(NATO Science Series 361) J. N. Reddy (auth.), Carlos A. Mota Soares, Cristóvão M. Mota Soares, Manuel J. M. Freitas (eds.)-Mechanics of Composite Materials and Structures-Springer Netherlands (1999).pdf

Mechanics of Composite Materials and Structures
NATO Science Series

A Series presenting the results of activities sponsored by the NATO Science
Committee. The Series is published by lOS Press and Kluwer Academic
Publishers, in conjunction with the NATO Scientific Affairs Division.
A. Life Sciences lOS Press
B. Physics Kluwer Academic Publishers
C. Mathematical and Physical Sciences Kluwer Academic Publishers
D. Behavioural and Social Sciences Kluwer Academic Publishers
E. Applied Sciences Kluwer Academic Publishers
F. Computer and Systems Sciences lOS Press
1. Disarmament Technologies Kluwer Academic Publishers

2. Environmental Security Kluwer Academic Publishers
3. High Technology Kluwer Academic Publishers
4. Science and Technology Policy lOS Press
5. Computer Networking lOS Press
NATO-PCO-DATA BASE
The NATO Science Series continues the series of books published formerly in the NATO ASI
Series. An electronic index to the NATO ASI Series provides full bibliographical references (with
keywords and/or abstracts) to more than 50000 contributions from internatonal scientists
published in all sections of the NATO ASI Series.
Access to the NATO-PCO-DATA BASE is possible via CD-ROM "NATO-PCO-DATA BASE" with
user-friendly retrieval software in English, French and German (WTV GmbH and DATAWARE
Technologies Inc. 1989).
The CD-ROM of the NATO ASI Series can be ordered from: PCO, Overijse, Belgium
Series E: Mathematical and Physical Sciences - Vol. 361

Mechanics of Composite
Materials and Structures
edited by
Carlos A. Mota Soares

Cristov80 M. Mota Soares
and
Manuel J.M. Freitas

Department of Mechanical Engineering,
Instituto Superior Tecnico,
Technical University of Lisbon,
Portugal
..
Springer-Science+Business Media, B.V.
Proceedings of the NATO Advanced Study Institute on
Tr6ia, Portugal
July 12-24, 1998
A C.l. P. Catalogue record for this book is available from the Library of Congress.
ISBN 978-0-7923-5871-8 ISBN 978-94-011-4489-6 (eBook)

DOI 10.1007/978-94-011-4489-6
Printed on acid-free paper
Ali Rights Reserved

© 1999 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 1999
Softcover reprint ofthe hardcover 1st edition 1999
No part of the material protected by this copyright notice may be reproduced or utilized in
any form or by any means, electronic or mechanical, including photocopying, recording or
by any information storage and retrieval system, without written permission from the
copyright owner.
NATO ADVANCED STUDY INSTITUTE

Troia, Portugal
12-24 July, 1998
Main Sponsor
NATO - North Atlantic Treaty Organization, Scientific and Environmental Affairs
Division
Other Sponsors
ONREUR - Office of Naval Research International Field Office
ADTRANZ Portugal
FCT - Fun~ para a Ciencia e Tecnologia
FLAD - Fund~ Luso-Americana para 0 Desenvolvimento
NSF - National Science Foundation
SOOAPAL - Sociedade Grafica de pajA
BPA - Banco Portugues do Atlantico
ITS - Junta de Turismo de Setilbal
Rank Xerox Portugal
1ST - Instituto Superior Tecnico
IDMEC - Instituto de Engenharia Mecinica
Director
Prof. Carlos A. Mota Soares, Technical University of Lisbon, Portugal
Organizing Committee
Prof. Carlos A. Mota Soares, Technical University of Lisbon, Portugal
Prot: T.W. Chou, University of Delaware, USA
Prot: J.N. Reddy, University of Texas A&M, USA
Dr. J.R. Starnes, NASA, USA
Prof. J.O. Williams, University of London, U.K.
Local Organizing Committee

Prot: Carlos A. Mota Soares, Technical University of Lisbon, Portugal
Prot: Joaquim Infimte Barbosa, Nautical School Infante Dom Henrique, Portugal
Prof. Manuel J.M. Freitas, Technical University of Lisbon, Portugal
Prot: Crist6vao M. Mota Soares, Technical University of Lisbon, Portugal
CONTENTS
Preface ix
List of Participants xi
List of Contributors xix
Theory and Analysis of Laminated Composite Plates

J.N. Reddy
Micromechanics Analysis of Composite Materials

A.S.D. Wang 81
Predictim of the Effective Damage Properties and Failure Properties of Non Linear
Anisotropic Discontinuous Reinforced Composites
K. DaTim, J. Fitoussi, G. Guo and D. Baptiste 131
Mechanics of Tbree-Dimmsiooal Textile Structural Composites: Processing

Tsu-Wei Cllouand TimothyD. Kostar 151
Mechanics ofTbree-Dimmsiooal Textile Structw"al Composites: Analysis

Tsu-Wei Cllou and Timothy D. Kostar 163
Mechanics ofThree-Dimmsional Textile Structural Composites: Performance Modeling

Tsu-Wei 01011 and Kishore Pochiraju 173
Testing and Modeling of IneIastic Behavior of Fibrous Composites

C.T. Sm 183
Characterizatim of Strain Rate-Depmdmt Behavior in Polymeric Composites

C.T.Sm 195
A Treatmmt of Interfacial Cracks in the Presmce of Frictim

C.T. Sm and W. Qian 205
Impact and High Rate Testing of Composites: An Overview

B.RK.BlackmanandJ.G. Williams 215
Impact and High Rate Testing of Composites: High Rate Delamination Testing
B.RK. Blackman and J.G. Williams 225
Composite for Marine Applications

Part 1. Testing of Materials and Structures for Surface Vessels
P. Davies 235
Composite for Marine Applications

Part 2. Underwater Structures
P. Davies and P. Cllauchot 249
Examples Co Non-Destructive Testing of Glass- and Carbon-Fiber-Reinforced

Polymer-Matrix Composites
A.I. BrIDlner, 1. Neuenschwander 261
Towards Standardised InterIaminar Fracture Tests for Unidirectional Fiber-Reinforced

Polymer-Matrix Composites
A.J. BrIDlner 279
Advantages and Disadvantages of Thermoset and Thermoplastic Matrices for Continuous

Fibre Composites
A.Cervenka 289
viii
Hygrothennal Ageing of Glass Reinforced Composite Pipes

A.Cervenka 299
Composite Pipes Based on Thermoplastic Matrices Reinforced by Continuous Fibres

A.Cervenka 309
Impact Damage of Composite Structures

M. de Freitas and L. Reis 319
Remarks 00 Linear and Nonlinear Variational Design Sensitivity Analysis

F.-J. Barthold 331
Remarks 00 Computing the Energy Release Rate and its Sensitivities

F.-J. Barthold and S. Mesecke 341
Some Remarks 00 Robust Design

F.-J. Barthold and S. Mesecke 351
Optimization of Thin-WaIled Structures

E. Stein and F.-J. Barthold 361
Theory and Computation of Multilayer Composites

E. Stein and J. Tessmer 369
Theory and Computation of Damage and Failure of Composites

E. SteinandJ. Tessmer 379
Optimal Design of Composite Structures with Integrated Piezoelectric Laminae

Victor M Franco Correia, Crist6vio M. Mota Soares and Carlos A. Mota Soares 389
Optimization of Composite Structures Using Response Surface Approximations

R T. Haftka, R Vitali and B.V. Sankar 409
Genetic Algorithms for Optimization of Composite Laminates

R T. Haftka 431
Identification Techniques in Composite Laminates

Pauli Pedersen 443
Sensitivity Analysis and Inverse Problems for Laminates and Materials

Pauli Pedersen 453
Asymptotic Approach to Optimal Structural Design with Brittle-Fracture ConsIraints

Part I: Prototype Problem
N.V. Banichuk 465
Asymptotic Approach to Optimal Structural Design with Brittle-Fracture Constraints

Part 2: Deterministic and Stochastic Problems
N.V. Banichuk 477
Some Composite Structures Experiences at NASA Langley Research Center

James H Starnes, Jr. and Damodar R Ambur 489
Prospects os Smart Structures for Future Airaaft

K. W. Dittrich 509
PREFACE
This book contains the edited version of the lectures presented at the NATO Advanced Study Institute
on Mechanics of Composite Materials and Structures, held in Tr6ia, Portugal, 12-24th of July 1998 and
organized by IDMEC - Institute of Mechanical Engineering, Instituto Superior Tecnico, Technical
University of Lisbon. The Institute was attended by 102 participants from 21 countries, including
leading scientists and engineering staff from universities, research institutions and industry and Ph.D.
students. Some participants presented invited and contributed papers during the Institute and almost all
participated actively in discussions on scientific and technological aspects during the Institute.
The Advanced Study Institute provided the first opportunity for internationally recognized researchers
and engineers to present the foundations, current state of art, recent development and directions of
research of all essential techniques related to mechanics of composite materials and structures. The
Institute provided a forum for interaction among eminent scientists and engineers from different
schools of thought and young researchers, leading to future increase in the international collaboration
in research in composites materials and structures.
The Institute addressed the foundations and current state of art of the mechanics of composite materials
and structures. Special emphasis was placed on classic and recent developed theories of composite
laminated beams, plates and shells, micromechanics, impact and damage analysis, mechanics of textile
structural composites, high strain rate testing and non-destructive testing of composite materials and
structures. Topics of growing importance were addressed, such as: numerical methods and
optimization, identffication and damage monitoring, and homogenization techniques for composite
materials and structures. The state of the art for modeling of smart composite materials, optimal
design with advanced materials, software codes and industrial applications was also presented.
The edited version of the majority of papers presented at the Advanced Study Institute will be
published this year in a Special Issue of the Journal of Computers and Structures and in a current issue
of the Journal of Mechanics of Composite Materials and Structures.
Without the sponsorship and financial support of the Scientific and Environmental Affairs Division of
NATO, the Advanced Study Institute and this book would not have been possible. The financial
support of all other sponsors contributed decisively to the success of the Institute.
The Editors deeply appreciated all the advice and help in organizing the Institute given by Prof. Luis
Veiga da Cunha of the Scientific and Environmental Affairs Division of NATO. The Editors are also
indebted to all members of the organizing committee (prof. T.W. Chou, Prof. J.N. Reddy, Dr. J.H.
Starnes and Prof. J.G. Williams) for all the outstanding work that led to a very successful Institute. We
are also grateful to all authors for their effort in writing the lectures and papers in time, allowing this
book to be published as planned.
Special thanks to IDMEC staff, Mrs. Andrea Marques and Mr. Amandio Rebelo, for their effort in
administrative planning and support of the Institute.
We are grateful to our families, for all the support given during the organization of the Institute and of
this book.
Lisbon, April 1999

Crist6vao M. Mota Soares
Manuel J.M. Freitas
ix
LIST OF PARTICIPANTS
Lectures:
Nickolay Banichuk
The Institute for Problems in Mechanics, 101, Prospect Vemadskogo
117526 Moscow, Russia
Didier Baptiste
Lab. Microstruc. M6canique des Materiaux, LMJ, 151 - Bould. de I'Hllpital, ENSAM
75640 Paris Cedex 13, France
Frau-JOIeph Barthold
Institut flIr Baumechanik \Uld Numerische, Mechanik, Universitllt Hannover
Applestrasse 9A, 30167 Hannover, Germany
Bamber R.K- BIaekmaa

Department of Mechanical Engineering, Imperial College
Exhibition Road, London SW7 2BX, United Kingdom
Aadreas J. Bnmner
EMPA, Oberlandslrasse 129
CH-8600 DObendorf; Switzerland
Tony Cervenka
University of Manchester & UMIST, Manchester Materials Science Centre
Grosvenor Street, Manchester MI 1HS, United Kingdom
Tau-Wei Chou
Department of Mechanical Engineering, Univ. of Delaware, 126, Spencer Laboratory
Newark, Delaware 19716-3140, U.S.A
Peter Davies
IFREMER - Laboratoire Materiaux Marins, IFREMER-DITlIGOIMM
BP-70, 29280 Plouzane, France
Raphael Haftka
Univ. of Florida, Dept. Aerospace En808., P.O.Box 116250.231 Aerospace Building
Gainesville, FL 32611-6250, U.S.A
Noboru Kikuehi
University of Michigan, Dept. Mech. En808. Applied Mech.
Computational Mech. Lab., Ann Arbor, MI 48109-2215, U.S.A.
Pauli Pedersen
Technical UniversityofDerunark, Dept. of Solid Mechanics, Building 404
DK-2800 Lyngby, Derunark
J.N.Reddy
Dept. of Mechanical Engineering, Texas A&M University, College Station
Texas 77843-3123, U.S.A.
Carlos A. Mots Soares

IDMECIIST, Av. Rovisco Pais
1049-00 1 Lisboa, Portugal
James Stsrnes
NASA Langley Res. Ctr •• Mail Stop 190
Hampton, Virginia, VA 23681-2199. U.S.A
xi
xii
Erwin Stein
Universitl!t Hannover, Inst. of Civil Engng. and Numerical Mechanics,
Applestrasse 9A, D - 30167 Hannover, Gennany
C. T.Suo
Purdue University, School of Aeronautics & Astronautics,
West Lafayette, IN 47907-1282, U.S.A.
Albert S. Wang
Drexel University, College of Engineering, Dept. Mechanical Engng. & Mechanics
32 nd & Chestnut Streets, Philadelphia, PA 19104, U.S.A.
CarlZweben
Lockheed Martin Missiles & Space, Advanced Technology Manager
230 Mall Blvd. King of Prussia, PA 19406, U.S.A.
Participants:
MemisAcar
Loughborough University, Depart. Mechanical Engineering
Loughborough Leics, LEI I 3TU, United Kingdom
Robert D. Adams
University ofBristoI, Dept Mech. Engng.
Queen's Building, University Walk, Bristol BSS I TR, United Kingdom
Georges Akhras
Royal Military College of Canada, Civil Engineering Dept.
P.O.Box 17000, STN Forces, Kingston, Ontario, ON K7K 7B4, Canada
NuriAkkas
Middle East Technical University, Department of Engineering Sciences
Ankara, 06531, Turkey
Aurelio Araujo
IDMEClIST, Av. Roviseo Pais
1049-001 Lisboa, Portugal
Mette Bagge
Technical University of Denmark, Department of Solid Mechanics
Building 404, DK - 2800 Lyngby, Denmark
William M. Banks
University ofStrathclyde, Department of Mechanical Engineering
James Weir Building, 75 Montrose Street, Glasgow G I IXJ, United Kingdom
Joaquim Infante Barbosa

Escola Nautica Infante D. Henrique, Av. Bonneville Franco
Pa~ de Arcos, 2780 Oeiras, Portugal
Maria Helena Barros

Faculdade Ciencias e Tecoologia, Universidade de Coimbra - DEC
Pinhal de Marrocos, 3049 Coimbra Codex, Portugal
Ayech Benjeddou
Cooservatoir Nationale des Arts et Metiers, Struct. Mech. & Systems Lab.
2, Rue Conte, F - 75003 Paris, France
Asghar Bhatti
University of Iowa, Dept. of Civil and Environmental Engng.
Iowa City, Iowa 52242, U.S.A.
x.iii
Jan Blacbut
The University of Liverpool, Dept. of Mech. Engineering
Applied Mechanics Division, Liverpool L69 3BX, United Kingdom
Bostjan Brank
Siovenian National Building & Civil Eng. Inst., Dimiceva ul. 12
SI-IOOO Ljubljana, Slovenia
Micbacl Bruyneel
University of Liege, LTAS - Structures Aerospatiales Bat C3
Rue Ernest Solvay, 21,4000 Liege, Belgiwn
Henry R. Busby
The Ohio State University, Department of Mechanical Engineering
206 W.J8thAvenue, Colwnbus,OH43210-1I07, U.S.A.
Albert H. Cardon
Free University of Brussels (VUB), Dept. ofMecltanics of Materials & Constructions
(TW), Pleinlaan, 2, 1050 Brussels, Belgiwn
Vidor 1. Cben
The Boeing Company, 2401 E. Wardlow Road, MIS C078-0315
Long Beach, CA 90807-5309, U.S.A.
JoHCirne
Universidade de Coimbra, FCTUC - Polo IIIDEM
Pinhal de Marroc:os, , 3030 Coimbra, Portugal
Wdoro Pinto Correia

Escola Nautica Infimte D. Henrique
Av. Bonneville Franc:o I'a9o de Arc:os, 2780 Oeiras, Portugal
Mic:hae\ A. Crisfield
Imperial College Science, Tech. & Medicine, Depart. of Aeronautics,
Prince Consort Road, London, SW7 2BY, United Kingdom
Subbend. Datta
Department of Mechanical Engineering, University of Colorado
Boulder, CO 80309-0427, U.S.A.
Marina DavydO\'ll
Institute of Continuous Media Mechanics, Urals Branch, Russian Academy Sciences
I, Acad. Korolev Str., 614013, Penn, Russia
ZDZ8D8 Dimitrovori
IDMECIIST, Av. Rovisc:o Pais
1049-00 1 Lisboa, Portugal
Kay W. Diuncb
Daimler-Benz Aerospace AG, Military Airaafts
D-81663 MOnchen, Germany
Piyusb Datta
US Army Cold Regions Research and Engineering Laboratory
72 Lyme Road, Hanover, NH 03755, U.S.A.
RaUl Fangueiro
Universidade do Minho, Departamento de Engenharia TSxtii
Campus de Azun!m, 4800 GuimaIies, Portugal
Constantina Filion
European Commission, Joint Research Centre, Scientific Office
P.O.Box 2,1755 Zg. Petten, The Netherlands
xiv
Victor Reis Franco

Escola N{lUtica Infante D. Henrique Av. Bonneville Franco,
Pa,.o de Arcos, 2780 Oeiras, Portugal
Mary Frecker
Pennsylvania State University, Dept of Mechanical Engineering
233 Reber Bldg., SUniversity Park, PA 16802-1412, U.S.A.
Manuel Freitas
Technical University of Lisbon, Department of Mechanical Engineering
Av. Rovisco Pais, 1049-001 Lisboa, Portugal
Dimitri V. Georgievskii
Mechanical and Mathematical Faculty, Moscow State University
Moscow, 119899, Russia
Mustafa Glden
Izmir Institute ofTechnology, Gaziosmanpasa Bulvari
N°. 16, Cankaya, Izmir, Turkey
Jose Miranda Guedes

IDMECIIST, Av. Rovisco Pais
Laurent Guillaumat
L.A.M.E.F. - E.N.S.A.M., Esplanada des Arts et Metiers
33405 Talence Cedex, France
HuHong
Universidade do Minho, Departamento de Engenharia Textil
Campus de Azurem ,4800 Guimaries, Portugal
Matthias Hllrmann
Institute of Structura1 Mechanics, University of Stuttgart
Pfaffenwaldring 7,0-70550 Stuttgart, Germany
DavidHui
University of New Orleans, Dept Mechanical Engineering,
Lakefront, New Orleans, LA 70148, U.S.A.
Han-Koo Jeong
Department of Ship Science, University of Southampton
Highfield, Southampton SO 17 1BJ, United Kingdom
Ramazau Karakuzu
Dokuz EylW University, Faculty of Engineering
Dept. of Mechanical Engineering, 35100, Bomova-Yzmir, Turkey
FraukKo
Drexel University, Department of Materials, Eng., Fibrous Materials Research Center
31st & Market Sts., Philadelphia, PA 19104, U.S.A.
Michal Kocvara
Institute of Applied Mathematics, University ofErlangen!Nuremberg
Martensstrasse 3, D-91 058 Erlangen, Germany
Stavros Kourkoulis
National Technical Univ. of Athens, Eng. Sci. Dept., Section of Mechanics
5, Heroes Po1ytechnion Av., Theocaris BId., Zografou Campus, 15773, Greece
Rog6rio Leal
Universidade de Coimbra, FCTUC - P610 IIIDEM
Pinhal de Marrocos, 3030 Coimbra, Portugal
xv
Cbi Tsieb Liu

OL-AC Air Force Res. Laboratory / PRSM, 10 E. Saturn Blvd.
Edwards AFB CA 93524-7680, U.S.A.
Gaspar Macbado
Universidade do Minho, Departamento de Maternatica
Campus de Azun!m , 4800 Guimariies, Portugal
Erdogan Madenci
The University of Arizona, College of Engin. & Mines, Depart. of Aerospace& Mechanical Engin.
Aero Building, Tucson, Arizona 85721, U.S.A.
Antonio Torres Marques

INEGIIFaculdade de Engenharia do Porto, Rua dos Bragas
4099 Porto Codex, Portugal
JamesMasoo
University of Notre Dame, Aerospace & Mech. Engineering
365 Fitzpatrick Hall, Notre Dame, Indiana, IN 46556-5637, U.S.A.
Ales Mihelic
Ministry of Science & Technology, Republic of Slovenia
Siovenska 50, SI-1000 Ljubljana, Slovenia
Jose M. S. Moita
Universidade do Algarve, Dept Eng". Mec., Escola Superior de Tecnologia
Campus da Penha, Estrada da Penha, 8000 FARO, Portugal
A1eksanderMuc
Institute of Mechanics & Machines Design, Cracow University of Technology
Ul. Warszawska24, 31-155 Krak6w, Poland
Jennifer Myers
The Ohio State University, Dept. Civil, Environ. Eng. & Geodetic Science
470 Hitchcock Hall, 2070 Neil Avenue, Columbus, Ohio 43210, U.S.A.
Maria Augusta Neto

Universidade de Coimbra, FCTUC - P610 IIIDEM
Pinhal de Marrocos, 3030 Coimbra, Portugal
Miguel Matos Neves

IDMEClIST, Av. Rovisco Pais
Philippe Olivier
Lab. Genie Mecanique de Toulouse, Institut. Univ. Tech. Paul Sabatier
50 Chernin des Maraicher, 31077 Toulouse Cedex 4, France
Vladimir Palmov
St. Petersburg Technical University, Depart. Mechanics and Control Processes
195251 Politechnicheskaya 29, St Petersburg, Russia
Eduardo Ribeiro Pereira

Instituto Superior Tc!a:licolDECivil, Av. Rovisco Pais
1096 Lisboa Codex, Portugal
Boris E. Pobedria
Composite Mechanics Department, Moscow State University
Moscow, 119899, Russia
y uri S. Protasov
Bauman Moscow State Technical University, Office for International Connections
2-nd Bauman Str. 5, 107005 Moscow, Russia
xvi
Maria Amelia Ramos Loja

Escola Nautica Infante D. Henrique, Av. Bonneville Franco
Paryo de Arcos, 2780 Oeiras, Portugal
Luis Reis
Pedro Ribeiro
University of Southampton, Institute of Sound and Vibration Research
Highfield Southampton, S017 IBJ, United Kingdom
Rolands Rikards
Riga Technical University, Faculty of Civil Engineering, Specialized Inst. Computer Anal. Structures,
I, Kalku Street LV-1658, Riga, Latvia
Helder Rodrigues
1049-00 I Lisboa, Portugal
Raimund Rolfes
German Aerospace Center, DLR InstitOt fllr Strukturmechanik,
Postfach 3267, 0-38022 Braunsdiweig, Germany
Farzad Rostam-Abadi
U.S. Army TARDECrrACOM, 4343 Robinwood Royal Oak
Michigan 48073, U.S.A.
George Rozvany
Essen University, Structural Design
FB 10, P.O.Box 103764, D-45711 Essen, Germany
Jose Santos
Vasily V. Saurin
Russian Academy of Science, The Institute for Problems in Mechanics
101 Prospect, Vernandskogo, 117526 Moscow, Russia
Mariam SAnchez de Ia Sierra

Universidad Carlos III, C1Butarque, IS
28911 Leganes, Madrid, Spain
Arlindo Silva
Manuel Gon~alves da Silva

Universidade Nova de Lisboa, FCTIDEC
Quinta da Torre, 2825 Monte da Caparica, Portugal
Anand V. Singh
The University of Western Ontario, Dept. of Mechanical Engineering
Faculty of Engineering Science, London, Ontario, N6A 5B9, Canada
Ivica Smojver
University of Zagreb, Faculty of Mechanical and Naval Architecture
Ivana Lucica 5, HR-lOOOO Zagreb, Croatia
Crist6viio M. Mota Soares

xvii
Win Somboonsong
Drexel University, Dept. of Materials Engineering, Eng. Fibrous Materials Research Center
31st & Market Sts., Philadelphia, PA 19104, U.S.A.
Alzal Suleman
IDMECIIST, Av. RovisCD Pais
Andras Szekeres
Technical University of Budapest, Dept Engineering Mechanics,
Muegyetem 5, 1-3 Budapest, H-IIII, Hungary
EwaTurska
IFrR - Polish Academy of Sciences, Ul. Swietokrzyska, 21
00-049 Warsaw, Poland
Geoffrey Turvey
Lancaster University, Engng. Dept, Faculty of Applied Sciences,
Lancaster LAI 4YR, Bailrigg, United Kingdom
CIuistiaD Van VoaoIe

Free University of Brussels (VUB), Dept. of Mechanics of Materials & Constructions
(TW), Pleinlaan, 2, 1050 Brussels, Belgimn
06vier Vermaat
University of Liege, LTAS - Structures Aerospatiales Bat C3
Rue Ernest Solvay, 21, 4000 Liege, Belgimn
KnysztofW--..n
IFrR - Polish Academy of Sciences, Ul. Swietokrzyska, 21
00-049 Warsaw, Poland
William E. Wolfe
Dept. Civil, Environ. Eng. & Geodetic Science, The Ohio State University
470 HitdJ.cock Hall, 2070 Neil Avenue, Columbus, Ohio 43210, U.S.A.
LIST OF CONTRIBUTORS
Damodar R. Ambur
NASA Langley Res. Ctr., Mail Stop 190
Hampton, Virginia, VA 23681-2199, U.S.A.
Nickolay Banichuk
The Institute for Problems in Mechanics, 10 I, Prospect Vernadskogo
117526 Moscow, Russia
Didier Baptiste
Lab. Microstruc. M6canique des Matl!riaux, LM3, lSI - Bould. de I'HOpital, ENSAM
75640 Paris Cedex 13, FI3IIce
Fraoz..Joseph Barthold
Institut flIr Baumechanik md Numerische, Mechanik, Universitit Hannover
Bamber R.K. BIadmwI

Aadreas J. Bnumer
EMPA, Oberlandslrasse 129
CH-8600 DObendorf; Switzerland
Tony Cervenka
University of Manchester & UMlST, Manchester Materials Science Centre
Grosvenor Street, Manchester Ml 1HS, United Kingdom
P.Chaochot
IFREMER - Laboratoire Man!riaux Marins, IFREMER-DITIlGOIMM
BP-70, 29280 Plouzane, FI3IIce
TSD-Wei Chou
Department of Mechanical Enginet2"ing, Univ. of Delaware, 126, Spencer Laboratory
Newark, Delaware 19716-3140, U.S.A.
Peter Davies
IFREMER - Laboratoire Materiaux Marins, IFREMER-DITIIGOIMM
BP-70, 29280 Plouzane, FI3IIce
K.Derrien
Lab. Microstruc. M6canique des Materiaux, LM3, lSI - Bould. de I'Hapital, ENSAM
Kay W. Dittrich
Daimler-Benz Aerospace AG, Military Aircrafts
0-81663 MOochen, Germany
J. Fitoassi
Lab. Microstruc. M6canique des Materiaux, LM3, 151 - Bould. de I'H/\pita\, ENSAM
Victor Reis Franco

Escola Nautica Infante D. Henrique Av. Bonneville Franco,
Pa\Xl de Arcos, 2780 Oeiras, Portugal
Manuel Freitas
xix
xx
G.Guo
Lab. Microstruc. Mecanique des Mateciaux, LM3, 151- Bould. de I'HOpital, ENSAM
75640 Paris Cedex 13, France
Raphael Haftka
Univ. of Florida, Dept. Aerospace Engng., P.O. Box 116250, 231 Aerospace Building
Gainesville, FL 32611-6250, U.S.A.
Timoty D. Kostar
Foster-Miller, Inc.
303 Bear Hill Road, Waltham, MA 02154, U.S.A.
S.Mesecke
Institut fllr Bawnechanik und Nwnerische, Mechanik, Universitat Hannover
J. Neuensdrwander
EMPA, Oberlandstrasse 129
CH-8600 Dllbendort; Switzerland
Pauli PedeneD
Technical University of Denmark, Dept. of Solid Mechanics, Building 404
DK-2800 Lyagby, Denmark
Klshore Poehiraja
Stevens Institute of Technology
Hoboken, NI07030, U.S.A.
W.Qian
J.N.Reddy
Dept. of Mechanical Engineering, Texas A&M University, College Station
Texas 77843-3123, U.S.A.
LuisReIs
B.V.Sankar
Univ. of Florida, Dept. Aerospace Engng., P.O.Box 116250,231 Aerospace Building
Gainesville, FL 32611-6250, U.S.A.

