075mm surfaces 0.76 mm notches

Remaining Life Ratio (Ratio of factored CI lives)

100%

C.A. 1.52 mm notches, 0.075 mm surfaces 70%

60%

50%

40%

30%

20%

90%

80%

70%

60%

50%

40%

30%

20%

10%

10%

0% 0%

0% 0%

20%

40%

60%

80%

100%

120%

140%

Expended CI Life at the peak stress location (In percentage of the factored CI life)

10%

20%

30%

40%

50%

60%

70%

80%

90% 100% 110%

160%

Expended Life at the Depth of Blend (In percentage of the factored CI life)

Fig. 8 Results of surface renewal test program.

Based on the results obtained, it was seen that the fatigue reset provided was not significantly different for constant amplitude or variable amplitude loading. The pre-mod fatigue damage calculated as shown in this paper is a unifying parameter to determine the remaining fatigue life, regardless of the type of loading used. Life predictions for the tested series were done using Miner’s rule and using the relationship shown above in Figure 8. Results of both life predictions compared with the experimental data are shown in Figure 9. The y axis presents the ratio of the safe life obtained for the blended coupons (including pre-cycling) over the safe life that would have been obtained if no surface renewal would have been completed. One can see that experimental data shows in most cases a great life improvement. Miner’s rule only provides a slight improvement caused by the lower stress level at the blend location. The new rule is still mostly conservative but less than Miner’s rule.

Fatigue Life Improvement Using In-situ Robotic Processes

837

Safe Life including mod / Safe-life withou mod

1.80 1.60 1.40 1.20

Experimental

1.00

Predicted (New rule) Predicted (Miner)

0.80 0.60 0.40 0.20 C.A.7

V.A.5

V.A.6

C.A. 10

V.A. 7

C.A. 13

C.A.4

C.A.6

C.A. 11 C.A. 12

Fig. 9 Life prediction results.

4 Conclusions Various successful applications of in-situ robotic surface treatments were developed on the CF-18. However, predicting the benefit of a material removal modification is often a challenge. This paper presents the results of an experimental program showing the benefit of blending aluminum parts after several levels of fatigue exposure. Based on the data available, a methodology to establish the benefit of a surface renewal modification is provided. The methodology provided is a simple tool valid for safe-life calculations and should be limited to the range of parameters covered by this data. A probabilistic approach to surface renewal issues would represent a good improvement but would require great efforts to implement. In conclusion, it can be stated that robotic surface improvement technologies have been developed to address specific aircraft structural issues. Coupon tests, component tests and in-service data have shown that these techniques provide efficient solutions to address military and commercial aircraft structural issues.

Acknowledgments We would like to thank the Department of National Defence for funding and supporting the surface renewal work. Special thanks to Capt. David Chown and Major Sébastien Thibault for the support they have provided us. We also would like to acknowledge the work of Marc Sova and his colleagues from the Quality Engineering Test establishment (QETE) for preparing and testing the specimens.

838

Z. Hajjar and B. Leblanc

References [1] Molent, L., Barter, S., Main, B.: Life assessment and repair of fatigue damaged high strength aluminium alloy structure using a peening rework method. Eng. Fail Anal. 15(1-2), 62–82 (2008) [2] Sharp, P.K., Liu, Q., Barter, S.A., Baburamani, P., Clark, G.: Fatigue Life Recovery in Aluminium alloy aircraft structure. Fatigue Fract Engng. Mater Struct. 25, 99–110 (2002) [3] Kioua, H., Forgues, S.: In-Situ Robotic Shot Peening for the Fatigue Life Improvement of Aircraft Structures. In: Proceedings of the 2003 Aerospace Technology and Innovation Conference, Canadian Aerospace and Space Institute, CASI (2003) [4] Takemoto, T., Jing, K.L., Tsakalakos, T., Weissmann, S., Kramer, I.R.: The importance of Surface Layer on fatigue Behavior of a Ti-6Al-4V Alloy. Metallurgical and Materials transactions A 14(1), 127–132 (1983) [5] Jeelani, S., Scott, M.A.: How surface damage removal affects fatigue Life. Int. J. Fatigue 10(4), 257–260 (1988)

26th ICAF Symposium – Montreal, 1-3 June 2011 Investigations into the Fatigue Enhancement Provided by the Hole Cold Expansion Process Using Accurate 3D FEA Simulations and Fatigue Testing S.J. Houghton, S.K. Campbell, and A.D. James Defence Technology Agency, Auckland, New Zealand [email protected]

Abstract. Enhancing the fatigue performance of aging aircraft structures is of significant concern for military and civil operators worldwide. One such method involves cold expanding fastener holes to exploit residual compressive stresses in the region surrounding the holes. The beneficial effect derived from this process depends on the magnitude and distribution of the residual stress surrounding each hole, therefore accurate identification of residual stress profiles is critical to the evaluation of the life of aircraft structure containing cold expanded fastener holes. A 3-D Finite Element (FE) simulation of the hole cold expansion process was created. Advances in FE technology allowed this simulation to closely represent the physical expansion induced by the commercial split-sleeve process. The simulation included a post-expansion loading step that applied a remote tensile load to the cold expanded hole to estimate the interaction of the residual stresses and stress concentration of the open hole. The FE simulations indicated a significant 3-D variation of the residual stress field, with notable variations through the thickness of the specimen. The magnitude of compressive residual stress was lowest at the mandrel entry face for the cold expansion process. It increased to a maximum at the mid-plane of the specimen, before decreasing to an intermediate value at the mandrel exit face. A threshold value of remote tensile load was identified, below which the residual stress field remained compressive at the bore of the hole. Over this threshold, the application of a remote tensile load created tensile stresses at the fatigue critical plane on the bore of the hole. A constant amplitude fatigue testing programme was then conducted to ascertain the correlation of predicted residual stress fields and the fatigue properties of cold expanded fastener holes. For constant amplitude fatigue loading below the tensile threshold, small fatigue cracks initiated and then arrested, with no crack growth after 10 Million fatigue cycles. There was good correlation between the size and shape of these cracks and the residual stress field predicted by the FE simulation. For constant amplitude fatigue loading above the threshold, similarly shaped fatigue cracks initiate and continue to propagate. Fractographic analysis showed initial crack growth was from the mandrel entry face. Propagation within the zone of compressive residual stress zone was complex. Once cracks had passed through the residual compressive stress zone, they rapidly transitioned to corner cracks and then to through cracks, which propagated to failure.

840

S.J. Houghton, S.K. Campbell, and A.D. James

1 Introduction Enhancing the fatigue performance of aging aircraft structures is of significant concern for military and civil operators worldwide. One such method that has gained common acceptance involves cold expanding fastener holes [1]. This process creates a region of residual compressive stresses near each fastener hole. These compressive residual stresses can be exploited to enhance fatigue performance in two ways. Firstly, they lower the peak tensile stress of fatigue cycles at the fastener hole, thus delaying the formation of fatigue damage. Secondly, once a crack like feature has nucleated at the fastener hole, the compressive residual stresses apply an additional closure force to inhibit crack growth. A number of methods can be employed to cold expand a fastener hole. One method that has gained widespread usage in the aeronautical industry involves placing a lubricated steel sleeve in the fastener hole [2; 3; 4]. The sleeve contains a longitudinal split. An oversized tapered mandrel is then drawn through the sleeve (and fastener hole). This process, known as split sleeve cold expansion, has been successfully commercialised. The split sleeve process is shown diagrammatically in Figure 1.

Fig. 1 Schematic diagram of the FTI split sleeve cold expansion process. Figure taken from [2].

The effectiveness of the split sleeve cold expansion process has been extensively studied, both in terms of the residual stress field created and its effect on the fatigue performance of aerospace materials [1; 2]. The residual stress field remaining after cold expanding a hole is dependent on the non-linear stress-strain

Investigations into the Fatigue Enhancement

841

response of the material being expanded. Early analytical and numerical predictions of the residual stress field often employed a bi-linear stress-strain relationship combined with an isotropic hardening model. These analyses would also typically utilise 2D analysis techniques, employing plane stress conditions. Axisymmetric analyses were also conducted in an attempt to consider the 3D nature of the problem. In recent times, 3D finite element models have been developed to predict the residual stress field. Most of the axisymmetric models and early 3D finite element models assumed a constant radial expansion of the fastener hole. In constant radial expansion models, the surface of the bore of the fastener hole was considered to expand at the same constant rate during the expansion process [5; 6]. Historically, good correlation between the plane stress 2D predictions and the residual stresses at the middle plane of axisymmetric or 3D constant radial expansion models have been reported. The constant radial expansion models predict lower residual stresses at the surfaces of the component being expanded, however the residual stress profile predicted by these models is invariably symmetric about the mid-thickness plane. Test evidence suggests that fatigue failures from cold expanded holes typically initiate from the mandrel entry face. This suggests that the residual stresses are lower at the mandrel entry face [6; 2]. Testing of cold expanded fastener holes invariably demonstrates a significant increase in the fatigue life of the component being tested. Improvement ratios of between 3 and 10 on the baseline (non-expanded) holes are commonly reported [7; 2]. Experimental data suggests that the life improvement ratio increases as the magnitude of the applied cyclic loads decreases. These data suggest the existence of an endurance limit for cold expanded fastener holes. The apparent existence of an endurance limit for cold expanded holes means that very little test data (to failure) exists for cyclic loads near the apparent test limit. Rather, a significant portion of the test data was either conducted at high loads, or stopped at a “runout” criterion. Despite the extensive analytical and experimental studies into the effect of hole cold expansion on the fatigue performance of aerospace materials, little progress has been made in developing models that can accurately predict these effects. The lack of quantitative predictive models has resulted in it being difficult to attain regulatory approval to “take credit” for the effects of hole cold expansion without extensive fatigue testing programmes. Effectively, regulatory credit for the effects of hole cold expansion can only be achieved during a full scale fatigue test of the component containing the cold expanded fastener holes. The work described in this paper involves applying Finite Element simulations of the cold expansion process, in an effort to gain an improved understanding of the residual stress fields created by the process. Constant amplitude fatigue testing was conducted to ascertain if there is any correlation between the residual stress field and fatigue failure modes of cold expanded holes. The testing was conducted at load levels similar to the apparent endurance limit for the material tested.

842

S.J. Houghton, S.K. Campbell, and A.D. James

2 Finite Element Simulation of the Cold Expansion Process The process of cold expanding a dogbone fatigue test coupon was simulated using the commercial Finite Element Code ABAQUS 6.9EF. For computational efficiency, the simulation only considered the working section of the test coupon, and exploited symmetry so that only one quarter of the coupon was modelled. Using symmetry in this manner means that the effects of the split in the sleeve are not fully captured. However, it has been shown experimentally that while the split in the sleeve does affect the residual stress field generated, the residual stresses are maximal at a plane oriented 90o from the plane of the split. Moreover, the experimental results show that on this rotated plane, the residual stresses are essentially symmetrical. Given that the process specification for cold working calls for the split in the sleeve to be oriented on a plane 90o from the fatigue critical plane, it was considered acceptable to employ quarter symmetry conditions. For the simulations, the test coupon was assumed to be manufactured from Aluminium Alloy 7075-T6. The working section was a nominal 2 inches (50.8mm) and a thickness of 0.25” (6.35mm). The hole to be expanded was centrally located with a final diameter of nominally 0.25” (6.35mm). The mandrel was modelled as a non-deformable rigid body, and both one quarter of the sleeve and one quarter of the test coupon were modelled as deformable solid bodies, using 20 node isoparametric quadratic brick elements (ABAQUS element type C3D20R [8]). The simulation included contact between the mandrel and sleeve, as well as between the sleeve and test coupon. Symmetry constraints were applied to the appropriate planes of the quarter sleeve and quarter plate components. The reaction of the sleeve against the tooling was simulated by restraining the sleeve. An attempt to model the flare in the sleeve at the mandrel exit face (which reacts against the tooling nosecap) encountered significant computational instability. Therefore, the sleeve was modelled as straight. Finally one end of the plate was fixed in space. The final computational mesh and boundary conditions used in the FE model are shown in Figures 2 and 3. Historically, FE simulations involving complex contact have exhibited sensitivity to mesh density. Typically, the contact pressures have converged at a slower rate (with respect to mesh density) than direct stresses. A mesh convergence study was employed to verify that the final mesh is adequate to resolve both the direct stresses in the test coupon and the contact pressures between the mandrel, sleeve and test coupon. A number of different contact conditions were simulated. These conditions ranged from ideal frictionless sliding contact between all surfaces to inclusion of the effects of friction between the surfaces by specifying a co-efficient of friction. The cold expansion process was simulated by prescribing a displacement to the mandrel. The mandrel displacement rate was assumed to be slow (in comparison to the speed of sound for aluminium), so that each displacement increment in the solution sequence was considered as quasi-static. The final frictional co-efficient was selected so that the average reaction force for the mandrel was similar to the reported pulling force reported by the vendor of a commonly used commercial cold expansion tooling set.

Investigations into the Fatigue Enhancement

843

The simulation included both geometric (large displacement and full green strain tensor) and material non-linearity. ABAQUS supports a number of material constitutive relationships. The most common of which is an isotropic material that exhibits isotropic hardening. However, preliminary investigations into simulating the cold expansion process using a constant radial expansion indicated that isotropic hardening laws do not adequately capture the reversed (compressive) yielding characteristics exhibited during the expansion process. A kinematic hardening law was found to be more appropriate. ABAQUS does not easily cater for the input of piecewise linearisation of stress-strain curves for materials exhibiting kinematic hardening (however, the input of a piecewise linear isotropic stress strain curve is relatively straightforward) [8]. The stress-strain curve used in the FE analysis was derived from four test tensile tests of AA7075-T651 coupons. A bilinear approximation of the stress-strain curve employing a kinematic hardening model was used. The measured stress vs plastic strain curve and the bi-linear approximation used for the FE simulations are shown in Figure 4 [9].

Fig. 2 Final Finite Element mesh for the simulation of the cold expansion process. The inset shows the refined area surrounding the mandrel and sleeve.

844

S.J. Houghton, S.K. Campbell, and A.D. James z

Prescribed mandrel displacement in z direction

y Sleeve displacement fixed in z direction

y

Sy mmetry con stra int - n oda l disp lace men fixe d in x di rection o n both pla te a nd sl eev

Fully fixed surface - displacements fixed in x, y & z directions

x Symmetry constraint - nodal displacements fixed in y direction on both plate and sleeve

Fig. 3 Boundary conditions applied during the finite element simulation.

Investigations into the Fatigue Enhancement

845

Fig. 4 The measured strss v plastic strain curve for Aluminium 7075-T651 and the BiLinear approximation used in the analysis. Note that this curve does not include the elastic component of strain, which is characterised by the elastic modulus (E) of the material.

3 Fatigue Testing Programme The FE simulations were supplemented by a fatigue testing programme. A number of open hole dogbone fatigue test coupons were manufactured from a single sheet of AA7075-T651. The expansion of the holes was performed by certified aeronautical maintenance operators, using a commercial tooling set and process specification. The final step of the cold expansion process was to perform a final ream on the expanded hole. The fastener hole bores and the surface of the test coupon near the holes were lightly polished. The mandrel entry face for each of the cold expanded coupons was recorded. A series of constant amplitude fatigue tests was conducted using these coupons. The tests were conducted using an Instron 1345 servo-hydraulic test machine. The coupons were held in the test frame by aligned hydraulic wedge grips. Constant amplitude fatigue loads with a stress ratio (R) of 0.1 were applied to the coupons at a frequency of 10 Hz. After every 2000 fatigue cycles, a block of 1000 cycles at a stress ratio of 0.9 was applied to provide contrasting marker bands on

846

S.J. Houghton, S.K. Campbell, and A.D. James

the fracture surface. The applied marker load spectrum was based on the work of Barter and Wanhill [10]. Three maximum stress levels were selected for the test programme. These were selected to represent typical maximal stress levels for lower center wing spar caps for large military transport aircraft [11]. For such components, limit load conditions, typically induce gross section stresses in the order of 22 ksi (152 MPa). The selected load levels were 18, 22 and 24 ksi (124,152,165 MPa). Historically, fatigue tests of cold work coupons at low load levels did not cause fatigue failures [2]. Rather, the tests were halted at a runout condition. In an effort to attain failures, runout for this programme was initially defined as 40 Million fatigue cycles, however as testing progressed this was deemed excessive, and progressively reduced to 20 and finally 10 Million cycles. No fatigue failures were observed in any of the coupons that were subjected to between 10 and 40 Million cycles. Coupons that reached runout were placed in a universal tensile test machine and loaded (in tension) to static failure. All failed coupons were subjected to fractographic examination. The fractographic inspections included both optical and electron microscopy.

4 Finite Element Simulation Results The circumferential component of the residual stress field as a function of distance from the hole is shown in Figure 5. These profiles, generated at the mandrel entry, mid surface and mandrel exit faces are the residual stresses from the cold expansion process, without any applied tensile loads. A significant variation of the residual stress profile through the thickness indicates a strong influence of the direction of mandrel motion. The peak circumferential compressive stress was observed at the mid-thickness of the plate and the minimum occurred at the entry face. This indicated that the residual stresses at the entry face will provide the least fatigue enhancement. After the application of remote tensile loads, the magnitude of compressive residual stress at the fatigue critical plane was reduced. The circumferential component of the stress field near the fastener hole, after cold expansion and the application of three remote stress levels (18, 22, 24 ksi or 124,152,165 MPa) are shown in Figure 6. Clearly at an applied remote loading of 22 ksi (152 MPa), the circumferential component of the stress field is fully tensile, implying that any cyclic loading exceeding this level has the potential to induce fatigue damage. A comparison of the stress field contours after cold expansion with no applied load an applied remote loading of 24 ksi (165 MPa) are shown in Figure 7.

Investigations into the Fatigue Enhancement

847

Fig. 5 Circumferential Stress Distribution predicted by FE Simulation. The three curves show the Mandrel Entry face, mid surface and mandrel exit face residual stress distributions after the cold expansion process with no applied remote load.

848

S.J. Houghton, S.K. Campbell, and A.D. James

Fig. 6 Circumferential Stress Distribution predicted by FE Simulation at the mandrel entry face. The three curves show three different levels of remote load, no load, 22 ksi and 24 ksi.

Fig. 7 Contour plots of the circumferential stress field after the cold expansion process (Right) and after expansioon and a remote load of 24 ksi/ 165 MPa (Left). Note the there is a scale change between the two images. In both images the top face is the mandrel entry face.

Investigations into the Fatigue Enhancement

849

5 Fatigue Test Observations and Fractography Fatigue failures were not observed for coupons that were subjected to cyclic stresses with a stress amplitude of 18 ksi (124 MPa). Only one fatigue failure was observed at a stress amplitude of 22 ksi (152 MPa), with all other tests at this load level reaching runout. One of the tests conducted with a stress amplitude of 24 ksi (165 MPa) reached runout. All other tests at this load level failed at a fatigue crack. Coupons that reached runout were loaded in tension until failure and a fractographic investigation of these coupons was conducted. The fractographic investigation showed that in each case multiple cracks had initiated on both sides of the hole near the critical fatigue plane. These cracks grew at a highly retarded rate, and could effectively be described as having arrested at depths of between 0.0050.03” (0.127-0.762mm). Typical arrested cracks for each of the load levels tested are shown in Figures 8-10.

Fig. 8 Fracture surface of an arrested crack after the application of 40 Million fatigue cycles with a stress amplitude of 18 ksi/ 124 MPa. The upper surface is the mandrel entry face.

850

S.J. Houghton, S.K. Campbell, and A.D. James

Fig. 9 Fracture surface of an arrested crack after the application of 20 Million fatigue cycles with a stress amplitude of 22 ksi/ 152 MPa. The upper surface is the mandrel entry face.

Fig. 10 Fracture surface of an arrested crack after the application of 10 Million fatigue cycles with a stress amplitude of 24 ksi/ 165 MPa. The upper surface of the lower fracture surfaceis the mandrel entry face.

Investigations into the Fatigue Enhancement

851

For applied load levels of 24ksi (165 MPa), a crack would form at the bore of the hole having a similar profile to those observed in the non-propagating tests. However, at the mandrel entry face, the crack would propagate much like a corner crack. Crack growth rates were relatively low. Eventually as the portion of the crack near the surface extended beyond the residual stress field, it tended to grow more quickly than the crack growing along the bore of the hole. As soon as the defect near the hole bore reached the opposing surface providing a through crack condition, the defect then rapidly grew to failure. Figure 11, shows an SEM image of the fracture surface of one of the coupons that failed under fatigue loads. The initial crack front, final crack front and the crack front propagation path (as determined by observation of the marker bands) are annotated.

Fig. 11 SEM Image of the fracture surface for a coupon subjected to cyclic loads with a stress amplitude of 24 ksi. The initial crack front, final crack tip and locations of marker bands have been highlighted. The test coupon failed after the application of 1,485,000.

The fatigue tests conducted with a cyclic load level of 22 ksi (152 MPa) produced both propagating and non-propagating cracks. The non-propagating cracks exhibited all of the same features as those at the 18 ksi (124 MPa) load level. Similarly, the propagating cracks exhibited the same features as those at the 24 ksi (165 MPa) level.

6 Observations The FEA simulation indicated that an applied load of approximately 22ksi (152 MPa) represents a threshold where the residual compressive stress is eventually overcome by the tensile stress at the mandrel entry surface. This tensile stress field means that any cyclic component of loading above this level has the potential to develop and propagate fatigue cracks. The stress fields in the centre and at the mandrel exit face of the plate remained compressive until the crack had developed enough for load redistribution to occur, and crack growth from the exit face became possible. This load redistribution in the presence of cracks was not simulated in this study.

852

S.J. Houghton, S.K. Campbell, and A.D. James

One of the aims of this study was to ascertain if there is a correlation between the stress fields predicted by finite element simulation of the cold expansion process (and the subsequent application of remote loading) and the fatigue properties of the cold expanded hole. The best evidence of such a correlation is shown in Figure 12. In this image, the circumferential stress field for a cold expanded hole with an applied remote load of 24 ksi (165 MPa) has been overlaid on the SEM image of the fracture surface from a test coupon that was subjected to cyclic loads with an amplitude of 24 ksi (165 MPa). The successive form of the crack fronts can be related to the magnitude of the local stress field. The effect of the restraint provided by the residual compressive stress in the material near the hole bore is clearly evident in the early phases of crack growth.

Fig. 12 SEM Image of the fracture surface for a test with a stress amplitude of 24 ksi/165 MPa overlaid with the circumferential stress field predicted by the FE Simulation with an applied remote stress of 24 ksi/165 MPa.

7 Conclusions The stress distributions derived from the FEA and the fracture analysis results from fatigue testing have together provided insights into the behaviour of fatigue cracks in the presence of residual stress fields induced by hole cold expansion. This knowledge will assist in the ongoing management of aircraft structure containing cold expanded fastener holes. The FEA and fatigue testing confirmed the mandrel entry face as the critical fatigue location. Unfortunately for a number of aerospace applications this represents the least inspectable location. The testing results also confirmed the existence of an effective fatigue endurance limit. The theoretical implication from this

Investigations into the Fatigue Enhancement

853

study is that an aircraft component containing only cold expanded open holes could be taken to limit loads millions of times without forming large fatigue cracks. However, this work considered only open holes subjected to constant amplitude cyclic loading. In real aircraft, the fastener holes will be filled with a fastener that may transfer loads, and alter the internal load distribution. Fastener hole edge distances may also be less than optimum and surface damage near cold expanded fastener holes may provide sites for crack initiation away from the bore of the holes. Moreover, typical aircraft components are subjected to spectrum fatigue loads which may include substantial underloads causing local yielding to modify the residual stresses around cold expanded fastener holes. Further work using the tools described in this paper is needed to evaluate these variables using appropriate test cases.

References [1] McClung, R.C.: A literature survey on the stability and significance of residual stresses during fatigue. Fatigue & Fracture of Engineering Materials & Structures 30, 173–205 (2007) [2] Stow, K.: SPAR Aerospace Fatigue Improvement Modification Cold Expansion Research Reort for RNZAF C-130 Programme. Fatigue Technology Inc. FTI Technical Report # 195386 (2006) [3] Fatigue Technology Inc. FTI Process Specification 8101D - Cold Expansion o Holes Using the Standard Split Sleeve System and Countersink Cold Expansion, CsCX (2002) [4] West Coast industries. Engineering Handout, Split Sleeve Coldworking Holes. WCIEH-9201-4.1 [5] Houghton, S.J.: Finite Element Analysis of the Cold Expansion of Aircraft Fastener Holes. Auckland: Defence Technology Agency, DTA Report 296 (2010) [6] Kokaly, M., Ransom, J.S., Restis, J.H., Reid, L.: Observations and Analysis of Fatigue Crack Growth from Cold Expanded Holes. In: Proceedings of the 8th Joint NASA/FAA/DoD Conference on Aging Aircraft, Pam Springs, California (2005) [7] Zvyagintsev, V.: RNZAF-FIM3-01 Rev 0 Ch 0 Fatigue Improvement Modification Verification Data. L3 Communications Spar Aerospace (2007) [8] Dassault Systemes Simulia Corp. ABAQUS User Documentaion Version 6.9-EF (2009) [9] Houghton, S.J.: Identifying Accurate Material Sress Strain Curves for Non-Linear FEA Using Tensile Testing. Auckland: Defence Technology Agency, DTA Technical Note 2009/6 (2009) [10] Barter, S.A., Wanhil, R.J.H.: Marker Loads for Quantitative Fractorgraphy (QF) of Fatigue in Aerospace Alloys. National Aerospace Laboratory, Netherlands, NLR-TR2008-644 (2008) [11] RNZAF. Hercules Aircfrat C-130H Structural Repair Manual, NZAP 6211.001-3 (1997)

26th ICAF Symposium – Montreal, 1-3 June 2011 Characterisation of Fatigue and Crack Propagation in Laser Shock Peened Open Hole 7075-T73 Aluminium Specimens G. Ivetic1, I. Meneghin1, E. Troiani1, G. Molinari1, A. Lanciotti2, V. Ristori2, J.L. Ocaña3, M. Morales3, J.A. Porro3, C. Polese4, and A.M. Venter5 1

University of Bologna, Forlì, Italy 2 University of Pisa, Pisa, Italy 3 Polytechnic University of Madrid, Madrid, Spain 4 University of the Witwatersrand, Johannesburg, South Africa 5 South African Nuclear Energy Corporation, Pretoria, South Africa

Abstract. The goal of this research activity is to evaluate the capability of Laser Shock Peening (LSP) technology to improve fatigue life in open-hole aluminium specimens. Thin, dog-bone specimens were LSP treated in direct ablation mode and subsequently tested. The obtained results have not proven the advantage of LSP technology over traditional residual stress insertion techniques around openholes, such as cold working. Therefore, the focus of the activity was moved towards understanding the causes of the observed fatigue life reduction.