IDMEClIST, Av. Roviseo Pais
Crist6vlo M. Mota Soares

James R Starnes, Jr.

NASA Langley Res. Ctr., Mail Stop 190
Hampton, Virginia, VA 23681-2199, U.S.A.
EnriDStein
Universitat Hannover, Inst. of Civil Engng. and Nwnerical Mechanics,
Applestrasse 9A, D - 30167 Hannover, Germany
C. T.SDD
xxi
J. Tessmer
Universitlit Hannover, Inst. of Civil Engng. and Nwnerical Mechanics,
Applestrasse 9A, 0 - 30167 Hannover, Germany
R. Vita6
Univ. of Florida, Dept. Aerospace Engng., P.O.Box 116250,231 Aerospace Building
Gainesville, FL 32611~250, U.S.A.
Albert S. Wang
Drexel University, College of Engineering, Dept. Mechanical Engng. & Mechanics
32 nd & Chestnut Streets, Philadelphia, PA 19104, U.S.A.
J.G. Williams
THEORY AND ANALYSIS OF LAMINATED
COMPOSITE PLATES
J. N. REDDY
Department of Mechanical Engineering
Texas A&M University
College Station, Texas 77843-3123 USA
e-mail: jnreddy@reddy20.tamu.edu
1. Introduction
Composite materials consist of two or more materials which together
produce desirable properties that may not be achieved with any of the
constituents alone. Fiber-reinforced composite materials, for example,
consist of high strength and high modulus fibers in a matrix material.
Reinforced steel bars embedded in concrete provide an example of fiber-
reinforced composites. In these Lv.nposites, fibers are the principal load-
carrying members, and the matrix material keeps the fibers together, acts
as a load-transfer medium between fibers, and protects fibers from being
exposed to the environment (e.g., moisture, humidity, etc.).
Composite materials are commonly formed in three different types:
(1) fibrous composites, which consist of fibers of one material in a matrix
material of another; (2) particulate composites, which are composed of
macro size particles of one material in a matrix of another; and (3)
laminated composites, which are made of layers of different materials,
including composites of the first two types. The particles and matrix in
particulate composites can be either metallic or nonmetallic. Thus there
exist four possible combinations: metallic in nonmetallic, nonmetallic in
metallic, nonmetallic in nonmetallic, and metallic in metallic.
A lamina or ply is a typical sheet of composite material. It represents
a fundamental building block. A fiber-reinforced lamina consists of
many fibers embedded in a matrix material, which can be a metal like
aluminum, or a nonmetal like a thermoset or thermoplastic polymer. Often,
coupling (chemical) agents and fillers are added to improve the bonding
2
between fibers and matrix material and increase toughness. The fibers
can be continuous or discontinuous, woven, unidirectional, bidirectional, or
randomly distributed. Unidirectional fiber-reinforced laminae exhibit the
highest strength and modulus in the direction of the fibers, but they have
very low strength and modulus in the direction transverse to the fibers.
A poor bonding between a fiber and matrix results in poor transverse
properties and failures in the form of fiber pull out, fiber breakage, and fiber
buckling. Discontinuous fiber-reinforced composites have lower strength
and modulus than continuous fiber-reinforced composites.
A laminate is a collection of laminae stacked to achieve the desired
stiffness and thickness. For example, unidirectional fiber-reinforced laminae
can be stacked so that the fibers in each lamina are oriented in the same or
different directions (see Figure 1.1). The sequence of various orientations
+8
-8
Figure 1.1. A laminate made up of laminae of different fiber orientations.

3
of a fiber-reinforced composite layer in a laminate is termed the lamination

scheme or stacking sequence. The layers are usually bonded together with
the same matrix material as that in a lamina. A unidirectional laminate
(Le., all laminae have the same fiber orientation) will be very strong along
the fiber direction and weak in the transverse direction. The laminate will
be weak in shear also. If a laminate has layers with fibers oriented at 30° or
45°, it can take shear loads. The lamination scheme and material properties
of individual lamina provide an added flexibility to designers to tailor the
stiffness and strength of the laminate to match the structural requirements.
Laminated composite structures also have disadvantages. Because of the
mismatch of material properties between layers, the shear stresses produced
between the layers, especially at the edges of a laminate, may cause
delamination. Also, during manufacturing of laminates, material defects
such as interlaminar voids, delamination, incorrect orientation, damaged
fibers, and variation in thickness may be introduced. It is impossible to
eliminate manufacturing derects altogether. Therefore, analysis and design
procedures should account for the defects.
2. Constitutive Equations of a Lamina

In this section we study the mechanical behavior of a typical lamina,
which is the basic building block of a composite laminate. In formulating
the constitutive equations of a lamina we assume that (1) a lamina
is a continuum, i.e., no gaps or empty spaces exist and (2) a lamina
behaves as a linear elastic material. The first assumption amounts to
considering the macromechanical behavior of a lamina. If fiber-matrix
debonding and fiber breakage, for example, are to be included in the
formulation of the constitutive equations of a lamina, then we must consider
the micromechanics approach, which treats the constituent materials as
continua and accounts for the mechanical behavior of the constituents
and possibly their interactions. The second assumption implies that the
generalized Hooke's law is valid. It should be noted that both assumptions
can be removed if we were to develop micromechanical constitutive models
for inelastic (e.g., plastic, viscoelastic, etc.) behavior of a lamina.
Consider a laminated plate of total thickness h, composed of N
orthotropic layers with the principal material coordinates of the kth lamina
oriented at an angle Ok to the laminate coordinate x. Some of these layers
can be for actuation or sensing purpose (i.e., piezoelectric layers or shape
memory materials). The xy-plane is taken to be the undeformed midplane
no of the laminate, and the z-axis is taken positive upward from the
midplane. The kth layer is located between the points z = Zk and Z = zk+l
in the thickness direction.
4
The linear constitutive relations (Le., generalized Hooke's law) for the
kth orthotropic (piezoelectric) lamina in the principal material coordinates
of a lamina are
!~lr °rr-al~T)
Ql1 Ql2 0 0
Ql2 Q22 0 0 o C2 - a2 l:1T
= 0 0 Q44 0 o C4
0 0 0 Q55 o C5
0 0 0 0 Q66 C6
(k)
0 e31
u:r
0 0 e32
[ 00
- e24 0 (2.1)
e~5 0 0
0 0
where Q~;) are the plane stress-reduced stiffnesses and e~;) are the
piezoelectric moduli of the kth lamina, (O"i,ci,Ci) are the stress, strain, and
electric field components, respect! vely, referred to the material coordinate
system (Xb X2, X3), al and a2 are the coefficients of thermal expansion
along the Xl and X2 directions, respectively, and l:1T is the temperature
increment from a reference state, l:1T = T - To. For layers other than
piezoelectric layers, the part containing the piezoelectric moduli e~;) should
be omitted. The coefficients Q~;) are known in terms of the engineering
constants of the kth layer:
(2.2)
and the piezoelectric moduli are known in terms of the dielectric constants
d;,j and elastic stiffnesses as
~ ][1
0 0 0
[Qll
0
~l]
Ql2
0 e31]
e32 Ql2 Q22 0 0 0 d32
o
[e!, e24 = 0 0 Q44 0 d24 0
0 o 0 0 0 Q55 0 0
0 o 0 0 0 0 Q66 0 0 0
(2.3)
Since the laminate is made of several orthotropic layers, with their
material axes oriented arbitrarily with respect to the laminate coordinates,
5
the constitutive equations of each layer must be transformed to the laminate
coordinates (x, y, z). The transformed stress-strain relations relate the
stresses ((J"xx, (J"yy, (J"xy) to the strains (cxx, Cyy, 'Yxy) and components of the
electric displacement vector (£x, £y, £z) in the laminate coordinates
(2.4)
where the layer label k is omitted, and Oij are the transformed elastic
stiffnesses, a1, a2 and a6 are the transformed thermal coefficients of
expansion, and eij are the transformed piezoelectric moduli:
011 = Q11 cos4 () + 2( Q12 + 2Q66) sin 2 () cos 2 () + Q22 sin4 ()

012 = (Q11 + Q22 - 4Q66) sin2 () cos 2 () + Q12(sin4 () + cos4 ())
022 = Q11 sin4 () + 2( Q12 + 2Q66} ;,;in2 () cos 2 () + Q22 cos4 ()
016 - Q12 - 2Q66) sin () cos3 () + (Q12 - Q22 + 2Q66) sin3 () cos ()
= (Q11
026 - Q12 - 2Q66) sin3 () cos () + (Q12 - Q22 + 2Q66) sin () (;OS3 ()
= (Q11
066 = (Q11 + Q22 - 2Q12 - 2Q66) sin 2 () cos 2 () + Q66(sin4 () + cos4 ())
- 2 2
Q44 = Q44 cos () + Q55 sin ()
045 = (Q55 - Q44) cos () sin ()
-
Q55 = Q55 cos2 () + Q44 sin 2 () (2.5)
a1= a1 cos 2 () + a2 sin 2 ()
a2= al sin 2 () + a2 cos 2 ()
a6= 2( al - a2) sin () cos () (2.6)
e3l = e3l cos () + e32 sin ()
2 2
e32 = e31 sin2 () + e32 cos 2 ()
e36 = (e31 - e32) sin () cos () (2.7)
Here () is the angle measured counterclockwise from the x-coordinate to

the Xl -coordinate. We define
(2.8)
6
3. Classical and First-Order Shear Deformation Theories

of Laminated Plates
3.1. INTRODUCTION
The objective of this chapter is to summarize the governing equations of
two commonly used laminate plate theories, namely the classical and first-
order shear deformation theories. To provide a background for the theories
discussed in this chapter, an overview of the laminate plate theories is
presented first. The equivalent single-layer (ESL) theories are derived
from the 3-D elasticity theory by making suitable assumptions concerning
the kinematics of deformation or the stress state through the thickness
of the laminate. Here we present a review of the equivalent single-
layer theories. The equivalent single-layer laminate theories are those in
which a heterogeneous laminated plate is treated as a statically equivalent,
single layer having an anisotropic constitutive behavior, reducing the 3-D
continuum problem to a :l-D problem. The ESL theories are developed
by assuming the form of the displacement field or stress field as a linear
combination of unknown function", dnd the thickness coordinate [1-5].
The simplest ESL laminate theory is the classical laminated plate theory
(CLPT), which is an extension of the Kirchhoff (classical) plate theory to
laminated composite plates. The Kirchhoff hypothesis implies that straight
lines normal to the midplane of the plate before deformation remain straight
and normal to the midsurface after deformation, and consequently, both
transverse shear and transverse normal effects are neglected. In most
cases these assumptions hold. However, for thick laminates and laminates
with high degree of anisotropy, the transverse deformation effects can be
significant.
The next theory in the hierarchy of ESL laminate theories is the first-
order shear deformation theory (FSDT), which extends the kinematics of
the CLPT by including a gross transverse shear deformation in its kinematic
assumptions, i.e., the transverse shear strain is assumed to be constant with
respect to the thickness coordinate. Inclusion of this rudimentary form of
shear deformation allows the normality restriction of the classical laminate
theory to be relaxed. The first-order shear deformation theory requires
shear correction factors, which are difficult to determine for arbitrarily
laminated composite structures. The shear correction factors depend not
only on the lamination and geometric parameters, but also on the loading
and boundary conditions.
Second- and higher-order ESL laminated plate theories use higher-order
polynomials in the expansion of the displacement components through the
7
thickness of the laminate. The higher-order theories introduce additional

unknowns that are often difficult to interpret in physical terms. For
example, the third-order laminate theory (see Reddy [5,6]) with transverse
inextensibility is based on the displacement field
u(x, y, z, t) = uo(x, y, t) + z<Px(x, y, t) + z3 (- 3~2) (<Px + 88:

0)
v(x, y, z, t) = vo(x,y, t) + z<py(x, y, t) + z3 (- 3~2) (<py + 8~0)

w(x,y,z,t) = wo(x,y,t) (3.1)
The displacement field accommodates quadratic variation of transverse

shear strains (and hence stresses) and vanishing of transverse shear stresses
on the top and bottom of a general laminate composed of monoclinic layers.
Thus there is no need to use shear correction factors in a third-order theory.
Complete derivations of the governing equations of the third-order theory
of Reddy and its analytical and finite element solutions are presented in
Chapter 11 of Reddy [5].
In addition to their inherent simplicity and low computational
cost, the ESL models often provide sufficiently accurate description
of global response for thin to moderately thick laminates, e.g., gross
deflections, critical buckling loads, and fundamental vibration frequencies
and associated mode shapes. Of the ESL theories, the FSDT with
transverse extensibility appears to provide the best compromise of solution
accuracy, economy, and simplicity. However, the ESL models have
limitations that prevent them from being used to solve the whole spectrum
of composite laminate problems. First, the accuracy of the global response
predicted by the ESL models deteriorates as the laminate becomes thicker.
Second, the ESL models are often incapable of accurately describing the
state of stress and strain at the ply level near geometric and material
discontinuities or near regions of intense loading - the areas where accurate
stresses are needed most. In such cases, 3-D theories or multiple model
approaches are required.
3.2. CLASSICAL LAMINATE PLATE THEORY (CLPT)
Consider a plate of total thickness h composed of N orthotropic layers with

the principal material coordinates (x~, x~, x~) of the kth lamina oriented
at an angle (h to the laminate coordinate, x. Although not necessary,
it is convenient to take the xy-plane of the problem in the undeformed
midplane no of the laminate (see Figure 3.1). The z-axis is taken positive
8
upward from the midplane. The kth layer is located between the points
Z = Zk and Z = zk+1 in the thickness direction. The total domain no
of the laminate is the tensor product of no x (-h/2, h/2). In general, the
bourrdary of Wo is a curved surface r, with outward normal ft = nxex+nye y.
y
x
h..
2
h..
2
Figure 3.1. Coordinate system and layer numbering used for a typical
laminated plate.
The Kirchhoff hypothesis requires the displacement field (u, v, w) to be

of the form
u{x,y,z,t) =
awo
uo{x,y,t) - z ox
owo
v{x, y, z, t) = vo{x, y, t) - z 8y
w{x,y,z,t) = wo{x,y,t) (3.2)
9
where (Uo, vo, wo) are the displacements along the coordinate lines of a
material point on the xy-plane. Note that the form of the displacement
field (3.2) allows a reduction of the 3-D problem to one of studying the
deformation of the reference plane z = 0 (or midplane). Once the midplane
displacements (uo, Vo, wo) are known, the displacements (u, v, w) of any
arbitrary point (x, y, z) in the body can be determined using Eq. (3.2).
The strains associated with the displacement field (3.2) can be
computed using either the nonlinear strain-displacement relations or the
linear strain-displacement relations. For small strains and moderate
rotations case (i.e., the von Karman strains) the strain-displacement
relations of the classical plate theory are given by
(O)} {(1)
cxx }
= c~g + z c~f
cxx cxx
{ cyy }
{ (3.3)
"/xy 'V( ) 'V( )
/xy IXY
{cO} = {:~
"/xy
} = {
~
8y
: : :
+~
ax
~:~:
+ ~~
ax 8y
}, {c1} = { :t~ } -2g=~ }
"/xy
= {
aQ
x y
(3.4)
where (0) (0) (0))
( cxx, h b . d ((1) (1) (1))
are t e mem rane strams, an cxx , cyy , "/xy are the
cyy ,"/xy
flexural (bending) strains, known as the curvatures.
Once the displacements (uo, vo, wo) of the midplane are known, strains
at any point (x, y, z) in the plate can be computed using Eqs. (3.3) and
(3.4). Note from Eq. (3.3) that all strain components vary linearly through
the laminate thickness, and they are independent of the material variations
through the laminate thickness. Note that stresses are also linear through
the thickness of each layer; however, they will have different linear variation
in different material layers when Q~7) change from layer to layer. If we
assume that the temperature increment varies linearly, consistent with the
mechanical strains, we can write
(3.5)
and the total strains are of the form in Eq. (3.3) with
(0) - Q xx TO }
cxx cxx -Qxx T1 }
{(1)
{co} -_ { cyy
(0)
- QyyrD , {c
1 _
}- (1)
cyy - QyyT
1
(3.6)
(0)
"/xy - 2Q xy T 0 (1)
,,/xy - 2Qxy T1
10
The governing equations of motion of the classical plate theory

theory are obtained using the dynamic version of the principle of virtual
displacements (for details, see Reddy [5]). They are
(3.7)
(3.8)
(3.9)
where
(3.10)
(3.11)
(3.12)
(3.13)
The quantities (Nxx , N yy , N xy ) are called the in-plane force resultants, and
(Mxx, M yy , Mxy) the moment resultants, Qn the transverse force resultant,
and (10, II, 12) are the mass moments of inertia. The terms involving 12 are
called rotary (or rotatory) inertia terms, and are often neglected. The term
can contribute to higher-order vibration modes.
The boundary conditions of the theory are also obtained from the
virtual work principle. To present the boundary conditions on a curved
boundary whose normal is not necessarily parrallel to the plate coordinate
axes, we must relate the displacements and stress and moment resultants
in the plate coordinate system (x, y, z) to the coordinate system (n, s, r)
on the curved boundary whose unit outward normal vector ft is oriented
at an angle () from the x-axis. The direction cosines of the normal vector
11
are nx = cos () and ny = sin (). Hence, the transformation between the
displacements (uo n , uo s ) and (uo, vo) is given by
(3.14)
Similarly, the normal and tangential derivatives (WO,n, wo,s) are related to
the derivatives (wo,x, WO,y) by
(3.15)
The stresses (0"nn, 0"ns) are related to (0" xx, 0"yy, 0"xy) by the transformation
(3.16)
Hence we have
(3.17)
(3.18)
The force (natural) boundary conditions are then given by
N nn - Nnn = 0, N ns - Nns = 0, Vn - On = 0, Mnn - Mnn =0 (3.19)
on r u , where
8Mns
Vn == Qn +---a;- (3.20)
(3.21a)
(3.21b)
12
The boundary condition, Vn = On, is known as the Kirchhoff free-edge

condition. In summary, the boundary conditions of the classical laminated
plate theory involve specifying the following generalized displacements and
forces on a boundary with unit outward normal n:
awo .
Un, Us, wo, an (essentIal)
(3.22)
The initial conditions of the theory involve specifying the values of the
displacements and their first derivatives with respect to time at t = 0:
(3.23)
where variables with superscript '0' denotes values at time t = O. We note
that both the displacement and velOcities must be specified.
As a special case, one can obtain the equations of equilibrium from
Eqs. (3.7)-(3.9) by setting all terms involving time derivatives to zero.
For linear analysis, we set N = 0 and P = 0, in addition to setting the
nonlinear terms in the strain-displacement equations to zero.
The force resultants are related to the membrane and flexural starins
by
(o)} Bll
{ Cxx {C1c1Jf }
{
Nxx}
N yy
N xy
= c1~ + [B12
-v(o) BI6
B12
B22
B26
B16]
B26
B66 -v(I)
/xy /xy
~~:] {:t~ }
~3.24)
(o)}
Mxx }
{ Myy =
[Bll
B12
B16
B26]
exx
{e1~ Dll
+ [D12
Mxy BI6 B66 -v(o)
/xy
DI6 D66 -v(I)
/xy
(3.25)
where Aij are called extensional stiffnesses, Dij the bending stiffnesses, and
Bij the bending-extensional coupling stiffnesses, which are defined in terms
of the lamina stiffnesses Q~7) as

13
1 ~ -(k) 3 3
Dij = 3" L.. Qij (Zk+l - Zk) (3.26)
k=l
Note that (j's, and therefore A's, B's, and D's, are, in general, functions
of position (x, y).
For the nonisothermal case, the laminate constitutive equations (3.24)
and (3.25) become
where {NT} and {MT} are thermal force resultants
(3.28a)
(3.28b)
Relations similar to (3.27) and (3.28a,b) can be written for hygroscopic

(Le., moisture) effects.
3.3. FIRST-ORDER SHEAR DEFORMATION PLATE THEORY
In the first-order shear deformation laminated plate theory (FSDT), the

Kirchhoff hypothesis is relaxed by removing the third part; Le., the
transverse normals do not remain perpendicular to the midsurface after
deformation. This amounts to including transverse shear strains in the
theory. The inextensibility of transverse normals requires that w not be a
function of the thickness coordinate, z.
The displacement field of the first-order theory is of the form
u(x, y, Z, t) = uo(x, y, t) + z</Jx(x, y, t)

v(x, y, Z, t) = vo(x, y, t) + UPy(x, y, t)
w(x, y, Z, t) = wo(x, y, t) (3.29)
where (uo, va, wo, ¢x, ¢y) are unknown functions to be determined. As
before, (uo, vo, wo) denote the displacements of a point on the plane Z = O.
Note that
8u 8v_A.
8z = ¢x, 8z - 'l'y (3.30)
14
which indicate that 4>x and 4>y are the rotations of a transverse normal about
the y- and x-axes, respectively. The notation that 4>x denotes the rotation
of a transverse normal about the y-axis and 4>y denotes the rotation about
the x-axis may be confusing to some, and they do not follow the right-hand
rule. However, the notation has been used extensively in the literature, and
we will not depart from it. The quantities (uo, vo, Wo, 4>x, 4>y) will be called
the generalized displacements. For thin plates, Le., when the plate in-plane
characteristic dimension to thickness ratio is on the order 50 or greater, the
rotation functions 4>x and 4>y should approach the respective slopes of the
transverse deflection:
A. __ 8wo A. _ _ 8wo
'Px - 8x' 'Py - 8y (3.31)
The nonlinear strains associated with the displacement field (3.29) are
(0) (1) ~+1(~)2 !!!&.
exx exx 8x 2 8x 8x
(0) (1)
r ~+1(~)2
rxx) eyy eyy eyy 8y 2 8y !!!hi.

'Yyz
'Yxz
=
(0)
'Yyz
(0)
'Yxz
+Z (1)
'Yyz
(1)
'Yxz
(1)
=1 ~+4>.
~+4>
8x x
+z ca'
0
!!!&'+!!!hi.
'Yxy (0)
'Yxy 'Yxy ~+~+~~ 8y 8x
8y 8x 8x 8y
(3.32)
The governing equations of the first-order theory are also derived using
the dynamic version of the principle of virtual displacements, and they are
8N + 8N _
xx xy 8 24>x
J, 8 2uo +I (3.33a)
--a;- ---ay - 0 8t2
8t2 1
8Nxy 8Nyy _ J, 82vo I 8 2¢y

--a;- + ---ay - 0 8t2 + 1 8t2 (3.33b)
8Qx 8Qy N _ J, 8 2wo (3.33c)

8x + By + + q - 0 8t2
8Mxx 8Mxy _ Q _ J. 8 24>x I 8 2uo (3.33d)
8x 8y +x - 2 8t2 + 1 8t2
8Mxy 8Myy _ Q _ J. 8 24>y I 82vo (3.33e)

8x + 8y y - 2 8t2 + 1 8t2
where the force and moment resultants are the same as defined in Eq.
(3.10), and (Qx, Qy) are the transverse force resultants
h
{ QX}=Kj'2 {O"XZ}dZ=K[A44 A45]
{ (o)}
'Yy~ (3.34)
Qy -~ O"yz A45 A55 'Y£)
15
where A 44 , ~5, and A55 are the transverse shear stiffnesses

N
" " - (k) - (k) - (k)
(A44, A45 , A55) = ~(Q44 ,Q45 ,Q55 )(Zk+l - Zk) (3.35)
k=l
and K denotes the shear correction factor, which is introduced to account

for the discrepancy between the actual shear forces and those due to the
constant stress state predicted by the first-order theory.
The generalized displacements and forces of the theory are
primary variables:
secondary variables: N nn , N ns , Qn, M nn , Mns (3.36)
(3.37)
Note that there are five pairs of generalized displacements and forces, as
can be seen from Eq. (3.36). Only one element of each of the pairs
(un,Nnn ), (us, N ns ), (wQ,Qn), (¢ Mnn) , and (¢s,Mns ) may be specified
at a boundary point.
4. Analytical Solutions of Rectangular Laminates

4.1. INTRODUCTION
The governing equations of composite laminates according to the classical
first-order shear deformation laminate theories were presented in sections
3.2 and 3.3, respectively. These equations can be solved either analytically
or numerically for the generalized displacements and strains. Stresses
can be determined using either the constitutive equations or the 3-D
equilibrium equations expressed in terms of stresses. Analytical solutions
can be developed for rectangular laminates with certain lamination schemes
and boundary conditions. The Navier solution procedure can be used
to determine the deflections, buckling loads, or natural frequencies of
antisymmetric laminates with all edges simply supported, and the Levy
solution procedure can be used to determine the solutions of antisymmetric
laminates with two parallel edges simply supported while the other two
having any boundary conditions. Here we discuss the N avier solutions to
bring out the effect of lamination scheme, loading, and shear deformation
on static deflections and stresses, critical buckling loads, and natural
frequencies for the following two cases of problems:
1. Simply supported (SS-l) antisymmetric cross-ply laminates, i.e.,
laminates with A 16 = A 26 = B 16 = B26 = D 16 = D26 = o.
16
2. Simply supported (SS-2) antisymmetric angle-ply laminates, Le.,

laminates with A 16 = A26 = Bll = B12 = B22 = B66 = D 16 =
D26 = o.
The two types of simply supported boundary conditions are shown in Figure
4.1 for CLPT. For FSDT, ow%x should be replaced with </>x and Owo/Oy
should be replaced with </>y. The details of the Navier solutions can be
found in the book by Reddy [5].
Y! N yy =Myy = 0 Y! Nxy : Myy =O

u = w =~ = o
----,----
.---__________________-:"1 v =w =-t'- =o
v =0 :----------- -- - - --- ... -
""'- .. .-...- ...:...... . .. .. ......,
.. -..-.,.... ..
v=O u=O u .0
w-O : w=O w =O : w =0
.au.
dz
=0 I:
I: SS-1 -t- =0 t aO SS-2 !' .au.
dz =0
Nn = ol: No: = 0 Nxy =0 ; Nxy=O
Mn = 01: Mu =O Mn=O : Mn=O
l~· ------- --------- -- ~
u=w=-:=o x v=w=-:=O x
Nyy =Myy=O Nxy =Myy=O
Figure 4.1. Two types of simply supported boundary conditions.
4.2. BENDING RESPONSE

First we discuss the results for antisymmetric cross-ply laminates (all B ij ,
except for Bn = -B22, are zero, and A 16 = A 26 = D 16 = D26 = 0)
according to the classical laminated plate theory.
17
Figures 4.2. and 4.3 show the effect of bending-stretching coupling

and plate aspect ratio on the transverse deflection ill = woE2h3/(qob4)
and normal stresses i'txx = [h 2/(qob2)Juxx (a/2, b/2, zo) for a fixed z = zoo
The material properties used are Ed E2 = 25, G 12 = G 13 = 0.5E2,
and 1112 = 0.25. The magnitude of deflections and stresses of symmetric
laminates (0/90/90/0) are about two to three times that of antisymmetric
(0/90/0/90) laminates for alb > 1. For uniformly distributed load there
corresponds an aspect ratio, around alb = 2.25 for (0/90h and alb =
3.5 for (0/90)8, for which the deflection is the maximum of all aspect
ratios. The effect of coupling is to increase the deflections and stresses.
The coupling coefficients Bij decrease in magnitude (hence the effect of
coupling decreases) with the increase in the number of layers (for the same
total thickness of the plate) in antisymmetric cross-ply laminates. The
nondimensionalized center deflection ill = woE2h3/(qob4) versus the aspect
ratio alb is shown in Figure 4.4 for (0/90)k (k = 1,2,3,4) laminates under
sinusoidal transverse loading (El = 25E2, G 12 = G13 = 0.5E2, G23 =
0.2E2, 1112 = 0.25). The nondimensionalized deflections of the six-layer
and eight-layer plates approach tL limiting case of an orthotropic plate.
It should be noted that in all cases the total thickness of each laminate
is the same. Thus, a two-layer (0/90) plate, for example, is of the same
total thickness as the eight-layer plate. Alternatively, the two-layer (0/90)
plate can be viewed as an eight-layer plate with four layers of 0° and four
layers of 90° orientations: (0/0/0/0/90/90/90/90); whereas the eight-layer
plate is of the lamination scheme (0/90/0/90/0/90/0/90); all layers of the
same thickness.
The dependence of the coupling effect on the modulus ratio is
illustrated in Figure 4.5, where the maximum nondimensionalized deflection
is plotted against the modulus ratio EdE2 (G 12 = G 13 = 0.5E2, and
1112 = 0.25) for the sinusoidal load. The solution rapidly approaches that
of an orthotropic plate for increasing number of layers.
Figures 4.6 and 4.7 show the distribution of the nondimensionalized
maximum normal stress (Txx = u xx (a/2, b/2, z)(h 2/qOb2) and transverse
shear stress (Txz = uxz(a, b/2, z)(h/qob), computed using the 3-D
equilibrium equations, through the thickness of two-layer and eight-
layer antisymmetric cross-ply laminates under sinusoidal loading (a/b =
1, a/h = 100, El = 25E2, G12 = G 13 = 0.5E2, 1112 = 0.25). The
two-layer plates experience larger stresses than eight-layer plates, and
the stress concentration is reduced in the latter. Thus, the effect of the
bending-stretching coupling in two-layer plates on stresses is to increase
the magnitude of stresses.
18
0.050 -rT"T"TTT"TTT"T"TTT"TTT"rrTTTTT1rrrTTTTl"TrrTTTTTTT"T"TTTTTT"rn
0.040
I~
~ 0.030
....o
~
Q)
tffi 0.020
Cl
0.010 --ss""L-
o.oOO--=i-r"""'~TTTT"TTT"rrTTTTT1rTTTTTT1"TrrTTTTTTTTTTTTTT1rr
o 1 2 3 4 5
alb
Figtlre 4.2. Nondimensionalized cel r transverse deflection ('III) versus plate

aspect ratio (a/b) of simply supported (88-1) laminates (CLPT).
0.00
.......
~ -0.20
~
I
~ -0.40
~ft
-0.60
1
~
I~~
.b
00
rif -0.80
til
Q)
-1.00
- - (0190190/0)
- - (0190/0/90)
- - - (0190)
1
-1.20
0 1 2 3 4 5
alb
Figure 4.3. Nondimensionalized normal stress (O"xx(a/2, b/2, -h/2)) versus

plate aspect ratio (a/b) for simply supported (88-1) laminates
(CLPT).
19
0.030 ----r--rrrrrTTTTTTTlnTTTTTTTIT1nTTTTTTTTTITTrrnTTTTTTTTl
( Sinusoidal load ) (0/90)

0.025
1;:3 0.020
g-
.....
..., 0.015
al
rg
t:I 0.010
o 1 2 3 4 5
alb
Figure 4.4. Nondimensionalized center transverse deflection versus aspect

ratio for simply supported cross-ply laminates (CLPT).
0.025
( SinuMidallo,') ~
1;:3
I:r
~
.....
...,0 0.015
'-I
(\)
~
(\)
t:I 0.010
~
0.005 (0/90)2
orthotropic plate
o 5 10 15 20 25 30 35 40 45
E11E2
Figure 4.5. N ondimensionalized maximum transverse deflection versus

modulus ratio for simply supported cross-ply laminates (CLPT).
20
0.50
~
........ 0.30
t\I
Qf
~
.....I=i 0.10
'E0
0
C.)
rIl
rIl -0.10
al
~
.....
C.)
,..Q
E-4
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8

Stress, Oxx
Figure 4· 6. Maximum normal str.,,,,,, versus plate thickness for simply

supported cross-ply laminates under sinusoidal load (CLPT).
0.50
~
........ 0.30
t\I
al n
~
I=i 0.10
~
'"'
0
0
C.)
rIl
rIl -0.10
al
~ C.)
:E -0.30
E-4
-0.50
0.0 0.1 0.2 0.3 0.4
Stress, Oxz
Figure 4.7. Maximum shear stress (iTxz) from equilibrium versus plate
thickness (zjh) for simply supported cross-ply laminates under
sinusoidal load (CLPT).
21
The effect of shear deformation on deflections and stresses is discussed

next. Figures 2.4.8 and 2.4.9 show the maximum nondimensionalized
deflections w = wo(E2 h 3 /b 4 qo) of simply supported, square, symmetric
and antisymmetric laminates under sinusoidally distributed load (SSL) and
uniformly distributed load (UDL), for different side-to-thickness ratios. The
material properties used are El = 25E2, G 12 = G 13 = 0.5E2, G 23 = 0.2E2,
V12 = 0.25, and K = 5/6. The results clearly show the diminishing effect of
transverse shear deformation on deflections, the effect being negligible for
side-to-thickness ratios larger than 20. The nondimensionalized quantities
in the classical laminate theory are independent of the side-to-thickness
ratio. The difference between the deflections predicted by the first-order
shear deformation theory and classical plate theory increases with the ratio
h/a.
The nondimensionalized stresses, i'txz = (Jxz(h/bqo) and i'tyz =
(Jyz(h/bqo), computed using the constitutive relations and equilibrium
equations are presented in Figures 4.10 and 4.11. In the case of i'txz, the
equilibrium equations predict the maximum stress to be at the midplane of
the plate, while the constitutive equations predict maximum stress in the
outer layers, which is the correct stress variation when compared to the 3D
result (not shown here).
0.020
0.018 (- - - - - Classical plate theo~
0.016
13
a
.....
0
0.014
~ 0.012 (0/90/90/0), UDL

<I)
~
<I)
0.010
Cl
0.008
0.006
(0/90/90/0), SSL
o 10 20 30 40 50 60 70 80 90 100
a/h
Figure 4.8. Nondimensionalized center transverse deflection (w) versus side-

to-thickness ratio (a/h) for simply supported (88-1) symmetric
cross-ply square laminates.
22
:.~J"""''''('~~~'~~''~=:'=;'']
~ (0/90). UDL '
I~ 0.020
....~g- 0.015
~ 0.010
(0°), UDL
0.005
~ (0°), SSL -l
0.000 11 iiiIi , , , ! 1 , , , I ' i , , I \ i i i I ' , i i i i , , I I ' iiI Iii 1 1 I' 1 1 ~
_ 10 20 30 40 50 60 70 80 90 100
alh
Figure 4.9. Nondimensionalized center transverse deflection (ill) versus side·

to-thickness ratio (a/h) for simply supported (88-1) orthotropic
and antisymmetric cross-ply (0/90) laminates under sinusoidall~
distributed transverse loading.
We now turn our attention to antisymmetric angle-ply laminates

First we discuss the results obtained with the classical laminate platE
theory. Figure 4.12 contains a plot of the nondimensionalized deflectioI
ill = WO(E2h3/qob4) versus plate aspect ratio a/h for simply supporte<
(88-2) antisymmetric angle-ply laminates (-45/45)k (k = 1,2,3,4) unde:
sinusoidal load. Figure 4.13 contains the plot of ill as a function of thE
lamination angle 0 for square laminates (-O/O)k under sinusoidal load. ThE
material properties used in these calculations are Ed E2 = 25, G12 = 013 =
0.5E2, and £112 = 0.25. It is clear that the effect of bending-stretchinl
coupling is to increase the deflection and decrease the stress. The couplinl
is quite significant for antisymmetric laminates (-454/454), but it decrease:
very rapidly when the layers are rearranged as (-45/45k The effect 0
coupling is significant for all modulus ratios except for those close to unit~
(results of the nondimensionalized deflection versus modulus ratio are no
included here). The reader may consult the book by Reddy [5] for thesl
results.
23
From
...
constitutive
relations
~- - -
FSDT,aJh=4
FSDT, aJh=10
From - - - CLPT, aJh=100
---~-
• I equilibrium
equations
!4-.<>A--.....,J..._,......L.Q._- '" -?
I
I I
-0.50 -4-1~~~~~TTTTT""T'·~~~irr,~<!>rrt't-r,-,,-j
0.0 0.1 0.2 0.3 0.4 0.5
Stress, crxz (O,b/2,z)
Figure 4.10. 8hear stress (0'xz) versus thickness (z j h) for simply supported
(88-1) symmetric cross-ply laminates.
0.50~r-r-,-G~-r6).TTTTTT Iii i i i i i i i
..:::
........ 0.30
t\I
Q)
'til
.S 0.10 ... _--_ .._.
[-
~
"0 FSDT,alh=4
Jo.<
o
-=_=-=.
oC) - - FSDT, alh=10
CLPT, alh=100
gj -0.10
Q)
~
~ -0.30
From
:---+====-- constitutive
relations
-0.50 T""T"·-,-,r,-~.,,,,-r-rr1
0.00 0.05 0.10 0.15 0.20 0.25 0.30

Stress, cryz (aI2,O,z)
Figure 4.11. 8hear stress (O'yz) versus thickness (zjh) for simply supported
(88-1) symmetric cross-ply laminates.
24
0.040 "OTTTrTTTTrTTTTTTrTTTTTTlrTTTTTTlrTTTTTT1TTTTTT1TTTrn
0.030
I~
I=f
.....
...,0
u 0.020
Q.l
~
Cl
0.010
0.000 --+"~~TTTTTT1"TTTTTrrrTTTTTTTT1TTTTTT1ITTTTTTT"TTT1
o 1 2 3 4 5
alb
Figure 4.12. Maximum deflect .1 versus plate aspect ratio for angle-ply
(-45/45)n (n = 1,2,3,4) laminates under sinusoidal load.
(--&9)
0.007
I~ 0.006
§
13 0.005
~ 0.004
:::1"""""""""
o 5 ro M ,':"""""""""
W M ~
Angle,
~
e
~ ,~
~
Figure 4.13. Maximum deflection versus lamination angle for angle-ply

(-O/O)n (n = 1,2,4) laminates under sinusoidal load.
25
Lastly, we consider the effect of transverse shear deformation on
deflections of antisymmetric angle-ply laminates. Figure 4.14 contains
plots of the nondimensionalized deflection as a function of the lamination
angle for two- and eight-layer antisymmetric angle-ply laminates (alb = 1;
EdE2 = 25, C 12 = C 13 = 0.5E2, V12 = 0.25, K = 5/6), (alh = 10),
(-f) I () I -()I ... ), subjected to sinusoidally distributed transverse load. The
effect of bending-stretching is significant in two-layer laminates.
0.009
0.008
(--&9)
13
0.007
J--------- __ _
~:f
........
0 0.006
()
CD
0.005
~
Cl
0.004
---
0.003
---
o 5 W ~ W ~ W ~ ~ ~
Angle, 9
Figure 4.14. Nondimensionalized center transverse deflection w versus

lamination angle () for simply supported (88-2) antisymmetric
angle-ply (-()I())n (n = 1,4) square laminates (alh = 10).
4.3 BUCKLING RESPONSE

When a plate is subjected to in-plane compressive forces, Nxx < 0 and
Nyy < 0, Nxy = 0, and if the forces are sufficiently small, the equilibrium
of the plate is stable (see Figure 4.15). The plate remains flat until a
certain load is reached. At that load, called the buckling load, the stable
state of the plate is disturbed and the plate seeks an alternative equilibrium
configuration accompanied by a change in the load-deflection behavior. The
phenomenon of changing the equilibrium configuration at the same load
26
and without drastic changes in deformation is termed bifurcation. The
load-deflection curve for buckled plates is often bilinear. The magnitude of
the buckling load depends on geometry, material properties, as well as on
the buckling mode shape. Here we determine the critical buckling loads of
simply supported specially orthotropic plates using the Navier method.
Figure 4.15. Buckling of a plate under in-plane compressive edge forces

0 0
(Nxx - -Nxx , Nyy - -Nyy ).
A _ A _
For buckling analysis, we assume that the only applied loads are the
in-plane forces and all other mechanical and thermal loads are zero:
Nyy
= -No, = -kNo, = -A-
A A
N xx N yy k (4.1)
N xx
For this case, the problem reduces to an eigenvalue problem and the
minimum eigenvalue is the critical buckling load, Ng.. Figure 4.16 shows
the effect of stacking sequence, plate aspect ratio, and modulus ratio
on nondimensionalized critical buckling loads N = Ncr(b 2/1[2 D22) of
rectangular laminates under uniform compression. The following material
properties were used: material 1: Ed E2 = 25, G 12 = G13 = 0.5E2, 1/12 =
0.25 and material 2: Ed E2 = 40, G12 = G 13 = 0.5E2, 1/12 = 0.25. The
nondimensionalized buckling load decreases for antisymmetric laminates as
the modulus ratio increases.
27
6·°1 " ,, I" " I" , ,I' ,,,I' " ,I" " I" ,]
~o- ~
~
1:<;
'"g~ 4.0
o
.-
(0/90)2, material 1
.....~ 3.0
(0/90)4, material 2
~
&l 2.0
(0/90)2, material 2 ~
1.0 ~
~ (0/90) , material 2 ~
0.0 l iii iii iii 1 iii iii iii Iii iii iii j ! iii j l
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
alb
Figure 4.16. Nondimensionalized buckling (if) load versus plate aspect

ratio (a/b) for simply supported (88-1) antisymmetric
cross-ply laminates (0/90)n under uniaxial compression
(CLPT).
Figure 4.17 shows the effect of transverse shear deformation on critical

buckling loads of symmetric (0/90/90/0) square laminates under uniaxial
and biaxial compression, while Figure 4.18 shows the effect of transverse
shear deformation and bending-extensional coupling on critical buckling
loads. The material properties used are: Ed E2 = 25, GI2 = G I3 = O.5E2,
G23 = 0.2E2, VI2 = 0.25. The eight-layer antisymmetric cross-ply plate
behaves much like an orthotropic plate.
Plots of nondimensionalized critical buckling loads versus plate aspect
ratio (a/b) for simply supported (88-2) angle-ply laminates (45/ - 45h
under uniaxial compressive in-plane loads are presented in Figure 4.19
for Ed E2 = 40, G I2 = G I3 = 0.5E2, and VI2 = 0.25. The buckling
mode associated with the critical buckling load is (m, n) = (1,1) for
alb ::; 1.4, (m, n) = (2,1) for 1.5 ::; alb ::; 2.4, (m, n) = (3,1) for
2.5::; a/b::; 3.4, (m, n) = (4,1) for 3.5 ::; a/b::; 4.4, and (m, n) = (5,1) for
4.5 ::; alb ::; 5. The effect of bending-stretching coupling is the most for
two-layer laminates, and the orthotropic solution is rapidly approached as
the number of plies is increased.
28
25.0
20.0 uniaxial
compression
I~
-
1f
0
.....~
15.0
=8~ 10.0
!Xl
5.0
0.0
0 10 20 30 40 50 60 70 80 90 100
alh
Figure 4.17. Buckling load (N) "_• .3US side-to-thickness ratio a/h for simpl)
supported symmetric cross-ply (0/90/90/0) square laminates.
25.0
20.0
I~
"ift
-.....
0
~
15.0
~~ 10.0
!Xl
5.0
0.0 -ttTTnTTTTTnTTTrTTTTTTCrTTTrrTlm>TTnTTTrTTTTTTrrl
o 10 20 30 40 50 60 70 80 90 100
alh
Figure 4.18. Buckling load (N) versus side-to-thickness ratio a/h for simpl)
supported antisymmetric cross-ply (0/90)n square laminates.
29
::1':""""""""';""""""""""""""~
'~ 200~ ~
-
o
~ 150
~u
~ 100 orthotropic plate
p:)
~ - -- - - - - - - - j - - - - - - - - - -
50 (-45/45)2 (-45/45)4
O~"'I""I""I""I~~~~~~:'I""I""I""I
0.0 1.0 2.0 3.0 4.0 5.0
alb
Figure 4.19. Buckling load (N = N cr b2 I E 2 h 3 ) versus plate aspect ratio

(alb) of antisymmetric angle-ply laminates under uniaxial
compressive edge load.
N ondimensionalized critical buckling loads versus the lamination angle

for uniaxial compression (k = 0) and biaxial compression (k = 1) of
antisymmetric angle-ply square laminates are shown in Figures 4.20 and
4.21, respectively, for EI = 40E2, G I2 = 0.5E2, and VI2 = 0.25. The
plots shown in Figure 4.21 are symmetric about () = 45°. Note that, once
again, the bending-stretching coupling severely reduces the buckling load
for the two-layer plate. The effect is negligible for eight or more layers. The
buckling load is the maximum for () = 45°.
Figures 4.22 and 4.23 show the effect of shear deformation, number of
layers (bending-extensional coupling), and the lamination angle on critical
buckling loads of antisymmetric angle-ply square laminates under uniaxial
compressive loads (alb = 1, alh = 10, EdE2 = 25, G I 2 = G I3 = 0.5E2,
VI2 = 0.25). The side-to-thickness ratio for the laminates in Figure 2.4.23
is taken to be alh = 10.
30
80
70
..£ orthotropic plate
"
60
I~
~ft 50
.....
0
bJ) 40
.....~
~::s 30
III
20 (-EII9)
10
0
0 20 30 40 50 60 70 80 90
Angle, 9
Figure 4.20. Buckling load eN) versus lamination angle (0) of antisymmetric
angle-ply square laminates under uniaxial compression.
_1: orthotropic plate

... (-EII9)4
30
I~
]'
.s
~ 20
.....
I 10 (--&9)
o w ~ w ~ W 00 ro ~ w
Angle, 9
Figure 4.21. Buckling load (N) versus lamination angle (0) of antisymmetric
angle-ply square laminates under biaxial compression.
31
50.0~~~~~TTTT"TT""..-rT1TTTTTT""rrTTTTT-rTTrTT1rrTT
45.0
__ J_________________
CLPT
_
40.0
n=8, (-45/45)4
12; 35.0
.-d'
til
......0
b.O
.....~
~;j
o::l
10.0
5.0
0.0
0 10 20 30 40 50 60 70 80 90 100
a/h
eN) versus side-to-thickness ratio (a/h) for simply

Figure 4.22. Buckling load
supported (88-2) antisymmetric angle-ply (-45/45)n square
laminates.
15.0
, , '- - ~- - - - - - - - -oj
1
(W-B)
o 5 10 15 20 25 30 35 40 45
Angle, e
Figure 4.23. Buckling load (N) versus lamination angle (0) for simply
supported (88-2) antisymmetric angle-ply (-0 jO)n laminates.
32
4.4. NATURAL VIBRATION RESPONSE

For free vibration, all applied loads and the in-plane forces are set to zero,
and we assume a periodic solution of the generalized displacements. This
problem too reduces to an eigenvalue problem.
Figure 4.24 shows a plot of fundamental frequency w versus aspect
ratio alb for symmetric (0/90)8 cross-ply and antisymmetric (0/90h cross-
ply laminates. The material properties used are Ed E2 = 40, G12 = G13 =
0.6E2, and 1112 = 0.25. Figure 4.25 shows the effect of coupling between
bending and extension on the fundamental frequencies of antisymmetric
cross-ply laminates. The material properties used are E1/ E2 = 25, G12 =
G13 = 0.5E2, and 1112 = 0.25. With an increase in the number of layers, the
frequencies approach those of the orthotropic plate. The bending-stretching
coupling has the effect of lowering the vibration frequencies. For example,
the two-layer plate has vibration frequencies about 40 percent lower than
those of eight-layer antisymmetric laminate or orthotropic plate with the
same total thickness.
Figure 4.26 shows the effect or uansverse shear deformation and rotary
inertia on fundamental natural frequencies of orthotropic and symmetric
cross-ply (0/90/90/0) square plates (Ed E2 = 25, G 12 = G 13 = 0.5E2'
G23 = 0.2E2, 1112 = 0.25). The symmetric cross-ply plate behaves much
like an orthotropic plate. The effect of rotary inertia is negligible in FSDT
and therefore not shown in the figure.
Figure 4.27 shows the effect of transverse shear deformation, bending-
extensional coupling, and rotary inertia on fundamental natural frequencies
(EdE2 = 25, G12 = G 13 = 0.5E2, G23 = 0.2E2, 1112 = 0.25). The eight-
layer antisymmetric cross-ply plate behaves much like an orthotropic plate.
The effect of rotary inertia is negligible in FSDT and therefore not shown
in the figure.
Nondimensionalized fundamental frequencies w = w(b 2/7r 2hlphi D22
of graphit~epoxy composites with Ed E2 = 40, G 12/ E2 = 0.5,1112 = 0.25
and alb = 1 are shown as a function of lamination angle in Figure 4.28.
The bending-stretching coupling due to the presence of B16 and B26 has
the effect of lowering the frequencies. The coupling is the maximum for
two-layer plates, and it rapidly decreases with increasing number of layers.
At f) = 45°, the fundamental frequency of the two-layer plate is about 40
percent lower than that of the eight-layer laminate. The same conclusions
hold for results presented in Figure 4.29. The effect of coupling is significant
for all modulus ratios, and the difference between the two-layer solution and
orthotropic solution increases with modulus ratio.
33
1O.0-'-'-""'-'-rTT1rTT"T-rrTTTTTTTT"Trrrn--rrrTTTTTTrrnrrnTl
9.0
8.0
7.0
6.0
5.0 \
\
\
4.0
~
3.0
2.0
1.0 -----------
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5
alb
Figure 4.24. Nondimensionalized fundamental frequency (w) versus plate

aspect ratio (a/b) for cross-ply laminates.
18
:>;
tJ 3.0
s::
2.0
s::<U
"C
~
s:: 1.0
::s (0190)3
~
(0/90)
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0
alb
Figure 4.25. Nondimensionalized fundamental frequency (w) versus plate

aspect ratio (a/b) for anti symmetric cross-ply laminates.
34
17
18 15
:>;
u 0·, FSDT, IROT .. O
s:: 13
Q)
6-
Q)
11
cl:: 0·, CLPI', mOT .. o
al
~
s::Q) 9
"C
~ 7
s::
::s
~ 5
3
0 .~
20 30 40 50 60 70 80 90 100
alh
Figure 4.26. Fundamental frequency (w) versus side-to-thickness ratio (a/b)

for simply supported (88-1), orthotropic and symmetric cross-
ply (0/90/90/0) laminates.
-- -------\-------(0I90)4, FSDT, WRI
(0190), CLPI', WRI
(0190), FSDT, WRI
3~~~~~~nT~~nnTrr~~TITrrnrl
o 10 20 30 40 50 60 70 80 90 100
alh
Figure 4.27. Fundamental frequency (w) versus side-to-thickness ratio for
simply supported antisymmetric cross-ply (0/90) laminates.
35
(-&6)
10.0
5.0 j ,
~
~
1 1
0.0 1""1' "'! 1" 1 I IIII I 1 I I I I 1 1 "! 1111I " 11111111
o 5 ro M M ~ W M ~ ~
Angle, e
Figure 4.28. Fundamental frequency versus lamination angle (0) of

antisymmetric angle-ply square laminates.
18 25.0
-I
5.0 ~
1
0.0 1",
o
'I I I I I
5 ro
I IIII I 1 "
~
1I 1"
M
1 I"
~
11I
W M
" 11I 1111I 1111I 1
~ ~
"~
W
Figure 4.29. Fundamental frequency (w) versus modulus ratio EI/E2 of

antisymmetric angle-ply square laminates.
36
Numerical results for two-layer (-45/45) and eight-layer (-45/45)4

plates with Ed E2 = 25, GI2 = GI3 = O.5E2, G23 = 0.2E2' lII2 = 0.25,
and K = 5/6 are given as a function of side-to-thickness ratio in Figure 4.30
and as a function of lamination angle in Figure 4.31. The effect of bending-
stretching coupling (Le., BI6 and B26), transverse shear deformation (Le.,
exz i= 0 and eyz i= 0), and rotary inertia is to lower the fundamental
frequencies. As the number of layers increases, the coupling decreases. The
effect of shear deformation decreases with increasing values of a/h. This
decrease is slower for eight-layer plates than two-layer plates. The effect of
rotary inertia is negligible in FSDT, whereas it is significant in CLPT only
for very thick plates.
4.5. SUMMARY
Based on the analytical solutions presented above, we can make the
following general observations.
• The presence of bending-extensional coupling in a laminate generally
reduces the effective stiffnes, : and hence increases deflections and
reduces buckling loads and natural frequencies. The coupling also
increases the period of oscillation in the transient problems. The
coupling is the most significant in two-layer laminates, and it decreases
gradually as the number of layers is increased for fixed total thickness.
• The presence of twist-curvature coupling in a laminate also has
the effect of increasing deflections, decreasing buckling loads, and
decreasing natural frequencies. The coupling dies out as the number
of layers is increased for fixed total thickness.
• The effects of bending-stretching coupling and twist-curvature coupling
on deflections, buckling loads, and natural frequencies of general
laminates, for example, unsymmetric laminates, can only be assessed by
specific studies. Such laminates can be analyzed only with approximate
methods of analysis.
• In general, the bending-twisting coupling in symmetrically laminated
plates has the effect of increasing deflections and decreasing buckling
loads and natural frequencies of vibration.
• The bending-extensional coupling as well as transverse shear
deformation in a laminate generally reduce the effective stiffnesses and
hence increase deflections and reduce buckling loads and reduce natural
frequencies. The coupling is the most significant in two-layer laminates,
and it decreases gradually as the number of layers is increased for fixed
total thickness.
37
8-layer, (-45/45)4
(Rotary inertia included)

- - - FSDT j
~ - - - CLPT ~
o 1""1""1""1""1""1""1""1""1""1",0
o 10 20 30 40 50 60 70 80 90 100
a/h
Figure 4.30. Frequency w versus sicte-to-thickness ratio for simply supported