1 Introduction Research motivation The present research programme evaluates the capability of the Laser Shock Peening technology to improve fatigue life in aeronautical structures by introducing compressive residual stresses around fastener holes in thin-walled structures representative of typical aircraft components. A possible advantage of LSP is that it can treat a larger portion of material, potentially influencing the propagation of a crack, as well. The results of this experimental campaign will permit to compare LSP technology on 7075-T73 with cold working and stress wave technology on the same material [1]. Unfortunately, the limited number of available specimens (19 in total) made it difficult to evaluate in an exhaustive way the effect of the sequence of operations on the specimens. The idea was to perform LSP both on the open-hole specimens and on the pristine ones, where the hole would be realized after the LSP treatment: later on, a comparison between the compressive residual stresses would tell if and how much of these stresses were released by the drilling after the LSP treatment. The choice was to use the specimens with open hole + LSP operation sequence, because of suspected release of residual stresses in case of LSP + open-hole

856

G. Ivetic et al.

sequence of operations. It was deemed that LSP + open-hole set-up would potentially produce inferior fatigue behaviour. Laser Shock Peening Laser shock peening (LSP) is a relatively recent surface treatment technique, which has been successfully applied to improve fatigue performances of metallic components [2], [3]. The LSP treatment produces mechanical shock waves by the means of high pressure plasma (order of GPa) created on the surface of the treated specimen, as the consequence of high-power density laser irradiation. A detailed description of the process can be found in [4], [5]. Compared with the more traditional methods for insertion of residual stresses, such as shot peening, LSP can introduce compressive residual stresses underneath the surface of the treated material that are several times deeper [2], thus having an advantage in the terms of longer nucleation and initial propagation periods of cracks present in mechanical components.

2 Experimental Activity and Results Specimens The specimens used for the present research were dog-bone specimens, obtained from 2.3 mm thick lamina of Al 7075-T73 (Figure 1) using a CNC machine. A total of 19 specimens was treated with LSP. Laser set-up When it comes to LSP treatment of thin aluminium specimens with open holes, a very limited number of published works exists [6], [7]. This may be due to the fact that high energy lasers that are generally used for the LSP treatment (usually applied on titanium and steel) are not suitable for treatment of thin aluminium sheets without previous optimization of the process. Therefore, in order to avoid detrimental effects of the treatment on an Al specimen, it is important to choose carefully the setup of the process. The laser used for LSP treatment for this experimental activity has relatively low output energy (the lasers from [6] and [7] are high energy lasers), and it is suitable for treatment of thin Al specimens.

Characterisation of Fatigue and Crack Propagation in LSP Open Hole

857

Fig. 1 Geometry of the specimen.

In addition, the previous experience of the researchers of Centro Laser of the Polytechnic University of Madrid in the application of LSP technology on Al alloys [8], [9] made it possible to identify the right parameters for this research activity. Selection of laser parameters In LSP treatment, the characteristics of the introduced residual stresses depend on the laser parameters. It is important to define: Laser power density - compressive stresses increase with the increase of laser power; Laser wavelength - different peak pressures are developed at different wavelengths; Peen size - superficial residual stresses increase with the size of the impact; Pulse Duration - reducing the pulse duration reduces the depth of the compressive residual stresses through the thickness; Overlapping of laser peens - increasing the number of layers increases the value of compressive residual stresses; Coating – the treatment can be performed with or without the use of a protective coating. The laser used for this research activity has the characteristics summarized in Table 1: Table 1 Laser properties used for LSP treatment.

Laser type [-] Nd-YAG

Wavelength [nm] 1064

Output energy [J] 2.8 (10% loss)

Output energy [ns] 9

Laser frequency [Hz] 10

Considering that the values in Table I are fixed, the only parameters that can be varied were peen size, the presence of protective coating and the overlapping of peens.

858

G. Ivetic et al.

Peen size. The spot size chosen was 1.5 mm in diameter. There are two main reasons for using such a small spot: • •

In order to obtain relatively high laser power densities (expressed in the terms of GW/cm2) necessary for successful LSP treatment, small beam spot surface is required, given the relatively low laser energy. The absence of a protective layer (coating) does not permit the use of large spots (with the same quantity of power density) because this could damage the surface of the treated specimen.

Coating. There are two basic approaches in LSP technology: direct (without absorbent coating [10]) and indirect ablation (with absorbent coating [6], [7]). The ablative layer has the function of protecting the surface of the treated specimen, since the direct exposition to the high temperature plasma can produce surface damage and increase surface roughness. The major difference between direct and indirect ablation lies in the fact that when no coating is used, the only way to avoid large surface damage is to use lower laser powers combined with very small impact size and high density of impacts, increasing the overlapping of the laser peen spots. Given the fact that the laser available had relatively small output energy, the best setup for the laser is the one with small peen diameter and consequently with a high overlapping rate. Therefore, it was decided to work in the direct ablation mode, since the great overlapping would increase both the costs and the time of the procedure if the protective painting needed to be applied after every shot. Overlapping. Given the relatively small peen diameter, it was necessary to set the appropriate density of laser peens (overlapping rate) in order to obtain significant residual stresses on the surface of the treated specimen. It was decided to verify two different settings: • •

625 spots per cm2 900 spots per cm2

Previous experimental activities on Al alloys performed at Centro Laser have shown that the optimum range lies between these two values: spot densities under 625 spots per cm2 would not introduce significant residual stresses in the material, while increasing the spot density over 900 spots per cm2 could cause excessive deformation of the specimen or even significant damage to the surface of the specimen. Both of the treatments were performed on the same specimen and on one side only, serving for purposes of residual stress measurement. In Figure 2, the specimen after the LSP treatment can be seen. It is interesting to observe in the figure the pattern of the laser beam together with the spot orientation (black lines and ellipsis), as well as the entering and exiting point of the beam. These points are a feature of the used laser and the non-homogeneity introduced is obviously undesirable for subsequent fatigue tests purposes. However, it can be easily avoided by starting and finishing the laser beam pattern outside the specimen.

Characterisation of Fatigue and Crack Propagation in LSP Open Hole

859

Fig. 2 Specimen used for process setup.

Residual stress measurement This specimen was used for hole drilling measurement of residual stresses. A possible problem with this method is that its use on thin specimens is limited and after certain depths, correction formulas need to be applied in order to obtain an accurate estimation of residual stresses present in the material. The results of the residual stresses measurement can be seen in Figure 3. The hook-shaped profile of residual stresses could be explained by: •

•

•

The LSP treatment was realized in a direct ablation mode, where no thermal protective material is used. In this case, there is a thermal effect on the surface of the specimen and as a result, tensile stresses may occur after cooling of the material reducing the amount of compressive residual stresses. The spot dimension and its circular shape affect the distribution of residual stresses. Such a small spot dimension creates a spherical shock wave (while a bigger one would create a planar one), with a complex shock wave interaction at the surface of the specimen that could affect the compressive residual stress in this area. This effect could be avoided or reduced by using a different spot shape (e.g. square); unfortunately with the laser at our disposal this was not possible. The accuracy of hole drilling method is limited when applied on thin specimens, therefore the measurement could be affected by this inaccuracy.

It can be noted that even though the setting with 625 spots per cm2 causes deeper compressive residual stresses than the 900 spots per cm2 setting (0.6 mm vs 0.45 mm), the maximum induced residual stresses are lower (229 MPa vs 332 MPa). It is important to consider that the hole drilling method is accurate only near the surface, so the measured depth of compressive residual stresses might be not so accurate, considering the low thickness of the specimen. Therefore, it is deemed better to base the settings on the value of the residual stress (closer to the surface), more than on the reached depth of compressive residual stresses. Given all these considerations, the set-up of 900 spots per cm2 was the one chosen for the treatment of remaining specimens.

860

G. Ivetic et al.

Fig. 3 Residual stresses introduced in the treated specimen measured using hole drilling method.

Fixing When a thin panel is laser peened, it is usually fixed to an immovable backing plate, in order to prevent the unwanted wave reflection on the back side of the plate. Moreover, thin specimens are subjected to deformation in the peened zone, so they must be fixed very rigidly to the backing plate in order to avoid vibrations. This rigid fixing can cause undesirable local variation in residual stresses introduced in the specimen. In order to avoid fixing problems without encountering undesired wave reflection, no backing plate was used in combination with a short impulse times that in fact ensure less shock reflection [11]. When treating the specimen on one side only, its bending was clearly visible (Figure 4a): this is obviously unacceptable. However, the specimens were treated on both sides in order to ensure the symmetry of the stress field. So, after treating on both sides, the specimen re-established its original shape (Figure 4b).

a) One side treatment

b) Two side treatment Fig. 4 Deformation of the LSP treated specimen.

Characterisation of Fatigue and Crack Propagation in LSP Open Hole

861

Roughness measurement Since it was decided to perform LSP in direct ablation mode, it is important to control the state of the surface after the treatment. The thermal effect could indeed damage the surface causing premature fatigue crack initiation and this effect is certainly to be avoided. Thanks to a confocal microscope, it was possible to obtain a three-dimensional map of the roughness of the specimen in the LSP treated zone and in the not treated one (Figure 5). There is an 82 μm drop in z-coordinate direction between the base material (38 μm) and the LSP treated zone (-44 μm). This was expected since the plasma pressure on the treated area reaches very high values (order of several GPa) and the consequent plastic deformation is relevant. In addition, the LSP treatment is performed in the direct ablation mode, so a certain quantity of material was removed during the process. Nevertheless the increase in roughness is of limited extension: from 1.1 to 3.6 μm which is surely lower than the roughness increase of a shot peened surface. That is why the negative effect of the increased roughness due to LSP treatment is expected to be limited as well. These statements are based on the observations reported in [12]. Table 2 gives a comparison between mean (Ra) and peak (Rt) roughness values: ones measured in this research activity and ones from [12]. The table shows that results relative to the roughness after LSP, obtained in this research, are comparable to the ones reported in the literature. The slightly higher values of the roughness measured in this research are probably due to different laser settings and shot overlay. Table 2 Comparison of roughness results.

Material condition 7075 as milled 7075 LSP 3 shots 7075 SP 125% 7075 as milled 7075 LSP 900 shots/cm2

Ra [μm] 0.6 1.3 5.7 1.1 3.6

Rt [μm] 5.2 11 42 7.9 15.6

Reference [12] Present experimental activity

Fig. 5 3D roughness map, the edge of the treated zone is shown.

862

G. Ivetic et al.

Fatigue Tests The specimens were prepared prior to fatigue tests by attaching additional tabs to the clamped zone of the specimen in order to avoid fretting fatigue damage. The maximum loads chosen for tests are selected in base of previous work [1] in order to be able to make a direct comparison of different technologies. However, the results relative to LSP technology have turned to be very disappointing, performing three times worse than the baseline in the terms of fatigue lives of tested specimens at σmax of 160 MPa and R=0.1. In order to exclude the roughness effects, the inner side of the hole was polished with diamond paste and the test was repeated, but the obtained result was practically the same. Finally, besides polishing the inner side of the hole, the entire LSP treated region was polished to as-milled state (measured Ra= 0.4 μm), but there was still no change between the results. Figures 6 and 7 show the surface state before and after polishing while Table 3 summarizes the results obtained.

Fig. 6 Inner side of the hole, before (left) and after (right) polishing.

Fig. 7 LSP treated zone, before (left) and after (right) polishing.

Characterisation of Fatigue and Crack Propagation in LSP Open Hole

863

Table 3 Fatigue lives with different open-hole treatments, at σmax 160 MPa and R=0.1

Treatment Baseline [1] LSP as treated LSP polished hole LSP polished treated zone Cold working [1] Stress Wave [1]

Fatigue life (cycles) 105 3.47x104 3.45 x104 3.21 x104 6.5x105 8.5x106

Difference 1x 0.3x 0.3x 0.3x 6.5x 85x

The fracture surfaces of the tested specimens are reported in Table 4, together with testing conditions (all tests at R=0.1) and fatigue lives obtained. Four out of five fracture surfaces show a clearly visible step at mid-thickness (black circles), which could possibly be attributed to a peak of tensile residual stresses present. The fatigue tests were stopped after eight tests due to clear negative trend of obtained results. It was decided that the remaining specimens, originally dedicated to propagation tests, should be used for additional residual stress measurements, in order to determine whether these peaks of tensile stresses were in fact the cause of the observed premature fatigue failure. Table 4 Fracture surfaces for different surface conditions.

As peened

Polished hole

Polished hole

Hole + LSP area polished

Hole +LSP area polished

σmax=200MPa Failure:14504

σmax=140MPa Failure:67276

σmax=160MPa Failure:34579

σmax=180MPa Failure:24135

σmax=160MPa Failure:32094

Additional residual stress measurements Last minute beam time has been granted at the synchrotron facility of Elettra Trieste in order to measure residual stresses at the edge of the treated hole. Even though the obtained results were not conclusive and additional measurements are necessary in order to draw more precise conclusions, some preliminary considerations can be illustrated. The qualitative comparison between the treated specimen and the base line suggests that treated specimen is more compressed in the surface, but this trend reverts in depth, where the treated specimen contains more tensile stresses than the base line. This would in fact explain the observed

864

G. Ivetic et al.

behaviour, but, as previously stated, these are just preliminary results and more complete picture can be obtained only after additional measurements, scheduled in the future.

3 Results Discussion Since the effect of fatigue life increase was not encountered, the focus of this investigation has moved towards understanding the causes of this result. Some of the possible causes of encountered fatigue life decrease are: • •

•

•

Increased roughness – to be excluded, since polishing the inner side of the hole and the entire LSP treated area did not produce any difference in fatigue lives. Laser power density of a too high magnitude – the measurement of residual stresses on the LSP treated area without the hole did not show any abnormalities or tensile stress peaks that might explain the premature failure of tested specimens. Spall damage – This phenomenon can occur when a shock wave, reflected from the free back side of the treated panel, intersects with the primary shock wave [13]. This effect can create a tensile zone in the intersection area and can cause material damage. However, this effect is probably to be excluded since laser pulse duration of 9 ns ensures reflections of elastic waves only. The sequence of operation open hole + LSP might have caused peaks in tensile stresses at the edge of the hole – Laser and plasma interaction in the presence of a sharp edge could have caused an effect of plasma lens [14], in which the laser beam passes through the created plasma and is being focused on the inner side of the hole. This phenomenon is illustrated schematically in Figure 8.

4 Conclusions The obvious result of this experimental campaign is that LSP technology, even if very efficient in treatment of metallic materials, can indeed introduce detrimental effects in treated specimens. It is therefore crucial to determine correct parameters for the process, taking into account geometric effects, as well. Additional consideration is related to the laser setup used; while LSP in [6] and [7] is performed using high energy lasers and big laser peens, it is important to remember that a low energy laser with small peen size and direct ablation mode has been used. From the results obtained in this research, the following conclusions can be drawn: •

LSP needs to be optimized for every application used, specially when it comes to low thickness specimens

Characterisation of Fatigue and Crack Propagation in LSP Open Hole

• • • •

865

In the investigated configuration, LSP turned to have detrimental effects to the treated specimen Additional investigations have excluded roughness effects as the cause of the premature fatigue failure Obtained results suggest that the tensile residual stresses are the cause of shorter fatigue lives, but this assumption needs to be confirmed by additional measurement Future investigations will be directed towards understanding the importance of the sequence of operations (first LSP and than open hole)

Fig. 8 Plasma lens effect. a) Ideally, laser beam partially hits the edge of the hole and partially passes through b) Plasma is created next to the edge of the hole c) The created plasma “spills” over the edge of the hole. d) Laser beam, being focused by “spilled” plasma towards the inner side of the hole, creates a local tensile hot-spot.

Acknowledgements The authors wish to acknowledge the support received from the European Science Foundation (ESF) within the activity “Super-intense laser-matter interactions”. Also, the authors would like to express immense gratitude to Dr. Andrea Lausi and Dr. Jasper Plaisier from Elettra Trieste Synchrotron MCX beamline for their great help received in residual stress measurement. Thanks go to Mr. Paolo Proli of the “MaSTeR Lab” laboratory of the University of Bologna for his assistance during the preparation of the specimens. Special thanks go to Mr. Francois Prinsloo of the CSIR National Laser Centre, South Africa, for his valuable advices and suggestions.

866

G. Ivetic et al.

References [1] Boni, L., Lanciotti, A., Polese, C.: Durability and Damage Tolerance of Aircraft Structures: Metals vs. Composites. In: Lazzeri, L., Salvetti, A. (eds.) Proceedings of the 24th ICAF Symposium, vol. II, pp. 832–845. Pacini, Pisa (2007) [2] Hammersley, G., Hackel, L.A., Harris, F.: Opt. Laser Eng. 34, 327–337 (2000) [3] Heckenberger, U., Hombergsmeier, E., Holzinger, V., von Bestenbostel, W.: Advances in Laser Shock Peening theory and practice around the world: present solutions and future challenges. In: Ivetic, G. (ed.) Proceedings of the 2nd International Conference on Laser Peening, pp. 22–33. Emerald Group Publishing Ltd, Bradford (2011) [4] Kruusing, A.: Handbook of Liquids-Assisted Laser Processing. Elsevier Science & Technology, Amsterdam (2007) [5] Ding, K., Ye, L.: Laser shock peening: Performance and process simulation. Woodhead Publishing Ltd., Cambridge (2006) [6] Yang, J.-M., Her, Y.C., Han, N., Clauer, A.H.: Mater. Sci. Eng. A 298, 296–299 (2001) [7] Zhang, Y.K., Ren, X.D., Zhou, J.Z., Lu, J.Z., Zhou, L.C.: Ṁater. Design 30(7), 2769–2773 (2009) [8] Ocaña, J.L., Molpeceres, C., Porro, J.A., Gomez, G., Morales, M.: App. Surf. Sci. 238(1-4), 501–505 (2004) [9] Rubio-Gonzalez, C., Ocaña, J.L., Gomez-Rosas, G., Molpeceres, C., Paredes, M., Banderas, A., Porro, J.A., Morales, M.: Mater. Sci. Eng. A 386(1-2), 291–295 (2004) [10] Sano, Y., Mukai, N., Okazaki, K., Obata, M.: Nucl. Instrum. Methods Phys. Res. B 121, 432–436 (1997) [11] Ivetic, G.: Surf. Eng. (2009), doi: 0.1179/026708409X12490360425846 [12] Peyre, P., Fabbro, R., Merrien, P., And Lieurade, H.P.: Mater. Sci. Eng. A 210, 102–113 (1996) [13] Eliezer, S.: The interaction of high-power lasers with plasmas. Institute of Physics Publishing, Bristol (2002) [14] Chen, P., Cline, D., Craddock, W., Decker, F.J., Iverson, R., Katsouleas, T., Kwok, P., Leemans, W., Masuda, S., Meyerhofer, D.D., Nakajima, K., Ogata, A., Raimondi, P., Sessler, A., Walz, D., Weidemann, A.: Nucl. Instrum. Meth. A 410(3), 407–417 (1998)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Damage Tolerance of Titanium Alloy Rotorcraft Components: Advantages and Challenges X. Li, B.R. Krasnowski, and W.P. Green Bell Helicopter Textron, Fort Worth, TX, USA

Abstract. A damage tolerance (DT) testing method is necessary to support DT design and certification and to simplify flight test monitoring of titanium yoke flexure components. A building block testing program was conducted. While the validity of the basic approach was confirmed by the testing results, a number of aspects have been revealed that are of primary importance to designing and certifying titanium yoke flexure components in particular, and any titanium rotorcraft component in general. In common practice, the through-crack data are used as material properties for crack growth analysis, although the majority of crack growth in rotorcraft components is observed to occur in part-through mode. The application of through-crack data to surface crack of a titanium part needs to be evaluated when taking the surface conditions into account. This is because the crack growth behavior of titanium components was found to be sensitive to microstructure changes and residual stresses that are induced during machining and surface treatment. Underestimating these factors and using through-crack data to predict the part-through crack growth can lead to an incorrect conclusion for a titanium component. Furthermore, unevenness of part-through crack growth between depth and surface growth needs to be considered differently for the basic material in terms of material form and heat treatment, and for the components with machined and shot-peened surfaces. This paper presents considerations supported by testing that lead to improvements in testing and analysis of rotorcraft titanium components and will allow for better use of this material for rotorcraft components.

1 Introduction Damage tolerance (DT) technologies have been developed to address structural integrity issues of fixed-wing aircraft and aircraft engines. Yet it has remained challenging to apply the DT philosophy to rotorcraft structural components due to their unique and more demanding issues. One of the concerns is on the dimensions and weight of the components that are subject to the DT design. The FAA, JAR, and military DT requirements for single load path rotorcraft components oftentimes result in an increase in dimensions and a substantial increase in weight, which is in conflict with good design practices. Both of these factors can lead to lowering performance of a new rotorcraft, which are designed subject to the new *

Oral presentation.

868

X. Li, B.R. Krasnowski, and W.P. Green

rules, below performance of the existing rotorcraft that were designed subject to the old rules. Material solutions were thus sought to accommodate this conflict. Titanium alloy Ti-6Al-4V has a high toughness-to-density index, KIC/γ, and threshold-to-density index, ΔKI,TH/γ, as compared to the generally used materials for rotorcraft components, such as 15-5PH and 4340 steels, or 7075T-73Al alloy, as shown in Table I. Advantages of using a titanium alloy have been revealed as the following [1-3]: a) High values of ΔKI,TH and KIC give a wide range of crack growth and large critical crack growth that with improved NDI methods enables long enough inspection intervals without weight penalty; b) Part-through near threshold growing cracks in Ti-6Al-4V show high ΔKI,TH as compared to near threshold growing cracks in steels and aluminum alloys; c) Part-through crack growth in Ti-6Al-4V shows a non-symmetrical and ragged crack front which indicates higher crack growth resistance in titanium than in steel and aluminum; d) Part-through crack growth in the parts made of Ti-6Al-4V can be further improved by appropriate surface treatments; e) Ti-6Al-4V is corrosion resistant in the majority environments for rotorcraft application. Use of a titanium alloy may therefore provide a material solution to lower weight and decrease dimensions, and allow meeting DT requirements without sacrificing performance to an unacceptable level. Table 1 Crack Growth Indices for Aerospace Metals.

2 Problems and Approaches While the titanium alloy reveals advantages for a matieral solution to meet the DT requirements without sacrificing weight panelty, several disadvantages exist that hinder the use of the Ti-6Al-4V for rotorcraft component DT design. The disadvantages are outlined below. i. ii. iii. iv.

High sensitivity to fretting and subsequent fretting cracks High sensitivity to machining methods Premature failure from subsurface origins Large scatter of crack growth data in complex stress fields.

DT of Titanium Alloy Rotorcraft Components: Advantages and Challenges

869

In order to use Ti-6Al-4V alloy to its full potential, the DT design tools of titanium alloy need to be develped that includs analysis methods, set of specific design data, and test requirements, which address the abovementioned concerns and support the DT design to meet the FAA/JAR/Military DT requirements without weight penalty as compared to other commonly used materials. In recent years rotocraft manufacturing and research communities have come with a number of approaches and studies to meet these requorements [4-10]. Effort has been made with a building block approach to develop a DT certification testing method for Titanium yoke component, as shown in Figure 1.

Fig. 1 Building Block Approach for Titanium Alloy DT Test and Analysis.

With this approach, analysis and testing are accommondated at each step for verification and refinement. At the coupon level, the through-crack data obtained from C(T) tests were compared with the data from Kb-Bar tests for surface cracks and Square-Bar tests for corner cracks. Further generalization of crack growth data is obtained in flexure element tests. The flexure element is a bar-element consisting of fillet radii that are representative of the yoke. The flexure elements are tested in tension and in bending with baseline and shot-peened surface conditions. This paper presents the results with regard to the limit of performance and indicated need for further R&D efforts to fully utilize Ti-6Al-4V and other titanium alloys for DT rotorcraft parts.