(88-2) antisymmetric angle-ply (-45/45)n square laminates.
20
19
18
~ 18
~
~
--- - CLPT
,. ,. /
,. ,.
Q) 17
;j
C"
Q) 16 ~,.,.,.""
~ :: - - ,. 8-layer, (6'~)4
til 15
....
~
Q) 14 "
~
"C
13
~
;j 12
~
11
10
0 5 10 15 20 25 30 35 40 45
Angle, e
Figure 4.31. Frequency (w) versus lamination angle 0 for simply supported
(88-2) antisymmetric angle-ply (-0 /O)n square laminates.
38
5. Finite Element Analysis of Composite Plates
5.1. INTRODUCTION
In section 4, the results of the Navier solutions to the equations of
composite plates were discussed for rectangular geometries. However,
analytical solutions to these equations cannot be developed when complex
geometries, arbitrary boundary conditions, or nonlinearities are involved.
Therefore, one must resort to approximate methods of analysis that
are capable of solving such problems. The finite element method is
a powerful computational technique for the solution of differential and
integral equations that arise in various fields of engineering and applied
science [5-8]. The basic idea of the finite element method is to view
a given domain as an assemblage of simple geometric shapes, called
finite elements, for which it is possible to systematically generate the
approximation functions needed in the solution of differential equations
by any of the variation~ and weighted-residual methods. The ability
to represent domains with irregular geometries by a collection of finite
elements makes the method a . 'uable practical tool for the solution
of boundary, initial, and eigenvalue problems arising in various fields of
engineering. For a detailed introduction to the finite element method, the
reader is advised to consult References 7 and 8.
Here we present finite element models of the first-order shear
deformation theory. Finite element models of classical plate theory can
be found in Reference 1. In the interest of brevity, only the major steps of
the development are included here (see Reddy [7] for details).
5.2. LINEAR FINITE ELEMENT MODELS OF FSDT

The virtual work statement of the first-order theory contains, at the most,
only the first derivatives of the dependent variables (uo, vo, wo, cPx, cPy).
Therefore, they can all be approximated using the Lagrange interpolation
functions. In principle, the sets (uo, vo), wo, and (cPx, cPy) can be
approximated with differing degrees of functions. For simplicity, we use
the same interpolation for all variables:
m
uo(x, y, t) = L Uj (t)-rpj (x, y) (5.1a)
j=l
m
vo(x, y, t) = L Vj(t)-rpj(x, y) (5.1b)
j=l
39
n
Wo(x, y, t) = L Wj(t)1fJj(x, y) (5.2)
j=1
P
<Px(x, y, t) = L SJ(t)1fJj(x, y) (5.3a)
j=1
P
<Py(x, y, t) = L S;(t)1fJj(x, y) (5.3b)
j=1
where 1fJj are Lagrange interpolation functions. In general, (uo, va), wo, and
(<Px, <py) may be interpolated with different degree of interpolation. One can
use linear, quadratic, or higher-order interpolations of these sets.
Substituting Eqs. (5.1)-(5.3) for (uo, va, wo, <Px, <py) into the virtual
work statements of FSDT, we obtain the following finite element model:
[Kl2]
[K22]
[K 23 )T
[K24)T
[K 25 f
IO[M] [0]
[0] Io[M]
[
+ [0] [0] (5.4a)
It[M] [0]
[0] It[M]
or
(5.4b)
where the coefficients of the submatrices [K a,8] and [Ma,8] and vectors {Fa}
are defined for (a, {3 = 1,2, ... ,5) by the expressions
K i·Ia =
J
1(
Oe
81fJi N 1
- a 81fJiN,a)
8x J · +
d d
8y- 6·J X Y
K~'!- =
~J
r
Joe (8'IjJi N,a.
8x 6J
+ 8'IjJi N a.)
8y 2J
dxdy
K~'!- =
~J
r
Joe (81fJiQa.
8x IJ
+ 81fJiQa.)
8y 2J
dxdy
K ij4a = r
Joe (88xWM Ija + 81fJi Ma ot,eQa )
8y 6j + Ij 'Pi
dxdy
K ij5a = r
Joe (81fJi Ma
8x 6j + 81fJi Ma ot,eQa)
8y 2j + 'Pi 2j dxdy
40
The coefficients Nn, Mn, and QJj for a = 1,2, """,5 and I = 1,2,6 are
given by
I a'ljJj a'ljJj 2 a'ljJj a'ljJj

N lj = An ax + Al6 ay , N lj = Al2 ay + Al6 ax
4 a'ljJj a'ljJj 5 a'ljJj a'ljJj
N lj = Bn ax + Bl6 ay , N lj = Bl2 ay + Bl6 ax
I a'ljJj a1//,! 2 a'ljJj a'ljJj
N 2j = Al2 ax + A 26 ay , N 2j = A22 ay + A26 ax
I a'ljJj a'ljJj
M lj = Bn ax + Bl6 ay ,
4 a'ljJj iNj
M lj = Dn ax + Dl6 ay ,
I a'ljJj a'ljJj
M 2j = Bl2 ax + B 26 ay ,
4 a'ljJj a'ljJj
M 2j = Dl2 ax + D26 ay ,
I a'ljJj a'ljJj
M6j = Bl6 ax + B66 ay ,
4 a'ljJj a'ljJj
M6j = D16 ax + D66 ay ,
41
Nrj = 0, N?j = 0, NJj = 0, Mrj = 0, M?j = 0, MJj =a (5.6a)
Qij = 0, Q~j = 0, Q~j = 0, Q~j =a

3 o'I/Jj o'I/Jj 3 o'I/Jj o'I/Jj
Qlj = A55 ox + A45 oy' Q2j = A45 ox + A44 By
Qtj = A55'I/Jj, Qfj = +A45'I/Jj, Q~j = A45'I/Jj, Q~j = A44'I/Jj (5.6b)
(5.7)
where NIx, MIx, etc. are the thermal force and moment resultants.
The displacement-based CO plate bending element of Eq. (5.4) is often
referred to in the finite element literature as the Mindlin plate element due
to the fact that it is based on the so-called Mindlin plate theory, which is
labeled in this book as the first-order shear deformation plate theory. When
the bilinear rectangular element is used for all generalized displacements,
the element stiffness matrices are of the order 20 x 20; and for the nine-node
quadratic element they are 45 x 45 (see Figure 5.1). Equation (5.6) can
be simplified for static bending, buckling, natural vibration, and transient
analyses.
The CO-plate bending elements based on the first-order shear
deformation plate theory are among the simplest available in the literature.
They are expected, in theory, to give the thin plate theory solution when
the side-to-thickness ratio ajh is very large (ajh ~ 100). Unfortunately,
when lower-order (quadratic or less) equal interpolation of the transverse
deflection and rotations is used, the elements do not accurately represent
42
the bending behavior as the side-to-thickness ratio of the element becomes

large (Le., thin plate limit). For thin plates, the shearing strains Cxz and
cyz are required to vanish, and the plate elements based on the first-order
theory become excessively stiff, yielding displacements that are too small
compared to the true solution. This type of behavior is known as shear
locking.
Shear locking is due to the inability of shear deformable elements to
accurately model the bending within an element under a state of zero
transverse shearing strain. When thin plates are analyzed by the shear
deformable elements, the energy due to transverse shear strains must
vanish. Numerically this is equivalent to requiring the product of the
shear stiffness matrix and the displacement vector be zero. Therefore, in
order to obtain a nontrivial solution, the shear stiffness matrix must be
singular. One way to achieve the singularity of the transverse shear stiffness
matrix is to use an order "f numerical integration lower than is necessary to
evaluate the integrals exactly. Thus, reduced integration of transverse shear
stiffnesses (Le., all coefficients in Kf/ that contain A 44 , A 45 , and A55) is
necessary.
y
1 2
1 5 2
Uo, Vo, Wo, <1>,:1 <l>y (at each node)

Figure 5.1. Linear and quadratic Lagrange rectangular elements for FSDT.
43
5.3. POSTCOMPUTATION OF STRESSES

Once the nodal values of generalized displacements (uo, vo, wo, ¢x, ¢y) have
been obtained by solving the assembled equations of a problem, the strains
are evaluated in each element by differentiating the displacements.
Since the displacements are only continuous across the element
boundaries, strain continuity across the boundaries is not ensured. That is,
along a boundary common to two elements, the strains and hence stresses
take different values on the two sides of the interface. However, strains and
hence stresses are continuous within an element.
As noted earlier, the strains and stresses are the most accurate if they
are computed at the (N - 1) x (N - 1) Gauss points, where N x N is the
Gauss quadrature rule used to evaluate the bending stiffness coefficients.
For example, the linear rectangular plate bending element of the first-order
theory requires 2 x 2 integration to evaluate the bending stiffnesses exactly.
Then the one-point integration should be used to evaluate the transverse
shear stiffness coefficients, strains. and stresses. Similarly, for a quadratic
rectangular element the reduced integration rule is the 2 x 2 Gauss rule.
Since the displacements in the finite element models are referred to
the global coordinates (x, y, z), the stresses are computed in the global
coordinates using the relations (when thermal effects are not included) at
the Barlow points (i.e., reduced integration points)
where
r'r
{Yyy
{Yxy
{YXZ
{Yyz
Qu Ql2 Ql6
Ql2 Q22 Q26
Q16
0
0
Q26
0
0
Q66
0
a
0
0
0
Q55
Q45
o
o
o
C145
Q44
rT"j
eyy
TXY
TXZ
TYz
(5.8)
ili!:JJ. ~
exx ax ax
!b!IJ. ~
eyy ay ay
~+¢
ay y +z a (5.9)
TYZ
TXZ
~+¢x 0
TXY ili!:JJ.+!b!IJ. !l.h+~
ay ax ay ax
If stresses and strains are required in the lamina principal material
coordinates, for example, to check for failures, the strains and stresses
44
of Eqs. (5.8) and (5.9) should be transformed to material coordinates
:!: )
associated with each layer using the transformation relations. Alternatively,
the strains can be transformed using
o o sin Ok cosBk
o o - sin Ok cos Ok
o 'Yyz
o 'Yxz
COS 2 Ok - sin2 Ok 'Yxy
(5.10)
and then the lamina constitutive equations are used to compute the stresses:
o o
o o
Q44 o (5.11)
o Q55
o o
5.4. NONLINEAR FINITE ELEMENT MODEL

5.4.1. Introduction
Historically, two distinct approaches have been followed in developing
nonlinear finite element models of structural problems. Each of the
approaches can be used to develop the displacement type, force type, or
mixed/hybrid type finite element models. The first approach is based on a
laminate theory, in which the 3-D elasticity equations are reduced to 2-D
equations through certain kinematic assumptions, as described in section 2.
In the nonlinear formulation based on a laminate theory, the geometry of
the structure is assumed to remain unchanged during the loading, and the
geometric nonlinearity in the form of the von Karman strains is included.
For small strains and moderately large deflections and rotations, these
assumptions yield good results. We shall term the elements based on such
assumptions as laminated elements.
The second approach is based on the 3-D continuum formulation, where
any kinematic assumptions are directly introduced through the spatial finite
element approximations. Full nonlinear strains or only the von Karman
nonlinear strains are included as desired, and the equations are derived
in an incremental form directly. The formulation accounts for geometric
changes that occurred during the previous increment of loading. Thus, the
geometry is updated between load increments. Finite elements based on
this formulation are called continuum elements.
45
The major objective of this section is to present the displacement finite

element models of the first-order shear deformation plate theory (FSDT)
when the von Karman nonlinear strains are accounted for, i.e., develop
nonlinear laminated plate elements. The shear deformable nonlinear finite
element model is used to study nonlinear bending and post buckling of
laminated structures.
5.4. 2. Nonlinear stiffness coefficients

When the von Karman nonlinear strains are accounted for in the finite
element model, the resulting finite element equations (5.4) will be nonlinear.
The nonlinear finite element model has the form
(5.12)
The elements of the stiffness matrix [Ke] and mass matrix [Me] are given
in Eqs. (5.5)-{5.7). The additional stiffness coefficients [Ra3] (a = 1,2,3)
are
(5.13)
for a = 1,2,6. The same expressions hold for the first-order plate theory
with '¢i replaced by ,¢i, because Wo is also approximated by the Lagrange
functions '¢i in FSDT. We note that the nonlinearity of the coefficients
[K 13 ], [K 23], [K31], [K 32 ], and [K33] is solely due to the transverse
deflection, woo In addition, the stiffness matrix [Ke] is not symmetric for
the nonlinear case because
a = 1,2 (5.14)
46
The fully discretized equations are given by
(5.15)
6. Finite Element Solutions

6.1. NONLINEAR BENDING OF PLATES
Figures 6.1 and 6.2 contain the load-deflection plots for a simply supported
orthotropic square plate and a clamped cross-ply (0/90/90/0) square
laminate, respectively. In both cases, a uniformly distributed transverse
load is used. The geometry and material properties used are given below.
Orthotropic plate:
a = b = 12 in., h = 0.138 in.

El =3 X 106 p(, E2 = 1.28 X 106 psi
G 12 = G23 = G 13 = 0.37 X 106 psi
V12 = V23 = V13 = 0.32 (6.1)
Cross-ply laminate:
a = b = 12 in., h = 0.096 in.

El = 1.8282 X 106 psi, E2 = 1.8315 X 106 psi
G 12 = G 13 = G23 = 0.3125 X 106 psi
V12 = V23 = V13 = 0.2395 (6.2)
For the simply supported orthotropic plate, the shear deformation

element yields results that are in excellent agreement with the experimental
data of Zaghloul and Kennedy [11,12]' while the classical plate bending
element underpredicts deflections even for this thin plate (a/h is about
120). For the clamped cross ply laminate, the finite element results are
not in as good agreement with the experimental data. This is believed
to be due to possible differences between the actual material properties
and boundary conditions and those used in the finite element model.
It is much more difficult to simulate fully clamped boundary conditions
(uo = Vo = Wo = <Px = <Py = 0 on all edges) in an experiment.
47
0.50
""'
,:::: - - - Experimental
.....
'-' - - - CLPT
0.40
0 ~ FSDT
;:s
,::::~
.....
0
4->
0.30
U
Q)
~
Q)
"C
Q) 0.20
00
I-<
Q)
>
00
,:::: 0.10
(1j
I-<
~
0.0 0.4 U.t! 1.2 1.6 2.0

Pressure, qo (psi)
Figure 6.1. Load-deflection curves (qo vs. wo) for simply supported
(88-1) orthotropic square plates under uniformly distributed
transverse load qo.
""' - - - Experimental
.....
,::::
'-' - - - CLPT
0.40
0
;:s ~FSDT
/
Ii
.....
0
4->
0.30 /
u Linear
Q)
~
Q)
"C
Q)
0.20
00
I-<
Q)
>
00
,:::: 0.10
(1j
I-<
~
0.0 0.4 0.8 1.2 1.6 2.0

Pressure, qo (psi)
Figure 6.2. Load-deflection curves (qo vs. wo) for clamped cross-
ply (0/90/90/0) square plates under uniformly distributed
transverse load qo.
48
Next, results of large-deflection bending of two-layer and six-layer cross-
ply and angle-ply rectangular plates (a/h = 40) are presented. The plates
are assumed to be clamped (uo = Vo = Wo = <Px = <Py = 0) and subjected
to uniformly distributed transverse load. The following material properties
of a lamina are assumed:
EI = 40E2, Gl2 = G13 = 0.6E2, G23 = 0.5E2, lIl2 = 0.25 (6.3)

A three degrees of freedom model is used. Figures 6.3 and 6.4 contain
plots of nondimensionalized center deflection versus the load parameter
P = 1O-4qo{2b/h)4/E2 for cross-ply and angle-ply laminates, respectively.
The degree of nonlinearity exhibited by the six-layer plates is very small.
Since the bending-stretching coupling is significant in two-layer plates and
the in-plane displacements are omitted in the model, the results for two-
layer plates should not be considered as typical. They are included only
for comparison.
6.2. NONLINEAR RESPONSE DUE TO IN-PLANE COMPRESSIVE

LOADS
Another interesting characteristic of composite laminates is their behavior
under compressive loads [13,14J. Most often the critical buckling loads are
determined through an eigenvalue analysis. The critical buckling loads can
also be determined from a geometric nonlinear analysis, where the critical
buckling load is taken to be the so-called limit load, if one exists.
First we consider an angle-ply (45/-45) laminate with the following
geometric parameters and material properties:
a = b = 1,000 mm, h = 2 mm
lIl2 = lIl3 = 0.24, lI23 = 0.49 (6.4)

For SS-2 type simply supported boundary conditions, the uniaxial buckling
load can be determined analytically. Here we use 4 x 4 mesh of nine-node
elements in the full plate to determine the critical buckling load, and the
same mesh is used to determine the nonlinear response under applied in-
plane compressive load Nyy = >'oN~y, where >'0 is the critical buckling
load determined from the eigenvalue analysis. Figure 6.5 contains a plot
of the maximum out-of-plane deflection Wo (mm) versus load parameter >.
(N~ = 10.85) N/m). It is clear the load-deflection curve exhibits a limit
point, which is the same as the critical buckling load determined from the
eigenvalue analysis.
49
35.0
30.0
---6- alb = 2 } two-layer
=
--0- alb 1.5 (0190)
~ 25.0 --0- alb =1 alh =40
0
:3
a
.~
20.0 --*- alb =2
___ alb = 1.5
} six-layer
(0f!I0I()I9{V0I9)
~ ___ alb=l alh=40
Q) 15.0
Ii:::I
Q)
0 10.0
5.0
0.0
0 200 400 600 800 1000
Load,P
Figure 6.3. Load-deflection curves (I' vs. wo/h) for clamped two- and six-
layer antisymmetric cre. _.ply plates (a/h = 40) under uniformly
distributed transverse load qo.
25.0
two-layer (451-45)
20.0
~ - 0 - alb = 1.5 } alh =
0 - 0 - alb=l 40
:3
a
.~
15.0
six-layer
+>
(45145145145145145)
1Il 10.0 -_ alb = 1.5 }alh=40
Ii:::I
Q) alb=l
0
5.0
Load,P
Figure 6.4. Load-deflection curves (p vs. wo/h) for clamped two- and six-
layer antisymmetric angle-ply (45/-45/-· .) plates (a/h = 40)
under uniformly distributed transverse load qo.
50
2.0
o~
:c;
.......
~~ 1.5
II
«
i Q)
1.0
IS
~
s:l. 0.5 (451-45), SS-2, full plate
'"C
.stIS - - Nonlinear analysis

o
Eigenvalue analysis
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6

Maximum deflection, Wo (mm)
Figure 6.5. The load-deflection curve (A vs. wo) of a simply supported (88-
2) two-layer antisymmetric angle-ply (45/-45) laminate under
uniformly distributed in-plane compressive edge load.
3.0 -,--r-r.,..,-.-"...,...,..,,....,""-.-,,--,-,-.-.,,~---;
:c;
o~
2.5
.......
~~
II
2.0 , .
« II
: (0l90I••• )
II
- - - 2 layers
~ 1.5 1/ 4 layers
•....... 6layers
Slayers
(1.0 {.
'"C i'e o• 0 • Fundamental
.stIS
0.5
eigenvalue
0.0 -+-,.--,,,,...,-,-,.--,,,-,-,-,-,,,,,...,-,-,.--,-rl
0.0 1.0 2.0 3.0 4.0 5.0
Maximum deflection, Wo (mm)
Figure 6.6. The load-deflection curve (A vs. wo) of a simply supported (88-
1) antisymmetric cross-ply (0/90;"') laminate under uniformly
distributed in-plane compressive edge load.
51
Next we consider antisymmetric cross-ply laminates. The geometry

and materials properties used are the same as those used for the angle-ply
laminate. The SS-1 type simply supported boundary conditions and 2 x 2
mesh of nine-node elements in a quarter plate are used to determine the
critical buckling loads No by eigenvalue analysis and load-deflection curves
in the nonlinear analysis under in-plane load Nyy = .AoN3y (N3y = 3.25
N/m). Figure 6.6 contains load-deflection curves for two-, four-, six-, and
eight-layer laminates. The critical buckling loads are indicated on the load-
deflection curves for comparison. Unlike the angle-ply laminates, the cross-
ply laminates do not exhibit clear limit-load behavior.
6.3. NONLINEAR RESPONSE OF ANTISYMMETRIC CROSS-PLY

LAMINATED PLATE STRIPS
Unlike isotropic metallic plates, composite plates exhibit quite different
nonlinear behavior [13-15]. For example, the geometric nonlinear effects
could be very significant even at small loads and deflections, depending on
the lamination scheme and boundary conditions. To illustrate the point
we analyze an antisymmetric cross-ply square laminate (90/0) with two
opposite edges pinned (uo = 0) or hinged (uo i= 0) and the other two edges
free, and subjected to uniformly distributed transverse load.
The geometry, finite element mesh, and boundary conditions for the
pinned-pinned and hinged-hinged cases are shown in Figure 6.7. The
material properties and geometric parameters used are
El = 20 msi, E2 = 1.4 msi, V12 = 0.3, G 12 = G 13 = G23 = 0.7 msi

a = 9 in., b = 1.5 in., h = 0.4 in. (6.5)
For hinged-hinged boundary conditions, the plate strip is essentially in

pure bending and hence the axial force N xx = o. Therefore, the solution is
independent of the sign of the applied load. For a pinned plate strip, the
axial force N xx is not zero; it is
duo
N xx = All [ dx +"2
1 (dWo)
dx
2] + Ell d<px
dx ' Bll < 0 (6.6)
For small values of the positive load, the expression containing the
All coefficient is small compared to the expression containing the Ell
coefficient, which is negative for 0 < x < 4.5. Hence, N xx is compressive and
increases the transverse deflection - analogous to the transverse deflection
of a plate strip under an in-plane compressive load and a transverse load.
52
I' a ·1 I' a 'I

z
illy =0
T
b
I
I
I Uo=O
I' 110=0
--"
I I wo=O
1 ;1 I
1--% ..
;
~ al:!--l
Figure 6.7. Geometry, loading, and 1..')undary conditions used for cylindrical
bending of a cross-ply plate strip.
~o Linear .
2.0 "., ..
3
s:f q=-qo
....~
o
1.5 .MH+UHi
L ... .. W.(9010)A
~
~ LO
~ q=-qo
....... .. f.········ ..... .
~ 0.5
0.0 -¥.,.--r.,---.-r-r-r-. . ..,---.-r-r-r-. . ..---r-r-r-r-. . .--.1

0.00 0.01 0.02 0.03 0.04 0.05
Load, qo (lbslin)
Figure 6.8. Load-deflection curves (qO vs. wO) for cylindrical bending of
a cross-ply (90/0) plate strip with pinned-pinned edges.
53
Thus, the nonlinear solution is larger than the linear solution for small
values of the load. As the load is increased, the Au expression becomes
larger than the Bu expression, and N xx becomes positive. This stiffens
the structure and the nonlinear solution becomes smaller than the linear
solution. The load deflection curves for the first few load steps are shown
in Figure 6.8. For a negative load, Bu1f: is positive, and the two terms in
N xx add up; this yields a larger axial force and therefore a stiffer structure
than for the positive load case. Therefore the deflection is lower than that
for the case of positive load.
6.4. POSTBUCKLING RESPONSE AND PROGRESSIVE FAILURE

OF COMPOSITE PANELS IN COMPRESSION
Because of low moduli and strengths in transverse directions compared to
that of in-plane directions, composite laminates may fail due to transverse
stresses. Indeed, it is found that (see [16,17]) composite laminates loaded
in compression fail due to high interlaminar stresses. Therefore, shear
deformable plate and shell elements are needed to provide information
regarding the through-thickness suength of composite structures. Insight
gained by using these elements may aid in the characterization of failure
modes of composite panels. Here we discuss a study by Engelstad, Reddy,
and Knight [17] of the post buckling response of several graphite-epoxy
panels loaded in axial compression. The study makes comparisons between
the experimentally obtained and analytically determined postbuckling
response of composite panels.
The post buckling and failure characteristics of flat, rectangular
graphite-epoxy panels, with and without holes, and loaded in axial
compression have been examined in an experimental study by Starnes
and Rouse [16]. The panels were fabricated from commercially available
unidirectional Thornel 300 graphite-fiber tapes preimpregnated with 4500 K
cure Narmco 5208 thermosetting epoxy resin. Typical lamina properties for
this graphite-epoxy system are: El = 131.0 GPa (19,000 ksi), E2 = 13.0
GPa (1,890 ksi), G 12 = 3.4 GPa (930 ksi), V12 = 0.38 , hk = 0.14 mm
(0.0055 in.), where hk denotes ply thickness. Each panel was loaded in axial
compression using a 1.33 MN (300 kips) capacity hydraulic testing machine.
The loaded ends of the panels were clamped by fixtures during testing
and the unloaded edges were simply supported by knife-edge restraints to
prevent the panels from buckling as wide columns. Most panels exhibited
post buckling strength and failed along a nodal line of the buckling mode
in a transverse shear failure mode (see [16]). A different failure mode was
observed for some of the 24-ply panels with holes. These panels failed along
a transverse line passing through the hole, and failed soon after buckling.
54
Here we analyze the panel without a hole, denoted C4 in Reference

16 (see Figure 6.9). The finite element results are compared with the
experimental results of Reference 16. Panel C4 is 50.8 cm by 17.8 cm (20.0
in. long and 7.0 in. wide), 24-ply laminate, [±45/02/ ± 45/02/ ± 45/0/90]8
(orthotropic). Panel C4 was observed in the test to buckle into two
longitudinal half-waves and one transverse half-wave.
I-- 7.0 in --I
.S
o t--
ci
C"I
Figure 6.9. Geometry and finite elements meshes of the C4 composite panel
used in the postbuckling study.
Finite element models of these panels were developed in Reference 96

using nine-node quadratic shell elements of Chao and Reddy [18], which are
based on continuum formulation of a laminated 3-D shell element, denoted
9CR. The incremental equations of a continuous medium are formulated
using the principle of virtual displacements and the total Lagrangian
description. The continuum shell element is obtained from the three-
dimensional solid element by imposing the same two constraints as in
FSDT: (1) straight lines normal to the midsurface before deformation
remain straight, but not normal, after deformation; (2) the transverse
normal components of strain, and hence stress, are ignored in the
development. These assumptions reduce a 3-D theory into a 2-D theory,
55
and the resulting shell element is called a 3-D degenerated shell element.
The nonlinear formulation admits large displacements and rotations of the
shell element and small strains, since the thickness does not change and the
normal does not distort. The final incremental equations of equilibrium for
an element are of the form
(6.7)
where {8~} is the vector of incremental nodal displacements, ([KLJ, [KNL])
are the linear and nonlinear parts of the stiffness matrix, and {F} is the
force vector:
[KLJ = [ [BLJT[C][BLJ dv
joe
[KNLJ = [ [BNL][S][BNLJ dv
joe
{F: = [ [BL]{B} dv (6.8)
joe
In these equations, [BLJ and fRIVLJ are linear and nonlinear strain-
displacement transformation matrices, [C] is the constitutive elasticity
matrix, [SJ and {B} are the matrix and vector of second Piola-Kirchhoff
stresses, and {R} is the external load vector. All matrix elements refer
to the deformed state and are measured with respect to the original
undeformed configuration.
To evaluate the integrals in Eq. (6.8) (e.g., [KL], [KNLJ, and {F}),
we use Gauss quadrature in the surface directions of the shell, but
explicit integration in the thickness direction. Thus the thickness direction
integration for matrices [KLJ and [KNLJ gives the following laminate
stiffnesses:
p
[AJ = ~=rc'Jk ((HI - (k)
k=l
p
[BJ =~ 2)c'Jk((~+1 - (~)

k=l
P
[DJ =~ I)C/t((~+1 - (~) (6.9)
k=l
Here (k is the thickness coordinate of the bottom of the kth lamina, P is
the number of laminae, [c'Jk is the constitutive matrix for the kth lamina
in the principal material coordinates, which has the form
[c'J = f([Q], fJ) (6.10)
56
where Qij are the plane stress-reduced elastic coefficients in the material
coordinates and () is the fiber orientation angle. For an orthotropic lamina,
Qij can be expressed as
(6.11)
where El, lIl2, 1I2l, E2, G13, G23, and G12 are the lamina material properties
and K is the shear correction factor, which is taken to be 5/6.
The finite element model used in Reference 17 consisted of six elements
per buckle half-wave in each direction. Hence, the finite element model of
Panel C4 consists of 12 nine-node quadrilateral elements along the panel
length. Figure 6.9 shows the model used for the C4 specimen. The total
numbers of 9 node quadrilateral elements in the finite element model is 72.
In order to proceed beyond th ~ritical buckling point in the analysis of
each panel, an initial geometric imperfection, typically the same shape as
the first linear buckling mode, was assumed in the finite element analysis.
The amplitude of each mode was selected to be 1-5% of the total laminate
thickness. This allows efficient progress past the critical buckling point,
but does not affect the results in the postbuckling range.
The maximum stress and Tsai-Wu failure criteria are used (see [19-
21]). In the maximum stress criterion, failure is assumed to occur if any
one of the following conditions are satisfied:
(6.12)
where (0"1, 0"2, 0"3) are the normal stress components, (0"4, 0"5, 0"6) are
shear stress components, (XT' YT, ZT) are the lamina normal strengths
in tension (T) along the (1, 2, 3) directions, and (R, S, T) are the shear
strengths in the (23, 13, 12) planes, respectively. When (0"1, 0"2, 0"3) are
compressive, they should be compared with (Xc, Yc, Zc), which are
normal strengths in compression (C) along the (1, 2, 3) principal material
directions, respectively.
The Tsai- Wu criterion is given by
6 6 6
F == L FiO"i + L L FijO"iO"j ;::: 1 (6.13)
i=l i=l j=l
57
where
where (Ji denote the stress components referred to the principal material
coordinates.
In reality, laminate failure occurs due to propagation of damage as the
load is increased. To model this effect, a progressive failure approach is
used in the nonlinear finite element analysis. At each load step, Gauss
point stresses are used in the selected failure criterion. If failure occurred
at a Gauss point, a modificatioh uf the lamina properties was made at
that Gauss point, which results in reduced stiffnesses [AJ, [B], and [D] of
the laminate. For example, for the maximum stress criterion, if the (Jl
stress exceeds the longitudinal tensile strength XT, then the longitudinal
modulus El at that point is reduced to zero. For the Tsai-Wu criterion,
if failure occurs, then the following expressions are used to determine the
failure mode:
HI = FWI + Hw~, H2 = F2(J2 + F22(J~

H4 = F44(J~, H5 = F55(J~, H6 = F66(Ji (6.15)
The largest Hi term is selected as the dominant failure mode and the
corresponding modulus is reduced to zero. Thus HI corresponds to the
modulus E I , H2 to E2, H4 to G23, Hs to G13, and H6 to G23. As a
consequence of this reduction, engineering material properties are updated
as failure progresses. An outline of the steps used in the analysis is given
below.
1. After nonlinear iterative displacement convergence is achieved,
calculate stresses in the global (x, y, z) coordinates at the middle of
each layer at each Gauss point.
2. Transform the stresses to the principal material coordinates.
3. Compute the failure index, F.
4. If failure occurs (Le., F 2: 1),
58
(a) identify the maximum Hi,
(b) reduce the appropriate lamina moduli at that Gauss point, and
(c) recompute laminate stiffnesses and restart the nonlinear analysis
at the same load step (Le., return to Step 1).
5. If no failure occurs, proceed to the next load step.
The end shortening of the panel is monitored as in a compression test. The
failure load is defined to be that load for which the panel undergoes large
end shortening for small increments of load.
Comparison between test results from Reference 16 and finite element
results from Reference 97 for Panel C4 are shown in Figure 6.10. The
figure shows end shortening UQ, normalized by the analytical end shortening
Uer at buckling (Figure 6.10a)j out-of-plane deflection Wo near a point of
maximum deflection, normalized by the panel thickness h (Figure 6.lOb)j
and the longitudinal surface strains e near a point of maximum out-of-plane
deflection, normalized by the analytical buckling strain eer. These are all
shown as functions of the applieu. load P, normalized by the theoretical
buckling load Per. The circles in the figure represent test data, and the
curves denote the results of the nonlinear finite element analyses. These
experimental and finite element results agree well up to failure of the panel.
The post buckling response exhibits large out-of-plane deflections (nearly
three times the panel thickness; see Figure 6.lOb) and high longitudinal
strains from front and back surfaces (nearly three times the analytical
buckling strain; see Figure 6.10c).
Figure 6.11 contains the distribution of the axial stress U xx in the third
layer of the laminate (a 0° ply) at panel midlength for three values of
the applied load. At the buckling load, the axial stress is nearly uniform
across the panel. Although the axial stress is large, the values are well
below the material allowable values in tension XT = 1400 MPa (203 ksi)
and compression Xc = 1138 MPa (165 ksi). The contour plot of axial
stress U xx over the entire panel in this 0° ply for an applied load of 2.1Per
indicates (not shown here; see Reference 17) that high compressive axial
stresses occur along the longitudinal edges of the panel.
Figure 6.12 shows the distribution of the transverse shear stress in the
third layer of the laminate (a 0° ply) at panel midlength for three values of
the applied load. The solid curves represent the transverse shear stress
distributions obtained using the constitutive relations, and the dashed
curves denote the transverse shearing stress distributions obtained from
the equilibrium equations. Both methods give very similar results. At the
59
buckling load, the peak transverse shear stress occurs near the center of
the panel. After buckling, the transverse shear stresses (J'xz redistribute
towards the edges of the panel. The peak values of the transverse shear
stress (J'xz approach the material allowable value of S = T = 62 MPa
(9 ksi) for P = 2.1Pcr , indicating the panel failure due to transverse shear
stress. A contour plot (not shown here) of the distribution of the transverse
shear stress (J'xz over the entire panel in this 0° ply for an applied load of
P = 2.1Pcr indicates that high transverse shear stresses occur along the
buckling-mode nodal line.
This failure mode can be further explained through a close examination
of the Green-Lagrange strain component (3.29)
exz = ~ (8Ul + 8U3 + 8 Ul 8 Ul + 8U2 8U2 + 8U3 8U3) (6.16)

2 8z 8x 8x 8z 8x 8z 8x 8z
in conjunction with the displacement field of the first-order shear
deformation theory
Ul(X,y,Z) =uo(x,y) +z¢x

U2(X, y, z) =vo(x, y) + Z¢y
U3(X, y, z) =wo(x, y) (6.17)
Substituting of the displacements from Eq. (6.17) into the strain in Eq.
(6.16) and noting that ¢y is zero along a buckling-mode nodal line, we
obtain
exz
1(¢x + 8wo
="2 8uo)
8x + 8x ¢x (6.18)
The quantity ~ (out-of-plane deflection gradient) is largest along a

buckling-mode nodal line and the quantity ~ (related to the membrane
strain) is largest along the panel edges. A similar examination of the other
transverse shearing strain eyz leads to the conclusion that the transverse
shearing strain exz is the dominant one.
Figures 6.13a and 6.13b present the progressive failure results for Panel
C4, using the maximum stress and Tsai-Wu failure criteria, respectively.
In addition to the strengths already mentioned, the other allowables used
are transverse tensile strength, XT = 80.9 MPa (11.7 ksi); transverse
compressive strength, Xc = 189.0 MPa (27.4 ksi); and in-plane shear
strength, T = 69.0 MPa (10.0 ksi). At some point in the analysis a dramatic
change in slope indicates an inability of the panel to support additional
load. This location is identified as the failure load.
60
2.5
Uo constant}
.
p ~ 2.0
Wo= dWo =0
dx
"'
I:l.,.
fil~
1.5
~
El
tIS
1.0
uo=wo=O} ~
1=1.
dWo =0 't:I
ax tIS 0.5 ( - Analysis )
oS o Test
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4
(a) End shortening, uolu cr
p 2.5
.
~ 2.0
Wo
"'
I:l.,.
fil~
1.5
~
El
/ ftIS 1.0
1=1.
't:I Analysis )
tIS 0.5
0 Test
...::I
0.0
(b) 0.0 0.5 1.0 1.5 2.0 2.5
Maximum deflection, wol h
2.5
.
~ 2.0
"'
I:l.,.
~ 1.6
~
1=1.
1.0
'"g 0.6
0
...::I
0.0
(c) -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1
Strain, e leer
Figure 6.10. Postbuckling response characteristics of panel C4.

61
(MPa)
20
PIP,,=2.1 100
10
~ -10
'O'a -20
.,.,"
....,..
-30
Q)
-40
<a -50
~ -60
-70 -500
-80
0.0 0.2 0.4 0.6 0.8 1.0
ylb
Figure 6.11. Axial stress, (/xx, distributions in the third layer from the surface
(0° ply) of panel C4.
(MPa)
1.0
0.0
-1.0
~
-10
-2.0
'0'4 -3.0
.,ui -4.0
b., -30
..
t\I
Q)
-5.0
-6.0 -40
..d
rJl
-7.0
-50
-8.0
-60
-9.0
0.0 0.2 0.4 0.6 0.8 1.0
ylb
Figure 6.12. Transverse shear stress, (/xz, distributions in the third layer from
the surface (0° ply) of panel C4.
62
Figures 6.13a and 6.13b show that the Tsai-Wu criterion estimates the
experimental failure more closely than the maximum stress criterion. This
is attributed to the presence of stress interaction terms in the Tsai-Wu
criterion failure index.
. 2.5 - - Ultimate failure

~
....... - - - - - - - - - - - - - - - - - - - - - - - First-ply failure
~ 2.0
J-oft
.BQ) 1.5
8
til
J-o
til 1.0
Po. - - Progressive failure
'"C
til 0.5 o Test
0
...:I • Test failure
0.0 ~:-.-r..-r"...rr-r""". . .rr-r--rT."..;-nrTT..-rrr-rr-r-,-,-. . .rrl
(a) Maximum stress 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0
failure criterion End shortening, uolucr
. - - - - - - - - - - - - - - - - - - - - - _. Ultimate failure
~ 2.0
....... - - - - - - - - - First-ply failure
~
J-oft
~ 1.5
~
:: 1.0
Po. - - Progressive failure
1 ...:I
0.5 o Test
• Test failure
0.0 2.0 4.0 6.0 8.0 10.0 12.0

(b) Tsai-Wu
failure criterion End shortening, uolucr
Figure 6.13. Progressive failure results of panel C4.

63
7. On Laminated Composite Plates with Integrated

Sensors and Actuators
7.1 INTRODUCTION
7.1.1. Background
In the last two decades, the subject area of smart/intelligent materials and
structures has experienced tremondous growth in terms of research and
development. Numerous conferences, workshops, and journals dedicated
to smart materials and structures stand testimony to this growth. The
vastness of the literature is obvious considering the interdisciplinary nature
of the subject. Physicists, mathematicians, and engineers from aerospace,
chemical, civil, electrical, materials, and mechanical engineering fields
are all involved in some part of the development of smart materials and
structural systems. One reason for this activity is that it may be possible
to create certain types of structures and systems capable of adapting
to or correcting for changing operating conditions. The advantage of
incorporating these special types of materials into the structure is that the
sensing and actuating mechanism becomes part of the structure by sensing
and actuating strains directly. These types of mechanisms are referred to
as strain sensing and actuating (SSA).
The technological implications of this class of materials are immense:
structures that monitor their own health, particularly useful in remote
operations, process monitoring, vibration isolation and control, and medical
applications, to name only a few. On the threshold of 21st century,
we are entering into research and development of the next generation
of smart materials and structural systems. The next generation of
smart material systems will feature thermo-electro-mechanical coupling,
functionality, intelligence, and miniaturization (down to nano length
scales). With the advent of these new generation of materials, reliability
and integrity of these systems become central issues. These systems operate
under varying conditions - they span the whole spectrum of magneto-
electro-thermomechanical conditions. These conditions could vary from
low to high temperatures, low to high pressures, low to high load levels,
low to high strain levels, and low to high electric and magnetic fields.
Such operating environments pose serious problems to the design and
maintenance of the smart structural systems. Experimental investigations
of both the smart material and the structural system, though possible, are
prohibitively expensive, and therefore they should be complemented with
theoretical analyses.
64
Here, a theoretical formulation based on the classical plate theory is

presented for the analysis of laminated composite plates with integrated
sensors and actuators and subjected to both mechanical and electrical
loadings [54]. A simple negative velocity feed back control algorithm
coupling the direct and converse piezoelectric effects is used to actively
control the dynamic response of an integrated structure through closed
loop control. The following review of literature provides a background for
the present study.
7.1.2. Review of Literature

The phrase smart structural system refers to a wide variety of active
material and passive structural systems. For instance, a sufficiently general
system is a composite (beam, plate, shell, or any other fundamental form)
with embedded or surface mounted piezoelectric or electrostrictive patches,
or even layers of active materials in a laminated system. The literature
contains many definitions of smart and intelligent structures. We prefer to
follow Newnham's definitions [22~ The structures with surface mounted
or embedded sensors and actuators, that have the capability to sense and
take corrective action are referred to as smart structures. The feedback
circuitry linking sensing and actuating is external to the sensor and actuator
components. In fact, this precisely distinguishes a smart structural system
from an intelligent structural system. Intelligent structural systems involve
smart components in which the functions of sensing, feedback control, and
actuating are all integrated. This type of system finds applications in
aircraft wings, helicopter rotors, weapon systems, automobiles, and so on.
Essentially, a smart structural system is a multi-functional unit. It has
a load (electrical, thermal, magnetic, or mechanical) bearing part which is
usually passive, and an active material part that performs the operations of
sensing and actuating. For example, vibration amplitudes in a flexible plate
structure may be suppressed using the sensing and actuation capabilities
of piezoelectric or piezoceramic films by bonding them to the surfaces of
the plate. As the plate deforms due to external applied loads, the bonded
piezoelectric film (sensor) also deforms, and due to its constitutive behavior,
it develops a surface charge proportional to the applied force. The charge
may be processed by a control system, which supplies an appropriate
voltage to the piezolectric film (actuator) that induces a counteractive
deformation to the plate structure and the amplitudes of vibrations are
suppressed.
Piezoelectric materials, electrostrictive materials, magnetostrictive
materials, shape memory alloys, and electro-rheological fluids are some
65
of the smart materials available today. Among the currently available
sensors and actuators, the smallest ones are of the order of few millimeters.
However, progress towards intelligent structures requires us to develop
smart material systems that are of the order of a few microns. The
reduction in size has tremendous technological benefits; however, clear
understanding of reliability and system integrity are vital to the efficient
and optimum use of these material systems. As dimensions get smaller,
induced electro-thermo-mechanical fields get larger. Therefore, the
convenience of linearity in modeling should be abandoned and material
and geometric nonlinearities should be accounted for.
Piezoelectricity is a phenomenon in which some materials develop
polarization upon application of strains [23]. This phenomenon is observed
in materials that have a non-centrosymmetric crystal structure. Examples
of piezoelectric materials are Rochelle salt, quartz, and the most popular
one, Lead Zirconate Titanate or PZT (Pb (Zr,Ti) 0 3 ).
Piezoelectric materials exhibit a linear relationship between the electric
field and strains for low field va' ~s (up to 100 Vjmm). However, the
relationship is nonlinear for large fields, and the material exhibits hysteresis
[24]. Furthermore, piezoelectric materials show dielectric aging and hence
lack reproducibility of strains, i. e., a drift from zero state of strain is
observed under cyclic electric field conditions [25].
Electrostriction is a second-order effect observed in all dielectrics where
an applied uniform electric field induces a strain, but a reversal of the field
does not alter the strain [23]. That is, the strain is proportional to the
square of the electric field. An electrostrictive material that is being used
increasingly is Lead Magnesium Niobate or PMN (Pb (Mg 1/ 3 Nb 2 / 3 ) 0 3 ).
These materials have a nonlinear field-strain relationship, but exhibit very
little, if any, hysteresis [26]. Additionally, these materials have excellent
zero strain reproducibility in cyclic electric field conditions [26]. However,
strains induced by electrostriction are comparable to piezoelectricity only
at large electric fields. But the voltage can be reduced to the piezoelectric
level for an electrostrictive material with a high dielectric constant [26].
Table 7.1 summarizes the characteristics of piezoelectric and electrostrictive
materials.
The active composite elements are available in two forms: multi-layered
active materials and multi-phase active materials. The multi-layered active
materials are stacks of active materials wherein it is possible to achieve
large fields for moderate voltage input levels [30-35]. A typical multilayer
piezoelectric actuator is shown in Figure 7.1.
66
Table 7.1. Characteristics of Piezoelectric and Electrostrictive Materials.
Piezoelectric Materials
• At low electric fields, the strain-electric field relationship is linear.
• At large electric fields, the relationship is nonlinear, and hysteresis
appears under cyclic field conditions.
• Exhibits drift from zero strain state under cyclic field conditions due
to dielectric aging.
• In the linear regime, the piezoelectric coefficient is constant and hence
cannot be electrically tuned with a bias field.
• Has been studied extensively in the last few years, and several analysis
tools have emerged; experimental characterization has been done
extensively.
• Currently still popular because of lower power requirements at high
frequencies, relatively lower pnces, and above all an understanding of
the component integrity at low fields.
Electrostrictive Materials
• The strain-electric field relationship is nonlinear.
• Shows very little, if any, hysteresis.
• Has excellent reproducibility of zero strain state.
• Requires a very high dielectric constant to generate a large
electromechanical coupling.
• Nonlinear relation between the strain and field can be used to tune the
dielectric constant; therefore, the electromechanical coefficient can be
tuned over a wide range, changing from inactive to extremely active
states.
• While the constitutive relationship has been in existence in the
literature for quite a while [26-29]' modeling and analysis of the
behavior of electrostrictive materials under various loading conditions
have not been carried out. Further understanding of reliability and
structural integrity of smart structural systems with electrostrictive
materials is necessary.
67
Figure 7.1. A multilayer actuator.
The multi-layer composite system is also amenable to miniaturization

as it is possible to decrease the thickness of active material layers.
However, accounting for nonlinearity in modeling is essential. Considerable
research and development has been carried out in piezoelectric multi-layered
ceramics. Active vibration control using these multi-layered piezoceramic
configurations has been investigated [33-35].
Multi-phase composite systems consist of an active phase embedded
in a passive matrix (for example, PZT rods in a polymer matrix [36],
as shown in Figure 7.2). Different connectivity schemes are possible in
a two-phase system. The classification based on connectivity of active
and passive phases was developed by Newnham, et al. [37]. Literature
abounds with papers on piezoceramic rods (active) embedded in polymer
matrix (passive), forming a 1-3 connectivity composite system [36]. Finite
element analysis of 1-3 composite systems has also been carried out
[38,39]. These multi-phase composites are most often useful when there
are conflicting requirements on the active material system. However, for
reliability, mechanics studies relating to active/passive material bonding
interactions are necessary. Certain receptor cells in the cochlea of the inner
ear undergo displacements five orders of magnitude more than what the
best piezoelectric ceramic can produce at the same level of applied voltage
and dimensionality.
68
; - - - Matrix (Polymer)
PZTRods
Figure 7.2. A schematic of a 1-3 connectivity composite.
The mechanics of smart material systems involves coupling between

electric, magnetic, thermal, and mechanical effects. In addition to
this coupling, it may be necessary to account for geometric and
material nonlinearities. For example, an electromechanical transducer is
characterized by five important ~-operties [22]: the resonant frequency,
acoustic impedance, mechanical damping coefficient, electromechanical
coupling coefficient, and the electric impedance. If nonlinear electroelastic
equations are included in the model, some or all of these properties can be
tuned; for instance, in an electrostrictive material, the electromechanical
coupling coefficient can be tuned with a bias field. In order to tune the
first fundamental resonant frequency of the transducer, thin rubber layers
are introduced in a multilayer PZT laminate [40]. The thin rubber layers
necessitate the use of nonlinear elastic relations.
Toupin [41] was the first one to consider the nonlinearity in
electro elastic formulations. Knops [42] presented a two-dimensional theory
of electrostriction and solved a simplified boundary value problem using
complex potentials. Tiersten [27,30] has derived rotationally invariant
nonlinear thermo electro elastic equations and stressed the importance of
including nonlinear terms in the constitutive relations.
Analytical studies have largely concentrated on fracture of
piezoceramics and electrostrictive ceramics. Cracks emanating from
electrode junctions, debonding in active/passive composite systems, and
delamination in multi-layered stacking are some of the analytical problems
that have been considered in the literature [43,44]. Studies of laminated
structures with piezoelectric actuators have been primarily devoted to
either vibration suppression or static bending [33-35, 46-54]. In most of
these studies, straight beams have been analyzed using the Euler-Bernoulli
or Timoshenko beam theories and plates/shells using CLPT and FSDT.
69
The beam, plate, and shell theories were modified to allow for plies with
induced strains. Most of these studies did not account for geometric or
material nonlinearities and transient response. Here a general formulation
for laminated composite plates with piezoelectric actuators and sensors
is presented using CLPT, and it takes into account the thermo-electro-
mechanical coupling, von Karman type geometric nonlinearity, and time
dependency. The Navier solutions for the linear case and the finite element
formulation of the theory are presented in Reddy [54].
7.2. THEORETICAL FORMULATION USING CLPT

7.2.1. Equations of Motion
The equations of motion derived in section 3.2 account for thermal (and
hence, moisture) and piezoelectric effects when the force and moment
resultants are calculated using appropriate constitutive equations. In
these formulations, it is to be understood that the material properties are
independent of temperature and electric fields, and that the temperature
T and electric displacement vec' ~ £ are known functions of position.
Thus temperature and electric fields enter the formulation only through
constitutive equations [see Eq. (2.4)].
The equations of motion of the classical laminate plate theory are [see
Eqs. (3.7)-(3.9)]
(7.1)
(7.2)
(7.3)
All of the quantities appearing in Eqs. (7.1)-(7.3) were already defined

earlier in section 3.2. Note that the force and moment resultants are
related to the mechanical, thermal, and piezoelectric strains through the
constitutive equations (2.4) and the definitions of Ns and Ms in Eq. (3.10).
Since laminated plates are more general than isotropic and orthotropic
plates, the developments of this section are also applicable to isotropic or
orthotropic plates with piezoelectric layers. We will consider the laminate
constitutive equations next.
70
7.2.2. Laminate Constitutive Equations
Here we relate the force and moment resultants to the strains. By definition
(3.10) and constitutive relations (2.4), we have
where {NT} and {N P } are thermal and electric force resultants
(7.5)
(7.6)
Na is the number of actuating layers, and

(7.7)
Similarly, we have
B12
B22
B16]
B26 cW }+ [DU
{c~?J D12
B26 B66 ",(0)
(zy
D16
-{~}-{~~}
M;'y M~
(7.8)
where {MT} and {M P } are thermal and electric moment resultants
(7.9)
(7.10)
71
Note that Q's, and therefore A's, B's, and D's, are, in general, functions
of position (x,y). Equations (7.4) and (7.8) can be written in a compact
form as
{N}} _ [[A]
{ {M} (7.11)
- [B]
Suppose that the temperature and electric fields vary linearly within
kth layer
b.T = Tf(x, y, t)'l/i~(z) + T~(x, y, t)'l/i~(z) (7.12a)

£z = £f(x, y, t)'l/i~(z) + £~(x, y, t)'l/i~(z) (7.12b)
where 'l/if are the linear interpolation functions of the kth layer
(7.13)
and Tf and TIJ denote the temperature at the bottom and top surfaces
of the kth layer. Similar notation is used for the electric field. Then the
thermal and electrical forces and moments can be evaluated as
where (i = 1,2,6)
(7.15a)
(7.15b)
72
are the thermal stiffnesses, and
(7.16a)
(7.16b)
are the piezoelectric stiffnesses, and Na < N is the number of actuator

layers. If the electric field intensity is constant across each lamina, then
one may approximate it as
(7.17)
where Vk is the applied voltage across the kth layer and hk is the thickness
of the layer.
7.3. SENSOR AND ACTUATOR EQUATIONS

7.3.1. Electro-Mechanical Coupling
The electro-mechanical coupling is a two-way coupling. The effect of
applied electrical field on stresses is, as indicated by Eq. (2.5), is analogous
to the effect of the temperature field. The components Di of electrical
displacement vector are related to the components of strains and electrical
field by
o O](k){Cl} [El1 E~2 ~] (k) {~~ }(k)

o o e2 + 0
o C6 0 o E33 E3
(7.18)
where eij are piezoelectric stiffnesses and Eij are dielectric coefficients. The
transformed equations are
o 1(k) {
Dx }(k) _ [ 0 o o cxx } [~11 !~~ _~ 1(k) {~: }(k)
{ Dy - 0 o cyy - E12
Dz €31 €36 1xy 0 o E33 Ez
(7.19)
This electromechanical equation provides the starting point for the
derivation of the sensor and actuator equations.
73
7.3.2. Sensor Equations

According to the Gauss law, the closed circuit charge measured through
the electrodes of the kth layer is
(7.20)
Figure 7.3. Lamina with surface electrodes; St= top surface electrode, Sb=
bottom surface electrode.
where S I = Sb n St is the intersection of the electrode surfaces on both

sides of the lamina (see Figure 7.3) and subscript's' denotes sensor.
The electrodes are assumed to be placed on the transverse surfaces with
the poling direction z. Hence, only the component Dz of the electric
displacement vector is nonzero. When Eq. (7.19) is applied to sensors
where the converse piezoelectric effect is negligible and the external applied
charge is zero, the sensed charge can be calculated from Eq. (7.19) as
(7.21)
The total charge in the laminate is calculated by summing over the number
of sensor layers, Ns:
(7.22)
74
Substituting Eq. (7.21) into (7.22), we obtain
Qs (t ) -- !2 ~
k=l
~
1 [-
81
e31cxx
(0)
+ e32 yy + e36'Yxy
- C(0) - (0)
k (- (1)
+Zc e31cxx + e32
-
Cyy + e36'Yxy
(1) _ (1»)] (k)
dxdy (7.23)
where z~ = O.5(Zk + Zk+1)'

We note that Eq. (7.23) was obtained by setting the externally applied
fields £(0) and £(1) to zero, because when the surface electrodes are short-
circuited on both sides of the piezoelectric lamina, the terms containing
the externally applied electric field intensities £~1) and £~2) should be set
to zero. The resulting equation is the closed-circuit charge sensor equation,
which relates the output signal to the plate deformation.
The current let) on the surface of the sensor is given by
1(" = dQs (7.24)

, dt
When the sensors are used as strain rate tensors, the current can be
converted to the open circuit sensor voltage output Vs by [33-35]
Vs(t) = Gcl(t) = Gcd~s (7.25)
where G c is the gain of the current amplifier.