870

X. Li, B.R. Krasnowski, and W.P. Green

3 Results Coupon Tests The majority of crack growth in rotorcraft components is observed to occur in part-through surface crack mode. On the other hand, in common practice the through-crack data are used as material properties for crack growth analysis, including analysis for a surface crack. This approximation is proven to be applicable to components made of steel and aluminium alloys. However, the use of throughcrack data for surface crack analysis of a titanium part needs to be evaluated when taking the surface conditions into account. Figures 2 to 4 show the da/dN and dc/dN data for Ti-6Al-4V obtained from C(T), Kb-Bar, and Square-Bar tests. C(T) testing data are for the through-crack, Kb-Bar for the part-through-crack, and Square-Bar for the corner-crack. The differences between FCG behavior of the through-crack and the part-through-crack can be observed in Figures 2 and 3, where the threshold ΔKI,TH for the part-through-crack growth is larger than that for the through-crack growth, indicating that the titanium alloy has higher resistance to growing a surface crack than a through-crack. Further examination of the coupons found that this “resistance” can be attributed to the compressive residual stress existing in the surface layer that resists crack growth in the surface area. Since the test specimens were as-machined coupons, this residual stress must have been induced during the machining process. In other words, the titanium alloy is sensitive to the machining process (i.e. machining parameters) for the surface residual stress and subsequently for the FCG behavior. Existence of the residual stress and microstructure changes of the as-machined titanium coupons can be related to the large scatter of the FCG data for surface crack Kb-Bar and corner crack Square-Bar as shown in Figures 3 and 4. This is again evidence that the machining process induced microstructure changes and residual stress imposed uncontrolled surface conditions, subsequently causing variation of the FCG data for these two non-through-crack growth tests.

Fig. 2 da/dN Summary Data of Through-Crack C(T) Tests.

DT of Titanium Alloy Rotorcraft Components: Advantages and Challenges

871

Fig. 3 da/dN Summary Data of Surface Crack Kb-Bar Tests.

Fig. 4 da/dN Summary Data of Corner-Crack Square-Bar Tests.

Flexure Element Testing Results Differences between shot-peened and unpeened FCG data. Flexure elements were tested under either tension or three-point-bending load. Two part-through-crack configurations were considered, i.e., surface crack and corner crack. In the fillet radius area near the center of the specimens, an EDM notch was induced on the center surface to simulate the surface crack growth, and the notch was induced at corner to simulate corner crack growth, as illustrated in Figure 5. In Figure 5 “SC” denotes the surface crack and “CC” denotes the corner crack. Figure 5 shows a-N curves obtained from the tension element tests under stress ratio R = 0. It is exhibited that the surface crack grew faster than the corner crack for the as-machined baseline specimens; however, with the shot-peened condition, the surface crack turned around to grow much slower than the un-peened, indicating that introduction of residual stress has a strong influence on the surface crack growth.

872

X. Li, B.R. Krasnowski, and W.P. Green

A comparison of baseline and shot-peened surface crack cases shows an increase of more than 2.7 times in the number of cycles to failure due to shotpeening, which would result in a weight savings of 28% for this case.

Fig. 5 Tension Element a-N Summary Data: Baseline vs. Shot-Peened, Testing vs. Analyses.

Differences between tension and bending FCG data. Figure 6 shows a comparison of the FCG behavior between tension and bending tests for the un-peened specimens for a surface crack. The stress ratio for the bending tests was R = 0.05, compared to the stress ratio of the tension tests, R = 0. It should be noted that with the same magnitude of tensile stress applied to the notch location, the fatigue growth of the surface crack under bending loading is about four to five times slower than under tension loading. The difference can be attributed to the difference in the stress intensity factor (SIF) distribution along the crack front and the crack front flattening as illustrated in Figure 7. These features are actually advantages for damage tolerance of titanium flexure components, leading to more cycles and easier inspection of a longer surface crack.

Fig. 6 a-N Curves of Surface Crack Flexure Elements: Bending vs. Tension.

DT of Titanium Alloy Rotorcraft Components: Advantages and Challenges

873

In particular, Figure 7 shows the crack front of a part-through-crack for both loading cases, i.e., tension and bending. The crack front departs from the semielliptic shape that is usually assumed for analysis, and shifts into a shape smaller than a half-ellipse. The same behavior of crack shape change can be observed in the presence of the stress gradient caused by surface treatments [11].

(a)

(b)

Fig. 7 Crack Shape of Part-Through Crack Observed on the Crack Surface: (a) Under Tension Load and (b) Under Bending Load.

Differences between AFGROW, NASGRO® and testing FCG data. a-N curves obtained from the simplified NASGRO® and AFGROW crack growth analyses are also shown in Figure 5. These analyses were subjected to the same loading conditions as those applied to the tests but used the through-crack data for the material FCG property inputs. The standard AFGROW and NASGRO® crack growth analyses were performed for the corner crack in the fillet radius. In order to closely simulate crack growth in the real stress fields, the analyses took the nonuniform stress distribution into account. The predictions, however, are still less than 40% of the actual test cycles. This means that the weight of the component designed based on such analyses will accordingly increase. In other words, a more accurate model that would predict the test data could save over 25% weight in this case. Premature failures (dent, subsurface crack, rough machining). During testing of the baseline and shot-peened element specimens, some specimens failed prematurally at a location other than the notch, away from stress concentration areas (i.e., fillet radius). This behavior can be attributed to various factors such as suspected rough machining, dents on the surface, or subsurface origins, revealing higher sensitivity to these factors as compared to the EDM notch at the highest stress location. These premature failures were subjected to fractographic evaluation to determine their failure origins. Figure 8 shows the fractograph photo of the premature failure of a tension element caused by rough machining, Figure 9 is the fractograph photo of a premature failure developed from a dent, and Figure 10 is the fractograph photos of premature failure that started from a subsurface origin.

874

X. Li, B.R. Krasnowski, and W.P. Green

(a)

(b) Fig. 8 Fractographic Photos of Premature Failure of Tension Element. (a) Fluorescent Penetrant Inspection of Rough Machined Surface with Multiple Secondary Cracks. (b) Crack Surface of an Open Secondary Crack.

Fig. 9 Fractographic Photo of Premature Failure from Dent.

DT of Titanium Alloy Rotorcraft Components: Advantages and Challenges

875

Fig. 10 Fractographic Photo of Premature Failure from Subsurface Origin.

4 Concluding Remarks Part-through crack growth is the dominant stage, in terms of durability, of the crack growth in rotorcraft titanium components, such as yokes. FCG behaviors of through-cracks and part-through-cracks are different for the titanium alloy; therefore, using through-crack FCG properties to analyze part-through crack growth may result in incorrect prediction of crack growth. The titanium alloy is sensitive to the machining process, e.g., an improper machining parameter setting may induce residual stress and microstructure changes that initiate and grow surface cracks. For components made of titanium alloy, crack growth behavior was found to be sensitive to the residual stresses and microstructure changes that were either generated during machining or induced via surface treatment. Unlike the tensile residual stress of as-machined specimens that results in faster crack growth on the surface, shot-peening induces compressive residual stress that increases the resistance to a crack growing in the surface layer. Underestimating these residual stresses and using through-crack data to predict the part-through crack growth can lead to an incorrect conclusion for a titanium component.

Acknowledgements This project was funded by the Vertical Lift Consortium (VLC), formerly the Center for Rotorcraft Innovation and the National Rotorcraft Technology Center (NRTC), U.S. Army Aviation and Missile Research, Development and Engineering Center (AMRDEC) under Technology Investment Agreement W911W6-062-0002, entitled National Rotorcraft Technology Center Research Program. The authors would like to acknowledge that this research and development was accomplished with the support and guidance of the NRTC and VLC. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of the AMRDEC or the U.S. Government. The U.S. Government is authorized to reproduce and distribute reprints for Government purposes notwithstanding any copyright notation thereon.

876

X. Li, B.R. Krasnowski, and W.P. Green

References [1] Lütjering, G., And Nilliams, J.C.: Titanium, pp. 1–77. Springer, Heidelberg (2003) [2] Helm, D., Roder, O.: Recent Titanium Research and Development in Germany. In: Proceedings of the 11th World Conference on Titanium, Kyoto, June 3-7 (2007) [3] Oberwinkler, B., Lettner, A., Eichlseder, W.: International Journal of Fatigue 33(5), 710–718 (2011) [4] Helicopter, B., Boeing, Sikorsky.: Implementation of Rotorcraft Damage Tolerance: Technical Issues, Challenges, and Approaches, RITA Interim Technical Report (2003) [5] Cronkhite, J., et al.: In: Proceedings of the American Helicopter Society 56th Annual Forum, pp. 980–992. American Helicopter Society, Alexandria (2000) [6] Everett Jr., R. A., Elber, W.: In: Proceedings of the American Helicopter Society 54th Annual Forum, pp. 145–156. American Helicopter Society, Alexandria (1998) [7] Le, D.: A Roadmap for Damage Tolerance Implementation in Rotorcraft. Presentation at the Workshop on Fatigue Design of Helicopters, Pisa, Italy (2002) [8] Krasnowski, B.R.: In: Proceedings of the 24th NATO RTO Meeting, pp. 7-1 – 7-8, NATO RTO MP-24, Neuilly-Sur-Seine, France (2000) [9] Le, D.: Research to Improve Rotorcraft Structural Integrity, Reliability, and Safety. Presentation at the International Society of Science and Applied Technologies on Safety and Reliability Assessment, Las Vegas (1999) [10] Le, D., Kanninen, M.: Residual Stress Effects on Fatigue and Fracture Testing and Incorporation of Results into Design. In: Bunch, J.O., Mitchell, M.R. (eds.) Proceedings of the ASTM Symposium on Residual Stress, ASTM International, West Conshohocken, PA (2007) [11] Green, W.P., Krasnowski, B.R., Li, X.: Towards Weight Saving for Damage Tolerant Masts and Drive Shafts. In: Proceedings of the 26th ICAF Conference, Montréal, Canada, June 1-3 (2011)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

NH90 Qualification According to Damage Tolerance Alain Struzik Eurocopter

Abstract. The NH90 is the first military helicopter in the world to be qualified according to the new Damage Tolerance requirements, FAR 29 amendment 31. For all Principal Structural Elements1, a Retirement Time and repetitive Inspection Intervals, approved by the Authorities, are included in the Airworthiness Limitation Section of the Instruction for Continued Airworthiness. The Retirement Time is based on the conventional “Safe Life” concept. The repetitive Inspection Interval is established by using one of the three offered concepts (“Flaw Tolerance”, “Crack Tolerance” and “Multiple Load Path”). The advantage of this pragmatic approach is to improve what exists today, i.e. the substantiation of repetitive Inspection Intervals with full-scale tests and/or analysis, formerly based on in-service experience. This paper presents the necessary research program undertaken by EUROCOPTER to deal with these new concepts and the full qualification process for some dynamic components. It is concluded that the Damage Tolerance philosophy is viable and should contribute to enhance the flight safety of helicopters.

1 Introduction The new generation military helicopter NH90 is qualified in accordance with FAR 29, Amendment 31. As such, the NH90 complies with FAR 29.571 “Fatigue Evaluation of Structure” as introduced by Amendment 28. A safety analysis was conducted to identify all PSEs and these parts were then qualified by applying one of three available Damage Tolerance concepts: “Flaw Tolerance”, “Crack Tolerance”, or “Multiple Load Path”. In 2005 the NH90 has thus become the first helicopter in the world to be qualified according to these new Damage Tolerance requirements. First of all the NH90 helicopter is presented in the following chapter. Afterwards Damage Tolerance concept is explained briefly. The approved methodology used for the NH90 qualification according to Damage Tolerance is described subsequently. The necessary research program undertaken by Eurocopter to support this methodology is provided. Then examples are given to show the qualification of the dynamic components. At last, the influence on the accident rate of the * 1

Oral presentation. Principal Structural Elements (PSE) are structural elements that contribute significantly to the carrying of flight or ground loads and whose failure due to fatigue can lead to the catastrophic failure of the helicopter.

878

A. Struzik

application of Damage Tolerance regulation is discussed. In the end, it is concluded that the Damage Tolerance approach is viable and should contribute to the enhancement of flight safety of helicopters.

2 Presentation of the Nh90 The NH90 is a new generation military, medium class, helicopter (MTOW2 11,000 kg3) and is operated by a single pilot and a crew of 2. It is a twin-engine helicopter, equipped with Roll-Royce/Turboméca/MTU RTM322 or General Electric GE T700 depending on the choice of the Customers. It features Fly-By-Wire controls, a corrosion free and crashworthy carbon fibre fuselage with low radar signature, and is offered with 2 cabin sizes (standard (1.58m) and high (1.82m)). Depending on the version/variant it can be equipped with a rear ramp, automatic tail and blade folding, and de-iced main and rear rotors for operations in Continuous Icing Condition as required by DEF-STAN 000-970 regulations. Thanks to its innovative design, modern technology and systems as well as Man-Machine Interface characteristics4 , the NH90 is able to perform tactical transport (version TTH5), naval (NFH6), SAR7, and “utility” missions by day or night and in adverse weather conditions (-40°C up to ISA + 35°C, rain, snow, wind and hail).

(Photo EUROCOPTER – Patrick Penna) Fig. 1 TTH version – High cabin.

2

MTOW = Maximum Take Off Weight. The NH90 may be operated at higher weight up to 11,400 kg under some conditions. 4 Including Night Vision Goggles (NGV), Forward Looking InfraRed (FLIR), Weather Radar, Digital Map Generator, and Helmet Mounted Sight and Display. 5 TTH = Tactical Transport Helicopter. 6 NFH = NATO Frigate Helicopter. 7 SAR = Search And Rescue. 3

NH90 Qualification According to Damage Tolerance

879

(Photo EUROCOPTER – Patrick Penna) Fig. 2 NFH version – Standard cabin – Folding phase.

NAHEMA8 is the NATO Agency that represents the initial four participating Nations (France, Italy, Germany, and the Netherlands). Portugal joined the Agency in 2001. It controls the overall execution of the programme, is responsible for the qualification of the NH90 weapon system, and is the interface to the contractor for the negotiation, placing and administration of the Prime Contracts. NHIndustries9 is the joint venture created by Agusta (32%), Eurocopter (62.5%) and Fokker Aerostructures (5.5%) to carry out the NH90 industrial programme management. NHIndustries responsibilities cover the design and development, the production, the marketing and sales, and the in-service support for the NH90 all over the world. NHIndustries signed the NH90 Design-and-Development contract with NAHEMA on the 1st of September 1992. The first TTH version entered service in 2006 and the first NFH version entered service in 2010. Currently the NH90 back-log consists of 529 firm orders, 122 options. The NH90 has been selected by 19 Armed Forces from 14 Countries (France, Italy, Germany, the Netherlands, Portugal, Greece, Finland, Norway, Sweden, Sultanate of Oman, Australia, New Zealand, Spain and Belgium). 44 NH90 helicopters are already in service today. The NH90 is becoming the true reference for the Armed Forces worldwide10.

3 Damage Tolerance Concept The term "Damage Tolerance" means here the evaluation considering the effects of both fatigue and expected damages. In this context, "Damage Tolerance" does not exclusively relate to "Crack Growth", as it is traditionally used. 8

NAHEMA = NATO Helicopter Management Agency. NHIndustries = NATO Helicopter Industries. 10 More information may be found on website www.NHIndustries.com 9

880

A. Struzik

FAR 29 requirements (Amendment 28) Based on the work of an ARAC11 group in the mid-eighties, new civil FAR 29 regulations were introduced to increase helicopter safety levels by mandating proof of Damage Tolerance as per Amendment 28, dated 27 November 1989. Furthermore, an accompanying Advisory Circular (AC 29 MG11) was issued in 1995. Note that the Amendment 31, the certification basis stipulated in the NH90 contract is identical to Amendment 28, for as far as FAR 29.571 Fatigue evaluation of structure is concerned. The Damage Tolerance approach was developed to mitigate cracking problems affecting components with pre-existing manufacturing deficiencies (e.g. scratch, flaw, burr, crack, etc...) or service induced damage (impact, scratch, loss of bolt torque, wear, corrosion, fretting corrosion, etc...) that were the root causes of fatigue failure. The Damage Tolerance approach is based on the assumption that a fatigue crack in a component can be safely detected by the operators, through inspections, before it grows to the extent where the component can no longer carry limit loads. It is now required to consider the effects of environment, intrinsic / discrete flaws and accidental damages in the fatigue evaluation, unless it is established that this cannot be achieved within the limitations of geometry, inspectability or good design practice for a particular structure. (A conventional Safe Life approach should be used in this case). Two concepts were proposed to fulfil the Damage Tolerance requirements (Figure 3): • Enhanced (Flaw Tolerant) Safe Life • Fail Safe (Single or Multiple Load Path) or a combination thereof.

Fig. 3 Damage tolerance design options with the details of Fail Safe design methodology.

11

ARAC = Aviation Rulemaking Advisory Committee

NH90 Qualification According to Damage Tolerance

881

These concepts are detailed below. Enhanced (Flaw Tolerant) Safe Life Concept. "Enhanced Safe Life" (also named "Flaw Tolerant Safe Life" by ref. [6]) is understood as the capability of a flawed structure to sustain, without measurable flaw growth, the spectrum of operating loads expected during the service life of the rotorcraft or during an established replacement time. Fail Safe Concept (Single or Multiple Load Path design). "Fail Safe" is understood as the capability of a structure with a standard crack (Initial Quality Crack) or a detectable crack (using a prescribed inspection plan) to sustain the spectrum of operating loads expected during the Inspection Interval. Fail safe design can be provided through different concepts (Figure 3). Figure 4 (Single Load Path design) and Figure 5 (two active Multiple Load Path design) (extract from ref. [6]) explain how the repetitive Inspection Intervals are set (difference between the time when the damage becomes detectable and the time when the extent of the damage reaches the critical value for residual static strength).

Fig. 4 Definition of the Inspection Interval based on L212for Single Load Path design.

12

Repetitive Inspection Interval set at L2/4 (according to ref. [6]).

882

A. Struzik

Fig. 5 Definition of the Inspection Interval based on (L2 + Lr)13 for two active Multiple Load Path design.

Damage tolerance approach for the NH90 In 1989, a working group called TOGAA14 was commissioned by the FAA following the ALOHA AIRLINES-BOEING 737 incident (8 April 1988) as a result of aircraft ageing. This was the well-known incident where a high-time commercial airliner lost 18 feet of the upper fuselage before landing safely. The TOGAA mission was to review ageing-related issues and recommend corrective actions. Following discussions with fixed-wing aircraft manufacturers and FAA/JAA, the TOGAA working group proposed a new FAR 25.571 paragraph that introduced the Damage Tolerance approach (means here “Crack Tolerance”) into the civil regulations for fixed wing aircrafts. Beginning with fixed wing aircrafts, TOGAA then expanded to include engines and finally to rotorcrafts back in 1993. The TOGAA working group intended to recommend the elimination of enhanced safe-life and conventional safe life approaches for rotorcraft. The Damage Tolerance approach (means here “Crack Tolerance”) became the only choice. From 1993 onwards, the TOGAA group discussed with the helicopter manufacturers and requested that the RCWG15 provide TOGAA with a « White Paper » 13

Repetitive Inspection Intervals set at (L2+Lr) / 3 (according to ref. [6]). TOGAA = Technical Oversight Group for Ageing Aircraft. This group was composed of high level figures from the U.S. aerospace community. 15 RCWG = Rotorcraft Community Working Group. This group composed of representatives from the major helicopter companies in the US and Europe, from the US (FAA) and European (JAA) airworthiness authorities and operators, was appointed to facilitate communication with TOGAA. 14

NH90 Qualification According to Damage Tolerance

883

(cf. ref. [7]) on fatigue and Damage Tolerance that would form the basis for a revision of advisory circular 29 MG11 and, possibly, FAR 29.571, if duly justified. After a very constructive co-operation with the helicopter manufacturers, a harmonised methodology for fatigue and Damage Tolerance focusing on metals was found and a « White Paper » was prepared and submitted to TOGAA for comments. From 2000 to 2002, the 29WG16 was solicited to prepare (1) a revised Advisory Circular for the current harmonised JAR/FAR rule 29.571, (2) a proposed new harmonised JAR/FAR rule 29.571 and (3) Advisory Circular to support the new rule. In March 2010, a Notice of Proposed Rulemaking (NPRM) was published in the Federal Register and was officially open for comments. At the beginning of the NH90 design and development phase, a common methodology was prepared by the Industry and submitted to NAHEMA for approval. As the industrial partners (except Fokker Aerostructures) of the NH90 were also members of RCWG and 29WG, this methodology was based on the « White Paper » and the proposed new harmonised JAR/FAR rule 29.571 and associated Advisory Circular. This methodology requires the establishment of a conventional safe life (initiation of fatigue crack using as-manufactured17 components) and repetitive Inspection Intervals based on one of the three equally concepts (Flaw Tolerant, Crack Tolerance or Multiple Load Path). Although the wording may appear similar to the one used in the rule JAR/FAR 29.571, the approach is different and detailed below (these definitions are only for use in the context of this paper). Flaw Tolerance Concept. "Flaw Tolerance" is understood as the capability of flawed18 structures to sustain, without measurable flaw growth or fatigue crack initiation, the spectrum of operating loads expected during the established Inspection Interval. This repetitive Inspection Interval is a conservative fraction of the time to initiate a fatigue crack from a detectable flaw (see Figure 6).

16

29WG = Working Group tasked by ARAC to address the FAR29.571 "Damage tolerance and fatigue evaluation of the metallic structure. ARAC (Aviation Rulemaking Advisory Committee). 17 Condition of a component that is produced as a result of a nominal performance of manufacturing processes specified for that component. 18 A flaw is a localised defect or anomaly related to manufacturing or service use. In metals this includes corrosion, fretting, nicks, dents, scratches and gouges, ...In assemblies, this includes loss of bolt torque, ...

884

A. Struzik

Fig. 6 Definition of the Inspection Interval for Flaw Tolerance.

At the time of the periodic interval, the part may be retired without inspection, returned to service if no flaw is found, retired or repaired if a flaw is detected. The inspection generally is a detailed visual inspection and more (Non Destructive Examination) if a doubt exists. Crack Tolerance Concept. "Crack Tolerance" is understood as the capability of a single load path structure with a detectable (using a prescribed inspection plan) fatigue crack to sustain the spectrum of operating loads expected during the established Inspection Interval. This repetitive Inspection Interval is a conservative fraction of the time the time for a detectable crack to grow to critical size under limit load (see Figure 7).

Fig. 7 Definition of the Inspection interval for Crack Tolerance.

NH90 Qualification According to Damage Tolerance

885

At the time of periodic interval, the part may be retired without inspection, returned to service if no crack is found, retired or repaired if a crack is detected. The inspection will generally involve a Non Destructive Examination. Multiple Load Path Concept. "Multiple Load Path" is understood as the capability of a (N) Multiple Load Path structure with (n) detectable (using a prescribed inspection plan) failed load paths to sustain the spectrum of operating loads expected during the established Inspection Interval. This repetitive Inspection Interval is a conservative fraction of the time to initiate a fatigue crack in any remaining load path (when a primary load path is broken) as a result of loading redistribution (see Figure 8). This is a kind of safe-life approach for the remaining load path.

Fig. 8 Definition of the Inspection interval for Multiple Load Path structure (N = 2, n = 1).

If a failed load path is found at the time of periodic inspection, all components of the affected load path will be retired; but if no failed load path is found, the parts may be returned to service. If some parts are found with flaws, these parts may be retired or repaired individually. The inspection will generally involve a visual inspection to detect the failure of one load path. In this concept, full-scale fatigue tests are performed with the remaining load paths (N-n) with as-manufactured parts, and the Inspection Interval is based on the initiation of a fatigue crack in the remaining overloaded load path.

4 Eurocopter Research Program Eurocopter has undertaken significant research studies to improve its knowledge and experience concerning crack propagation and to constitute a material database (crack growth rate versus stress intensity factor range curves and fatigue curves

886

A. Struzik

with flaws). These studies were funded by Eurocopter itself and by the European Union under the BRITE/EURAM19 and 5th PCRD20 programs. Flaw tolerance First, the type and size of flaws encountered in service were identified from available literature and from Eurocopter's own experience (based on detailed investigation related to overhauls, major incidents21 and accidents22 in flight). The main flaws were identified to be scratch, impact, corrosion, fretting, wear and loss of tightening torque. The standard flaw size was defined to cover 90% of the flaw size distribution. As an example, for steels, the standard depths are shown to be 0.2 mm (≈ 0.008 in) for scratch, 0.25 mm (≈ 0.010 in) for impact and 0.3 mm (≈ 0.012 in) for corrosion pits (see Figure 9).

Scratch

Impact

Corrosion pits

Fig. 9 Standard flaws.

In practice, on the components to be tested in fatigue with flaws, the scratches were machined, and the impacts were applied through an impactor dropped from a pre-defined height or hit with a hammer. The corrosion was obtained by exposing the components (without corrosion protection) during 750 hours in a salt spray atmosphere. Fatigue tests were performed on specimens made of different kinds of material (steel, stainless steel, titanium, aluminium alloys and magnesium alloys) with these standard flaws (see Figure 10), in order to complete the material database with data for the materials used for the NH90. 19

BRITE EURAM - « DAMTOL » Contract n° BREU-0123 DAMageTOLerance on helicopter metallic parts (1990-1993). 20 th 5 Programme Cadre de Recherche et Développement - « ADMIRE » Contract n° G4RD-CT-2000-0396 Advanced Design concepts and Maintenance by Integrated Risk Evaluation for aerostructures (2001-2005). 21 Every malfunction which could interrupt, cancel or delay significantly the mission or endanger the crew (loss, failure or damage of critical safety components, use of emergency procedures (engine failure, abnormal heating that might start a fire). 22 Accident with loss of life, hull damage, full or partial destruction of the helicopter.

NH90 Qualification According to Damage Tolerance

887

Fig. 10 Typical fatigue curves (with and without flaw).