The sensor forces {N P } and moments {M P} (and {PP}) are then
evaluated by using
{&}= U} h.
(7.26)
in Eqs. (7.10) and (7.11). Here hs denotes the thickness of the sensor layer.
The forces and moments will be determined in terms of of the strain rates
in the sensor layers.
7.3.3 Actuator Equations

The actuator equations for piezoelectric actuator can be derived using
induced strain actuation. We assume that the piezoelectric actuator layer
does not have any applied stress. Then Eq. (7.19) gives the strains
developed by the electric field on the actuator layer
(7.27)
75
or
(7.28)
The stresses due to the actuator strains are
(7.29a)
or
Ui=Qijdkj£k (i,j=I,2,6; k=I,2,3) (7.29b)
The stress vector can be used to generate the actuator forces {N P } and
moments {M P }.
If the voltage applied to the actuator in the thickness direction (only)
is Va, then
{E}~ {1}
ha
(7.30)
where ha is the thickness of the a . lator layer. Then we have
(7.31)
where z~ is the distance from the midplane to the center of the actuator
layer.
8. References for Additional Reading

1. Ambartsumyan, S. A., Theory of Anisotropic Sheils, NASA Report TT
F-118 (1964).
2. Ambartsumyan, S. A., Theory of Anisotropic Plates, Izdat. Nauka,
Moskva (1967), English translation by Technomic, Stamford, CT (1969).
3. Jones, R. M., Mechanics of Composite Materials, Scripta Book Co.,
Washington, D.C. (1975).
4. Reddy, J. N. (ed.), Mechanics of Composite Materials. Selected Works
of Nicholas J. Pagano, Kluwer Academic Publishers, Netherlands (1994).
5. Reddy, J. N., Mechanics of Laminated Composite Plates, Theory and
Analysis, CRC Press, Boca Raton, FL (1997).
6. Reddy, J. N., Energy and Variational Methods in Applied Mechanics,
John Wiley, New York, 1984.
7. Reddy, J. N., An Introduction to the Finite Element Method, Second
Edition, McGraw-Hill, New York, 1993.
76
8. Zienkiewicz, O. C. and Taylor, R. L., The Finite Element Method, Vol. 1:
Linear Problems, McGraw-Hill, New York, 1989.
9. Reddy, J. N., "A Penalty Plate-Bending Element for the Analysis of
Laminated Anisotropic Composite Plates," International Journal for
Numerical Methods in Engineering, 15(8), 1187-1206 (1980).
10. Reddy, J. N., "Geometrically Nonlinear Transient Analysis of Laminated
Composite Plates," AIAA Journal, 21(4), 621-629 (1983).
11. Zaghloul, S. A. and Kennedy, J. B., "Nonlinear Behavior of Symmetrically
Laminated Plates," Journal of Applied Mechanics, 42, 234-236 (1975).
12. Zaghloul, S. A. and Kennedy, J. B., "Nonlinear Analysis of Unsymmetrically
Laminated Plates," Journal of Engineering Mechanics Division, ASCE,
101(EM3), 169-185 (1975).
13. Reddy, J. N., "On Refined Computational Models of Composite Laminates,"
International Journal for Numerical Methods in Engineering, 27, 361-
382 (1989).
14. Barbero, E. J. and Reddy, J. N., "Nonlinear Analysis of Composite
Laminates Using a Generalized ~aminate Plate Theory," AIAA Journal,
28(11),1987-1994 (1990).
15. Sun, C. T. and Chin, H., "Analysis of Asymmetric Composite Laminates,"
AIAA Journal, 26(6), 714-718 (1988).
16. Starnes, J. H., Jr. and Rouse, M., "Postbuckling and Failure
Characteristics of Selected Flat Rectangular Graphite-Epoxy Plates Loaded
in Compression," AIAA Paper No. 81-0543 (1981).
17. Engelstad, S. P., Reddy, J. N., and Knight, N. F., Jr., "Post buckling
Response and Failure Prediction of Graphite-Epoxy Plates Loaded in
Compression," AlA A Journal, 30(8), 2106-2113 (1992).
18. Chao, W. C. and Reddy, J. N., "Analysis of Laminated Composite Shells
Using a Degenerated 3-D Element," International Journal for Numerical
Methods in Engineering, 20, 1991-2007 (1984).
19. Tsai, S. W., "A Survey of Macroscopic Failure Criteria for Composite
Materials," J. Reinforced Plastics and Composites, 3, 40-62 (1984).
20. Hill, R., "A Theory of the Yielding and Plastic Flow of Anisotropic Metals,"
Proceedings of the Royal Society, Series A, 193, 281-297 (1948).
21. Wu, E. M., "Phenomenological Anisotropic Failure Criterion," Composite
Materials, 2, 353-431 (1974).
22. Newnham, R. E., "Ferroelectric Sensors and Actuators: Smart ceramics," in
Ferroelectric Ceramics (edited N. Setter, and E.L. Colla), 363-380 (1993).
23. Maugin, G. A., Continuum Mechanics of Electromagnetic Solids, North-
Holland, Amsterdam, The Netherlands (1988).
77
24. Uchino, K., "Electrostrictive Actuators: Materials and Applications," Cer.

Bull., 65, 647-652 (1986).
25. Cross, L. E. and Jang, S. J., "Electrostrictive Materials", in
R.C. Pohanka and P.L. Smith, "Recent Advances in Piezoelectric Ceramics,"
Electronic Ceramics, Properties, Devices, and Applications (edited by
L.M. Levinson), Marcel Dekker, Inc., New York, 129-137 (1988).
26. Nomura, S. and Uchino, K., "Electrostrictive Effect in Pb (Mg 1/ 3 Nb 2 / 3 )
03 - Type Materials," in Piezoelectricity (edited by G.W. Taylor, et al),
Gordon and Breach Science Publishers, New York, 151-166 (1985).
27. Tiersten, H. F., "On the Nonlinear Equations of Thermoelectroelasticity,"
International Journal of Engineering Science, 9, 587-604 (1971).
28. Baumhauer, J. C. and Tiersten, H. F., "Nonlinear Electroelastic Equations
for Small Fields Superposed on a Bias," Journal of Acoustical Society of
America, 54, 1017-1034 (1973).
29. Nelson, D. F., "Theory of Nonlinear Electroacoustics of Dielectric,
Piezoelectric, and Pyroelectric Crystals," Journal of Acoustical Society
of America, 63,1738-1748 (19.u).
30. Tiersten, H. F., "Electroelastic Equations for Electroded Thin Plates
Subject to Large Driving Voltages", Journal of Applied Physics, 14,3389-
3393 (1993).
31. Takahashi, S., "Multilayer Piezo-Ceramic Actuators and
Their Applications," in Ferroelectric Ceramics (edited by N. Setter and
E.L. Colla), 349-362 (1993).
32. Winzer, S. R., Shankar, N., and Ritter, A., "Designing Cofired Multilayer
ElectroStrictive Actuators for Reliability," Journal of Acoustical Society
of America, 72, 2246-2257 (1989).
33. Lee, C. K., "Theory of Laminated Piezoelectric Plates for the Design of
Distributed Sensors/Actuators. Part I: Governing Equations and Reciprocal
Relationships," Journal of Acoustical Society of America, 81, 1144-1158
(1990).
34. Mitchell, J. A. and Reddy, J. N., "A Refined Hybrid Plate Theory for
Composite Laminates with Piezoelectric Laminae," Journal of Solids and
Structures, 32(16), 2345-2367 (1995).
35. Reddy, J. N. and Mitchell, J. A., "On Refined Nonlinear Theories of
Laminated Composite Structures with Piezoelectric Laminae," Sadhana,
20, pp. 721-747 (1995).
36. Sottos, N. R., Li, L., Scott, W. R., and Ryan, M. J., "Micromechanical
Behavior of 1-3 Piezocomposites," in Smart Structures and Materials
(edited by V.K. Varadan), Albuquerque, NM, 87-96 (1993).
78
37. Newnham, R. E., Skinner, D. P., and Cross, 1. E., "Connectivity and
Piezoelectric - Pyroelectric Composites," Materials Research Bulletin, 13,
525-536 (1978).
38. Hossack, J. A. and Hayward, G., "Finite Element Analysis of 1-3 Composite
Transducers," IEEE Transanctions on Ultrasonics, Ferroelectrics, and
Frequency Control, 38, 618-629 (1991).
39. Hladky-Hennian, A. C. and Decarpigny, J. N., "Finite Element Modeling of
Active Periodic Structures: Application to 1-3 Piezocomposites," Journal
of Acoustical Society of America, 94, 621-635 (1993).
40. XU, Q. C. et al., "Nonlinear Multilayer Composite Transducers,"
Proceedings of the Seventh International Symposium on Applications of
Ferroelectrics, Urbana (1990).
41. Toupin, R. A., "The Elastic Dielectric," Archives of Rational Mechanics
and Analysis, 5, 849-915 (1956).
42. Knops, R. J., "Two-Dimensional Electrostriction," Quaterly Journal of
Mechanics Applied Mathematics, 16, 377-388 (1963).
43. Taylor, D. J., Newnham, R. -"'., and Carlson, W. B., "The Effect of
Electric Field on Mechanical Strain and Stress in Flawed Electroceramics,"
Ferroelectrics, 87, 81-84 (1988).
44. Yang, W. and Suo, Z., "Cracking in Ceramic Actuators Caused by
Electrostriction," Journal of Mechanics of Physics and Solids, 42, 649-
663 (1994).
45. Lee, C. K. and Moon, F. C., "Laminated Piezopolymer Plates for Torsion
and Bending Sensors and Actuators," Journal of the Acoustical Society of
America, 85, 2432-2439 (1989).
46. Tzou, H. S. and Gadre, M., "Theoetical Analysis of a Multi-Layered Thin
Shell Coupled with Piezoelectric Shell Actuators for Distributed Vibration
Controls," Journal of Sound and Vibration, 132,433-450 (1989).
47. Hwang, W. S. and Park, H. C., "Finite Element Modeling of Piezoelectric
Sensors and Actuators," AIAA Journal, 31, 930-937 (1993).
48. Suleman, A. and Venkayya, V. B., "A Simple Finite Element Formulation
for a Laminated Composite Plate with Piezoelectric Layers," Journal of
Intelligent Material Systems and Structures, 6, 776-782 (l995).
49. Koconis, D. B., Kollar, 1. P., and Springer, G. S., "Shape Control
of Composite Plates and Shells with Embedded Actuators: 1. Voltages
Specified," Journal of Composite Materials, 28, 415-458 (1994).
50. Miller, S. E., and Abramovich, H., "A Self-Sensing Piezolaminated Actuator
Model for Shells Using a First Order Shear Deformation Theory," Journal
of Intelligent Material Systems and Structures, 6, 624-638 (1995).
79
51. Chen, C. Q., Wang, X. M., and Shen, Y. P., "Finite Element Approach of
Vibration Control Using Self-Sensing Piezoelectric Actuators," Computers
€9 Structures, 60, 505-512 (1996).
52. Lee, C. K., "Piezoelectric Laminates: Theory and Experiment for
Distributed Sensors and Actuators," in Intelligent Structural Systems
(edited by H. S. Tzou and G. L. Anderson), Kluwer Academic Publishers,
Netherlands, 75-167 (1992).
53. Robbins, D. H. and Reddy, J. N., "Analysis of piezoelectrically actuated
beams using layer-wise displacement theory," Computers €9 Structures,
41(2),265-279 (1991).
54. Reddy, J. N., "On Laminated Composite Plates with Integrated Sensors and
Actuators," Engineering Structures, to appear.
About the Author
J. N. Reddy is the Distinguished Professor and holder of the Oscar S.

Wyatt Endowed Chair in the Depa. Hnent of Mechanical Engineering at Texas
A&M University, College Station, Texas. Prior to the current position, he was
Professor and holder of the Clifton C. Garvin Professorship in the Department
of Engineering Science and Mechanics at Virginia Polytechnic Institute and State
University (VPI&SU), Blacksburg, Virginia.
Dr. Reddy has authored over 200 journal papers and ten books on
the theory and finite element analysis of laminated composite plates and
shells, computational fluid mechanics, numerical heat transfer, and applied
mathematics. His seminal contributions to mechanics of composite materials
and structures are well known through his research on refined structural theories
and finite element models of laminated composite plates and shells.
Professor Reddy is the recipient of the Walter L. Huber Civil Engineering
Research Prize of the American Society of Civil Engineers, the Worcester
Reed Warner Medal and the Charles Russ Richards Memorial Award
of the American Society of Mechanical Engineers, the 1997 Archie Higdon
Distinguished Educator A ward from the American Society of Engineering
Education, and the 1998 Nathan M. Newmark Medal from the American
Society of Civil Engineers. Dr. Reddy is a Fellow of the American Academy
of Mechanics, the American Society of Civil Engineers, the American Society
of Mechanical Engineers, the U.S. Association of Computational Mechanics, the
International Association of Computational Mechanics, and the Aeronautical
Society of India. Dr. Reddy is the editor of the journal Mechanics of Composite
Materials and Structures, and serves on the editorial boards of over two dozen
other journals.
MICROMECHANICS ANALYSIS OF COMPOSITE MATERIALS
A. S. D. WANG
Drexel University
Philadelphia, PA 19104 USA
Foreword.
Micromechanics is a discipline that treats the mechanics of heterogeneous matrerials with

definitive microstructures. When applied to fibrous composites, the common approach is to
pass the physical material to an abstractive one; based on which analytical models are then
formulated, capable of describing the mechanics of the material under external load. In this
context, a "forward approach" and a "backward approach" exist. The former simplifies or
homogenizes the microstructure of the composite, while the latter attempts to restore a part
or the whole of the microstructure back into the abstracted composite. This paper offive (5)
connecting parts discusses the limitations in the foreward approach, and the needs and
challenges in the backward approach.
Part I. Composite Microstructures and Mechanics Modeling
Part I discusses the geometric nature of the microstructures in fibrous composites, the level
of abstraction necessary to render analytical models for their mechanical properties and the
consequences of microstructure abstraction. For clarity but without loss of generality.
laminates made of unidirectional fiber-reinforced plies are used in the discussions.
1. Multi-Level Microstructures.
Fiber-reinforced composites are a class of multi-phased materials that possess multi-level

microstructures. The physical detail in the microstructure depends on the dimensional scale
of observation; the description of the observed microstructure at a chosen dimensional level
will invariably involve many factors of material and geometrical origins.
Fig.l.l illustrates the many dimensional levels that a composite could be examined.
Suppose that the curved panel shown in Fig. I. 1a is a part of the fuselage in a small
airplane; the panel's dimension would measure in terms of meters. Consider the failure of
this panel under load; it may fail by a loss of stiffness globally. or by a loss of strength
81
82
locally; both may be precipitated by the presence of local weakening spots, such as at the
bolt-hole or at the lap-joint shown in Fig.l.lb.
To understand the mechanisms in the failure process at this level, it is necessary to
investigate the microstructure of the panel, especially near the bolt-hole or the lap-joint.
Suppose that the panel is made of laminated plies; the microstructure is one of layered
medium such as shown in Fig.1.lc. At this dimensional scale, all physical quantities of
size comparable to the thickness of the laminating plies should constitute the panel's
microstructure.
Of course, when viewed at the next dimensional scale, each of the plies will have a
microstructure such as depicted in Fig.1.l d. Here, a clear distinction can be made between
the fiber and the matrix phases; and the microstructure is endowed with a distribution of
fibers and other physical matters of size on the order of the fiber diameter.
Further focusing at anyone single fiber may reveal yet the minute details on the fiber-
matrix interface, e.g. the coating material on the fiber surface, the presence of microdefects
between neighboring fibers, etc., see Fig.l.le.
Clearly, one can go down to the next dimensional level, if needed, to investigate the
microstructure of the individual fiber or the physical nature of any material which may be
present in the fiber/matrix interface. Moreover, one can still go down to the atomistic scale
if necessary.
Structural Component Laboratory Coupon Fibe riMa tr ix/Interface/Flaw

( » 100 m < 1 - 100mm) <1 -100 "m)
1 >- 1-< >-1-<
GEOMETRIC LOOI LlZED- SUB LAMINATE- MICROFLAW- INTERFACE-
IMPERFECTION DAMAGE ZONE CRACKING D ISTRI BurlON MEOIANISMS
BI£KLlNG DAMAGE DELAMINATIJN. CRACK- DEBONDNG.

GROWlH ETC. COALESCENCE ETC
(a) (b) (c) (d) (e)
Figure 1.1 Dimensional regimes of microstructure in composite laminates
Suppose that the panel is a laminate made of a carbon fiber/polymer matrix ply system
and the smallest geometric factor of interest is to be defined at the fiber/matrix level. In that
case, the various identifiable geometric factors in the panel's microstructure are listed in
Table 1.
Note that these factors already span six or more decades of the dimensional scale.
While some factor are definitively measurable, many others are random in nature; and the
randomness is increasingly accute as the dimensional scale becomes smaller.
Now, given a composite material, the question is: what is the proper dimensional scale
at which an analytical model may be formulated so as to yield the relevant information
desired by the designer and/or the analysts? While this question is of extreme importance
83
from both a theoretical and practical point of view, the answer to the question is not easily
available.
TABLE 1. Geometric variables and

sizes in graphite-polymer laminates
Variables Size range
Fiberdiam. d =5-lOllm
Fiber volume fraction
Defects and voids l-lOd
Ply thickness t = 1O-50d
Number of plies =
n 4-50
Ply orientation/sequence
Laminate thickness h =4-50t
Laminate defects l-lOh
Laminate overall dim. » 100h
2. Microstructure Abstraction for Mechanics Modeling.

The microstructure in composites is almo~t "lways abstracted or simplied in the interest of
deriving a realistic mechanics model, regaraless the dimensional level of observation. On
the one hand, the abstraction is necessitated by the desires to describe the microstructure
precisely without ambiguity and to generalize a class of similar microstructures into one.
On the other hand, the abstracted microstructure must also contain the essential features of
the real materials and the unique geometries therein; so the resulting mechanics model may
yield reliable mechanical behaviors of the composites under applied load. However, these
are often mutually exclusive requirements, making a well-balanced mechanics modeling
effort difficult.
In this section, several well-developed mechanics modeling approaches are discussed in
light of their underlying microstructure abstraction and the limits of the resulting mechanics
models.
2.1. LAMINATE MODELS.
Laminates are most efficient structurally in the form of a plate, a curved panel or a shell. In
these structures, primary stresses develop in the plane of the lamination; the out-of-plane
stresses are generally secondary, except near local stress-risers such as a through-hole, a
cut-out, or the free-edge; see e.g. Fig. I. 1b.
There exist many mechanics models for laminates, depending on the level of
mathematical simplification rather than the level of microstructure abstraction. In this
regard, the classical laminated plate model, the various higher-order laminated plate models
and the three-dimensional laminate model are all based on a microstructure abstracted at the
ply-level as depicted in Fig. 1.2. Here, all the lamination models consider the unidirectional
ply as a heterogeneous material; when several plies are stacked together to form a laminate,
distinction of the individual plies (inhomogeneeeeity) then becomes observable. Thus, at
the heart of the ply-level abstraction is the assumption of ply homogeneity; i.e. distinction
between the fiber and matrix in the ply microstructure is ignored; or the ply is treated as one
84
without a microstructure al all. However, the effective properties of the ply must somehow
retained.
Unidirectional ply Lamination Laminate
Figure 1.2. Microstructure abstraction of laminated composites at the ply level

2.1.1. Ply-Elasticity
As aforementioned, the homogenized unidirectional ply material must retain the essential
properties that stem from the ply's microstructure. Ply Elasticity is one "forward" modeling
approach which serves as the basis for mar-' '.amination theories. Specifically, ply elasticity
assumes that the ply is a linearly elastic and homogeneous material characterizable by the
generalized Hooke's law in the theory of anisotropic elasticity [1]. The inclusion of material
anisotropy and evaluation of the effective material constants in the Hooke's law are
attempts to retain the effects of the ply microstructure.
For clarity, let (L,T,Z) be the principal material directions of the ply, with L in the fiber
direction, T transverse to fiber direction and z normal to the ply, see Fig. 1.2; let (x,y,z) be
the reference frame for the laminate, with x-y plane being the laminate midplane and z being
parallel to Z. The ply stress-strain relations expressed in the (L,T,Z) frame are given by the
generalized Hooke's law:
O"i =Cij Ej i,j = 1,2, .. 6. (Ll)
The ply stress-strain relations may include thermal effect due to a temperature change:
ij = 1,2, .. 6. (1.2)
where aj is the ply thermal expansion coefficients and l\T is the temperature drop from the
reference temperature To (usually , To is near the laminate curing temperature).
Material symmetry may be further assumed for the ply material [2]; either an orthogonal
symmetry may be assumed with respect to the (L,T,Z) frame or a transversely isotropic
symmetry may be assumed with (T-Z) being the plane of isotropy (in some cases, a
monoclinic symmetry with respect to the ply midplane may also be assumed). Material
symmetry reduces the number of the independent constants in Cij and aj-
The constants Cij and aj in (Ll) and (1.2) are known as the "effective" properties of
the ply; they suppose to reflect the properties of the ply microstructure. These constants can
85
be obtained from two approaches: either by direct measurement from experiment or by a

suitable micromechanics model that takees into account the ply microstructure. Discussions
on both approaches are briefly given in section 2.2.
2.1.2. Three-dimensional Laminate Model

At the outset, the plies in the laminate are assumed perfectly bonded together; thus, the
laminate displacements in the (x,y,z) frame can be expressed by the continuous functions:
ui (x,y,z) i = 1,2,3. (1.3)
the laminate strains are derived from the displacements as:
ei = (ui,j + uj,i)12 for i = 1,2,3

= (ui,j + Uj,i) for i = 4,5,6 (1.4)
The strain field in anyone ply can be obtained from the strains in (1.4) by a coordinate
transformation from (x,y,z) to (L,T,Z); and the stress field in that ply is given by (1.1) or
(1.2).
Equilibrium of stresses in each ply, matching of the interfacial stresses on each ply
interface and satisfaction of boundary conditions on the laminate surfaces will furnish a set
of governing relations in terms of the displacement functions Ui (x,y,z); solutions for Ui
(x,y,z) from the governing relations will in tum provide the ply strains via (1.4) and the ply
stresses via (1.1) or (1.2).
The 3-D elastic lamination model is logically simple only in appearance; it can become
analytically difficult for practical problems. In this regard, 3-D finite element methods are
commonly used to render numerical solutions when a 3-D stress analysis of the lamiante is
required.
2.1.3. The Classical Lamination Model

For laminates with relatively thin thickness, the inplane stresses are orders of magnitude
greater than the thickness-wise stresses. The ply stress-strain relations are essentially two
dimensional; the expression in (1.1) and (1.2) are then reduced, respectively, to:
0i = Qij Ej i,j = 1,2,6. ( 1.5)

OJ = Qij (Ej - a.j~T) j,j = 1,2,6. ( 1.6)
where Qij are the reduced stiffnesses:
Qij = Cij - (Ci3 Cj3)/C33 i,j = 1,2,6 ( 1.7)
The 2-D laminate displacements Ui reduce to the form:
ui (x,y,z) = uOi (x,y) + zfj (x,y) for i = 1,2

u3 (x,y,z) = u0 3 (x,y) ( 1.8)
while the laminate strain field is obtained from (1.8) in conjunction with (1.4); and the 2-D
stresses in anyone ply of the laminate are obtained using (1.5) or (1.6) whichever is
86
appropriate. It remains to determine the laminate displacements ui in (1.8) to complete the

formulation of the classical lamination theory.
The elastic response of the laminate as a whole is characterized by the integral effect of
the stresses and strains over the thickness, a treatment similar to the classical plate theory
[3]. Briefly, the plate force and plate deformation relations in the (x,y,z) frame are given
by:
i = 1,2,6 (1.9)
where Ni and Mi are the integrated plate forces and moments defined, respectively, as:
Ni = Icri dz Mi = Icri zdz i = 1,2,6 (1.10)
and the matrices Aij, Bij, and Dij are defined as:
Aij =JQij dz Bij =JQij zdz Dij = JQij z2dz i = 1,2,6 (1.11)
All integrals are carried over the thickness of the laminate.

In (1.9), eOj are strains in the laminate midplane and Kj are the curvatures of the
midplane; all are derivable from the displacements Ui in (1.8):
eOI =uOI,1 e0 2 =u02,2 e0 6 =uOI,2 + u02,1

KI =fl,1 K2 = f2,2 K6 = fl,2 + f2,1 (1.12)
The A, B, D matrices in (1.11) represent the properties of the laminate as an integrated

plate element; the effect of its ply materials and ply stacking structure on the global plate
behavior is retained through the thickness integrals. Physically, Aij measure the inplane
elastic plate stiffness; Dij measure the out-of-plane bending rigidity and Bij represent the
coupling between the inplane and bending deformations.
The field equations governing the displacements ui are derived from the equilibrium of
the plate element, on which the internal plate forces Ni, and moments Mi, and externally
applied surface tractions act. Complete solutions for the displacements ui are obtained by
satisfying the prescribed loading and displacement constraints on the laminate boundaries;
see, e.g. [2].
The classical laminated plate model is widely used due to its ease and versatility. It is
particularly useful for stiffness-dominant behaviors, such as membrane stretching, plate
bending, low frequency vibrations, etc. The laminates must be loaded within its elastic
range; or no significant failure in the laminate should occur.
However, the 2-D lamination model becomes inaccurate at locations such as along the
laminate boundary, around a bolt-hole, cut-out, sharp notches or delaminated interfaces in
the laminate interior. For in these localized regions, ply stresses are generally three
dimensional and with high gradient; a higher order lamiantion theory or the 3-D elasticity
model may become necessary.
2.1.4. Higher-Order Lamination Models

The higher-order lamination model is a reduced version of the 3-D ply elasticity lamination
model; but it is an improved version over the 2-D classical lamination model in that it can
87
recover a certain amount of the laminate thickness effect lost in the 2-D model. Here, the 3-
D laminate displacements ui (x,y,z) are expanded into a power series in the variable z:
ui = uOi (x,y) + zfi(x,y) + z2gi(x,y) + z3hi(x,y) + - - - - i=I,2,3. (1.13)
It is to be noted that the classical lamination model retains only the first two terms in u I
and U2, and the first term in u3, see (1.8); the higher-order lamination model, however,
would retain additional terms in each of Ui, see e.g. [4]. Thus, inclusion of higher order
terms in the model formulation will provide 3-D field solutions, including the interlaminar
stresses and strains that may occur in the laminate. In principle, if more terms in (1.13) are
retained in the higher order lamination model, the model would yield field solutions more
closer to the 3-D ply elasticity model; but the higher-order lamination model can become
just as analytically cumbersome as the 3-D ply elasticity lamination model.
2.2. MODELING AT THE FlBERlMATRIX SCALE.
In the above, a key material law used in formulating the various lamination models is the
generalized Hooke's law, (1.1) or (1.2), for the laminating plies. Implicit in this law is the
assumption that the elastic constants Cij and the thermal expansion coefficients CXj represent
the averaged or the effective response properties of the ply, which is now devoid of its
microstructure. It is thus a challenge of high order to obtain these or other ply properties
properly. The main challenge here is to realisticaly model the unidirectional ply at the
fiber/matrix level. In this context, a prop(.•• ~presentation of the ply microstructure at that
scale and an understanding of the response mechanisms therein are essentail.
However, even with the unidirectional ply system, the microstructure detail at the
fiber/matrix level can be rather perplex both materially and geometrically. As illustrated in
Fig. 1.3, the fiber-matrix composition may contain misaligned fibers, broken fibers, uneven
fiber size and distribution, fiber-matrix interface disbonds, voids and microcracks in the
matrix, just to mention a few. Moreover, these "anomalies" often exist in random and their
geometric details cannot be determined for sure.
Figure 1.3. Ply microstructure and the representative volume (RV)

For this reason, a proper abstraction or simplification of the ply microstructure becomes
necessary if any realistic micromechanics model is to be formulated. Generally, the extent
of microstructure abstraction varies with the intent of the analyst. In what follows, several
88
widely used microstructure abstraction approaches are discussed along with the associated
modeling limitations.
2.2.1. The Concept of Representative Volume

The concept of representative volume (RV) assumes that a small volume of the ply can be
identified and isolated such as shown by the insets in Fig. 1.3, and that the microstructure
in the RV is representative of that in the whole ply. Thus, by modeling the isolated RV with
a suitable micromechanics analysis, the averaged response properties of the RV such as Cij
and <Xj can be obtained in terms of the properties of the fiber and matrix materials along
with other possible elements that may be present in the RV. Since the RV is representive of
the ply, the locally determined effective properties are assumed to be the same throughout
the homogenized ply material.
This approach has been followed in characterizing composites at the fiber/matrix level
almost universally; several micromechanics models of varying degree of approaximation
are widely applied in practice, including the simple "rule-of-mixture".
2.2.2. The Rule of Mixture

The rule of mixture is the earliest and perhaps the most elementary model. Essentially, the
identified RV is a concentric cylinder of infinite length, in which the circular fiber is bonded
perfectly to a layer of the matrix. With this representation, there is no need to specify the
exact diameter of the fiber, nor the exact thickness of the matrix layer; the only geometric
parameter of importance is the fiber/matrix volume ratio, VfN m which matches the average
fiber/matrix volume ratio of the ply. Mecb~";cs analysis of the RV as an isolated free-body
is then performed with some "appropriate" stresses or strains specified onto the RV's outer
boundaries; so the stress and strain distributions in the fiber and matrix phases can be
obtained. Homogenization of the RV involves integration of the fiber/matrix stresses and
strains over the volume of the RV, which then results in a transversely isotropic material
with the (T,z) plane being the plane of isotropy.
In their most rudimentary forms, the five engineering elastic constants and the two
thermoelastic expansion coefficients in the principal material coordinates (L,T,z) are given
by:
EL = EfVf+ Em Vm ET = (V rlEr + V mlEm)-1
GLT = (Vf/Gf+ Vm/Gm)-l ~z= GrYr+GmVm
VLT =VfVf + Vm Vm (1.14)
and
<XL = (<xrEfV f+ <XmEm Vm)/(EfVf+ Em Vm)
<XT =<XrV f+ <Xm V m (1.15)
The above results are regarded as a first estimation of the ply properties, due obviously
to the lack of rigor in the micromechanics analysis. In fact, except for the axial modulus EL
and perhaps also <XL, the others are often inaccurate. Some refined mechanics analyses
have added correction factors to the transverse properties in (1.14) and (1.15), in order to
augment the accuracy of these properties to some extent (see, e.g. [5]).
2.2.3. Bounding by Variational Mechanics.

Note that the RV in the form of a concentric cylinder is an isolated free-body from the
larger ply material. As such, the outer boundary of the RV is subjected to unknown surface
89
tractions from its surrounding material. This fact was not observed in the rule-of-mixture
analysis discussed above.
The premise of the bounding method is to specify either tractions or displacement (or
strain) on the RV outer surface. In either case, the fiber/matix stress and strain distributions
in the RV can be analyzed in terms of the specified surface traction or displacement via a
variational procedure; see, e.g. Hashin, et. al. [6]. Now, if surface traction is specified,
minimization of the total strain energy in the RV will result in an over-estimate for the
displacement field; if surface displacement is specified, it will result in an under-estimate.
Thus, upon homogenization of the RV, the former yields a lower bound for the elastic
constants, while the latter yields an upper bound.
Because it is difficult to specify a near-the-truth surface agency distributions on the
boundaries of the RV, the bounds obtained for the composite properties can very widely
depending on the assumed surface agency distributions. Nonetheless, the bounding method
can provide a useful property range estimate for the laminating plies.
2.2.4. Self-Consistency Method

The essence of this method is to place the RV back into the ply material whose effective
properties are yet to be determined; see, e.g. Hill [7]. The model then consists of the RV
with the prescribed fiber/matrix microstructure and the surrounding ply material having no
microstructure. Now, assuming that all interfaces are perfectly bonded and that the effective
properties of the homogenized ply are given apriori, the fiber/matrix stress and strain
distributions in the RV can be computed when appropriate traction or displacement agencies
are prescribed at the remote boundaries of the model. But, after the RV is homogenized, its
effective properties are then obtained; ane. _.e latter must match those assumed previously
for the ply material. If not, iterations of the process may hopefully lead to a converged
comparison.
This approach avoids arbitrary or improper specification of the surface agencies for the
RV, since the tractions on the outer surface of the R V are a part of the solutions under this
scheme.
Several variations of this approach are possible. For instance, there may exist a finite
layer of interphase substance between the fiber and the matrix; or, in some composite
systems, the RV may contain more than one fiber or even some forms of microflaws; see,
e.g. Pagano [8]. Although the general procedure in the self-consistent method remains
applicable for most of these possible variations in the representation of the RV, it invariably
causes complications in the mechanics analysis; frequently, numerical methods such as the
2-D and 3-D finite element procedures are employed.
2.2.5. Experimental Methods

It is to be noted that although micromechanics modeling of composites at the fiber/matrix
level such as discussed above is useful and often necessary, it is not the only way to obtain
the effective properties for the composite. For instance, the ply properties such as Cij and
aj in (1.1) or (1.2) are frequently measured directly from experiment by testing the ply
material in the bulk; see, e.g. [9]. In a sense, direct experiment avoids the many difficulties
and the unwarranted assumptions that had to be made in the micromechanics models; when
conducted properly, it can also provide the opportunity to examine the actual microstructure
and the response mechanisms therein under the applied load. On the other hand, in order to
obtain the desired properties in a appropriate manner, knowledge from the micromechanics
analysis can help in the design of the experiment and in the reduction of the recorded data.
Finally, it is noted that experiment can provide a certain strength property of the
composite while the micromechanics analysis often cannot. The problem of the strength of
90
composites is an ever more perplexing subject both experimetnally and analytically. This is
to be elaborated in various details in the reminder of this article.
Part II. Concepts and Criteria for Failures ill Composite Materials
Discussions will be focused on the basic assumptions made in formulating the various
conventional failure theories for composites. As most of the failure theories are based on
the effective stresses stemming from the abstracted composite, they do not and cannot
account for the true mechanisms of failure occuring in the composite microstructure. This,
then, results in severe inadequacy of the failure models. In order to facilitate discussions on
the specifics, the same class of laminates discussed in Part I will be used here again.
1. The Concept of Material Strength.
The concept of material strength is that a given material is endowed with a certain strength
which is an intrinsic property of its own. This concept has largely been derived from
engineering experience with metals, notably structural steels. Traditionally, these materials
are considered a homogeneous solid, as their microstructures are seen only at the grain
level, the scale of which measures in microns. Here, the assumption of homogeneity is
justifiable; and failure of the material is characterized by routine tests using the material in
the large bulk, without having to relate to the microstructure.
In this context, the failure modes observed at the global level can be broadly classified
in two extreme categories: the ductile and L._ brittle modes. The distinction between the two
modes can be readily made by testing the bulk material under simple tension as shown in
Fig.2.1.
Figure 2.1. Ductile and brittle material behaviors under simple tension.
In ductile failure, the stress-strain curve exhibits an noticeable departure from elasticity
when the applied stress reaches a critical value, cry. At this stress level, the material is said
to have yielded and a plastic defonnation process takes places. In some cases, the material
is still capable of carrying the applied load after yielding and can sustain large amount of
plastic deformation.
91
In brittle failure, on the other hand, the stress-strain curve breaks abruptly when the
applied stress reaches the critical value, crf. At this stress level, the material is said to have
failed by fracture without experiencing any plastic deformation.
Actually, the ductile and the brittle behaviors observed in tests at the global scale are a
manifestation of the deformation mechanisms occurring in the material microstructure. In
some metals, such as aluminum and mild steel, dislocations exist in the crystalline structure
as defects; one or more of the dislocations can move along a preferred slip plane under
some critical shear stress. This dislocation movement mechanism is the source of material
yielding and plastic deformation that are observed at the global scale [10]. In other solids,
such as ceramics and coarse-grained metallic alloys, crack-like defects are present on the
crystallographic planes or in the grain boundaries. These crack-like defects cause local
stress concentrations which can drive a crack-like defect to suddenly propagate into large
proportion when the local stress condition is critical. This crack propagation mechanism at
the local scale manifests itself as fracture of the material observed on the global scale.
In principal, the dislocation motions in the material microstructure can be modeled so as
to predict the yield strength cry of the material under various stress condition provided that
the microstructure, including the detail of its dislocations, of the material could be described
in modeling. This approach, however, is avoided in practice for obvious reasons. Instead,
tests on the bulk material such as shown in Fig.2.1 are designed and conducted to obtain
the yield strength cry directly and it is considered a constitutive property of the material.
As such, cry is often used in theoretical models predicting material failure if it fails in a
ductile mode. To give an example, the von· wises criterion states that
(2.1)
where crl, cr2 and cr3 are the principal stresses at the point where material failure is to be
predicted.
The von Mises criterion in (2.1) and others of similar forms imply yielding in the
material as the mechanism of failure; it works reasonably well with structural metals that
display ductility. The yield strength cry measured at the global scale is generally repeatable
under different stressing conditions even under compression; and it is usually independent
of the size or geometry of the specimen tested.
In brittle failures, it is also impractical to model the crack propagation mechanisms in
the microstructure for the same reasons. Tests conducted at the global scale, on the other
hand, do not provide a unique fracture strength crf. For instance, in tests such as shown in
Fig.2.1, the measured value for crf is known to vary with the specimen size, geometry,
surface preparation and other non-material factors introduced at the global level.
Furthermore, the fracture stress crf obtained in the test represent at best the tensile limiting
stress of the specimen tested; it cannot be taken as a constitutive material property, as it
could cry in ductile failures.
Unfortunately, the notion of material strength has proven to be so pervasive in practice
and deep-rooted in concept that it is sometimes applied indiscriminately, regardless the
brittle or ductile nature of the material failure. Moreover, the concept has been carried
forward to treat composites whose failure behaviors are mostly brittle in character. Even in
composites that do fail in ductile mode, the multi-leveled nature of their microstructures can
give rise to vastly different failure mechanisms at different dimensional levels and under
92
different loading conditions. Consequently, the conventional concept of material strength is

inadequate to treat a host of peculiar failure behaviors unseen in conventional materials.
2. Failure Criteria for Composites.
In this section, some failure models, that are widely used in treating the class of laminates
considered here, will be discussed. At the outset, it is pointed out that these models are
developed based on the traditional concept of strength. The emphasis here is to address the
limitations inherent in these models.
2.1. THE PLY STRENGTH CONSTANTS.
In the lamination theories discussed in Part I, the basic ply is homogenized; hence, it is
without a microstructure. The 2-D laminated plate model and the 3-D elasticity lamination
model are then used to provide the stresses in each ply; with the stress solutions, failure of
the ply is to be defined. Based on the material strength concept, the ply which is anisotropic
is then assumed to possess a set of strength constants. In principle, these constants can be
determined either by a mechanics model developed at the fiber-matrix level or by testing the
ply material in bulk at the global scale. The former approach is rarely followed in practice,
except for a rule-of-mixture estimate of the ply strength, see e.g. [II]. Instead, ply strength
constants are measured directly by testing the ply material in the bulk, see e.g. [9].
Within the confine of the 2-D laminated plate theory, each ply is in plane stress state;
then five independent ply failure modes are assumed: tensile and compressive failures in the
fiber direction; tensile and compressive ."ilures transverse to the fiber direction and
shearing failure in the ply plane.
Generally, the five ply strength constants are found by testing the ply material in the
bulk; thus
Xt - the limiting tensile stress crL in the fiber direction;
Xc - the limiting compressive stress crL in the fiber direction;
Yt - the limiting tensile stress crT transverse to the fiber direction;
Yc - the limiting compressive stress crT transverse to the fiber direction;
S - the limiting in-plane shear stress 'rLT of the ply in the L-T plane.
Experience with polymer matrix ply systems such as graphite-epoxy indicates it is not
always possible to determine these constants independently. Moreover, failure associated
with most of these constants are brittle in nature although some ductility has been observed
in inplane shear failures. Perhaps with the exception ofXt, test data exhibit unusually wide
scatter; and most are known to vary with the loading method, size and geometry of the test
specimen and the particular failure mode in which the test specimen is failed [9]. This is not
surprising; for the failure mechanisms of the ply material are profoundly influenced by its
microstructure, at least at the fiber-matrix interface scale.
Despite of this understanding, the five ply constants discussed above are routinely
determined along with the ply elastic constants and are widely used in failure criteria in the
same manner as the yielding stress cry is used in the von Mises criterion quoted in (2.1).
2.2. CRITERIA BASED ON POINT FAILURE CONCEPT
Ply failure criteria are developed in conjunction with laminate analysis; it is seldom applied
93
to the bulk ply material alone. In laminate failure analysis, it is theorized that one or more
plies may fail before the laminate; and failure of anyone ply is governed by the stress state
in that ply. Further, most theories assume that if the stresses at any point in the ply satisfy a
certain criterion, not only the material at that point fails the entire ply also fails. To
illustrate, the following specific criteria are cited for purpose of discussion on their
limitations.
2.2.1. Maximum Stress Criterion.

Let crL, crT and 'tLT be the inplane stresses at a point in the ply. The material at this point
and also the ply itself are deemed failed if anyone of these three stresses reaches or exceeds
the limiting value independently. Thus, the maximum stress criterion entails a set of five
independent conditions:
crL ~ Xt if crL is tensile IcrL I ~ Xc if crL is compressive

crT ~ Yt if crT is tensile IcrT I ~ Yc if crT is compressive
I'tLTI~ S (2.2)
Supposedly, this criterion also provides the failure mode of the ply. Namely, if crL~t
is satisfied alone, then the ply fails in tension; or if I'tLT I~ S is satisfied alone, then the ply
fails by inplane shearing. In this sense, failure of the ply is independently caused by one of
the three inplane stresses, whichever is pf("~~minant.
A variation of the maximum stress criterion is to compute the ply strains eL, eT and 'YLT
at the point from the ply stresses crL, crT and 'tLT. Then a set of five conditions, similar to
(2.2), can be written in terms of the ply strains; the corresponding limiting strains are found
in the tests along with the limiting stresses. This criterion is known as the maximum strain
criterion. It is similar to the maximum stress criterion, but can be different if the ply
material displays non-linearity before failure.
2.2.2. Interactive Stress Criteria

This assumes that the ply material at a point can fail even if none of the five conditions in
(2.2) is satisfied, especially when crL, crT and 'tLT are in the same order of magnitude. The
premise of this assumption stems from the stress interaction effect in the yielding criterion
of von Mises which is based on the stored distorsional energy. Extension of the von Mises
criterion to materials with anisotropic properties was first presented by Hill [12] and
applied to composite laminates by Azzi and Tsai [13]. Thus, for the ply under inplane
stresses, the so-called Tsai-Hill criterion:
(2.3)
where
X :: Xt if crL is tensile
:: Xc if crL is compressive
Y:: Yt if crT is tensile
:: Yc if crT is compressive.
94
It is to be noted that the basis of the Tasi-Hill criterion rests upon the assumption of
material yielding, an assumption not usually valid for ply failures in laminates even if the
failure exhibits some ductility. This later point will be further discussed later in section 3.
2.2.3. The Strength Potential Theory
In anisotropic materials, the distorsional and dilatational energies are not usually separable;
hence, the Tsai-Hill criterion for yielding is not purely associated with shear deformation. It
is, in fact, more of a mathematical than a physical proposal. Variations of the Tsai-Hill
proposal have led criteria of purely mathematical nature, without any association to specific
failure mode.
The Tsai-Wu criterion [14] assumes that an energy-like potential exists in the ply
material and that it is a function of stress alone. When, at a point in the ply, the stresses are
such that this potential function denoted by f(Oi) reaches a certain limiting value, the
material at that point is then assumed to have failed.
The specific form for the potential function f( 0i) is not known but is expandable into a
power series. Consistent with von Mises' or Hill's energy criterion, the expanded potential
function is truncated after the quadratic term:
f(oj) = Fo + Fi 0i + FijOiOj i,j = 1,2, .. 6 (2.4)
Note that fo =0 iff(oj} is positive definite and iff(Oi) is null at zero stress. Hence, the
general form for the Tsai-Wu criterion is:
Fi 0i + FijOiOj = 1 (2.5)
The criterion in (2.5) applies to plies under 2-D or 3-D stresses; and it is invariant to the
reference coordinate frame used. The constants Fi and Fij are to be determined from testing
the ply material in bulk. By virtue of material symmetry in the unidirectional ply, which is
assumed to be either transverse isotropic or orthotropic, there are 3 independent constants
in Fi and 9 in Fij if the ply is under 3-D stresses. The numbers reduce to 2 in Fi and 4 in Fij
if the ply is under 2-D inplane stresses.
In the 2-D stress case, five of the constants can be related to the ply strength constants
Xt, Xc, Yt, Yc and S obtained from testing the ply material in bulk:
Fl = 1IXt + 1IXc F2 =IIYt + IIYc Fil =-lI(XtXc)

F22 =-lI(YtYc) F66 =_lIS2 (2.6)
The sixth constant F12 cannot be determined uniquely by experiment, although it can
be bounded from above and below by an energy argument. Generally, the value ofF12 is
negligible and, if necessary, it can be estimated from inside the bounds.
If the ply is under 3-D stresses, the criterion is till valid only in principle. To determine
all the constants independently is not a routine matter; as far as it could be traced, it is not
clear whether or not all have been uniquely determined by testing the ply material in bilk.
In practice, criteria based on the strength potential concept are mostly limited to 2-D
stress cases.
At this point, it is to be noted that all of the above ply failure criteria depend on testing
the ply material in bulk. And, the ply strength constants, such as Xt, Xc, Yt, Yc and S
obtained in the 2-D case, are treated as a constitutive property of the ply. But, in fact, these
95
are at best the limiting stresses of the ply material tested in bulk. As it will be explained in
section 3, the failure constants obtained from testing the ply in bulk do not represent the
insitu ply strength when the ply is an integral part of the laminate.
2.3. PROGRESSIVE PLY FAILURES IN LAMINATES
As stated previously, the various ply failure criteria discussed above are used in failure
analysis of laminates. Failure of the laminate may thus be a process in which the weaker
ply fails first while the rest of the plies in laminate may still carry the load. The progressive
ply failure model assumes that when one ply fails, it no longer can carry the stress in it; so
it can be deleted from the lamination structure. Then, a new round of laminate analysis is
carried out to determine if any of the remaining plies will fail under the previously applied
load. If one ply does fail due to redistribution of the load, the failed ply is again deleted; if
no ply fails after load redistribution, the applied load is progressively increased until one
ply does fail. This procedure is repeated until the last ply of the laminate fails.
The above is known as the ply discount procedure [\5], since the failed plies are
deleted or discounted from the lamination structure in the successive rounds of laminate
failure analysis. Modifications of this procedure include treating the failed ply as having
yielded either in the elastic-perfect plastic sense [16] or in the elastic-strain hardening sense
[17]. In any case, the failed ply remains inside the laminate and continue to function as a
yielded ply.
The progressive ply failure concept discards the effects of ply-to-ply interaction at two
levels. First, in the 2-D laminated plate analysis, the stress matching conditions are not
satisfied so failure in the ply interface cam. __ be predicted; then, at the individual ply level,
the microstructure is lost in the ply homogenization process so interactions between the
microstructures of the neighboring plies cannot be recovered in the model as well.
3. Insitu Failures in Laminates
When inside the laminate, each ply is interacting with the neighboring plies through their
bonded interfaces. Thus, one ply is interacting with the other plies in order to preserve the
integrity of the lamination. The effects are reflected in several ply failure modes that are not
predictable by failure criteria based on ply strength.
To explain, consider two specific series of laminates that are laminated similarly and
loaded in the same manner. Specifically, the first is the [0190 n/0] series and the other is the
[25/-25/90 n/-25125] series, with n = 1,2,3, . When loaded under uniaxial tension as shown
on the left in Fig.2.2, the laminates in the first series developed intralaminar cracking,
while the laminates in the second series developed interlaminar cracking. These two failure
modes are rather unique in laminates; both do not occur in homogeneous materials.
As it will be explained next, failure mode in either series of laminates should be the
same if any of the ply failure criteria discussed previously is used for a prediction. But, in
fact, the failure mode developed in the first series is fundamentally different from that
developed in the second series; neither could be predicted by the ply failure criteria. Details
in these different failure modes are explained next in terms of their physical character and in
terms of the mechanisms that cause the respective failure mode.
3.1 INTRALAMINAR FAILURE
Suppose that the above mentioned two series of laminates are made of a polymer-matrix ply
system, specifically, graphite-epoxy and the width and length of the test specimens are
96
orders of magnitude greater than their thickness. Under the applied tensile load, see
Fig.2.2, each ply of the laminates develops inplane stresses 0"1, 0"2, and 0"6 in the laminate
interior and fully 3-D stresses near the laminate free edges. The former is readily computed
by the 2-D laminated plate model, while the latter is computed by the 3-D lamination theory
(see discussions in Part I).
Figure 2.2. 2-D ply stresses in [0/90n/OJ laminates under uniaxial tension
According to calculations for the [0/90n/0] series [18], the 3-D stresses in the laminate
free edge region are not predominant enough to cause initial failure, as do the 2-D stresses
in the laminate interior. As shown schematically in Fig.2.2, the stresses in the outer 00
layers are the fiber-wise stress (O"L)O and the transverse stress (O"T)0; both are tensile. The
stresses in the middle layer of n-900 plies are the fiber wise stress (O"L)90 and the transverse
tensile stress (O"T)90; the former is compressive while the latter is tensile.
In a laboratory test [19], it is found that the first failure in this laminate series occur in
the middle layer of n-900 plies; the mode of failure is in the form of a transverse crack, as
display in Fig. 2.3. On the global level, the crack is seen to occur suddenly and it severs
the 900 layer across the width of the specimen. The crack, however, is confined inside the
90 0 layer due to the constraining effect provided by the outside 00 plies. Hence, the
laminate remains intact under the applied tension.
Clearly, the transverse crack is precipitated by the tensile stresses (O"T)90; and perhaps it
is also assisted by the compressive (O"L)90 in that layer. According to test data reported in
[19], the insitu stress (O"T)90 at the first transverse crack is not a constant; it, in fact varies
with the thickness of the 900 layer, or the value of n. Specifically, the critical in situ stress
(O"T)90 decreases as n increases, as shown schematically on the upper right of Fig.2.3. The
97
asymptotic value of (<JT)90 at large n, however, tends to approach the tensile transverse
strength of the ply, Yt, tested with the ply material in bulk.
~-+---+---r __ ~ n
2 3
90~ 11.~.~.I.I.I.I.I
layer
Figure 2.3. Transverse cracks in {OI90n/Ojlaminates; n = 1,2,3.
This result is in conflict with the ply failllre criteria developed previously. For if the ply
material has a transverse tensile strength, Yt, then the critical insitu stress (<JT)90 at failure
would have been the same for all the laminates in this series; or, the critical (<JT)90 would
have been independent of the 900 layer thickness, the value of n.
The ply thickness dependent behavior of intralaminar cracking is a reflection of the
constraining effect of the stiff outer 0 0 plies on the freedom of deformation of the 900 layer
through their bonded interfaces. This constraining effect will be further explained later in
Part IV.
In addition to the thickness dependent behavior, test results have also shown that more
transverse cracks have formed in the 900 layer under increased laminate loading [19]. This
multiple cracking behavior at higher insitu ply stresses is a reflection of the 90 0 ply
microstructure which is interacting with the neighboring 00 plies through their interfaces.
All these interaction effects are not accounted for in the ply failure criteria. However,
the ply failure criteria do predict the first transverse crack at a much lower laminate loading,
which is on the safe side. Models that treat transverse cracking as a load-dependent process
are discussed in Part V.
Finally, it is remarked that, although intralaminar cracking is not usually suspected in
plies where the fibers are carrying the major load, ply-to-ply interactions can induce a
secondary tensile stress transverse to the fibers in such plies. In many laminates made of
epoxy based ply systems, this secondary transverse tensile stress can become predominant
in causing transverse cracking.
3.2 INTERLAMINAR FAILURE
As in the [0/90 n/0] series discussed above, transverse cracking of middle layer of n-90 0
plies in the [25/-25/90 n/-25/25] laminate series can also occur when loaded under uniaxial
98
tension. In this case, however, the 3-D stresses near the laminate free edge become more
dominant than the 2-D inplane stresses in the laminate interior. In particular, a localized
tensile interlaminar normal stress (O'z)90 exists in the thickness of the middle 900 layer near
the free edge; its magnitude rises sharply toward the free edge; see Fig.2.4.
/
o
interior free ed e __
Figure 2.4. Distribution of interlaminar normal stress in [45/-45/902/45/45] laminate
Thus, the first failure mode in the laminate under the applied load is a plane crack
parallel to the ply interface and along the length of the specimen, as shown in Fig.2.5. This
plane crack, known as free edge delamination, is initiated mainly by the tensile interlaminar
normal stress (O'z)90 near the laminate free edges.
"--r--I-t-~ n
2 3
90~ layer _-t:=::::;;~
Figure 2.5 Free edge delamination in [25/-25/90n/-25/25] laminates; n=l,2,3.
Curiously, the critical insitu stress that causes the onset of free edge delamination is
again varying with the thickness of the 900 layer, or the value of n [20]. Specifically, the
critical (O'z)90 in the 900 layer at the onset of delamination is found to decrease with the
increase of the thickness of the 90 0 layer, or the value of n; see display on the upper right
99
in Fig.2.5. The reason for this thickness dependence is rather subtle; but it can be attributed
to the variation in the laminate microstructure when the value of n changes.
Again, use of the ply failure criteria cannot predict the critical insitu stress (<Jz)90, nor
can it explain the thickness dependent nature of the critical onset stresses. Models for
treating free edge delamination are discussed in Part V.
Actually, in the [O/90 n/0] series, an interlaminar normal stress (<JZ)90 is also present
near the free edge; but its magnitude is much smaller than the tensile transverse stress
(<JT)90 in the laminate interior. This is due to the details in the lamination structure and the
ply-to-ply interactions thereof. In general, free edge stresses develop in laminates with the
neighboring plies having a dramatic mismatch in their compliance. In the [O/90n/0] series,
the outer 00 ply has a Poison ratio v 12 of about 0.3 compared to about 0.1 for the middle
900 layer. In the [25/-25/90 n/-25/25] series, on the other hand, Vl2 for the outer ±250 plies
as an integrated layer is nearly 0.9; hence the pronounced magnitude of (<J z)90 at the free
edge.
Though the failure modes in the two laminate series differ fundamentally from each
other, they differ only slightly in the geometric details in their respective lamination
structures.
Finally, it is remarked that <fpl"mination growth is more damaging to the laminates than
transverse cracking; without a crack arrest mechanism, delamination can grow in size under
increased laminate loading and render the laminate useless structurally. As a failure mode,
delamination can occur near any stress ri,--~ in the laminate in addition to the free edges;
smaller holes, cut-outs, surface scratches are examples of such stress risers. Even the
formation of transverse cracks can induce localized delamination at sites of the transverse
cracks; these localized delaminations can also growth into global proportion if without any
arresting mechanism.
3.3. THE EVOLVING MICROSTRUCTURE
A closer examination of the transverse cracking and delamination modes reveals readily that
these are two competing failure modes exists simultaneously at a material point under 3-D
stresses. If one of the stresses normal to the fibers or to the ply plane is the largest and is in
tension, then a transverse crack or a delamination can occur, respectively. If transverse
cracking occurs first, each such crack can be a local stress risers which may in turn induce
localized delamination. On the other hand, if a delamination crack occurs first, stress
concentration can develop at the crack tip region, which in turn can precipitate transverse
cracking; see [21].
Continuous formation of these competing damage modes makes the microstructure in
the laminates to evolve with the applied load. In this sense, the laminate microstructure is
not stationary; it is, in a sense, a load-history dependent property of the laminate in service.
This realization is important in dealing with failures in composites, especially when long
term effects such as fatigue failures are to be analyzed.
Part III. Multiple Cracking and Microstructure Evolution in Composites
Part III discusses two most important elements in the microstructure that influence the
mechanisms of failure: one is the random distribution of flaws or defects and the other the
100
network for internal load-transfer; interaction between the two elements can result in a host
of peculiar failure modes not seen in conventional structural materials.
1. Random Flaws in Composite Microstructures
Fiber-reinforced composites are a multi-phased material with multi-level microstructures.