Crack tolerance Particular tests were performed to establish crack growth curves (da/dn23 versus ΔK24), specifically near threshold, for different R25 ratios, in order to complete the material database with data for the materials used for the NH90 (see Figure 11).

Fig. 11 Typical crack growth curves.

23

da/dn = crack growth rate. ΔK = Stress Intensity Factor range during a loading cycle (Kmax-Kmin). 25 R = Kmin/Kmax. 24

888

A. Struzik

To evaluate the accuracy of the theory, methodology, tools and material database used by Eurocopter, analysis results were compared with experimental results. Analyses were performed using the in-house software package PROPAK, which has been derived from NASGROW/ESACRACK and PREFFAS (cf. ref. [1]) as developed by EADS-CCR26. Experimental results were obtained for two materials (aluminium alloy and titanium), simple geometry (with and without stress concentration factor kt = 1.6), under simple loading (CAL) 27 and complex loading (simplified Helix32 spectrum28), both in tension and bending. Hereafter 2 examples are provided that show the rather good correlation between prediction (time to propagate an initial circular 0.380 mm ( = 0.0015 in) radius crack to failure versus the maximum nominal stress) and test results (see Figure 12 and Figure 13).

Fig. 12 Specimen without stress concentration – Titanium Bending - CAL - R = 0.7.

26

CCR = Centre Commun de Recherche (Common Research Centre). CAL = Constant Amplitude Loading. 28 Helix is a standard loading sequence that relates to the main rotor of helicopters with a hinged (articulated) rotor (cf. ref [3], [4] and [5]). The purpose of this standard loading sequence is twofold. Firstly, it provides a convenient tool for gathering fatigue data under realistic loading, readily comparable with data obtained by other organisations. Secondly, it can be used to provide design data.Helix has been developed by a collaborative group consisting of MBB, IABG, LBF from Germany and NLR from the Netherlands.Helix32 is a shortened version of Helix. 27

NH90 Qualification According to Damage Tolerance

889

Fig. 13 Specimen with stress concentration – Aluminium Tension - Helix32.

5 Examples from Nh90 Qualification For as far as new designs are concerned, the Damage Tolerance aspects have to be considered at a very early design stage. The NH90 design is based on proven concepts, and the number of structural components and bearings is minimised by design practice (amongst others, through the use of a spherical elastomeric thrust bearing and supercritical tail rotor drive shaft). Lessons learned from in-service experience were utilised and accounted for to improve upon these known concepts. Some critical parts/functions have been designed more tolerant by using Multiple Load Path design. Examples are a Tail Gear Box attachment with 4 bolts instead of the 3 that are generally used and the Main Gear Box being supported by 4 instead of 3 struts. In addition, design efforts were made to prevent flaws or to mitigate their influences, at the component level itself. For example, critical components that are usually made of steel (rotor hub, sleeve, bolts, …) are now made of titanium, or stainless steel to prevent corrosion. Deposits resistant to fretting or wear have been used on most critical interfaces. Moreover, anti-shock paint has been applied on most components. NH90 achieved NAHEMA qualification in early 2006. However, in terms of stress and fatigue its qualification was complete by the end of 2005. Thus, the NH90 is the first military helicopter to be "Damage Tolerant" in the world. Hereafter 3 examples are presented that illustrate the Damage Tolerance qualification process. Lead-lag damper The four-blade main rotor is a SpheriflexTM design, with laminated elastomeric spherical bearing providing flapping, lead-lag and pitch variation functions through elastomeric deformation. The main rotor blades feature curved down and swept high-speed tips. The lead-lag damper is mounted between two adjacent sleeves (see Figure 14).

890

A. Struzik

(Photo EUROCOPTER – Patrick Penna) Fig. 14 NH90 Main Rotor Hub.

The lead-lag damper is loaded axially due to a relative blade motion as well as perpendicularly due to the centrifugal load. The outer part of the component is made of aluminium alloy and the inner part is of stainless steel. An elastomeric part is located in-between the two. Two lugs are screwed into the adapter at both ends. The inspection intervals are based on Flaw Tolerance (for inner and outer parts), on Crack Tolerance (for the elastomeric part), and on Multiple Load Path (for inner part/lug and outer part/lug links).They were established from fatigue tests on flawed components. Damages accounted for during qualification are impacts, scratches and corrosion (for outer part), as well as complete loss of tightening torque (see Figure 15).

(Photo EUROCOPTER) Fig. 15 Flaws on lead-lag damper.

The following table summarizes the retirement life and the Inspection Intervals, for the TTH version.

NH90 Qualification According to Damage Tolerance

891

Table 1 Retirement life and the Inspection Intervals for lead-lag damper.

Retirement Time

Damage Type scratch impact

10,000 h

corrosion

Visible Crack (in elastomeric part) Tightening torque (*) means > 10,000 h

Inspection Interval Maintenance Calculation Manual 900 h recomunlimited (*) mended 900 h unlimited (*) recommended 500 h 500 h mandatory 50 h

50 h mandatory

unlimited (*)

900 h recommended

TTH Sleeve The TTH Sleeve is loaded through the blades (lift and drag loads, centrifugal force, flapping and drag bending moments), as well as by loads coming from leadlag damper (leading and trailing edge), and from pitch rod (via the pitch horn). The sleeve is made of titanium. The inspection intervals are based on Flaw Tolerance (for sleeve), and on Multiple Load Path (for pitch horn/sleeve, spherical bearing/sleeve, leadlag damper/sleeve links). They were established from fatigue tests on flawed components. Damages accounted for during qualification are impacts and scratches (for sleeve), complete loss of tightening torque (spherical bearing/sleeve and lead-lag damper/sleeve links), and the loss of one bolt out of four of the pitch horn/sleeve link (see Figure 16).

892

A. Struzik

Fig. 16 Flaws on sleeve.

The following table summarizes the retirement life and the Inspection Intervals. Table 2 Retirement life and the Inspection Intervals for TTH Sleeve.

Retirement Time

Damage Type scratch impact

3,000 h Missing bolt Tightening torque (*) means > 10,000 h

Inspection Interval Maintenance Calculation Manual 900 h unlimited (*) recommended 900 h unlimited (*) recommended 900 h 2,150 h mandatory 900 h unlimited (*) recommended

SARIB fitting The SARIBTM 29suspension system is an anti-resonance isolation system, which consists of 4 individual units, equally spaced around the Main Gear Box (see Figure 17). The struts transmit the vertical static and dynamic main rotor loads to 29

SARIB = System Anti-Resonance Integrated in the Bar.

NH90 Qualification According to Damage Tolerance

893

the structure through the rigid part of the flexible beams, the SARIB fitting, and conical laminated bearing. The suspension system allows for small rotations of the Main Gear Box. The flexible beams are connected to the Main Gear Box via elastomeric bearings and provide the required elastic stiffness in a plane perpendicular to the mechanical deck. The flapping arms provide the link between the flapping masses and the flexible beams. The stiffness of these parts is chosen such as to have the best transfer of inertial loads coming from the flapping masses. The adjustment of the flapping masses, in combination with the geometry of the system, is optimised to reduce the transmissibility of the 4 per rev dynamic loads coming from the main rotor.

Fig. 17 SARIBTM suspension system.

The SARIB fitting is loaded by a static and dynamic strut load, in conjunction with a dynamic inertial load from the flapping mass. The SARIB fitting is made of titanium. The inspection intervals are based on Flaw Tolerance (for the SARIB fitting), and on Multiple Load Path (for SARIB fitting/structure link). They were established by combining fatigue tests using as-manufactured components with usage of flaw factors. Damages accounted for during qualification are impacts and scratches (for the SARIB fitting), and the loss of four bolts out of twenty four of the SARIB fitting/structure link (see Figure 18).

894

A. Struzik

Fig. 18 Flaws on SARIB fitting.

The following table summarizes the retirement life and the Inspection Intervals. Table 3 Retirement life and the Inspection Intervals for SARIB fitting.

Retirement Time

Damage Type Scratch, impact

unlimited Missing bolt

Inspection Interval Calculation Maintenance Manual 600 h 749 h mandatory unlimited (*)

900 h recommended

(*) unlimited means > 10,000 h Experience from the NH90 qualification Damage tolerance may be achieved by using the new proposed harmonised FAR/JAR29.571, as recommended by the 29WG. Inspection Intervals can be established, area by area, by selecting the most suitable concept amongst Flaw Tolerance, Crack Tolerance and Multiple Load Path. Finally, Inspection Intervals specified in the Maintenance Manual as required by the rule may either be mandatory or recommended.

6 Expected Influence of Damage Tolerance Beyond the strict application of the regulation, we can ask ourselves what the influence of Damage Tolerance approach on the present accident rate of helicopters will be. Root cause of accidents Helicopters are highly complex systems, tricky to pilot, and often used for demanding missions in hostile environments. Although dramatic improvements were

NH90 Qualification According to Damage Tolerance

895

achieved over the last 30 years, the turbine helicopter accident rate remains however much higher than that of large air carriers, and has been levelling off for some years (cf. ref. [2]). It is noteworthy that turbine helicopter accident rate per flight number is similar to large air carriers, if only transport of passengers is considered. In 2009, on a world-wide/all mission basis, the Eurocopter average rate of accidents is 28.4 per million flight hours (out of which 8.1 were fatal). For the last ten years (2000-2009), the analysis of the root causes of accidents in flight shows that (see Figure 19): •

79 % were due to "Operational conditions, human factors and environment". This includes poor estimation of distance with fixed or moving obstacle, poor piloting (no reaction to weather condition worsening, fuel shortage, non observance of flight manual limitations, wrong behaviour upon non catastrophic events or failure), non qualified pilots (helicopter type qualification, weather condition qualification) and pilot's physical inability to perform the required tasks.

•

10 % were due to "Incorrectly performed maintenance". This includes improper assembly, omitting components, not implementing a mandatory modification, polluted fuel, non detection of a clearly detectable damage

•

1.5 % was due to "Vehicle". This includes poor design, non conformity of component, and fatigue crack. 0.9 % was due to "Supplemental Type certificate". This includes all design modifications and/or optionals not qualified by Eurocopter.

•

Fig. 19 Root cause of accidents.

Another analysis performed on all the accidents within the Eurocopter fleet showed that the Damage Tolerance rule could have influenced about 20 accidents (over 43 million flight hours). It can be concluded that conventional safe life (fatigue tests on as-manufactured parts, in-flight load measurement, conservative usage spectrum and high load and life safety factors) is successful in providing a high safety level.

896

A. Struzik

This safety could be improved by applying a Damage Tolerance approach. Customer behaviour To date, neither the Authority nor manufacturers are able to predict how the application of Damage Tolerance principles will influence the state of mind of the users. Some dramatic events in the past have shown the possible consequences of an excess of trust in the state of the art. We just have to refer to the Titanic story, for example. As the ship was widely considered "unsinkable", a series of detrimental choices were made and clear warnings ignored, just because from the Captain to crew members, they simply did not think that serious problems could occur. Another point is related to the mandatory and recommended inspections. Will the customers still perform the recommended inspections, as they used to in the past, even under continuous economic pressure? Conclusion The new Damage Tolerance regulation should contribute to enhance the safety of helicopters. However, its true effect should be carefully monitored by the helicopter community (Customer, Authority and Manufacturer) over the next few years. In addition to the Damage Tolerance approach, other important improvements are also being developed by Eurocopter in an attempt to decrease the accident rate (for more details, report to ref [2]).

7 Conclusion This paper shows that Damage Tolerance may be achieved by using the new proposed harmonised FAR/JAR29.571, as recommended by the 29WG. Inspection Intervals can be established, area by area, by selecting the most suitable concept amongst Flaw Tolerance, Crack Tolerance and Multiple Load Path. In 2005, the NH90 became the first military helicopter in the world to be qualified according to these new FAR 29 Damage Tolerance requirements. The new Damage Tolerance regulation should contribute to enhance the safety of helicopters. However, its true effect should be carefully monitored by the helicopter community (Customer, Authority and Manufacturer) over the next few years.

Acknowledgment Special thanks to C. Aguilar-Grieder, E. Ahci, M. Polychroniadis, C. Juelfs, C. Giry and M. Soulhiard (Eurocopter), J.L Leman (NHIndustries), D. Adams (Sikorsky Aircraft Corporation), and S.J Turner.

NH90 Qualification According to Damage Tolerance

897

References [1] Foulquier, J., Aliaga, D., Lachaud, B.: Fatigue crack growth under in-service loadingDescription and evolution of PREFFAS model. In: ICAF, vol. 18 (1995) [2] Pouradier, J.M.: Helicopter flight safety enhancement: a Eurocopter continuing action. In: ICAS (2002) [3] Edwards, JP.R., Darts, J.: (eds.), Standardised Fatigue Loading Sequences for Helicopter rotors - Helix and Felix - Part 1: Background and fatigue evaluation. NLR TR 84043 U (1984); Also published as RAE TR 84084, LBF FB-167 PT 1, IABG B-TF 1425/1, ICAF Doc. No. 1441 [4] Edwards, P.R., Darts, J.: (eds.) Standardised Fatigue Loading Sequences for Helicopter rotors - Helix and Felix - Part 2: Final Definition of Helix and Felix. NLR TR 84043 U (1984); Also published as RAE TR 84085, LBF FB-167 PT 2,IABG B-TF 1425/2, ICAF Doc. No. 1442 [5] http://www.nlr.nl/images/genesis/reports/NLR-TR84043part1.pdf [6] Fatigue evaluation of transport category rotorcraft structure (including flaw tolerance) AC29MG11 issue on 30/09/1999 [7] Roesch, J., Adams, D., Krasnowski, B.: Rotorcraft Fatigue and Damage Tolerance. White paper presented to TOGAA, Lake Oswego, Oregon (January 1999); It is referred as FAA–2009–0413 in FAA documentation

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Damage Tolerance for Composite Parts Rupert Pfaller Eurocopter, Germany

Abstract. The EC135 was the first helicopter at ECD (Eurocopter Deutschland) to be qualified according to the Damage Tolerance requirements first known as ‘Special Condition’ of the German Luftfahrtbundesamt (LBA). Further investigations for other composite parts followed this approach. In addition to the traditional “Safe Life” approach Damage Tolerance concepts become more and more important to dynamically loaded parts. Due to the fact that composite parts are not established as metal parts the authorities focused on those components first concerning Damage Tolerance aspects. This paper presents the effort performed for Damage Tolerance aspects of composite parts. Also the continuous quality tracking is shown. It is concluded that Damage Tolerance has a long tradition for composites. In combination with the right design, composites show excellent Damage Tolerance performance.

1 Introduction One of the first applications of fiber reinforced composites has been rotor blades of helicopters. In 1967 the BO105, a product of the former helicopter division of MBB, now EUROCOPTER Deutschland GmbH, flew for the first time with full composite main and tail rotor blades. Later on some other projects followed to extend the usage of composites from secondary parts such as fairings to primary fuselage structures. At MBB a fiber composite fuselage of the BK117 was launched as research program in 1985 and is still flying up to now. With the start of EC135, a multipurpose light twin helicopter, the first regulations have arisen considering Damage Tolerance. From this time on more and more effort was invested to design and test composite and also metallic parts by taking into account Damage Tolerance features. Some examples for the evolution of the Damage Tolerance methodology in historical order are given here.

2 Damage Tolerance Features for the MBB-BO105 and BK117 Main Rotor Blades The first flight of the BO105 took place in 1967. This light helicopter was manufactured almost 1500 times. BK117, the next serial development, had its maiden flight in 1979 (see Fig. 1). *

Oral presentation.

900

R. Pfaller

Fig. 1 BO105 and BK117 Helicopters.

The worldwide first serial hingeless main rotor system was a key element of these helicopters using the advantage of the newly developed fiber glass technology. The principle of this design is shown in Fig. 2. Due to the high flexibility of the fiber glass epoxy material the flapping and lead lag hinges could be eliminated. The substitution of the hinges was a big step towards weight reduction and cost saving due to the reduction of parts. Although the BO105 and BK117 have been developed according to the old rules, Damage Tolerance features have been investigated. One reason for this was the development of the EC145 with the same rotor system. The blade is fixed to the rotor hub with a main and a secondary bolt as shown in Fig. 3. The critical failure at the blade component test with centrifugal load, flap and lead lag bending occurred always at the beginning of the radius at the lug area.

Pitch Bearing

Elastic blade neck as flapping and lead lag hinge

Fig. 2 Principle of Hingeless Main Rotor.

Damage Tolerance for Composite Parts

Beginning of f iber crack

901

Half of the fitting with cut fiber glass root end section

Main Bolt

Secondary Bolt Fig. 3 Main Rotor Blade BK117 with Fitting at Root End Section.

During the fatigue bending test performed with the BK117 blade root the blade was dismounted and checked several times (ca. 10x) with Computer Tomography (CT). With this non- destructive testing the delaminated and cracked area was determined. To categorize the delaminated area two projected areas of delamination have been regarded, which are parallel and perpendicular to the rotor plane. For the cracked area only the crack as it grows perpendicular to the unidirectional fibers was evaluated. The two main results from this investigation as shown in Fig. 4 are: • •

The crack growth rate is very low for a long time There is nearly no crack growth for a long period, while the delamination grows linearly. Regarding the CT pictures it is obvious that the crack stops at the horizontal symmetry plane of the lug. This is also the area where there is a clear delamination. In this plane, due to the manufacturing as two separated halves, some inhomogeneous areas are generated, which are responsible for the delamination. So it can be stated that the delamination works as a stop for the crack for many cycles. This was a principal behavior at the fatigue test for all test specimens.

902

R. Pfaller

Cross section

Delaminated area horizontal Delaminated area vertical Cracked area Delaminated area

Crack / Delamination growth

Load cycles

Fig. 4 Diagram with Growth of Delamination and Cracked Area versus Load Cycles.

To show the behavior of the blade with severe damages an artificial crack was applied with increased depth step by step. At the last step 60 % of the cross section area was cut away as shown in Fig. 5. The intention was to use the lead lag deviation at the blade tip as it is measured on ground runs for balancing of the rotor blades as supplementary indirect inspection method. It should be used together with the direct inspection method of removing the metallic fitting. As shown in the diagram in Fig. 6 clearly visible lead lag deviation is obtained at a certain pre- cut depth which is also depending on the rotor thrust. When the whirl tower test has been finished the cut lug was fatigue tested. At the end a residual strength test with Limit Load and an overload factor for humidity and temperature was successfully performed.

Damage Tolerance for Composite Parts

0

Lead lag deviation i [mm[

Unbalance at fitting bolt [mm]

Fig. 5 Demonstration of Maximum Cut at BK117 Main Rotor Blade Root Lug.

Fig. 6 Correlation of Cut Depth with the Lead Lag Deviation at Blade Tip.

903

904

R. Pfaller

3 Damage Tolerance Substantiation of the EC135 Main Rotor Blades A further development of the hingeless rotor was the main rotor of the EC135 helicopter without hinges and bearings (Fig. 7).

Fig. 7 EC135 Helicopter.

In this case also the torsion bearing to introduce pitch angles is replaced by an elastic element built of mainly unidirectional glass fibers with cruciform shape. The principle of the hingeless and bearingless rotor is explained in Fig. 8 below.

Elastic blade neck as flapping hinge Elastic blade neck as lead lag hinge

Soft torsion section for pitch angle

Fig. 8 Principle of Hingeless and Bearingless Main Rotor of the EC135.

Damage Tolerance for Composite Parts

905

The main design details of this blade concept are shown in Fig. 9. The EC135 (first flight in 1994) has been certified according to Joint Aviation Requirements JAR27 ‘Small Rotorcraft’. However, as the primary structure includes composite materials, the German airworthiness authority Luftfahrtbundesamt issued a Special Condition ‘Primary structures designed with composite material’ which had to be fulfilled additionally. The special condition addresses subjects like: • •

Demonstration of ultimate load capacity including consideration of manufacturing and impact damages Investigation of growth rate of damages that may occur from fatigue, corrosion, intrinsic defects, manufacturing defects or damages from discrete sources under repeated loads expected in service

Fig. 9 Design Features of the Bearingless EC135 Main Rotor Blade.

906

R. Pfaller

•

Fatigue evaluation for parts suitable or unsuitable for damage tolerance method and the related inspection procedures Residual strength requirements Consideration of the effects of environmental conditions and material variability Substantiation of bonded joints

• • •

For the substantiation of the dynamically loaded EC135 rotor blade the FlawTolerant Safe Life Method was used. In addition fail-safe features were incorporated into the design to ensure sufficient residual strength capability after flaw growth. The composite structures were pre- damaged with the help of impactors up to 25 Joule as cut- off level.

Fig. 10 Blade Section for Calibration of Impact Force.

Fig. 10 shows a part of a main rotor blade used to calibrate an impact leading to a so called Barely Visible Impact Damage (BVID). Visibility is normally defined by a 0.5 to 1 mm dent. Up to this size, the impact is assumed as intrinsic defect. It is also conservatively assumed that such a defect is located at a critical area and remains undetected during the complete helicopter life. Therefore, the right impact energy to generate a BVID is obtained with a section as shown in Fig. 10. Later, this impact energy will be applied on the test specimen. For thick parts even with high impact energies no visible damage can be obtained. Therefore the applied maximum impact energy has to be cut off at a reasonable level which has to be defined with the help of realistic and probable scenarios of possible impacts during manufacturing, maintenance and service life. Such an impact with a cut off level of 25 J is applied at the flexbeam (the inner part of the blade) and marked in Fig. 12.

Damage Tolerance for Composite Parts

907

Dynamic Strength Component Testing and Demonstration of Limit Load Capacity It is not possible to test a complete blade realistically at all possible load combinations in a testing machine. Therefore, the blade was subdivided into several components each of them being tested under its critical load conditions. For each test type several specimens with intrinsic, manufacturing and impact damages were tested at different load levels. The impact energy for flexbeam and control cuff was 25 J. This means a 2.5 kg impact mass falling from 1 m height. For all static tests with composites the influence of high temperature and moisture had to be taken into account according to the ‘Special Condition’. After the fatigue tests residual strength tests had to be performed. Limit load capacity for parts which show Damage Tolerance behavior was proven there, also including load amplification factors to simulate hot/wet conditions. The strength degradation applied as additional factor was determined by coupon tests. Summary of the (sub-) component tests taking into account the requirements of the ‘Special Condition’can be given as follows: 1. Component specimens - Specimens with intrinsic, manufacturing and impact damages 2. Tests - Separate component tests for critical areas - Constant amplitude tests at different load levels - Test monitoring - Documentation of: Type of damage Damage begin Size Location Growth rate 3. Residual strength test with pre- damaged specimens after fatigue test - Proof of Limit (Ultimate) Load capacity - Load amplification factor to simulate hot/wet conditions Fig. 11 shows a bending specimen of the flexbeam in its upper and lower test position. It is loaded by a centrifugal force of about 150 kN and simultaneously loaded by flapping and lead lag moments. At the left side the blade attachment area is clamped into a fork simulating the rotor hub. At the right side two hydraulic cylinders introduce the maximum transverse forces and flapping and lead-lag moments simultaneously. This test mainly simulates the load conditions between blade attachment and ‘flapping hinge’. The torsion capability of the flexbeam was proved in another test sequence. Fig. 12 shows the specimen unloaded and loaded. The cuff is almost completely removed. At the right side the blade attachment area of the flexbeam is clamped.

908

R. Pfaller

(To the left it is followed by the flat ‘flapping hinge’ and the torsion element with its slit cruciform cross section.) In Fig. 12 the specimen has a pretension of 150 kN and is twisted by 100°. This means a torsion angle of 2°/cm length of the torsion element. The specimen showed no failure, the test was only limited by the capacity of the testing machine. This test proved the outstanding qualities of the EC135 flexbeam.

Fig. 11 Flexbeam Bending Test (Overlaid Picture).

Impact energy 25J

Fig. 12 EC135 Flexbeam Torsion Test,Flexbeam Unloaded (above) Flexbeam Loaded by Centrifugal Force and Twisted by 100° (below).

Damage Tolerance for Composite Parts

909

4 Quality Assurance Methods Applied for Composite Rotor Blades At Eurocopter Deutschland Computed Tomography (CT) is used for the quality assurance of all rotor blades. For the EC135 blades CT was also used during the design phase and has been performed for each blade at the beginning of the serial production. CT is a very effective Non-Destructive Testing (NDT) method to check the quality of fiber composite parts. Damages or defects like cracks or waves in the laminate of at least 0.2 mm size can be detected. By the determination of special CT numbers the local material density can be established. Thus, it can be checked if dark spots in a cross section consist of resin or critical air inclusions. With the help of CT, the manufacturing quality of the EC135 blade was improved significantly. An example at an early development stage is shown in Fig. 13 below.

Fig. 13 Non-Destructive Testing of the EC135 Main Rotor Blade with Manufacturing Defects at an Early Development Stage by Means of Computed Tomography.

It is mentioned in the regulatory that intrinsic manufacturing damages have to be considered. After a certain production rate a destructive testing is required by internal quality assurance rules. These quality assurance tests are used to substantiate all known and unavoidable manufacturing defects by testing the lower end of the manufacturing quality. If some new and unknown phenomena occur all such blades are first restricted to be used for flight and stored. If the phenomenon is assumed to be principally uncritical and if it is also hard to avoid it, the worst will be tested after a certain amount of similar blades are produced. If this blade

910

R. Pfaller

shows enough fatigue capacity a rule for the limitation of such a phenomenon will be introduced. Certain conservatism for an acceptable defect size will always be taken into account. With this method and the help of CT as non- destructive testing intrinsic manufacturing defects are considered as much as possible.

5 Damage Tolerance Substantiation for Tail Drive Shaft Besides rotor blades some other parts can have a benefit to be designed as fiber reinforced plastic parts. For instance a part of the tail drive shaft can be designed in composite to a certain weight and stiffness, which could be important for dynamic reasons (Fig. 14). As for most carbon fiber reinforced parts impact together with compression loading is the dominating damage. Therefore, such damage is applied at each fatigue test. Impacts from 5 J to 30 J have been applied to check the visibility, but only at the inner side an effect of the impacts have been observed (Fig. 15).

Fig. 14 Position of Carbon Fiber Reinforced Tail Drive Shaft at Tail Boom.

Damage Tolerance for Composite Parts

911

Fig. 15 Carbon Fiber Reinforced Tail Drive Shaft at the Test Bench.

In Fig. 16 the torsion load applied during the test as constant amplitude versus the load cycles is shown. The S/N-curve expressed in amplitude values is represented by the following relation:

S A = S A∞ +

S A, ult − S A∞ ⎡⎛ log( N ) ⎞ β ⎤ exp ⎢⎜ ⎟ ⎥ ⎣⎢⎝ α ⎠ ⎦⎥

(1)

Where SA∞ is the endurance limit, SA,ult is the ultimate value, N is the number of cycles and α, β are the shape parameters for the adjustment of the curve [5]. For such SN curves used at ECD the static strength is linked to the dynamic strength, because the curve starts at N = 1 load cycle. The upper curve shows the mean curve and the lower shows the safe working curve which is reduced to guarantee certain survivability for the scatter due to material strength, manufacturing quality etc. Because the fatigue life of this carbon part is assumed to be very high or unlimited depending on the test level the test is stopped when the component reaches load cycles which guarantee sufficient fatigue life without fatigue damage. Then a residual strength test with ultimate load considering the influence of humidity and temperature is performed. Ultimate load had to be chosen because this part showed no sign of fatigue damage, which could be used as typical feature for an inspection interval. Static failure in conjunction with impact damage is the dominating critical load case too.

912

R. Pfaller

Torsion Moment Mt

Load cycles N Fig. 16 SN Curve of Carbon Fiber Reinforced Tail Drive Shaft.

The real lifetime is assumed to be much higher.

6 Damage Tolerance Substantiation for NH90 Torque Panel Another example for Damage Tolerance of fiber structures is given by the so called torque panel. The location of this part at the helicopter is shown in Fig. 17. It is located below the gear box and fulfils the following tasks.: • •

Distribution of the main rotor torque and the shear forces coming from the gear box and introduced with a waved titanium membrane. Compliance together with the waved metallic membrane for a certain ztranslation of the gear box. This translation of the gear box results from the vibration isolation system.

As support for the metallic membrane at the fatigue test of this part it survived the test without damage. For the static strength an ultimate test should be performed. At 1.1 x Limit Load buckling occurred but the load could be increased up to 2 x Limit Load without failure. This test was only limited by the capacity of the test rig. At the next trial the metallic membrane broke through and the test was stopped. At the fiber part no growth of some small damages could be detected. So, the fiber part showed an excellent Damage Tolerance behavior even for static loads and in post buckling conditions.

Damage Tolerance for Composite Parts

913

Fig. 17 NH90 with Position of Torque Panel.

Titanium membrane

CFC torque panel

Fig. 18 Torque Panel with Metallic Membrane and Test Rig after Buckling.

7 Conclusion During the last decades the former helicopter division of MBB, now Eurocopter Deutschland, has consequently developed the main rotor systems towards simplification, improved reliability, increased life, lower weight and reduced service and maintenance costs. It started with the hingeless rotor of the BO105 and continued with the bearingless rotor of the EC135. However, this became only possible by using the outstanding qualities of glass fiber composites with regard to strength and flexibility.

914

R. Pfaller

The low crack growth behavior of composite materials and improved methods of quality assurance, such as Computed Tomography (CT), reduce the life cycle costs and improve the structural safety of helicopters remarkably. It can be stated that sometimes even delamination becomes helpful when it works as crack arrest. Also for other parts of the dynamic system composites show their advantages. So for the tail rotor drive shaft of the EC135 the high fatigue life compensates the impact sensibility of carbon fiber parts. It has to be kept in mind that only very few parts get an impact during their service life but all of them are loaded in fatigue. For the torque panel of the NH90 it is shown that Damage Tolerance is also given for a completely post buckling condition.

Acknowledgment Special thanks to E. Ahci, H. Bansemir, S. Emmerling, K. Pfeifer.

References [1] Bansemir, H., Emmerling, E.: Fatigue Substantiation and Damage Tolerance Evaluation of Fiber Composite Helicopter Components .Applied Vehicle Technology Panel (AVT): Applications of Damage Tolerance Principles for Improved Airworthiness of Rotorcraft April 21-22, Corfu-Greece ECD-0096-99-PUB (1999) [2] Pfeifer, K., Bansemir, H.: The Damage Tolerant Design of the EC135 Bearingless Main Rotor. In: 24th European Rotorcraft Forum, Marseilles, France, September 15-17 (1998) [3] Oster, R.: Computed Tomography as a Nondestructive Test Method for Fiber Main Rotor Blades in Development, Series and Maintenance. In: 23rd European Rotorcraft Forum, Dresden, Germany, September 16-18 (1997) [4] Pfaller, R., Bansemir, H., Pfeifer, K.: Entwicklung schadenstoleranter Faserverbundstrukturen für Hubschrauber. DGLR Jahrestagung, Leipzig, ECD-012200-PUB (2000) [5] Och, F.: Fatigue Strength, AGARDograph No 292, Helicopter Fatigue Design Guide, ISBN 92-835-0341-4 (November 1983)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Towards Weight Savings for Damage Tolerant Masts and Driveshafts W.P. Green, B. R. Krasnowski, and X. Li Bell Helicopter Textron, Fort Worth, TX, USA

Abstract. The design of a rotorcraft main rotor mast or tail rotor driveshaft is critical to the sizing of many drive system parts, including gearboxes and attachments to the airframe. A reduction in diameter of these parts could result in significant weight savings for the aircraft. Due to the criticality of masts and driveshafts, a damage tolerant design is highly desirable. However, damage tolerance requirements for single load path structures tend to produce heavier parts than those designed using safe-life methods. To alleviate the weight impact without sacrificing safety, the damage tolerant design of the mast and driveshaft must consider all aspects of design. This may include non-planar crack propagation, asymmetrical part-through crack growth, and the effect of surface treatments such as shot peening, carburization, etc. The present research incorporates effects that slow crack propagation into fatigue crack growth analysis, supported by a building-block test program of coupon and element level specimens. The test results show that surface condition influences crack front shape and aspect ratio, and that complex geometry can cause asymmetrical and non-planar crack growth. A NASGRO® simulation of a fillet radius element specimen showed a good correlation with testing results for complex geometry without surface residual stress, and future work is proposed to conduct a fatigue crack growth simulation with the effects of typical surface treatments. Successful prediction of crack growth in masts and driveshafts may lead to substantial weight savings in those parts, and the potential for similar improvements in other affected drive system components.

1 Introduction Both the main rotor mast and tail rotor driveshaft are cores of the central systems of a helicopter. Their dimensions determine the dimensions of other systems, such as the transmission and control systems, and through them determine dimensions of the airframe parts to which they are attached. Due to their multiple functions, the mast and driveshaft have complex geometry, splines, flanges, bearing races, bushings and fillet radii; in addition, they are subjected to various heat treatments and surface treatments such as shot peening, carburizing, nitriding, and induction hardening, to improve their performance. *

Oral presentation.

916

W.P. Green, B.R. Krasnowski, and X. Li.

Motivation The FAA/JAR/military damage tolerance (DT) requirements for helicopter components have posed a challenge to the helicopter industry to change the safelife methods of substantiating critical components. These requirements lead to increased sizes of these components, which increases the weight of the whole helicopter, and decreases key performance parameters [1, 2]. The key challenge, therefore, is to minimize the size of the main rotor mast and tail rotor driveshaft by all means available, without reducing their airworthiness below required levels. In recent years, helicopter manufacturers and research organizations have come up with a number of approaches and studies to meet this challenge [3-9]. The field and fatigue test failures indicate a complex cracking pattern in main rotor masts and tail rotor driveshafts, starting from a surface crack that grows slowly and asymmetrically through the thickness. A typical crack is non-planar, its growth skewed with regard to the center axis. The part-through crack growth period is much longer than the through-crack growth, with very complex multiplanar critical failure. Taking into account all aspects of crack growth in the main rotor mast and tail rotor driveshaft, including surface treatments that change the part-through crack growth (such as aspect ratio, a/c), together with advanced NDI techniques and design support testing, the DT required dimensions of the mast/driveshaft may be minimized. This will lead to weight savings for main rotor and tail rotor drive system components, main rotor and tail rotor control system components attached to them, and airframe components. Approach The goal of the research is to develop a set of damage tolerant drive system design tools comprised of analysis methods, specific design data, and test requirements. When properly applied, these tools will assure that the design meets DT requirements without weight penalty, and will not lower helicopter performance relative to non-DT designs. A building block approach (Figure 1) is used to address each step in the testing and analysis process, including part-through cracks in complex geometry with a complex residual stress field, and complex loading.

Towards Weight Savings for Damage Tolerant Masts and Driveshafts

917

Fig. 1 Building Block Approach to Damage Tolerant Design.

Element-level testing of mast critical features, such as a fillet radius or spline, under torsion and shear loading, has been developed. This test data, combined with material properties and surface treatment effects determined by coupon-level testing, will lead to substantial weight savings and better use of advanced technologies in material production, heat treatments, surface treatments, inspections, and design.

2 Effects of Surface Treatments The lowest level of the building block approach includes two types of coupon specimens. ASTM standard middle-tension specimens were used to establish the material’s crack growth propagation properties and Kb-Bar specimens were used to characterize crack growth behavior for part-through surface cracks. The Kb-Bar specimen is simply an axially-loaded bar with a rectangular cross-section. A crack is grown from a semi-circular notch machined at the center of one side of the specimen. Kb-Bar specimens are of greater interest, due to their ability to demonstrate relative differences in part-through crack propagation behavior of materials with various surface treatments [10]. This is particularly pertinent to rotorcraft masts and driveshafts that may possess multiple surface treatments, such as shot peening and carburization, both of which are shown to influence crack propagation behavior in steel materials. Effect on Crack Front Profile One visible impact of surface condition on fatigue crack growth is the alteration of the crack front aspect ratio (a/c, where a is crack depth and 2c is the width of the crack on the surface). Schematic illustrations of the crack front shape for each of three common surface treatments are shown in Figure 2(a)-(c) for measured a and

918

W.P. Green, B.R. Krasnowski, and X. Li.

c dimensions. Figures 2(d)-(f) are the corresponding actual crack fronts for typical baseline, shot peened and carburized Kb-Bar specimens. The photographs of baseline and shot peened specimens are 4340 steel, and the carburized specimen is 9310 steel.

(a)

(b)

(c)

(d)

(e)

(f)

Fig. 2 Kb-Bar Crack Front Shapes for (a) and (d) Baseline, (b) and (e) Shot Peened, and (c) and (f) Carburized Specimens.

In Figure 2(a), the surface is in the baseline, or untreated, state. The semicircular initial flaw grows slightly faster along the surface, resulting in a gradually decreasing a/c ratio. The actual baseline fracture surface, Figure 2(d), is close to the expected semi-elliptical profile. The shot peened case, Figure 2(b), shows an opposite trend; the initial flaw grows faster into the depth, resulting in an increasing a/c ratio. As the crack grows larger, the influence of the shot peened surface layer diminishes, and the aspect ratio tends to behave more like the baseline case. Measured crack dimensions, however, do not predict the inward-shifted ellipse observed in actual shot peened specimens, as shown in Figure 2(e). Figure 2(c) represents fatigue crack growth with a carburized surface. The carburization process fundamentally alters the material properties at the surface, including fracture toughness [11], and turns the surface layer of the material brittle. This results in accelerated crack growth along the surface, and therefore a decreasing a/c ratio as the crack grows. This can be described as an outwardshifted ellipse, and is consistent with the actual carburized specimen shown in Figure 2(f). However, it should be noted that the real crack front transforms into a bell-shaped curve as it approaches the specimen edges. In summary, part-through crack characterization as a semi-ellipse is valid only for the baseline case. For the shot peened case, the semi-ellipse transforms into a larger-than-half elliptical section, and for the carburized case into a smaller-thanhalf elliptical section.

Towards Weight Savings for Damage Tolerant Masts and Driveshafts

919

Effect on Fatigue Crack Growth Pattern The variation in crack front aspect ratio (a/c) with crack depth (a) for three KbBar specimens is shown in Figure 3 for baseline, shot peened, and carburized surfaces. The baseline and shot peened specimens were 4340 steel, and the carburized specimen was 9310 steel. While carburization clearly has an effect on fatigue crack growth, the remainder of the paper will focus on baseline and shot peened surfaces for 4340 steel.

1.6 1.4

ShotPeened 1.2

a/c

1 0.8

Baseline

0.6

Carburized (9310)

0.4 0.2 0 0

0.5

1

1.5

2

2.5

3

3.5

a, mm

Fig. 3 Kb-Bar Crack Front Aspect Ratio (a/c).

The results presented in Figures 2 and 3 show limitations of the current surface crack growth approach, which assumes self-similar semi-elliptical growth.

3 Effects of Complex Geometry and Loading Another critical characteristic of masts and driveshafts is complex geometry. As with the surface treatments described above, this may also lead to conservative simplifications for analysis. Fillet radius element specimens were designed to incorporate several key features common to typical rotorcraft masts. These features include hollow tube geometry, shot peening, and a fillet radius outer diameter transition. The specimens were subjected to bending and torsion loads applied as an offset shear. The specimen and test setup are shown in Figures 4(a) and 4(b), respectively.

920

W.P. Green, B.R. Krasnowski, and X. Li.

(a)

(b) Fig. 4 (a) Fillet Radius Element Specimen and (b) Test Setup.

Torsion and bending loads were applied to the element specimens through an offset shear input. Two ratios of bending moment to torsion were tested: (1) moment equal to torsion (M = T), and (2) moment equal to twice the torsion (M = 2T), both at R = 0.05. A notch, located at the base of the fillet radius, was machined at the proper angle for each load ratio to facilitate growth in the appropriate plane (normal to the maximum principal stress field). The notch orientations for the M = T and M = 2T load cases are shown in Figure 5.

Towards Weight Savings for Damage Tolerant Masts and Driveshafts

M=T M = 2T

α 30° 15°

921

β 0° 0°

Fig. 5 Notch Angles for Fillet Element Specimen.

Effect on Crack Front Profile The influence of complex geometry is evident in Figure 6, particularly as highlighted within the box. This crack grew in a skew plane defined by angles α and β , defined in Figure 5, passing through the fillet radius transition. In the highlighted region, the beta factors are clearly different than those on the other half of the crack (right-hand side of the photo). This geometry-induced asymmetric growth behavior indicates that three-parameters (a, c1, c2), instead of two (a, 2c), are required for crack front measurement and analysis.

Fig. 6 Photo of Selected Marker Bands for Baseline Fillet Element Specimen (4340 Steel).

Effect on Fatigue Crack Growth Pattern Complex geometry and loading are often simplified for damage tolerance analysis, using the closest readily available model with a known SIF solution. However, simplified damage tolerance analysis may be overly conservative, and could offset any potential gains in weight savings. Figures 7(a) and 7(b) illustrate this for each

922

W.P. Green, B.R. Krasnowski, and X. Li.

of two load cases applied to the fillet element specimens. In each case, the simplified AFGROW and NASGRO® models, assuming self-similar growth of a semi-elliptical crack in a plane perpendicular to the specimen axis, predict about one-half of the actual cycles to failure of a crack that has grown in a non-selfsimilar fashion, in a skewed plane. A more accurate crack growth model would prevent the part from being overdesigned, and would, in this case, result in substantial weight savings. The estimated weight savings for the fillet radius element specimen is about 33% for constant outside diameter (OD) and variable inside diameter (ID), or about 20% for constant ID (variable OD). In the latter case, the OD would be reduced by about 7%, leading to potential dimensional reductions – and therefore weight reductions – for other drive system components as well.

6

6

5

5

4

4

AFGROW(Simple) NASGRO(Simple) NASGRO(SwRI)

CrackDepth(a),mm

CrackDepth(a),mm

TestData,M=2T

3

2

3

2

AFGROW(Simple) NASGRO(Simple)

1

1

NASGRO(SwRI) TestData,M=T 0

0 0

50,000

100,000

150,000

200,000

0

50,000

100,000

150,000

Cycles(N)

Cycles(N)

(a)

(b)

200,000

250,000

300,000

Fig. 7 a vs N for Selected Baseline Element Test Data, with AFGROW and NASGRO® Baseline Simulations: (a) M = T Load Case; (b) M = 2T Load Case (4340 Steel).

The a-N curves marked as “NASGRO (SwRI)” in Figure 7 were obtained from a contracted analysis by Southwest Research Institute (SwRI®). This NASGRO® fatigue crack growth (FCG) analysis accounted for the complex geometry of the baseline fillet element specimen. The model used for the simulation, which excludes any geometric features that were designed only to facilitate mechanical testing, is shown in Figure 8. The analysis considered KI only. In addition, the simulation assumed that the crack would remain planar, in the initial orientation, with the applied loading conditions the same as the laboratory testing. The simulation incorporated material property data for 4340 steel (180-200 ksi UTS) developed from the M(T) tests. No surface treatment effects were considered for this simulation.

Towards Weight Savings for Damage Tolerant Masts and Driveshafts

923

Fig. 8 Advanced NASGRO (SwRI) Analysis Geometry (Baseline).

The SwRI® NASGRO® simulation, as plotted in Figures 7(a) and 7(b), closely matches the baseline test data, which suggests that it can be a useful tool in achieving optimal weight savings.

4 Future Work A method for reliable fatigue crack growth simulation in critical drive system parts is the desired output of this research. To date, NASGRO® has demonstrated sufficient capability to simulate crack growth in complex geometries without consideration of surface residual stress. While significant work has been done to develop stress intensity factor solutions for cracks growing through a residual stress field, no crack growth simulations have been run for a case with all of the complexities described in this paper. Another NASGRO® simulation, incorporating both surface and geometric effects, under complex loading conditions, is needed to verify the software’s capability to simulate crack growth in realistic conditions. Another important milestone to validate the damage tolerance analysis is the completion of a full-scale component test with representative loading. This, the uppermost part of the building block approach (Figure 1), is a complex and expensive endeavor; however, a successful result would likely expedite the approval of the method detailed herein by the various certification authorities.

5 Conclusions and Recommendations •

The complex crack growth pattern in the main rotor mast and tail rotor driveshaft contributes to their high crack growth resistance.

924

W.P. Green, B.R. Krasnowski, and X. Li.

•

•

• •

Part-through crack growth is the dominant stage (in terms of duration) of crack growth in the main rotor mast and tail rotor driveshaft. Surface treatments, material selection, and special heat treatments can change the part-through crack growth pattern and increase crack growth resistance. Advanced crack growth analysis can predict, within prescribed application limits, the crack growth pattern in the main rotor mast and tail rotor driveshaft. The analysis can be expanded to other parts of the drive system such as shafts, adapters, couplings, etc., with additional testing. A parametric approach is needed to expand the methodology to a larger range of geometric parameters. Special attention needs to be given to non-inspectable sections of main rotor masts and tail rotor driveshafts, such as sections inside transmission cases or permanently connected. These areas should be designed by safelife methods to maintain an acceptable level of conservatism.

Acknowledgements This work is based on a project entitled “Application of RCDT Methodology to Rotorcraft Masts” that is under the FAA contract DTFACT-07-C-00009 RCDT. The project was supervised by Dr. John Bakuckas, FAA Program Manager, and the FAA RCDT Team with Ms. Traci Stadtmueller as the Contracting Officer’s Technical Representative (COTR).

References [1] Krasnowski, B.R.: J. Am. Helicopter Soc. 36(3), 13–22 (1991) [2] Krasnowski, B.R., Rotenberger, K.M., Spence, W.W.: J. Am. Helicopter Soc. 36(2), 52–60 (1991) [3] Helicopter, B., Boeing, Sikorsky.: Implementation of Rotorcraft Damage Tolerance: Technical Issues, Challenges, and Approaches, RITA Interim Technical Report (2003) [4] Cronkhite, J., et al.: In: Proceedings of the American Helicopter Society 56th Annual Forum, pp. 980–992, American Helicopter Society, Alexandria (2000) [5] Everett Jr., R. A., Elber, W.: In: Proceedings of the American Helicopter Society 54th Annual Forum, pp. 145–156, American Helicopter Society, Alexandria, VA (1998) [6] Le, D.: A Roadmap for Damage Tolerance Implementation in Rotorcraft. Presentation at the Workshop on Fatigue Design of Helicopters, Pisa, Italy (2002) [7] Krasnowski, B. R.: Application of Damage Tolerance Principles for Improved Airworthiness of Rotorcraft, Proceedings of the 24th NATO RTO Meeting, pp. 7-1 – 7-8, NATO RTO MP-24, Neuilly-Sur-Seine, France (2000) [8] Le, D.: Research to Improve Rotorcraft Structural Integrity, Reliability, and Safety. Presentation at the International Society of Science and Applied Technologies on Safety and Reliability Assessment, Las Vegas (1999)

Towards Weight Savings for Damage Tolerant Masts and Driveshafts

925

[9] Le, D., Kanninen, M.: Residual Stress Effects on Fatigue and Fracture Testing and Incorporation of Results into Design. In: Bunch, J.O., Mitchell, M.R. (eds.) Proceedings of the ASTM Symposium on Residual Stress, ASTM International, West Conshohocken, PA (2007) [10] Kearsey, R.M., Tsang, J., Lafleur, P., Au, P.: Characterisation of Semi-Elliptical Surface Crack Threshold Behaviour of Several Rotorcraft Alloys, NRC-CNRC Report, LTR-SMPL-2007-0164 (2007) [11] Krauss, G.: In: ASM Handbook, vol. 19, pp. 680–690 (1996)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Challenges in Damage Tolerance Approach for Dynamic Loaded Rotorcraft Components – From Risk Assessment to Optimal Inspection Planning Jack Zhao1 and David Adams2 1

Structural Methods and Prognostics 2 Ground Test Sikorsky Aircraft Corporation Stratford, CT 06516 USA [email protected]

Abstract. The use of Crack Growth Damage Tolerance as a substantiation methodology for helicopter dynamic components is receiving increased attention as a logical and viable improvement in fatigue reliability and structural integrity. It has seen only limited use in helicopters because the addition of difficult periodic inspections was seen as a significant burden to the operator. However the certifying agencies are moving towards the simultaneous use of both Safe-Life and Damage Tolerance methodologies on each component. In order to mitigate the cost issue, a means to optimize the inspection protocol using a risk-informed damage tolerance based fatigue reliability model and maintenance optimization tool is evaluated in this paper. It was desired to maintain the same “6-9’s” level of structural reliability for Damage Tolerance that is now the standard practice for safe-life substantiations. The newly developed fatigue reliability methodology incorporates the variabilities in initial crack size, crack growth rate, nondestructive inspections, flight loads, and the usage spectrum. The reliability model is further integrated with optimization technique for inspection planning. An example case using the crack propagation test result from a helicopter main rotor spindle is evaluated with the reliability model. The concept of DT risk assessment and optimal inspection planning, impact of NDI detection capability and repair quality on risk reduction, and importance of incorporating CBM logistic requirement are demonstrated. It is concluded that a fatigue reliability model for Damage Tolerance was successfully demonstrated and that it can be used to determine an optimized inspection protocol that reduces the operator’s inspection burden while providing the required 6-9’s level of fatigue reliability.

1 Introduction Damage Tolerance, specifically Crack Growth Damage Tolerance, has been successfully applied in a limited number of helicopter fatigue substantiations for *

Oral presentation.

928

J. Zhao and D. Adams

more than 50 years, although it was originally called “Fail-Safe” methodology. The number of applications is now increasing, driven by an increased emphasis on Damage Tolerance by civil and military certifying agencies. The FAA’s Amendment 28 to FAR 29.571 in 1989 provided that Fail-Safety (Damage Tolerance) was an equal-choice option to Safe Life as a substantiation methodology. And a pending new 29.571 will require implementation of both methods on every substantiated component. Damage Tolerance methodology relies on the assumption that the component exhibits some initial damage that subsequently grows progressively over a period of time prior to catastrophic failure. A successful damage tolerance design must be capable of: 1) predicting crack initiation; 2) accurate modelling of subsequent crack growth; and 3) adequate NDI methodology with suitable inspection schedule. The advantage of a Crack Growth Damage Tolerance method over Safe Life is that the cause of an initial crack or damage does not matter since the inspection program will detect the presence of whatever crack occurs before it becomes catastrophic, with a significant safety margin. The disadvantage is the cost of the inspection program in terms of the intrusive down time, man-hours, training, and equipment required. Damage Tolerance will not be accepted as a viable and desirable methodology unless its benefits are perceived to be worth its cost. There is, therefore, an opportunity to employ a reliability approach to determine an optimum inspection methodology – one that provides a required level of structural reliability but does not require unnecessary or too-frequent inspections. Conventional Approach to Crack Growth Substantiations Sikorsky’s methodology for the substantiation of flight-critical fatigue-loaded components is entirely empirical and was initially developed in the early 1960’s for aluminium spar main rotor blades. This substantiation, called “Blade Inspection Method”, or BIM, is still in use today on thousands of rotor blades. It is based on sensing a loss of internal gas pressure in the event of a spar crack, with the inspection interval based on a full-scale fatigue test program that fully characterized the crack growth behaviour under conservative maximum flight loads and severe usage. Sensing of the pressure loss is done by a special visual indicator at the blade root. The inspection interval is essentially a pre-flight visual inspection that was set at minimum 3 to 1 reduction in the test crack growth time from detection to failure. Inspections start at zero time. This method – conservative full-scale test determination of crack growth, demonstration of a field inspection method, determination of a failure point, and an inspection interval based on a fraction of the test time – is still in use today with a few developments. We now require a static test demonstration for critical crack size, we avoid the inclusion of any blunting effect in metals due to high fatigue test loads, we have employed the method in composites, and we have standard methodology for number of fatigue test specimens and the inspection interval reduction factor. The basic method is accepted by all of our civil and military certifying agencies as illustrated in the figure below from the FAA’s AC 29-2C MG-11.

Damage Tolerance Approach for Dynamic Loaded Rotorcraft Components

929

Fig. 1 Potential-to-Functional Failure Curve from NAVAIR 25-403.

A reliability determination has not been part of the current crack growth damage tolerance method. Because of the conservative treatments of the flight loads, the usage, and the test-based crack growth characteristics in the substantiation, the current method meets the generic requirement that failure is “extremely remote”, and this criteria has been achieved in 50 years of service. There has been a methodology development, Reference [3], called “Empirical Damage Tolerance”, which allows the determination of an inspection interval for a different load spectrum than was applied in the full-scale test program. This development is also useful in the reliability studies that follow and is described in more detail later. Reliability-Based Approach to Helicopter Damage Tolerance The work done to show the reliability of a Damage Tolerant approach for helicopter dynamic component fatigue is not extensive. One early effort did show that a 6-9’s level of reliability was achieved for a multiple load path case, Reference [9]. However a good starting point for reliability-based approach is ReliabilityCentered Maintenance (RCM) as described in NAVAIR 25-403, Reference [5]. The figure below illustrates the key points of RCM. This is a much more general methodology, referring to the decline in a functional capability to the point where the functionality is declared failed. The figure is generally known as a P-F curve. The P-F interval is the age interval (in flight hours, cycles, or calendar time) between the Potential Failure (some loss of functionality) becoming detectable (P) to the point of the defined functional failure (F). The inspection interval (I) is a defined fraction of the PF interval.

930

J. Zhao and D. Adams

(P)

(F)

Fig. 2 Potential-to-Functional Failure Curve from NAVAIR 25-403.

A reliability-based optimal inspection interval would provide a required predetermined level of structural reliability while minimizing the cost of conducting inspections too frequently. One simplified approach to the reliability is discussed in NAVAIR 25-403, where the Inspection Interval was initially determined by requiring that the projected probability of failure be reduced to less than or equal to the acceptable probability of failure. The interval of on-condition task, denoted as I, can be estimated by:

I=

PF n

(1)

where PF is the Potential-to-Functional failure interval and n denotes the number of inspections during P-F interval. In general, n can be determined by either safety requirements or cost optimization. For flight-critical components, the total risk considering the inspections shall not exceed the maximum acceptable risk, Therefore,

PT = (1 − θ ) ≤ Pacc n

where

(2)

Pacc is the maximum acceptable level of probability of function failure and

θ is probability of detecting a potential failure in one inspection assuming it exists. The equation above implicitly assumes the failure will always occur in the P-F interval and a constant detectability which is independent to the size of damage. The extreme condition satisfying the risk constraint occurs if the total risk equals to the maximum acceptable level. Accordingly, the number of inspections can be determined by n=

ln(Pacc ) ln(1 − θ )

(3)

Damage Tolerance Approach for Dynamic Loaded Rotorcraft Components

931

The approach outlined in Eq. 1-3 is based on assumption that a potential failure always exists within the P-F interval and is independent between inspections. As a result, the inspection interval may be too conservative, meaning too-frequent inspections, which does not meet our minimized cost objective. The basic RCM approach does not consider the failure mechanics or the scatter of failure progression. Often, the potential failure mode under consideration exhibits inherent randomness. This is particularly important for the failure modes associated with progressive damage accumulation such as crack initiation and growth, corrosion, and mechanical wear. To effectively address variability and uncertainty of damage progression and understand their impact on P-F interval, it is highly desirable to incorporate stochastic characterization of failure progression into the RCM process. In this paper, a new approach is proposed to establish a risk-based interval for on- condition tasks by incorporating a baseline probability of failure and a characteristic detectability for inspection capability. Generally, the probability of failure for a component under scheduled inspections can be expressed as the probability of a sequence of events, such as: n

p f = p0F + ∑ piG piND piF

(4)

i =1

F

Where, p0

is the probability of failure before the first inspection due to G

excessive damage progression; pi is the probability that damage will grow to a detectable limit right before the ith inspection

(i = 1,2, ", n) ; piND is

the

conditional probability that inspection will not be able to detect damage at the ith F

inspection given that damage exists, and pi is the conditional probability that un-detected damage at the ith inspection will further grow to failure before the next inspection [(i+1)th] or end of intended service life. Clearly, the probability of failure of these events depends on the probability of damage progress, inspection capability, the timing of inspection, and the number of inspections. Therefore, a more rigorous risk assessment of inspection planning requires comprehensive understanding of the physics of damage initiation, progression and associated randomness, as well as the mathematical model representing inspection capability, and advanced probabilistic methodology capable of performing complicated numerical simulation and assessment. Due to its simplicity for further implementation, the concept of P-F interval and procedure outlined in NAVAIR 00-25-403 serve as a good starting point for establishing a rough estimate of inspection interval. For the purpose of addressing inherent randomness of failure progression and to further facilitate quantitative risk assessment and management for CBM, a more rigorous approach incorporating physics-based damage accumulation model, inspection capability, and advanced probabilistic methodology is needed urgently.

932

J. Zhao and D. Adams

2 Challenges in a Damage Tolerance Approach In damage tolerance approach, structural integrity is ensured through a predictive crack growth model representing the true nature of damage progression, nondestructive inspections to eliminate excessive damaging, and proper repair and maintenance actions. Many factors affects the effectiveness, robustness, an accuracy of the damage tolerance approach, including validation of crack growth model, qualify capability of desirable NDI methods, developing optimal inspection plan, establishing repair limits and criteria for proper maintenance actions, and setting up rational level of target reliability for risk management. This paper discusses some of the aforementioned technical challenges associated with rotorcraft components and presents a stochastic methodology for predicting rotorcraft component fatigue lifetimes and optimal inspection intervals and assessing underlying risk. Prediction of fatigue crack growth behaviour Damage tolerance approach relies heavily on capability of a fracture mechanics (FM) model to accurately predict potential damage progression initiated at preidentified locations. Several commonly used FM software packages are available for such purpose, including NASGRO, AFGRO, and FASTRAN. They are developed based on linear elastic fracture mechanics and possess a rich library of stress intensity solutions for the commonly encountered structural configuration and geometric profile for anticipated crack growth. From time to time, more advanced fracture mechanics may be employed for more complicated crack growth behaviour, structural layups, and loading, if there is the stress intensity solutions do not exist. These advanced fracture mechanics tools, such as BEASY and FRANC-3D, engage boundary element based numerical procedure and simulation. Occasionally, the crack behaviour will also be observed and derived directly from crack growth testing at full component level, such as the empirical damage tolerance approach reported in Reference [3]. These approaches represent various levels of modelling and numerical simulation efforts to ensure adequacy of the fracture mechanics model building and accuracy of the predictive capability. For the purpose of qualifying a crack growth model for further DT application, model validation is critical important. There are several ways to achieve the goal. One engages seeded fault testing and the other is to compare the predicted results against the fielded cracking data for further correlation. Uncertainty modelling and quantification for DT approach Primarily, probabilistic uncertainty analysis and risk assessment involves modeling all of the fundamental quantities entering the problem, and also all uncertainties that arise from lack of knowledge in these quantities, which may affect failure of the component or system. These terms are referred to as basic variables including quantities of structural dimensions and material properties, yield stress and other ultimate response limitations, operating conditions and degradations, environmental and loading factors, etc. The sources of uncertainty in

Damage Tolerance Approach for Dynamic Loaded Rotorcraft Components

933

probabilistic analysis can be mainly classified into two categories as aleatory and epistemic uncertainties. Aleatory uncertainties refer to the natural randomness associated with an uncertain quantity, which is inherent in time, in space and measurements. This kind of uncertainty is quantified through the collection and analysis of data to fit to theoretical distributions and, since it is inherent, it cannot be reduced. Epistemic uncertainties reflect a lack of knowledge or information about a quantity, which can be considered in either model or statistical uncertainty-subdivisions. Modal uncertainties arise from simplifications and idealizations that are necessary to model the behavior in a reliability analysis, or from an inadequate understanding of the physical causes and effects. Statistical uncertainties are only due to a shortage of information, and originate from a lack of sufficiently large samples of input data. Statistical uncertainties can be reflected through either parameters of a distribution with a limited set of data or the type of a theoretical distribution to be chosen to fit to data. Since epistemic uncertainty is associated with a lack of knowledge and/or information it follows that it can be reduced through an increase in knowledge by gathering data for a longer period, taking more measurements or carrying out further tests, doing research, and by expert judgment. In order to consider these uncertainties in a structural analysis, appropriate uncertainty models are essential for performing reliability methods to estimate the probability of failure. As one of the key building blocks of a damage tolerance risk assessment and design process, all the sources of uncertainty and their statistical characteristics related to the key design variables must be identified, quantified and further integrated into probabilistic damage tolerance design system. It is well recognized that fatigue initiation and its subsequent crack growth is a random phenomenon. As depicted in Figure 3, various sources of uncertainties contribute to random fatigue and fracture process, including fatigue initiation time, micro-crack initiation and propagation, stress intensity threshold, crack growth rate, usage and loads, and inspection capability for product/in-service inspection.

UNCERTAINTIES ALEATORY OR CATEGORY 1

EPISTEMIC OR CATEGORY II

(IRREDUCIBLE) LOADS/USAGE

DEFECT GEOMETRY

MATERIAL PROPERTIES

(REDUCIBLE) Crack Growth (Manufacturing)

MODELING BIAS and ERROR

SENSOR ERROR

STATISTICAL DATA BIAS

HUMAN ERROR

(Manufacturing)

Defect Defect Defect Defect

Type Size Occurrence Rate Orientation/Shape

Basic Material Properties Fatigue Strength Fracture Toughness

Crack Growth Rate Threshold

Imperfect Knowledge Simplification Assumptions Numerical Precision

Systematic Bias Measurement Error False Alarm

Incomplete Information Insufficient Data

Fig. 3 Uncertainty Identification for Damage Tolerance Approach.

It is beyond the scope of this paper to provide comprehensive review of the statistical procedure for modelling of aforementioned sources of variability associated with DT assessment, details of statistical procedure, methodologies, and practices for DT uncertainty identification and modelling can be found in reference [10].

934

J. Zhao and D. Adams

Probabilistic risk assessment methodology Probabilistic methodologies have been widely applied for uncertainty quantification and associated risk assessment. Among the procedures developed for structural reliability assessment and failure probability prediction, a prominent position is held by simulation methods. The Monte Carlo simulation technique, as the basis of all simulation-based techniques, is the most widely applied numerical tool in probabilistic analysis. The convergent rate of the Monte Carlo estimator is appropriately measured by the coefficient of variation of the estimated probability of failure. In general, the basic Monte Carlo technique requires a large sample size to achieve accurate estimate of probability of failure. This becomes a major limitation for the practical application of basic Monte Carlo simulation in structural reliability applications involved in a small probability of failure. To address the challenge, the Importance Sampling technique has been developed and becomes the most prevalent approaches in the context of simulation-based methods for probabilistic analysis. In importance sampling scheme, instead of drawing random samples arbitrarily as the way implemented in a basic Monte Carlo simulation, the majority of the random samples are drawn from the region that contributes the most for the probability of failure. Several approaches can be employed to identify the important region; including 1) MPP obtained through first order reliability methods (FORM) or second order reliability methods (SORM) solution; 2) a priori estimate from pre-sampling; and 3) Markov Chain Monte Carlo simulation. In general, the efficiency of the Importance Sampling technique improves significantly with a large reduction of the variance of estimator, once the appropriate Importance Sampling density function is identified. In general, DT risk assessment requires generating and repeatable drawings of short-life samples. This requirement dictates the utilization of sampling based methodology in risk assessment and inspection optimization. As the alternative, a MCMC based algorithm has been developed to identified the important region followed by Adaptive Stratified Importance Sampling (ASIS) procedure for the purpose of fast PoF computing and short life sampling (Wu, et.al, 2010). Quantification of NDI capability Non-Destructive Inspection (NDI) has been widely used in engineering practice, for both laboratory and field conditions, to ensure adequate structural integrity. Various NDI techniques exist and each of them has its unique capabilities and limitations. Ideally, a perfect NDI should fully detect the presence of a flaw if its size exceeds the detection threshold. In reality, the ideal detection can never be achieved. Due to inherent variability associated with material properties, condition of the structure and its surrounding environment, durability and sensitivity of NDI equipment, field condition to perform inspection, and operator skills, the capability of a NDI is typically expressed by probability of detection (POD) model.

Damage Tolerance Approach for Dynamic Loaded Rotorcraft Components

935

Various forms of mathematic representations are available to describe capability of a NDI. Most commonly used POD models are the parametric ones, including Lognormal, Loglogistic, and Gamma distribution. Sometimes, a nonparametric POD model may be used if the detection data cannot be well fitted into the parametric models or lack of sufficient data for a parametric study. Often, a single POD model derived from limited fault detection data which may not be sufficient to represent anticipated fielded conditions. As a result, a lower confidence bound is introduced to address the potential limitation of sample size. In many engineering application, a characteristic crack size represents high detectability with high confidence is used as a indicator of NDI capability. The aforementioned POD modelling choices are illustrated in Figure 4. Single POD (Perfect Inspection) 1

0.9

Ideal POD 0.8 (Perfect Inspection) Mean POD with Lower Confidence Bound

Probability of Detection

0.7

0.6

0.5

0.4

0.3

Mean POD 0.2

95% confidence bound Ideal POD

0.1

Single POD 0 0

10

20

30

40

50

60

70

80

90

100

Crack Size

Fig. 4 Notional Sketch for POD Modelling Options.

Once the POD model is fully defined, the probability that growing crack is smaller than a pre-defined characteristic detection threshold can be expressed as a

PD (a ) = ∫ f ( x )POD( x )dx

(5)

ao

( )

()

where f x is the probability density function of growing crack, POD x represents the porbability of detection model, and ao is the lower detetion threshold for the POD model. With given PD a , the assciated statistical distribution of detections can be estimated by

( )

D(a ) = PD (a ) PD (ad ) where

ad is the upper detection limit for the POD model.

(6)

936

J. Zhao and D. Adams

Accordingly, the probability that growing crack is missed during inspection can be expressed as

PM (a ) =

a

∫ f (x )[1 − POD(x )]dx

(7)

ao

and statistical distribution of missed detections can be estimated by

M (a ) = PM (a ) PM (ad )

(8)

These equations are essential for evaluating the statistical distributions of the population of growing crack that are detected or missed during an inspection.

3 Risk-Based DT Assessment and Inspection Optimization The risk of structure failure of a rotorcraft fleet is heavily impacted by the size of the growing crack population at critical locations and the capability of nondestructive inspection methods applied to detect the crack at these locations. As discussed earlier, due to the inherent randomness associated with damage accumulation and aleatory and epistemic nature of inspection capability and process, advanced probability methodologies are employed to assess the underlying DT structural integrity and subsequent inspection optimization. Inspection event tree and maintenance options As depicted in Figure 5, the maintenance decision associated with inspection process exhibits complcated event tree stucture. In general, two outcomes, crack detection or non-detection, are expecetd at each inspection. If no crack is detected, no maintenace action would required and the structural integrity of the component will be eveluated at next inspection. There are two maintenance options are avilable If a crack is detected during the inspection. For the case that the detceted crack is small or structural redundency exist, repair minor damage may not be the most cost-effective decision. Instead, continuous monitoring of damage growth could be a vitble option. Otherwise, a decision of repair should be made if the detected crack is excesive or the component considered is on safety-critical path, for example, the dynamic components. Various levels of repairment could be enforced, including eplacement of the dmaged part with new “as-build” one, replacement with a used “defect-free” parts, or on-site repirement. Clearly, the levels of repairement may affect the quality of the repair. There is a need to incoroprate requality model in the DT risk asseessment and inspection planning process to understand its imapct on maintenance decision making.

Damage Tolerance Approach for Dynamic Loaded Rotorcraft Components

937

Fig. 5 Illustration of Inspection Event Tree.

As discussed previously, the population of growing crack size can be modelled through Eq. 5-8, assunming impcitly that the damaged parts once detected will be immediately removed from the population. As discussed earlier, the outcome of an inspection depends on crack growth rate, actual usage and applied load of the compoennt, inspection time, and NDI capability and limitation (POD/PFA). Intutively, occurrence of damge detection and assocaietd repair events increases with the increase of inspection interval and anticipated life span. The statistical distribution of crack size for the population of the repaired componets adds additional complexity to the crack size distribution after the inspeciton and repairment. As discussed in reference [6], the statisitcal disbution of crack size just after inspection is composed of three distinct populations, as expressed below,

Pr (a < A)After Repair

= Pr (a < A)Repaired + Pr (a < A)Monitoring + Pr (a < A)Non-Detectionr

(9)

The statistical distributons for each of the three populations can be further evaluated. For practical applications, evaluation of these statistical populations requires significant amount of numerical eforts Reliability allocation There is no specific requirement of target reliability for DT based design and assessment imposed by government regulatory bodies or consensus agreement reached in rotorcraft industry. For USAF military fixed wing aircraft, it is required that the maximum acceptable frequency of the loss of adequate structural rigidity, or proper structural functioning, or structural failure leading to loss of vehicle shall not exceed 10-7 occurrences per flight.

938

J. Zhao and D. Adams

As applied to the safe-life case, 6-9’s notion is a shorthand description of reliability meaning less than 1 component in 1 million will experience a catastrophic failure within its established retirement time. This is a reliability goal that was originally established by the US Army for a calibration of a basic fatigue substantiation methodology. It cannot be a specific requirement, because it cannot be proven, and because fatigue failures result from discrepancies in strength, loads, or usage that cannot be statistically characterized in advance. Achievement of the 6-9’s goal for Sikorsky safe-life methodology was first demonstrated in Reference [9] and later verified with an advanced reliability model in Reference [8]. It was shown that the 6-9’s are roughly comprised of 3 from strength, 2 from loads, 1 from usage, meaning that approximately 3-9’s of reliability are provided by the 3-sigma S-N working curve, 2-9’s from the “max measured” treatment of flight loads, and 1-9 from the “worst case” assumption for each element of the usage spectrum. For a Damage Tolerant approach it is suggested that 6-9’s is also a appropriate goal but should be defined as: less than 1 component in 1 million will fail with no mandatory retirement time. It is stated to be independent of retirement time because currently it is not required that a Damage Tolerant, or “on-condition”, component has to have a fixed retirement time. It is anticipated that in the future that both methodologies (inspections and retirement times) will be required on every substantiated component, however this added measure of conservatism will be most effective if the two methodologies are independent. In theory, there are numerous paths to achieve the 6-9’s in Damage Tolerance approach. One way to meet desirable level of reliability is to design and manufacture the part to eliminate cracking during its intended service life. This is essential the safe-life approach. On the other extreme, the part can be designed to tolerate excessive amount of cracking during the anticipated service life and risk of fatigue failure can be mitigated through rigorous inspections. Obviously, either approaches dictates significant life-cycle cost. A more balanced approach is needed to take the advantages of inspection with optimal cost allocation. In this study, it is also recommended that the allocation of the 6-9’s be 1 from usage, 2 from loads, as before, plus 1 from strength and 2 from inspections. The latter meaning 1-9 from a conservative treatment of the rate of crack growth, and 2-9’s from the probability of detection in an inspection program. Current crack growth substantiations assume that the life reduction taken on a full scale propagation test result encompasses both the variability in crack growth rate and missed inspections. Separating these factors will allow a rigorous understanding of both. Formulation of reliability based inspection optimization Considering the uncertainties assocaietd with both datamge progression and inspection capability, the DT inspection optimization is regarded as a Reliability Based Maintenance Optimization (RBMO) problem. In constrained optimization problem, we aim to determine suitable time for inspections, Ti (i = 1,2,..., n) , by maximizing the probability of detection

Damage Tolerance Approach for Dynamic Loaded Rotorcraft Components

Maximizing {Pr[a (Ti ) − aDetected (POD ) ≤ 0]} Where

939

(10)

a(Ti ) is the crack size at the time of ith inspection, a Detected (POD ) is the

detected crack size per scribed by the POD model considered. The optimization subjects to several reliability constraints. The first constratint imposed is to ensure acceotbale level of relaibility before performing the first inspection, such as

Pr[a(T1 ) > acritial ] ≤ PrAccept

(11)

Where a(T1 ) is the crack size at the time of the first inspection, acritial is the critical crack size defining DT failuere, and PrAccept is the accepatable probability of failure which is assumed to be 10-6 in this study. The second contraint is to enforce adequate level of relaibility given the inspections. This can be expressed as

Pr[a (T Ti ) > acritial ] ≤ PrAccept Where

(12)

a(T Ti ) is the growing crack size as a function of remaining time until

next inspection or reaching expected service life after ith inspection,

acritial is the

critical crack size defining DT failuere, and PrAccept is the accepatable probability of failure which is assumed to be 10-6 in this study. To achieve the subject optimal solution for inspection scheduling, the following strategy is recommended for effective damage tolerance risk management: • • •

•

Acceptable baseline probability of failure without inspection shall be around 10-3. This can be achieved through conservative usage spectrum, applied loads, and crack growth properties. The risk of fracture failure before any given inspection shall not exceed 10-6, in compliance to six nines’ reliability target for safe-life approach; The acceptable conditional probability of failure after an inspection shall be around 10-2 to 10-3. The associated reliability is attributed to the incorporation of inspection and can be adjusted by inspection capability, repair or replace criteria, and inspection planning (in terms of time of first inspection, number of inspection and associated inspection intervals). The total probability of failure considering inspection(s) shall not exceed 10-6. This will result in a target reliability level of 0.999999.

RBMO framework To establish the optimal inspection schedule, non-linear optimization methodology has been employed. One of the essential elements of the

940

J. Zhao and D. Adams

methodology is to satisfy the reliability constraints (Eq. 11 – 12) imposed to the problem. To assess the reliability requirements, the overall probability of failure as the outcomes of baseline risk assessment, inspections, repair / replacement, and other possible maintenance actions needs to evaluated. The concept of the overall probability of failure is straight forward and highlighted by Eq. 4. But in reality, the task is not a trivial one and full of technical challenges. To implemented aforementioned concepts, an integrated framework needs to be developed by incorporating a physics-based damage progression model, quantitative risk assessment capability, mathematical representing of inspection capabilities, repair quality models, and optimization methodology. This leads to the development and implementation of a reliability based maintenance optimization (RBMO) software tool. General Approach. As depicted in Figure 6, the RBMO framework engages a three stage approach. In the first stage, a baseline probability damage tolerance assessment is conducted to evaluate the risk of excessive crack growth for an anticipated component service life. In this stage, the inspections are not considered and risk of failure is primarily attributed to the inherent scatter of crack growth behaviour and anticipated variation of usage and applied load. Several numerically efficient methods, such as fast probability integration or importance sampling, can be utilized in this stage to facilitate fast probability computation and identify important domain in the problem design space. Further sampling of short life crack growth is performed in the identified important region and the result crack growth data is stored for further inspection trade –off study. In addition, the predicted failure probability prior to the inspections, as defined in Eq. 4, is estimated. In the second stage, the maintenance requirements and inspection capability for selected NDI are defined first. The numerical simulations are carried out to generated inspection outcomes based on defined POD model. Further comparisons are made between pre-compiled short life crack samples and possible inspection outcomes at any candidate time performing anticipated inspection. The risk of missed detection is calculated and the conditional probability that un-detected damage at current inspection will grow to failure before the next inspection is also evaluated. Most importantly, the reduction of failure probability due to the inspections are determined. The effect of repair quality is addressed in this stage via numerically efficeint recursive probabiluity integration (RPI) scheme. During the last stage, inspection optimization is performed as the outer loop of the second stage. All the candidate inspection times are evaluated and associated risk reduction due to inspections is calculated. The optimal solution is obtained when the maximum amount of risk reduction is achieved and resulting probability constraints are satisfied.

Damage Tolerance Approach for Dynamic Loaded Rotorcraft Components

941

Fig. 6 Three Stage Approach for RBMO.

Sampling strategy. To facilitate affordable numerical simulations, an Adaptive Stratified Importance Sampling (ASIS) methodology is developed and applied. The ASIS aims to take the advantage of efficient sampling in the most important region which contributes significantly to the failure probability. In this case, our focus is on the fatigue crack growth lives that fall short of the specified component crack growth life. Following the basis of fracture mechanics, the short life is mainly attributed to several random variables, including larger initial crack size, excessive high load, fast crack growth rate, and low stress intensity threshold. The essence of ASIS is to identify the aforementioned contributors and determine their combinations for short life sampling. This is achieved through a Markov Chain Monte Carlo (MCMC) algorithm which is specifically developed to rapidly search the design space and locates the potential candidates. By the aid of one-step memory of Markov Chain and a suitable accept-rejection criterion for short life, MCMC is capable of identify the critical region within several hundred limit state function calls. Once the importance region is identified, ASIS is followed with a very efficient sampling draws in the vicinity of the critical domain. The accuracy of the importance sampling is monitored via an adaptive scheme and controlled by a specified error bound. The details of the integrated, efficient, and versatile reliability-based maintenance optimization (RBMO) framework and aforementioned methodologies and algorithms, including RAM, MCMC, and ASIS can be found in Reference [7].

942

J. Zhao and D. Adams

4 Example Case To illustrate the concept and application of Damage Tolerance Risk Assessment and Inspection Optimization, the example of the CH-53E Spindle Lug crack growth fatigue substantiation is used, taken from Reference [3]. This component was subjected to a conventional crack growth substantiation in order to allow continued service of these components until a manufacturing change could be implemented to eliminate fretting caused by contact of the bonded liner with the lug bore. The full scale test program employed a conservative high envelope spectrum of edgewise bending load related to the lag damper forces. Crack initiation in the test was assisted by an EDM notch at the known origin site. Crack detection was by an ultrasonic method that could be used in service, and the crack size at the start of the propagation test was an easily detectable 1.12 mm. Crack growth was periodically measured in the test by use of acetate replicas. The failure point was declared when the crack covered virtually the entire lug cross section. The component and the lug crack are illustrated in Figure 7. The crack growth characteristic obtained in the spindle full-scale test program is shown in Figure 8 below as the blue data points. The red line is a curve fit to the data obtained with Empirical Damage Tolerance methods, described in reference [3]. EDT allows the determination of Da/DN relationships in the form of the Paris Law, using only test data (the crack growth curve and the test load spectrum). The basic EDT equation is:

Da DN

(

= F PV

a

)

φ

(13)

Where upper case D’s are used to denote that the result is empirical and to not imply knowledge of an infinitesimal crack growth. The term PV is a normalized vibratory load parameter which is usually set at 1.00 equal to the maximum half peak-to-peak spectrum load. a is the crack length. F is a crack growth rate parameter and φ is a curve shape parameter, both of which are determined by a trial and error fit to the experimental data.

Damage Tolerance Approach for Dynamic Loaded Rotorcraft Components

943

View Crack Length

EDM Notch

1.51”

Crack Front Shapes

1.19”

Fig. 7 CH-53E Main Rotor Spindle Lug Crack.

In the case of the CH-53E spindle, a shape change near at the end of the crack growth as depicted in Figure 8. As the result, the crack growth prompted formulation of 2 curve segment equations of the form of Eq. 13, as discussed in Reference [3].

Fig. 8 CH-53E Spindle Crack Propagation Test Result and Curve Fit.

These empirical crack growth equations can be numerically integrated with any load set, so that a different load spectrum than applied in the test program can be evaluated. This was the original intent of this method – to remove any “analysis”

944

J. Zhao and D. Adams

content from crack growth fatigue substantiation so that the conclusions are entirely based on an independent test program. One feature of the numerical integration – inclusion of a threshold below which no crack growth occurs – does require choosing a threshold in terms of da/dN value from a standard coupon test crack propagation Material Characterization (the da/dN curve). As an example of how the integration works, two very different load spectrums are considered. The first is the CH-53E damper moment load spectrum used to conduct the test program. Since the main rotor damper is a relatively “constant load” device, most of the cycles occur around 65% of the maximum load, which is a Ground-Air-Ground or GAG load. A control load, such as main rotor rotating pushrod load, exhibits a very different characteristic, with most of the cycles around 20% of the maximum (GAG) load. This illustrated in Figure 9, with both spectrums normalized to the same maximum load (PV = 1.0).

Fig. 9 Notional Comparison of Helicopter Flight Load Spectrums.

Integration Eq. 13 using the parameters defining two segments with these two spectrums has the primary result that the pushrod load crack propagation time is orders of magnitude longer because the basic load is so much lower. But of most interest here is the fact that the basic shape of the crack growth in service changes. This can be illustrated by normalizing the crack growth by the fraction of cycles to failure, as shown in Figure 10.

Damage Tolerance Approach for Dynamic Loaded Rotorcraft Components

945

Fig. 10 Normalized Crack Growth Behaviour at Different Load Spectrums.

This result shows how the pushrod load produces much more of a “hockey stick” shape, which is a very different reliability problem that the damper load spectrum. For example, a nominal threshold was included in the integrations for these two curves, and it can be seen that a different threshold choice could produce even more curvature in the pushrod load result. This is because that if the infrequent GAG load is the only load propagating the crack initially, many thousands of hours could elapse before the crack grows enough to exceed the detectability threshold, at which time the rate of crack growth can accelerate rapidly. This characteristic can provide the benefit of allowing a delay in the start of the inspection protocol. On the other hand, the damper load spectrum produces significant growth right away and could benefit from an improvement in the initial detectable crack size. Inspection capability and POD modelling In this case study, an ultrasonic technology (UT) has been considered. Its detectability was carefully studied and further validated under the anticipated field condition. To establish creditable detection limit, typical crack size of the interest was determined and full scale components embedded with the identified crack size at anticipated critical location were prepared. The UT method was applied onto the cracked specimens and outcomes of detection/non-detection were documented. The study revealed that the UT method possesses a high reliability of detecting a crack of 1.12 mm. Due to the lack of sufficient NDI demonstration data, a nonparametric POD model is suggested in this case study. The detection capability is modelled as a step function. To account for potential unknown unknowns, it is further assumed that there is a 90% probability of detection of a 1.12 mm crack. A

946

J. Zhao and D. Adams

sketch of the POD model is given in Figure 11. In the plot, the aforementioned POD is represented by POD Model 5C and 5D.

POD 5C/5D Non-Parametric Model

POD 5A Parametric Model

Fig. 11 Plot of Notional POD Models Considered.

To address the potential numerical challenge associated with the infinite slope of the step function, the POD models were slightly adjusted. In addition, a “detection cap” (a maximum POD that is not exceeded) is introduced in the models to address potential issue with miss-detection. For POD Model 5D, the detection cap is assumed to be 99.9% with associated crack size of 5.08 mm, while a cap of 97.5% is imposed to POD Model 5C. This implies that there is a 2.5 % of chance of miss – detection even at very large crack size. For further benchmarking purpose, a parametric model, POD Model 5A, is considered. The POD 5A represents a NDI with lower detectability and wider detection range, as compared to POD Model 5C and 5D. Baseline risk assessment To fully understand the effect of inspection on DT risk reduction, a baseline risk assessment is performed first. To illustrate the key concept and technical elements of the DT risk assessment and RBMO application, the following discussions will be focused on case based on pushrod load spectrum. In this example, five random variables (RVs), including initial crack size (Cini), rate of the first stage crack crack growth (F), stress intersity threshold (Thre), rate of the second stage crack crack growth (F2), scaling Factor for normalized load (PFAC), were considered. Introducing the fifth RV aims to address the variablity associaetd with the fatigue load anticipated in fielded operation. Additionally, variability of usage has been implicitly incorportaed in the load spectrum used, which is based on the concept of composite worst case (CWC) usage. Other crack

Damage Tolerance Approach for Dynamic Loaded Rotorcraft Components

947

growth parameters, for example the curve shape parameters for the two segments are assumed to be fixed values. Statistical distributions of the random variables considered are summarized in Table 1 and associated statistical characteristics were obtained from statistical analysis of material testing and load data. Table 1 Statistical Characterization of EDT Uncertainties Considered.

Variable Cini F Thre F2 PFAC

Description Initial flaw size Rate of the first stage crack crack growth Stress intersity threshold Rate of the second stage crack crack growth Scaling Factor for normalized load

Distribution Lognormal Lognormal Lognormal Lognormal Lognormal

In this study, moderate amunt of scatter is considered for the key RVs and coefficient of variation (CoV) of 15% is suggested for the purpose of demonstartion. The associaetd mean values of each RV can be determined from the nominal “design” value with associated percentile used. For example, the percentile of nominal value for crack growth related RVs, such as F, Thre, and F2 are assumed to be 50%, since average material bahavior is expected during the testing. The percentile of nominal value for initial flaw size is assumed to be 90% followed the common practice used in industry for characterization of inspection capability. Therefore, mean value of crack size is estimated to be 0.635 mm based on assumed 90th percentile value of 1.12 mm and a CoV of 15%. This mean value is in good agreement with a clear detectable defect size established from previous study. Similarly, the nominal value associetd with the nominal value of the scaling factor for fatigue load is assumed to be associated with 95th percentile. This assumpation is consistent to the notional six-nine’s fatigue relaibility approach used in the rotorcraft industry. It is also noted that the selection of a lognormal distribution for statistical characterization is based on past experience and supported by Goodness-of-Fit analyses for a limited amount of material characterization and flight test data. In a Damage Tolerance approach, the component retirement life is consiste of crack initiation life and crack growth life. For the spindle lug case, the crack growth life is estimated be of 9,665 hrs. To determine the fatigue crack growth (FCG) life, the average values for crack growth parameters, usage, load, initial crack size, and threshold are used. The trasition between crack initiation and growth is defined by the clear detectable defect with a mean value of 0.635 mm. Probabilistic analyses were carried out for the notional component crack growth life of 9,665 flight hrs and the associated probability of failure (PoF) without considering inspection is around 10-1. The higher than anticipated PoF is attributed to the fact that the EDT model was derived from experimental study which possesses average material behaviour. Therefore, it is reasonable to predict a high chance of failure. Extensive simulations of the example case with pushrod

948

J. Zhao and D. Adams

load have been performed to determine probability of failure as a function of component retirement life. The results are summarized and shown in Figure 12. As depicted in the figure, the PoF increases with the increase of retirement life. The desired six-nine’s reliability can be achieved at a retirement life of 1,500 hrs without inspection.

FCG Life of 9,665 Fight Hrs with PoF of 10-1 FCG Life of 1,500 Flight Hrs with PoF of 10-6

Fig. 12 Example Case Probability of Failure vs. Fatigue Crack Growth Life [Without Inspection].

In this study, ASIS methodology has been applied for fast probability computation. ASIS PoF calculation reaches convergence after 900 ASIS samples. For the purpose of further benchmarking, both FORM and ASIS methodology are applied in the baseline DT risk assessment. Figure 12 shows the comparison of PoF estimates between FORM and ASIS. In this case, great agreements between those two methodologies are observed at various levels of specified fatigue crack growth life. Inspection optimization Once the baseline DT risk assessment is performed, we are ready to proceed to investigate the effect of inspection planning in DT risk reduction and management. A short life crack growth database containing 5,000 data points generated in the baseline assessment is used in inspection optimization. To illustrate the scatter associated with crack growth, 200 randomly selected samples from the database is plotted. As depicted in Figure 13, the plotted short life a-vs.-N sampling data exhibits a tremendous amount of scatter with a lowest crack growth life of 3,000 hrs. To ensure adequate level of structural integrity, the first inspection shall be

Damage Tolerance Approach for Dynamic Loaded Rotorcraft Components

949

scheduled before the predicted first DT failure. In addition, the baseline risk assessment indicates there is 1.5075×10-6 chance that a crack would grow from the initial size of 0.635 mm to the critical size of 27.9 mm within 1,500 hours. Therefore, the first inspection shall be performed no later than 1,500 flight hours. In this study, we are assuming that the detected cracking component will be immediately repaired and restored to its original “as-built” condition. Once a new part is in service, it would take ten thousand hours to initial a crack for further growth.

Multiple Insp. Optimization Reference FCG Retirement Life 9,665 Flight Hrs

Significant Amount of Scatter in FCG

1200

Crack Size [mils]

1000

800

600

110 mils Crack Length with Detectability of 97.5%

400

P-F Interval of 4,250 Flight Hrs

200

0

0

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

Fatigue Growth Life [hrs] Fatigue Crack Crack Growth Life [Flight Hours]

Fig. 13 Plot of a-vs.-N Curves for Short Life Samples.

To ensure adequate level of structural integrity dictated by six-nine’s reliability requirement for a target component crack growth life of 9,665 flight hours, there are roughly 6 fixed interval inspections needed and the first inspection shall be performed prior to 1,500 flight hours.

950

J. Zhao and D. Adams

Fig. 14 Sensitivity Plot of Inspection Time Short on PoF [Six Optimal Inspections using POD Model 5A].

RBMO analysis has been performed to determine the optimal inspection intervals to maximize risk reduction for each of the three POD models considered. Figure 14 shows the sensitivity of inspection time on DT risk management at each of the six inspection events considered using POD Model 5A. The sensitivity plot shown in Figure 14 provides more insight into the design space for optimal inspection planning. The data points marked in red represent the top 10% candidates to perform the inspection, while the blue points are the bottom 10% candidates. The lowest points are the optimal choices for most efficient inspections. As shown in the plot, the first inspection can be effectively performed between 1,000 and 1,500 hours of crack growth time. Additional insepctions will be followed per suggestion from RBMO with a variate inspection interval ranging from 1,000 to 2,000 hours of addiotnal crack growth time measured from the ajencent inspeciton. The resultsing PoF with the optimal inspeciton schedule is around 8×10-7. To fully explore the design space of optimal inspection, extensive RBMO runs were conducted with different sceneries to study effect of number of inspections, Pod models, and difference between optimal and user specified fixed intervals on DT risk management. The outcomes of the study are presented in Figure 15 Figure 18. Effect of POD models As discussed earlier, three POD models are considered in this study. The POD Model 5C and 5D are developed to represent detection capability of the UT method sued for the subject spindle lug. These two models are essentially identical except different values of detection cap are assumed. In Model 5C, a 97.5% of detection cap is imposed at a crack size of 2.794 mm, while a detection cap of 99.9% is forced in Model 5D. Clearly, Model 5C is less capable as compared to Model 5D. The effect of POD models in terms of different detection caps on DT

Damage Tolerance Approach for Dynamic Loaded Rotorcraft Components

951

risk management is investigated and the results are depicted in Figure 15, where the PoF as a function of number of optimal inspections based on both PODs are plotted.

Baseline PoF of 10-1

1.0E+00 1.0EͲ01 1.0EͲ02 1.0EͲ03

PoFwOptInspUsingPOD5D

Target PoF of 10-6

1.0EͲ04

PoFwOptInspUsingPOD5C

PoF

1.0EͲ05

PoFw/tInsp(Baseline)

1.0EͲ06 1.0EͲ07

Number of Fixed Interval Inspections for PoF of 10-6

1.0EͲ08 1.0EͲ09 1.0EͲ10 1.0EͲ11 1.0EͲ12 1.0EͲ13 1.0EͲ14 0

5

10

15

20

NumberofOptimized Inspections

Fig. 15 Example Case PoF as a Function of Number of Optimal Inspection.

As anticipated, the PoF decreases with the increase of number optimal inspections. The benefit of Model 5D over the Model 5C is also highlighted. It is observed that two magnitude of risk reduction can be achieved through Model 5D once the number of optimal inspection exceeding 10. To ensure the desirable six nine’s reliability, five optimized inspections per suggested by RBMO is sufficient if POD Model 5C is used. If the detection cap can be further improved, as the case associated with Model 5D, the number of optimal inspection can be further reduced to four. User-defined fixed interval inspection Often, inspection planning is subjected to other maintenance/logistic requirements. Therefore, a fixed interval inspection plan might be preferred over the adjustable inspection scheduling. It is important to evaluate adequacy of user-specified inspections to ensure required structural integrity. First, we are considering POD Model 5A, which is introduced to represent a NDI technique possessing lower characteristic detection capability (a90) with wider spread than the other two POD models. Based on extensive simulations, Figure 16 highlights the effect of POD 5A, 5C and 5D on the DT risk reduction as a function of fixed inspection intervals varying from 500 hours to 5,000 hours. It is observed that there is essentially no difference on amount of risk reduction obtained using POD Model 5A and 5C. This finding illustrates the inadequacy of using a single characteristic value to represent NDI capability (for example, a90), as being used in the traditional approach for inspection scheduling. As anticipated, the dispersion of a POD model also affects the effectiveness of POD in DT risk management. The PoF associated

952

J. Zhao and D. Adams

with inspection interval of 1,500 hours is around 10-5. Considering additional one 9’s of reliability implicitly incorporated in the CWC usage, the actual PoF can be further adjusted to 10-6. Therefore inspection shall be performed every 1,500 hours to ensure the adequate level of structural integrity, if POD Model 5C is considered. This finding coincident with the inspection interval established through conventional approach, where the crack growth life using average crack growth property is 4,250 hours. Using a safety factor of 3.0, the inspection interval is determined as 1,400 hours. This value illustrates that the conventional methodology does produce 6-9’s reliability in this particular case.

Baseline PoF of 10-1 1.0E+00 1.0EͲ01

1,000 Flight Hrs FCG Life with PoF of 10-6 [POD 5A/5C]

1.0EͲ02 1.0EͲ03 1.0EͲ04

Target PoF of 10-6

1.0EͲ05

PoF

1.0EͲ06 1.0EͲ07 1.0EͲ08

PoFwInspUsingPOD5A

1.0EͲ09

1,500 Flight Hrs FCG Life with PoF of 10-6 [POD 5D]

1.0EͲ10 1.0EͲ11

PoFwInspUsingPOD5C PoFwInspUsingPOD5D

1.0EͲ12 PoFw/tInsp(Baseline)

1.0EͲ13 1.0EͲ14 0

1000

2000

3000

4000

5000

FixedInspection Interval[FlightHrs]

Fig. 16 Plot of PoF as a Function of Fixed Inspection Interval.

The difference of DT risk reduction due to the optimal inspection and userdefined fixed interval inspections are also studied using POD Model 5C. As shown in Figure 17, the optimal inspections are much more effective in risk reduction than the conventional fixed interval inspections. With five inspections, the optimal scheduling reduces the DT risk from 10-1 to 10-6, while the five fixed interval inspections yield a less effective result with a reduced PoF of 5×10-4.

Damage Tolerance Approach for Dynamic Loaded Rotorcraft Components

953

Baseline PoF of 10-1

1.0E+00 1.0EͲ01 1.0EͲ02 1.0EͲ03

Target PoF of 10-6

1.0EͲ04 1.0EͲ05

PoF

1.0EͲ06 1.0EͲ07 1.0EͲ08 1.0EͲ09 1.0EͲ10 1.0EͲ11

PoFwFixedInspUsingPOD5C

1.0EͲ12

Number Inspections for PoF of 10-6

PoFwOptInspUsingPOD5C

1.0EͲ13

PoFw/tInsp(Baseline)

1.0EͲ14 1.0EͲ15 0

5

10

15

20

NumberofFixedIntervalInspection

Fig. 17 Impact of Inspection Strategy on PoF as a Function ofNumber of Inspections.

The impact of missing-detection It is not a rare NDI does not detect a crack, even with a larger crack size. In this study, the possibility of miss-detection can modelled by a detection cap imposed at large crack size in POD model. As mentioned earlier, the POD Model 5C is introduced to investigate the effect of possible miss-detection. Intuitively, for the DT risk management purpose, more inspections may be required if the inspection capability is limited. Figure 18 reinforces the point. As shown in the figure, RBMO suggests placing the first and second inspections at almost the same time. This surprising arrangement is attributed to the detection cap associated with the POD model. For the POD Model 5C, inspection will always have 2.5% chance of miss-detection for even very large crack size. As the result, RBMO suggests to performing two independent inspections at almost the same time to enhance the detectability and achieve maximum reduction of risk of failure. The first inspection is placed right before the predicted first failure due to extreme load and material conditions followed by the second inspection in a very short interval. With two closely placed inspections, the change of miss-detection reduces to 0.0625% (2.5% × 2.5%). It seems counter-intuitive to place two inspections close together, but if there is a detectability cap, this practice can improve reliability

954

J. Zhao and D. Adams

PoF Reduction Attributed to First Two Insp

PoF at the Target Service Life is 2x10-7

First Inspection

Second Inspection

Stabilized PoF as the benefit from the subsequent Inspections'

Component Service Life [hrs]

Fig. 18 PoF Reduction as a Function of Inspection Time [Five Optimal Inspections Using POD Model 5C].

Procedure for inspection interval determination Based on the results of the case study presented in this section and our current understanding of the RBMO methodology, a procedure for performing DT risk assessment and optimal inspection scheduling is proposed and summarized as below. •

•

•

Development and validation of DT model Review design requirements and fielded Reliability and Maintainability data; Develop/review Finite Element (FE) / Fracture Mechanics (FM) models for further DT assessment; Perform deterministic FM assessment to correlate predictive results with legacy data, fielded experience, and teardown findings; Conduct additional DT coupon / component tests for further model validation, as needed; Selection of NDI and POD model qualification Identify candidate NDIs and quantify NDI capabilities (POD / PFA); Demonstrate / qualify NDI capability and POD model; DT uncertainty quantification and baseline risk assessment Compile material database and document statistical models for DT inputs;

Damage Tolerance Approach for Dynamic Loaded Rotorcraft Components

955

•

•

Perform baseline risk assessment to identify opportunity for reliability improvement via NDI and conduct POD trade-off studies to confirm for NDI / POD selection; Perform probabilistic sensitivity study to identify the key contributors and document potential impact of uncertainty parameters on underlying reliability assessment; Risk based inspection planning and optimization Perform preliminary study to identify optimal time for first inspection and number of inspections required; Perform trade-off study to identify impact of POD uncertainty and repair quality on improved structural reliability; Perform final inspection plan assessment incorporating R&M and logistic requirements and confirm desired reliability improvement; Documentation and decision Document entire analysis processes with maintenance requirements, assumptions, models, inputs, simulation strategy, standard work / best practice applied, outputs of baseline assessment, sensitivity analysis, inspection options and repair quality trade-off, and recommended inspection plan; Communicate recommended inspection planning and associated with mishap risk for leadership decision.

5 Summary and Conclusions 1.

An risk-based methodology Crack Growth Damage assessment and inspection optimization has been developed.

Tolerance

2.

A level of reliability equal to 6-9’s (less than 1 failure in 1 million components) is recommended for the Damage Tolerance case, same as the current Safe-Life standard.

3.

The Damage Tolerance reliability is determined independent from any reliability contribution from the crack initiation life, in order to provide the required reliability for any application.

4.

The Crack Growth Damage Tolerance reliability model considers the variabilities from the initial crack size, crack growth rate, non-destructive inspections, flight loads, and usage spectrum. The 6-9’s reliabilities are allocated 1 to usage, 2 to flight loads, 1 to crack growth, and 2 to inspections.

5.

An example case using the results from a crack growth fatigue test on a helicopter main rotor spindle was successfully evaluated using the reliability model and inspcetion optimization framework.

956

J. Zhao and D. Adams

6.

For the specific example case it was shown that the conventional lifereduction determination of an inspection interval provides more than 69’s reliability.

7.

An optimized inspection protocol was determined for the example case which results in fewer inspections while still providing 6-9’s of fatigue reliability.

The study presented in this paper paves a path for further technical development and offers opportunity for adaptation of the methodology and correlation with fielded experience and data. It opens the door for further integration of DT models, risk assessment techniques, and optimization methodology for life-cycle condition based maintenance and cost-benefit analysis.

References [1] Berens, A.P., Hovey, P.W., Skinn, D.A. (1991), In: Risk Analysis for Aging Aircraft Fleet, US Air Force Wright Laboratory Report, WL-TR-91-3066, vol. 1 (October 1991) [2] Ditlevsen, O., Madsen, H.O.: Structural Reliability Methods. John Wiley & Sons, New York (1996) [3] Adams, D.O., Tristch, D.: Empirical Damage Tolerance. In: Proceeding of the American Helicopter Society 61st Annual Forum, Grapevine Texas (June 2005) [4] Gallagher, J.P., Babish, C.A., Malas, J.: Damage Tolerant Risk Analysis Techniques for Evaluating the Structural Integrity of Aircraft Structures. In: Proceeding of the 11th International Conference on Fracture, Turin, Italy (March 2005) [5] Guidelines for the Naval Aviation Reliability – Centered Maintenance Process, NAVAIR 25-403, Department of Defence (2005) [6] Kulkarni, S.S., Achenbach, J.D.: Optimization of Inspection Schedule for a Surface – Breaking Crack Subject to Fatigue Loading. Probabilistic Engineering Mechanics 22, 301–312 (2007) [7] Wu, Y.-T., Zhao, J., Shiao, M., Millwater, H.: Efficient Methods for Probabilistic Damage Tolerance Inspection Optimization. In: Proceeding of the 51st AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, Orlando, Florida (April 2010) [8] Zhao, J., Adams, D.O.: Achieving Six-Nine’s Reliability Using an Advanced Fatigue Reliability Assessment Model. In: Proceeding of the American Helicopter Society 66th Annual Forum, Phoenix Arizona (May 2010) [9] Thompson, A.E., Adams, D.O.: A Structural Reliability Evaluation of Fail-Safe Helicopter Dynamic Components, AHS National Rotorcraft Structures Technical Specialist’s Meeting, Williamsburg, Virginia (October 1991) [10] Zhao, J.: Structures Technologies for Condition Bases Maintenance (2007-C-10-01.1P3), Final Technical Report Submitted to CRI for CRI/NRTC CBMT Program PY 2007-2008 (2008)

Damage Tolerance Approach for Dynamic Loaded Rotorcraft Components

957

[11] Wu, Y.-T., Shiao, M., Shin, Y., Stroud, W.J.: Reliability Based Damage Tolerance Methodology for Rotorcraft Structures. Transactions Journal of Materials and Manufacturing, paper 2004-01-0681 (July 2005) [12] Thompson, A.E., Adams, D.O.: A Computational Method for the Determination of Structural Reliability of Helicopter Dynamic Components. In: AHS Annual Forum, Washington, D.C. (May 1990) [13] Certification of Transport Category Rotorcraft, FAA AC 29-2C (2008) [14] Non-Destructive Evaluation System Reliability Assessment, US Mil-HBDK-1823A, Department of Defence (2009)

26th ICAF Symposium – Montreal, 1-3 June 2011 *

Improvements in Fatigue Evaluations of Helicopter Transmissions Ugo Mariani1, Rosanna Molinaro1, Sergio Sartori2, Giuseppe Gasparini2, and Carlo Gorla3 1

AgustaWestland, Fatigue AgustaWestland, Transmissions 3 Politecnico di Milano, Dept. Of Mechanical Engineering 2

Abstract. Helicopter transmission is a critical assembly which significantly contributes to helicopter performance, life cycle costs and safety. Any improvement in fatigue evaluation can therefore give relevant benefits to the customer. AgustaWestland and Politecnico di Milano have carried out several research and development activities which can be considered a comprehensive project toward improved fatigue evaluation of transmissions.

1 Introduction Fatigue evaluation of helicopter transmission was usually carried out during the design phase according to structural analysis and prototype experience. Final clearance was provided by the full scale certification test of the transmission system. This test has max torque applied for 10 million loading cycles. Torque is factored for scatter using 1.4 amplification for one test, which is the typical standard due to cost and duration involved. Common practice has been to adjust transient peak loads and improved power ratings beyond test values. Most of the approach is based on legacy data which do not reflect present materials, technology and analysis tools. Moreover full scale test results are run outs for the large majority of gears involved and we do not have evidence of the true fatigue strength and failure mode. Some years ago we have started some research and development programs to improve fatigue evaluation of helicopter gears. At the same time other projects related to HUMS (Health and Usage Monitoring System) proved capable for fatigue tracking of helicopters. We are now combining these results to achieve benefits for existing helicopter models and additional advances for future projects.

2 Test Methodology and Data Base The research was focused on some specific items starting from a consolidated test methodology to develop a reliable fatigue data base for gears covering both LCF *

Oral presentation.

960

U. Mariani et al.

and HCF (Low and High Cycle Fatigue). By these tests we could check material strength and technological improvements on gear representative elements. Many aspects of gear design and manufacturing must be controlled in order to obtain such result, including material cleanliness, case depth and hardness, tooth root shape and roughness and compressive residual stresses. A precise knowledge of the S-N curve shape is of great importance to address service life. A single tooth bending (STB) test procedure has been developed to optimally map gear design parameters. A test program on case-carburized gears has been performed in order to appreciate the influence of various technological parameters on fatigue resistance and to draw the curve shape up to the giga-cycle region. In the first phase, tests up to 10 million cycles have been performed on four test groups differing by material (VAR and VIM-VAR 9310, and VIM-VAR EX-53) and by manufacturing process (ground fillet versus un-ground fillet). In the second phase, the VIM-VAR 9310 ground fillet specimen has been tested up to 100 million cycles. All the gear types were shot peened. FEM analysis, strain gauge measurements and rating formula of AGMA standards are used to express test loads in terms of tooth root stresses. The test fixture was designed specifically for the present research program. By changing the length of the anvil on the left side, the position of the load along the flanks of the tooth can be varied, thus changing the stresses on the two loaded teeth. With an appropriate length of the anvil, the symmetric condition can be obtained, and the pin is used only for the positioning of the gear and can be removed for test. In this way, no load can be absorbed by the pin and the load and stress on the two teeth are the same. The tests have been performed on a mechanical resonance 60 kN Schenck pulsator, without the pin. Figures 1and 2 shows the test facility and the specimen during test. Failure mode is point out in Figure 3. A second test fixture was adjusted to a hydraulic rig to allow testing in LCF, since resonant Schenck has a transient loading during start up which is not suitable for short duration. This second rig allowed some complementary activities for FEM model validation by strain survey. The relation between the applied load and the tooth root stress has been investigated with three different approaches: AGMA standard, finite element analyses and strain gauge measurements (Figure 4). The fatigue limits obtained in the test program are much higher than those included in AGMA and ISO rating standards, but a direct comparison with that data is not meaningful because they are not specific to the aerospace applications and do not consider the influence of such parameters like shot peening. AgustaWestland rating procedures are based on the use of a continuous S-N shape curve S/SL = H + A x (N+C)B where S is the stress, N is the number of cycles, SL is the fatigue limit, and H, A, B and C are constants computed by best fit of test data.

Improvements in Fatigue Evaluations of Helicopter Transmissions

Fig. 1 HCF Test facility.

Fig. 2 Test fixture.

961

962

U. Mariani et aal.

Fig. 3 Failure mode.

Fig. 4 FEM Stress analyses.

In the very high cycle fatigue f tests on 9310 VIM-VAR, two failures occurred iin the range between 10 and d 100 million cycles. The results of the very high cycle tests confirm the curve sh hape determined with the ordinary tests and its asymptottic value (Figure 5).

Improvements in Fatigue Evaluations of Helicopter Transmissions

963

3G6320T00451 120000

100000

F [N]

80000

60000

40000

20000

0 1,00E+00 1,00E+01 1,00E+02 1,00E+03 1,00E+04 1,00E+05 1,00E+06 1,00E+07 1,00E+08 1,00E+09 Number of cycles Test

Test Very High Cycle Fatigue

GEAR 06

GEAR 05

Fig. 5 SN data, megacycle fatigue.

GEAR 05 2200

2000

σ [MPa]

1800

1600

1400

1200 1,00E+03

1,00E+04

1,00E+05

1,00E+06 451

N 551

1,00E+07 651

751

Fig. 6 Correlation of test data.

1,00E+08

1,00E+09

964

U. Mariani et al.

VIM-VAR EX53 and 9310 (both according to AgustaWestland proprietary specifications) have shown the highest values of fatigue resistance with a slightly higher figure for EX53. The fatigue limit of 9310 VIM-VAR with unground fillet is about 8 % lower while the fatigue limit of 9310 VAR (according to AMS6265) and form grinding is about 11 % lower (Figure 6). Data base was then improved with four different specimen groups of nitrided gears typically used in AgustaWestland transmissions. Two of them were in 32CDV13 with different heat treatment and case depth specifications and two were in Nitralloy N, respectively shot peened and unpeened. The fatigue limit of Nitralloy N is about 19% lower than the more performing specification of 32CVV13, but with the application of shot peening the difference is almost compensated. With the families tested up to now, the fatigue limits of nitrided gears are slightly lower than those of case hardened ones. The test have also pointed out a different behavior in the low cycle region, with test data of nitrided gears significantly lower than case hardened. This is well known in the scientific community and in accordance with the shape curves provided by rating standards.

3 Gear Teeth Loading and Fatigue Evaluation Bending fatigue tests are generally performed using an STBF (Single-Tooth Bending Fatigue) scheme rather than reproducing gear meshing. The data for actual running conditions can then be determined by means of an appropriate factor, which can be explained as a consequence of a different load ratio R and of statistical considerations depending on the number of teeth loaded (Figure 7). The load ratio R, which is defined as the minimum test load versus the maximum test load in a load cycle, is R = 0 in running gears and typically R = 0.1 in STBF tests. A limited block loading test program was carried out by hydraulic rig to check interaction of overloading with fatigue life of gears. We used torque spectrum data for a utility helicopter and scaled load amplitude to obtain damage equal to 1 with

Fig. 7 Engine and transmission monitoring.

Improvements in Fatigue Evaluations of Helicopter Transmissions

965

8 repetitions, about 2.4 million cycles. Miner’s summations were based on CAL loading tests on the same type of gear specimens. In six tests Miner’s damage was in a range from 1 to 3. Two additional tests provided abnormal results higher than 8. No failures occurred prior to Miner’s = 1.0. Although a limited set of test was carried out, applicability of Miner’s damage was confirmed to reassess gear fatigue lives in case of over torque loading or new power spectra. In order to improve the calculation of the fatigue load on each single gear meshing, on the basis of torque time histories, which are available thanks to the HUMS system described in the following, load spectra on each single meshing for the typical operating conditions can be reconstructed by tracking the torque flow though the transmission. By considering a time history of proper length, the stress spectra on gear teeth include the effect of the interaction of the torque fluctuation frequency with the mesh frequency of each reduction stage. Combining suitably scaled stress spectra with S-N shape curves, effective Miner’s damages can be therefore calculated for typical flight conditions and can be compared with the traditional approach based on the assumption of a constant torque value for each condition, thus neglecting the influence of the frequency effect.

4 Hums for Transmissions The above described activity is also exploited for improving the reliability and accuracy of the on-board HUMS systems. The Agusta Westland current view about these systems is summarized below: Status of the main functions of HUMS for Transmissions Function Health

Transmission system

Technology Status Techniques for health monitoring include vibration and debris analysis. Overall HUMS health monitoring is considered more mature than HUMS usage monitoring

Using vibration analysis techniques monitor the health of each gear, bearing and shaft in the transmission so as to identify impending transmission problems and their causes

Transmission vibration analysis is fairly mature and all commercially available HUMS provide this function.

Monitor the condition of every oil-wetted rubbing surface and report if excessive wear is detected.

Simple chip detectors have been in common use for a long period. Full-flow quantitative debris monitors are being integrated into HUMS (but sensor reliability issues have to be solved)

966

U. Mariani et al.

Function

Usage

Technology Status

Techniques applied in support of usage monitoring include event counting (flight hours, engine starts, GAG cycles …), torque monitoring, flight condition monitoring … While the simpler techniques have been widely used, the more complex techniques (eg flight condition monitoring) are less mature and difficulties have been experienced with the measurement of some parameters. Overall HUMS usage monitoring is considered less mature than HUMS health monitoring

Transmission system Monitor mechanical component (eg gear) usage and structural component (eg rotor masts) usage

Torque monitoring is provided in most HUMS and with appropriate software can be applied fir assessing damage accrual for fatigue lifelimited gears. The importance of usage monitoring is well recognized and much effort is being applied to improve the capability of HUMS in this area. This function is provided in most HUMS

Monitor torque exceedances

Fatigue tracking of transmission system Transmission Usage Monitoring TUM on board function is that more specifically and more relevant to gear tooth fatigue lives continuous tracking. TUM collects five torque spectra during flight: one for each engine and one for each rotor. Each torque spectrum is defined as a set of 36 torque intervals, where a time counter is recorded for each interval, expressed in seconds. The torque inputs are provided by the engine torquemeters and the tail drive shaft torquemeter to the SMC (System Management Computer). These inputs are used, in real time, to compute the main rotor torque estimate value. The TUM log files in the DTC (Data Transfer Cartridge) are then transferred to and decoded by the ground station computer, where the torque spectra are allocated to each monitored component by means of its torque path code, i.e. for each component a specific torque spectrum shall be calculated based on its position inside the transmission system and its nominal rotating speed. Some transmission system gears can be fatigue life-limited, depending on helicopter model and usage. Furthermore it is fairly common for gear durability to

Improvements in Fatigue Evaluations of Helicopter Transmissions

967

limit the engine power available to the rotor system over much of the helicopter operating envelope. All helicopters provide an indication of the level of torque developed by each engine and hence torque sensing is always included. Engine torque thus provides a direct load measurement parameter for main rotor gearbox gear although tail torque needs to be taken into account if not measured. Gear usage monitoring (via torque history monitoring) may provide a number of benefits (Figure 8) if transmission gears are life limited and therefore durability is a relevant parameter. These include: • • •

Avoidance of some gearbox removal which, without usage monitoring, would have been required on the basis of the uncertainty associated with pilot reporting of magnitude and duration of an overtorque. Performance enhancement by allowing normal torque limits to be exceed on the basis that the effects of such exceedances are monitored and taken into account. Life extension of individual gears if their lives are based on the measured severity of in service usage (Figure 9).

Improving in safety and in the avoidance of unnecessary gearbox removals can yield significant cost benefits for some helicopters. The economic benefit from event-type monitoring will be sub-divided into two main categories: - Benefit from flying time monitoring which means that service intervals and retirement times for fatigue life-limited components are tracked on the basis of manually logged times. Over-estimation of flying time can artificially increase the values of these parameters and results economic penalties. - Benefit from exceedance monitoring which takes into account that maintenance action is a function of the identified or assumed level and duration of the exceedance. Conventional methods of identified exceedances place reliance on reports made by the aircrew. The high safety standards adopted often result in expensive overhauls because, in the absence of more reliable data, the worst case may have to be assumed. Since events like gearbox over-torque can require expensive maintenance, an accurate recording of exceedances via HUMS would provide objective evidence to identify the characteristics of the event.

968

U. Mariani et al. MAIN ROTOR TORQUE 70.00

60.00

50.00

% Time

40.00

30.00

OVERTORQUE 20.00

10.00

0.00 AUTOROTATION ÷ GROUND OPERATION ÷ TAXING

DESCENT ÷ LEVEL FLIGHT ÷ REVERSAL

LANDING ÷ BANK TURN

DECELERATION ÷ HOVERING ÷ PULLUP/PUSH-OVER ÷ FLARE TO IGE/OGE

TAKE-OFF ÷ CLIMB ÷ ACCELERATION

-

-

0 ÷ 7100

7100 ÷ 12120

12120 ÷ 13007

13007 ÷ 14186

14186 ÷ 16267

16267 ÷ 18333

>18333

Main Rotor Torque band [Kgm]

Fig. 8 Main transmission torque.

10000 Potential safety risk

9000

Life consumption

8000

Current service life

Design life of component

7000 6000 5000

Additional use gained with monitoring

Severe usage

4000 3000 2000

Mild usage

1000 0 0

2000

4000

6000

8000

10000

12000

14000

Time in operation

Fig. 9 Effect on retirement of usage monitoring.

5 Conclusions This paper presents a summary of research activities aimed to improve fatigue evaluation of helicopter transmissions. A dedicated test methodology was developed to provide a reliable and cost effective data base. Improvements in HUMS performances have provided flight by flight tracking of power spectra and aircraft usage in service. Combining these capabilities will result in increased

Improvements in Fatigue Evaluations of Helicopter Transmissions

969

accuracy and confidence in fatigue evaluation, improving performances, reducing costs and supporting conditions based maintenance initiatives.

Acknowledgments This paper presents an overview of many AW projects on fatigue of helicopter transmissions jointly carried out or in progress at AgustaWestland and Politecnico di Milano. Several colleagues provided expertise on specific tasks as already reported in dedicated conferences and workshops on gears fatigue and HUMS. The authors wish to express their appreciation to: Francesco Rosa, Mauro Filippini and Franco Concli of Politecnico di Milano; Angela Cerrini, Giuseppe Ratti, Raul Rovellotti and Amos Curlante of AW Fatigue; Alberto La Fortezza, Alfredo De Nuntiis, Andrea Gabrielli, Simone Bertolotti of AW Trasmissions.

References [1] Gasparini, G., Mariani, U., Gorla, C., Filippini, M., Rosa, F.: Bending Fatigue Tests on Helicopter Case Carburized Gears: Influence of Material, Design and Manufacturing Parameters. In: Proceedings of AGMA Fall Technical Meeting (2008) [2] Cerrini, A., Gasparini, G., Mariani, U., Sartori, S., Gorla, C., Filippini, M., Rosa, F.: Bending Fatigue Tests on Helicopter Nitrided Gears: Influence of Different Materials and Manufacturing Parameters. In: Proceedings of VDI International Conference on Gears, Garching (D), October 4-6 (2010) [3] ANSI/AGMA 2101-D04 – Fundamental Rating Factors and Calculation Methods for Involute Spur and Helical Gear Teeth [4] ISO 6336-3, Second edition, Calculation of load capacity of spur and helical gears — Part 3: Calculation of tooth bending strength (2006) [5] McPherson, D.R., Rao, S.B.: Methodology for Translating Single-Tooth Bending Fatigue Data to be Comparable to Running Gear Data, in Gear Technology (2008) [6] Mariani, U., Molinaro, R.: Usage monitoring for a safer aviation. In: Proceedings of 4th EASA Symposium, Cologne (2010)

Author Index

Adams, David 927 Alderliesten, R.C. 105 Amsterdam, E. 199 Anandavel, K. 697 Andr´e, B. 655 Aoki, Yuichiro 561 Arnold, E.M. 685 Asakawa, M. 83 Backman, David 785 Bakuckas Jr., John G. 735 Balasubramani, P. 697 Ball, D.L. 265 Balter, J. 465 Barter, S.A. 199, 249 Bayles, R. 685 Baynham, J.M.W. 347 Beauvais, Yves 601 Benassi, L. 495 Benedictus, R. 105 Best, R. 61 Beumler, Th. 105 Bogdanov, S. 415 Bombardier, Yan 231, 335 Bos, M.J. 473 Botsev, Y. 465 Bourinet, J.-M. 399 Bucci, R.J. 265 Buchholz, R. 389 Cajani, Maurizio 585 Campbell, S.K. 839 Cenic, Nikola 539 Chattopadhyay, A.B. 325

Chen, Y.J. 811 Ciotola, Roberto 585 Clark, Graham 1, 625 Curtin, T.J. 347 Da Veiga, A. 359 David, J´er´emy 585 Delgrange, G. 93 Despujols, J. 655 Donald, J.K. 265 Doucet, J. 753 DuQuesnay, D.L. 671 Ehrstr¨ om, J.C.

93

Figueroa-Gordon, D. 753 Fitzpatrick, M.E. 753 Fleischer, Th. 61, 135 Franke, R. 61 Freed, Yuval 145 Fressinet, M. 427, 655 Fuchs, F. 427 Fujita, S. 83 Gadalinska, E. 277 Gali, S. 465 Galyon Dorman, Sarah E. Gasparini, Giuseppe 959 Glinka, G. 325, 415 Goldbach, S. 61 G´ orka, A. 167 Gorla, Carlo 959 Green, W.P. 867, 915 Gruner, J. 61

645

972 Gud’s, P. 465 Guertsman, V.Y. Gupta, N. 465

Author Index

635

Haberle, Till 539 Hajjar, Zahi 827 Hammond, Matthew J. 645 Handelman, A. 465 Heinimann, M. Boscolo M. 753 Hena-Zamal, H. 189 Herrmann, C. 529 Hilfer, G. 529 Hirano, Yoshiyasu 561 Hirose, Yasuo 507 Hoa, Suong V. 179 Hoa, S.V. 155, 189 Hoeppner, David W. 219 Houghton, S.J. 839 Huber, N. 811 Hu, W. 303, 615 Irving, P.E. 753 Ito, S. 443 Ivetic, G. 855 Ivetic, Goran 771 Iwahori, Yutaka 561 Jackson, P. 303 James, A.D. 839 James, M.A. 265 Janhunen, Harri 453 Jylh¨ a, Juha 453 Kabir, A. 155 Kaniowski, J. 277 Klimaszewski, S. 519 Koca´ nda, D. 167 Kok, L.J.J. 73 Korzeniewski, B. 277 Kothari, Rushabh 551 Krasnowski, B.R. 867, 915 Kressel, I. 465 Kurdelski, Marcin 519, 573 Kuroki, Hiroshi 207, 315 Kuwayama, K. 83 Laborde, A. 359 Lanciotti, A. 855 Leblanc, Benoit 827 Lee, Yongwon 645 Leski, Andrzej 519, 573

Liao, Min 231, 335 Liljedahl, D. 753 Li, X. 867, 915 Machida, S. 83 Mactabi, Roham 179 Madelpech, P. 427, 655 Makeev, Andrew 119 Mariani, Ugo 959 Mattrand, C. 399 McDonald, M. 199 Mellings, S.C. 347 Meneghin, Ivan 771, 855 Mikheevskiy, S. 415 Minakuchi, Shu 507 Miyazawa, K. 207 Molinari, Gianluca 771, 855 Molinaro, Rosanna 959 Moore, Gary 73 Morales, M. 855 Moran, P.J. 685 Morita, H. 207 Murooka, T. 207 Nagao, Yosuke 561 Nakamura, T. 83 Nesterenko, B.G. 39 Nesterenko, G.I. 39 Neumair, M. 495 Newman Jr., J.C. 289 Nikishkov, Yuri 119 Oca˜ na, J.L. 855 Ohnuki, Takeshi 561 Oikawa, K. 207 Okada, T. 83, 443 Okumura, I. 207 Oldersma, A. 473 Paroissien, E. 359 Patterson, Eann A. 785 Peters, C. 529 Pfaller, Rupert 899 Pillai, A.C.R. 465 Polese, C. 855 Porro, J.A. 855 Poston, Ken 73 Prakash, Raghu V. 697 Prasad, M.H. 465 Quilter, A.C.

375

Author Index Ramakrishnan, R. 289 Rans, C.D. 105 Reichard, S. 61 Reid, Len 797 Renaud, Guillaume 231, 335 Reymer, Piotr 573 Ridzewski, J. 61, 135 Ristori, V. 855 Rosca, Iosif D. 179 R¨ oßler, N. 529 Ruotsalainen, Marja 453 Rzepka, Sven 145 Sachse, M. 135 Salonen, Tuomo 453 Salvi, Jaco 585 Sartori, Sergio 959 Sathya, S. 465 Schirmacher, F. 135 Schubbe, J.J. 685 Sears, Thomas 785 Shekhter, A. 199 Shepherd, G. 753 Shigenari, Y. 207 Siljander, Aslak 453 Stefaniuk, M. 519 Stolz, C. 495 Struzik, Alain 877 Sugimoto, Sunao 443, 561 Sundaram, R. 465 Swift, Steve 27

973 Takeda, Nobuo 507 Tanaka, A. 207 Terada, H. 83 Th´eret, D. 399, 655 Tiong, Ung Hing 625 Torregosa, R.F. 615 Troiani, Enrico 771, 855 Tur, M. 465 Ueda, Yusuke 207, 315 Uz, M.V. 811 Valli`eres, G.M. 671 Ven¨ al¨ ainen, Ilkka 453 Venter, A.M. 855 Visa, Ari 453 Walker, K.F. 249 Walker, S.K. 375 Wallbrink, C. 303 Wanhill, R.J.H. 199 Westerman, Bud 735 Witek, Lucjan 721 Wronicz, W. 277 Yadav, A.K. 465 Yamashita, Yoichi Zapf, Bernd 539 Zasada, D. 167 Zhang, X. 753 Zhao, Jack 927

315

ICAF 2011 Structural Integrity: Influence of Efficiency and Green Imperatives

Temperature-Fatigue Interaction (European Structural Integrity Society)

Metallic Materials with High Structural Efficiency

Interpreting Imperatives

Interpreting Imperatives

Biaxial Multiaxial Fatigue and Fracture (European Structural Integrity Society)

Fracture Mechanics: Applications and Challenges (European Structural Integrity Society)

Biaxial Multiaxial Fatigue and Fracture (European Structural Integrity Society)

Influence Function Approach : Selected Topics of Structural Mechanics

Influence Function Approach : Selected Topics of Structural Mechanics

Green Communications and Networks: Proceedings of the International Conference on Green Communications and Networks (GCN 2011)

Multiaxial Fatigue & Fracture (European Structural Integrity Society)

Increasing Returns and Efficiency

Pressure Vessel Design: The Direct Route (Advances in Structural Integrity)

A Structural Theory of Social Influence (Structural Analysis in the Social Sciences)

Minimum Reinforcement in Concrete Members (European Structural Integrity Society)

A Structural Theory of Social Influence (Structural Analysis in the Social Sciences)

Influence

Vision: Coding and Efficiency

Aggregation, Efficiency, and Measurement (Studies in Productivity and Efficiency)

Green Networks. A Structural Analysis of the Italian Environmental Movement

Signal integrity

The Little Green Data Book 2011

Green Lantern (Vol 4) #60 Jan 2011

The Rise and Fall of Structural Marxism: Louis Althusser and His Influence

Influence: Science and Practice

Influence: Science and Practice

Influence Science and Practice

Multiscale Fatigue Crack Initiation and Propagation of Engineering Materials Structural Integrity

Application of Fracture Mechanics to Polymers, Adhesives and Composites, Volume 33 (European Structural Integrity Society)

Sphere Of Influence

ICAF 2011 Structural Integrity: Influence of Efficiency and Green Imperatives

Temperature-Fatigue Interaction (European Structural Integrity Society)

Metallic Materials with High Structural Efficiency

Interpreting Imperatives

Interpreting Imperatives

Biaxial Multiaxial Fatigue and Fracture (European Structural Integrity Society)

Fracture Mechanics: Applications and Challenges (European Structural Integrity Society)

Biaxial Multiaxial Fatigue and Fracture (European Structural Integrity Society)

Influence Function Approach : Selected Topics of Structural Mechanics

Influence Function Approach : Selected Topics of Structural Mechanics

Green Communications and Networks: Proceedings of the International Conference on Green Communications and Networks (GCN 2011)

Multiaxial Fatigue & Fracture (European Structural Integrity Society)

Increasing Returns and Efficiency

Pressure Vessel Design: The Direct Route (Advances in Structural Integrity)

A Structural Theory of Social Influence (Structural Analysis in the Social Sciences)

Minimum Reinforcement in Concrete Members (European Structural Integrity Society)

A Structural Theory of Social Influence (Structural Analysis in the Social Sciences)

Influence

Vision: Coding and Efficiency

Aggregation, Efficiency, and Measurement (Studies in Productivity and Efficiency)

Green Networks. A Structural Analysis of the Italian Environmental Movement

Signal integrity

The Little Green Data Book 2011

Green Lantern (Vol 4) #60 Jan 2011

The Rise and Fall of Structural Marxism: Louis Althusser and His Influence

Influence: Science and Practice

Influence: Science and Practice

Influence Science and Practice

Multiscale Fatigue Crack Initiation and Propagation of Engineering Materials Structural Integrity

Application of Fracture Mechanics to Polymers, Adhesives and Composites, Volume 33 (European Structural Integrity Society)

Sphere Of Influence

Recommend Documents