Flaws in the constituent materials and in their interfaces are introduced in random, albeit
unintentionally, at every stage of material processing, fabrication and handling. While the
existence of random flaws in composites is generally acknowledged, few failure models
actually include the presence of material flaws. Omission of flaws in any failure model
would make it devoid of the ability to capture the physical failure mechanism. This point
can be readily seen in the failure modes in a material when it is loaded in bulk and when it
is loaded as an integral part in composite.
1.1. FIBER FLAWS AND TENSll..E STRENGTH
In the legendary work of Griffith in 1921 [22], inherent material flaws are for the first time
hypothesized as the source for the rupture of solids. Supporting experiments have shown
that the mean tensile strength of a glass rod decreases with the increase of diameter, if the
length is kept constant; the strength decreases with the increase of length, if the diameter is
kept constant. The size and length effect on tensile strength is evident that minute flaws are
present on the surface andlor in the interior of the rods, since larger flaws are found
statistically in material of larger volume. The material flaw hypothesis is foundation for the
Griffith theory of rupture [23].
In this context, it is natural to appreciate the importance of material flaws on failure of
composites. To illustrate, consider a fiber of length Lo and let it be assumed that flaws on
the fiber surface andlor in the interior are present and these are distributed randomly along
the fiber length. Upon tensile loading of the fiber, stress concentration develops near each
of the flaws; when the local stress condition at one of the flaws becomes most critical, the
flaw will suddenly propagate and rupture the fiber in two halves.
The applied tensile stress on the fiber at the rupture event is commonly taken as the
tensile strength of the fiber. In truth, however, this strength is associated only with the
most critical of the flaws in the fiber. Hence, it is merely the lowest possible strength of the
fiber. In fact, if the two halves of the ruptured fiber are each loaded again, the tensile
strength of each would be higher than obtained first for the original fiber of length Lo. The
reason is that the flaws remaining in the ruptured fiber are all less critical than that caused
rupture of the original fiber of length Lo.
It may then be inferred that the fiber of length Lo possesses a flaw distribution and a
tensile strength distribution along its length; there is a one-to-one correspondence between
the flaw and the tensile strength distributions. Furthermore, if the flaw distribution is an
inherent property of the fiber, so is the fiber strength distribution.
1.2. FIBER FAll..URES WHEN EMBEDDED IN MATRIX
When the fiber is embedded inside another material, the tensile failure behavior of the fiber
is influenced by a host of factors that are external to the fiber. Consequently, the one-to-one
correspondence between the fiber flaw and fiber tensile strength distributions is no longer
available. Moreover, the fiber may now be capable of multiple ruptures at different insitu
stress levels.
101
Consider the experiment by Rosen [24] in 1964. A simple composite model of a single
fiber embedded in matrix is used to show multiple fractures in the fiber. The model is
loaded uniformly by a remote tensile load, as shown schematically in Fig.3.1; a series of
fiber fractures occurs in succession inside the matrix under increasing applied tension. The
multiple fiber failure mode is possible due to the built-in load-transfer mechanism in the
composite model. Namely, let the first fiber fracture be caused by the propagation of the
most critical flaw in the fiber; then the load carried by the fiber is taken by the surrounding
matrix material. Owing to the fiber-matrix interface bonding, this same load is transferred
back to the fiber again by an interface shear mechanism.
+++++++ # fiber fractures
insitu fiber stress

, " , "~ interface shear
""'"
Figure 3.1. The Rosen model of a single fiber embedded in matrix
To fully develop the load-transfer mechanism, a finite "shear-lag" zone, 0, is created;

so the fiber can be reloaded again in tension beyond this shear-lag zone. Since the flaws in
the broken fiber segments are now all less critical than the flaw at the first fracture, each of
the fiber segments is expected to break again but at a higher fiber stress than before. Thus,
the fiber can be fractured successively at higher and higher fiber stresses. On the right in
Fig.3.! is a schematic plot of the number of fiber fractures versus the insitu fiber stress.
Note that the number of fiber fractures approaches a limiting value after a certain fiber
stress. This is due directly to the size of the shear-lag zone, or the value of O.
Analysis of the Rosen model shows, however, that the value of 0 is a function of the
fiber diameter, the matrix material stiffness relative to the fiber's and the bonding strength
of the fiber-matrix interface. These are all factors external to the fiber; hence, the insitu fiber
fracture stress cannot be treated as an inherent property of the fiber itself.
The Rosen model provides a valuable insight into the relationships between the inherent
material flaws and the load-transfer mechanisms in composites; the conventional view of
material strength may operate only when a material is loaded in the bilk. The inherent flaws
present in the fiber, matrix and the interfaces are all part of the composite microstructure.
Together, they determine a unique sequence of failures when the composite is subjected to
a given loading condition. If this failure sequence is to be described by a failure model, the
required input must include, at least, the essence of the flaw distribution in the composite
microstructure.
102
Finally, as flaws fracture and local load-transfers occur in the composite under loading,
the microstructure evolves with the failure process; and as the microstructure evolves, the
failure mechanisms also change.
1.3. FLAWS IN UNIDIRECTIONAL PLY
The laminate series [0/90 n/0] under uniaxial tension considered in Part II, is somewhat
analogous to the Rosen model discussed above. The (90 0 )n layer in the middle of the
laminate is analogous to the single fiber in the Rosen model; variation of n is analogous to
variation of the fiber diameter. As discussed in Part II, the (90 0 )n layer develops multiple
transverse cracks when the laminate is under uniaxial tension; the nature of the transverse
cracking is analogous to the multiple fiber fractures in the Rosen model.
Thus, one may draw from the parallel that the (900 )n layer possesses a distribution of
random flaws; and the load-transfer mechanism by which these flaws become transverse
cracks is similar to that in the Rosen model. Namely, the load-transfer mechanism here is
provided by the bonding of the ply-to-ply interface and the load-carrying capacity of the 00
plies.
Similarly, if transverse cracking in the laminates is to be quantitatively modeled, input
to the model must include the flaw distribution in the (900 )n layer. Given the material flaw
hypothesis, this represents an iuuJrmountable challenge as the true identity of the flaws
would seem to be never and ever practically unavailable.
However, in some cases, the qualitative nature of the flaws may be inferred from a
physical reasoning. Specifically, the flaws mat cause transverse cracks in the (900 )n layer
when the laminate is under uniaxial tension must be those crack-like defects which exist
parallel to the fiber-matrix interfaces in the (90 0 )n layer; each of these crack-like defects is
subjected to a tensile normal stress operating in that layer. When the insitu tensile stress
near one of the flaws becomes most critical, the flaw will be propagated into a transverse
crack. This reasoning significantly simplifies the tasks in identifying the material flaws.
In this case, only the crack-like defects that are parallel to the fiber-matrix interfaces in
the (90 0 )n layer need to be identified, or more simply, represented in some fashion so that
they effect the transverse cracking failure mode observed at the laminate level. This then
same leads to the heuristic but useful concept of effective flaws first applied by Wang in
analyzing failures in composites [24].
2. The Concept of Effective Flaws
The concept of effective flaws is but a necessary alternative to grossly represent the crack-
like defects and their effects in causing failure in composites [25]. In a sense, the effective
flaws are like the effective elastic constants associated with ply elasticity. The effective
flaws are abstracted quantities, but are endowed with a certain unique property that reflect
the same effects on failure as the real flaws.
At this point, it is useful to recall that the mechanics models for composite laminates are
all based on ply elasticity, either the 2-D or the 3-D versions. There, the microstructure of
the basic ply is homogenized in order to avoid the impossible and perhaps unnecessary task
of including the ply microstructure in the lamination models. But then, the models so
formulated are devoid of the ability to predict failures inside the laminates based on failure
mechanisms. There is a need to restore the ply microstructure back into the laminate again,
including the inherent flaws. But to do so contradicts the very premise of ply elasticity in
the first place.
103
The dilemma may be partially circumvented by the concept of effective flaws. Namely,
instead of seeking the true identity of the flaws, a set of virtual flaws is assumed for the ply
material, provided that the virtual flaws can effect the same failure mode in the predictive
model, the model being formulated consistent with ply elasticity.
In this context, it is important to note that the effective elastic constants obtained by
testing the ply material in the bulk implicitly reflect the effect of the flaws in the ply; the
introduction of the effective flaws to the ply does not affect the values of the elastic ply
properties. In fact, the effective flaws become physical only at the instance when the flaw is
to become a transverse crack.
As the transverse crack is arrested by the outer laminating plies, the crack simply
becomes a new addition to the laminate microstructure; meanwhile, ply elasticity remains
operational.
2.1. THE EFFECTIVE INTRALAMINAR FLAWS
Considered the [0/9010] laminate under uniaxial tension as a specific example. Here, the
single 90 0 ply is said to contain a random distribution of crack-like defects that lie parallel
to the fibers, known as intralaminar flaws. The picture on the left in Fig.3.2 is a schematic
representation of the flaws distributed on a typical plane. When the 900 ply is an integral
part of the laminate under tension as shown, the flaws on that plane can act like small
cracks which may suddenly propagate and coalesce to start the formation of a transverse
crack. It is the at this incipient coalescence of the flaws that the identity of the effective flaw
on that plane is defined.
The picture on the right in Fig.3.2 is a SImplified representation of this abstractive flaw
on the same plane of the ply. In this case, the flaw is assumed to span the entire width of
the ply but is within the thickness of the ply; the size of the flaw is thus measured in the ply
thickness direction and its propagation is one-dimensional in the same direction. Of course,
any representation of the effective flaw is not unique; the only requirement is that it can
effect the same kind of failure mode in the simulated model. In this specific case, the size of
the effective flaw shown on the right in Fig.3.2 must be such that it propagates into a
transverse crack in the simulation model under the same ply stress condition as that in the
picture on the left in Fig.3.2 where flaw coalescence occurs and starts the formation of a
transverse crack at the same ply location in real test conditions ..
Intralaminar flaws on a plane The effective flaw on the same plane
Fig.3.2. Representations of the intralaminar flaws in unidirectional ply

104
The simplified version of the effective intralaminar flaws shown at the right in FigJ.2
can be characterized by the probability distribution functions flea) and f2(x), where a is the
size of the effect flaw on the plane located at the distance x along the ply. Both a and x are
independent random variables; and if the ply thickness is t then O<a<t; if the ply length in
the direction of loading is a unity, then and O<x<l. The meanings of these two random
variables are illustrated in Fig.3.3.
/---------,~ -"'- Ply thickness t
Fig. 3.3. Size and location distributions \ the simplified virtual flaw.
At this point, it remains to determine tJ.. o distribution functions fI(a) and f2(x) for the
90 0 ply. To that end, a simulation model is needed which takes the effective flaw
distributions fl (a) and f2(x), and mimics the transverse cracking events in a selected
laminate situation, for example, the [0/90/0] laminate loaded under uniaxial tension. A
concurrent experiment using [0/90/0] laminates is then conducted in which transverse
cracks are recorded as a function of the applied laminate tension. Correlation between the
simulated and test results provides the final identification of the flaw distribution functions
f1(a) and f2(x). The details of the latter are deferred to Part V.
2.2. VOLUME METRIC EFFECT ON EFFECTIVE FLAWS
Once the effective flaw distribution functions for the single 90 0 ply is described, the
effective flaw distribution in grouped 900 plies, (900 )n, can be estimated by a volumetric
rule.
To be specific, let the flaw size and location variables a and x be discrete valued;
assume that there are M effective flaws in a unit length of the 900 ply. Then, the flaw size
distribution is represented by the set {ail, with O<ai<t for all i= 1, M; similarly, the
corresponding location distribution is represented by {Xi}, with O<xi<l for all i=l, M. A
set ofM values is then generated randomly in the interval (0,1) with uniform probability;
these are assigned to be the location distribution {Xi}, i=l, M. The size distribution (ail is
assumed to follow a certain known form of distribution function, for example the normal,
the Weibull, etc. Without a specific reason, let the Weibull function be used here. Then,
fiCa) can be expressed as
f1(a) = (a/~) (aI~)a-1 exp[-(aI~)a] (3.1)
105
and the corresponding cumulative function is expressed by
(3.2)
where the parameters a and ~ are yet to be determined for the 900 ply.
For the discrete values of {ai}, each ai is determines as follows: substitute a by ai in
(3.2) and equate FI(ai) to the ith value of the M random numbers generated earlier in the
interval (0,1), which is also Xi in {xi}:
F1(ai) = Xi (3.3)
Assume a value for a (>0) and another for ~ (0 <~<t); then ai is computed from (3.3).
Now, in the [0190 n/0] laminates with n >1, the effective flaws in the (900 )n layer are
represented by the size and location distributions {an,i} and {xn,i}, respectively, with i=l,
M. The values of {an,i} and {xn,i} may be inferred from that of the single 900 ply by a
modified Weibull volumetric rule [26]:
xn,i = Xi ~ll,i = ai(n)21a (3.4)
Thus, for the [0190 n/0] laminate series with n>l, the effective flaw distribution for any
n>l can be defined explicitly, pending the selection of the parameters a, pand M from the
basic 90 0 ply.
It is to be noted that the determined effective flaw distributions fICa) and f2(x), or {ai}
and {Xi} are not unique; other functional forms may also work well in the simulation and
experiment correlation. The point here is that only three parameters (a, ~ and M) need to be
defined; and the effective flaws in a class of laminates are then defined.
2.3. INTERLAMINAR FLAWS
Laminates fail by delamination when interlaminar stresses are present and become critical.
These stresses exist in the ply-to-ply interfaces or on planes parallel to the interfaces; the 3-
D ply elasticity lamination model may be used to compute these stresses in specific laminate
situations. Failure models based on the interlaminar stresses alone do not seem to provide a
general criterion for delamination, which involves a characteristic onset and subsequent
growth. Failure models for free edge delamination onset and growth have been developed
based on fracture propagation [27]; and it suffices to say that most of these models are once
again built upon the concept of effective flaws.
At first, it is given within the material flaw hypothesis that inherent flaws are present in
the laminate ply interfaces and planes that are parallel to the interface. These are planar
crack-like defects normal to the laminate thickness direction, known as interlaminar flaws
such as schematically shown on the left in Fig.3.4. Clearly, these flaws are random in size
and shape and distribute throughout the laminate interior and near its free edges.
In the laminate interior, interlaminar stresses are not generally present unless stress
risers such as a small hole, a cutout, etc. are also present. On the free edges, however,
interlaminar stresses not only exist but also rise with a high gradient toward the free edge.
In particular, if the interlaminar normal stress crz is tensile, the flaws near the free edge can
106
propagate and coalesce into planar crack of a global proportion, thus defining the onset of
free edge delamination.
Interlaminar flaws on ply interface The effective interlaminar flaw on free edge
Figure 3.4 Representation of interlaminar flaws and effective edge flaw
In order to fonnulate a predictive model, the concept of effective flaws is then invoked.
In the case of free edge delamination, the distributed edge flaws are represented by a single
effective edge flaw in the intended ply interface as displayed on the right in Fig.3.4. Note
that this effective flaw extends uniformly along the laminate free edge; its size is
characterized by the random variable a. Jo, the effective flaw can turn itself into a
delamination of size a when the critical condition near the flaw is critical; this then defines
the onset of delamination. Upon onset, the delamination then propagates one-dimensionally
into the interior of the laminate if the applied load can sustain such propagation.
Clearly, the size a of the effective edge flaw must be such that its propagation predicted
by the simulation model correlates with the onset of delamination found in a concurrent
experiment.
Again, introduction of the effective interlaminar flaw into any ply interface in the
laminate does not affect the operation of the 3-D ply elasticity lamination model. The
identity of the flaw appears only in the simulation model in order to provide a criterion for
the onset of delamination. Once delamination occurs, it is then real and physical; it will be
treated as newly added part of the laminate microstructure. Ply elasticity continues to be
operative in that context.
Part IV Fracture Mechanics and Analysis Approaches for Composites
In fibrous composites, failure almost always begins with some local flaws which propagate
and/or coalesce under a certain critical loading condition. The exact modes of failure are
influenced, at least in part, by the load-transfer mechanisms inherent in the composite
microstructure. The methods of fracture mechanics are often found useful in simulating
such failure activities at the initial stage. In Part IV, the basic concepts of fracture
mechanics and the techniques that are applicable to fiber composites are presented and
discussed.
1. Fracture As A Material Failure Criterion.
Without a flaw, solid material can be fractured by breaking its atomic bonds in tension
107
[28]. FigA.1 depicts the two atoms (in a solid material) in equilibrium at the distance Xo;
force or stress is needed to alter the equilibrium state. Namely, a compressive stress is
needed to push the two atoms closer and a tensile stress is needed to pull them apart. This
stress-stretch relationship is shown on the right in FigA.I. Thus, elasticity of the solid is
provided by the atomic bonds in the neighborhood x/xo"" I; loss of elasticity occurs when
the tensile stress reaches and over the maximum, (Jm; at which point, the atomic bonds
break and the solid fractures. The shaded area under the stress-stretch curve is the strain
enrage stored in the deformed solid; this energy is absorbed as surface energy by the newly
created surfaces in the fractured solid.
Figure 4.1. The stress-strain relation in breaking the atomic bond.
The shaded area A under stress-stretch curve is approximated by:
(4.1)
while, for solids of regular atomic packing, the surface energy density y is approximated by
y= EXo/40 (4.2)
Equating (4.1) and (4.2), the maximum stress (Jm needed to break the atomic bond is:
(Jm ",El6 (4.3)
The expression in (4.3) is considered the theoretical tensile strength of a crystalline

solid without a flaw. But, most brittle materials fracture at tensile stresses less than 1110 of
the theoretical value. This, of course, is due to the presence of flaws.
1.1. THE ENERGY CRITERIA FOR FRACTURE.
The original Griffith model [23] assumes that a crack-like flaw already exists in a 2D
atomic lattice; see illustration in FigA.2. The pre-existing crack has a linear size a in the 2D
plane; under the applied tensile stress, the bond (the darkened line) between the two atoms
108
at the crack-tip will be the next to break. As this bond breaks, the crack length advances
stably with a small increment L'la; so a new surface 2L'la is created. At the same time, the
stress and strain fields near the crack-tip have changed; work (W) is done by the externally
applied stress and strain energy (U) is lost due to crack extension; and the total loss of
energy is absorbed by the newly created surfaces. Thus, a balance of energy yields
L'lW - L'lU = (2L'la)y (4.4)
11 '''''k-1ike fl,w ___
.....
.
aj • • • .....
- ..........
~
cr ~
• •
~
-
•• • ..... cr
T- • • • •
- .....
~
~
• • • • •
~L'la.
• • • • •
Figure 4.2. The Griffith model for rupture of atomic bonds
Equation (4.4) is the fundamental criterion governing the rupture, or more precisely
cleavage, of crystalline solids that contain a pre-existing crack-like flaw of known size. It
remains to express L'lW and L'lU in terms of the applied stress cr and the crack size a in order
to complete the execution of (4.4).
When generalized for homogeneous engineering materials, the Griffith model in
Fig.4.2 is replaced by an infinite plate under the uniform tensile plane stress cr as shown in
Fig.4.3. The pre-existing crack is then a small line crack of size 2a lying normal to the
applied tension cr. The critical condition for crack extension is governed by the same
criterion in (4.4) where L'la is replaced by the infinitesimal crack extension oa:
o(W - U)/oa = 2y (4.5)

109
Figure 4.3. The Griffith model: a small crack in an infinite plate.
To execute (4.5), it is necessary to express Wand U in terms of the applied tension a

and the crack size a. This, however, requires the complete solution of the stress, strain
and/or displacement fields for the plate under load; and, it is not always straight forward.
For a linearly elastic and isotropic infinite plate, these field solutions are obtained from the
plane stress theory of elasticity by Wester_ rrd [29]. In particular, along the crack line, or
the x-axis, the normal stress O'y and the vertical displacement v of the crack surface are
given by:
O'y = mu[-,i(x2 - a2)] for x > a; v = 2a,1(x 2 - a2)/E for x < a (4.6)
It is then argued that, for an infinite plate, the far-field displacements are essentially
unchanged during the infinitesimal crack extension da; so there is no external work done,
dW=O; but there is a loss of the stored strain energy dU associated with the crack extension
da.
By an inverse crack extension, or crack closure, the Westergaard solutions in (4.6) can
be used to compute the work done by closing the crack completely; this work is equivalent
to the cumulative strain energy loss U as the crack extends from the size a=O to the current
size of 2a. Thus,
a
-U = 2 J (0'/2) vex) dx (4.7)
Substituting (4.6) into (4.7) yields
(4.8)
and from (4.5), the critical stress for stable crack extension is then given by:
O'cr = -,i(2Eyl1ta) (4.9)

110
1.2. THE ENERGY RELEASE RATE.
It is noted that the result in (4.9) is based on the assumption that the crack extension in the
plate involves breaking of atomic bonds, or cleavage in the crystalline structure; so the loss
of strain energy is converted totally to surface energy during the crack extension. In that
context, the term d(W -U)/da in (4.5) is the driving force, while the quantity 2y denotes the
material resistance to crack extension.
In engineering materials, such as metals, metallic alloys and fibrous composites, brittle
crack extension often follow weakened grain boundaries, bi-material interfaces or paths of
least resistance. There, the energy lost during crack extension is absorbed by mechanisms
other than breaking of the atomic bonds. In that case, following a similar argument
articulated by Orowan [30], the surface energy yin (4.5) can be replaced by an effective
surface energy: Yeff.
Conventionally, the left-hand side of (4.5) is termed the energy release rate, denoted
by
G(O",a) == d(W - U)/da (4.10)
while the right-hand side of (4.5) is called the critical energy release rate, denoted by
Gc == 2Yeff (4.11)
Note that, owing to heuristic nature of Yeff, G c can only be an experiment-adjusted

material quantity.
1.3. THE STRESS INTENSITY FACTOR.
In the Griffith cracked plate model, Fig.4.3, the tensile stress O"y along the crack-line is
primarily responsible for crack opening. Note that O"y given in (4.6) reduces to the far-field
tensile stress 0" away from the crack-tip; but it is unbounded near the crack-tip. Or,
mathematically, the solution for O"y contains a singularity at x==a. To evaluate the nature of
the singularity, let r==(x-a) be the distance measured from the crack tip; then the stress O"y
and the crack surface displacement v near the crack tip are obtained from (4.6); in the limIt
of x ~a one obtains
O"y (x) == KI'I/(2m) for r >0 or x>a
vex) == 2(KIE)'I/(-r/21t) for -a<r<O (4.12)
where
K == O"'I/(1ta) (4.13)
The quantity K is known as the stress intensity factor; it is a scaling factor for the stress
which is singular-valued at r==0, with the singularity strength of 1/'l/r.
Note that K contains exclusively the applied stress 0" and the crack size a; though it does
not have a physically meaningful dimension, it is nevertheless a quantity unique to the
III
mathematical characteristics of the crack-tip stress field. This point will be elaborated later
in section 1.2.
104. RELATION BETWEEN G AND K.
The energy release rate expressed in (4.10) is a unique quantity characterizing the exchange
of energy at the crack-tip during crack extension. As in K, G is also an explicit function of
the applied stress cr and the crack size a. By means of the crack-closure technique outline
by Irwin [31], (4.10) can be rewritten in the form:
a+ila
Gila=il(W-U)= 2/ (cr /2)v(x-ila)dx (4.14)
a y
In (4.14), ila is infinitesimal; so the near-field cry and v expressed in (4.12) are
substituted into (4.14). Carrying out the integral, it then yields the relation between G and
K:
(4.15)
Thus, K2/E also represents the crack extension driving force. If G=Gc is a material
criterion for crack extension, so should K =Kc; hence, by deduction, the term
Kc ="(EGc) (4.16)
can be regarded a material quantity in the same context as Gc .
At this point, it is perhaps appropriate to remarked that fracture is basically a tensile

brittle failure phenomenon; it involves extension of a pre-existing crack. When the crack
extends, new surfaces are created; so an exchange of energy takes place during crack
extension. This is the basis for the energy criterion.
The difficulty is in the execution of the terms in the criterion; usually, it requires a
complete solution of the stress and strain fields of solids which may contain one or more
propagating cracks. The difficulty becomes increasingly acute if the material body has a
complex boundary, or is part of the material microstructure. Yet. the basic argument for
fracture remains independent of these difficulties.
Much effort in the past are devoted in developing techniques to compute either G or K,
given a fracture situation. Some of the techniques are useful in simulating brittle failures in
composites; these are briefly discussed in the next section.
2. Techniques in Linear Elastic Fracture Mechanics.

The linear elastic fracture mechanics is an application of the plane elasticity theory to a solid
body that contains a propagating crack. It focuses at first on the stress state near the crack-
tip region. FigAA shows the general layout of the problem, where the elastic body is
under plane stress or plane strain; only the region containing the pre-existing crack is
considered.
\12
mode-I
y
mode-II
mode-III
Figure 4.4. Three possible crack extension modes.

Let the body of material be homogeneous and isotropic and the crack extends self-
similarly in three possible modes; see Fig.4.4. Mode-I is a tensile opening; mode-II and
mode-III involve sliding shear and anti-plane shear, respectively.
The plane stresses of the 2D body are derivable from the stress function 'l'(x,y) which
is governed by the bi-harmonic equation:
(4.17)
A general solution for 'l'(x,y) may be found by the method of complex functions [32].
In terms of the local coordinates (r,e) shown in Fig.4.4, the solution for 'l'(x,y) provides
the in-plane stresses of the general form:
<Ji oerA fi(e) i= 1,2,6. (4.18)
where the exponent A is an eigen value and fi(e) is the associated eigen function; both are
determined by requiring the stress solution (4.18) to satisfy only the stress-free condition
of the crack surfaces:
on e=±7tl2 (4.19)
which yields the admissible eigen values
A=-1/2, 0, nl2, ... n =1,2,3, ... (4.20)

113
Thus, for each value of A, there is an eigen stress and the associated eigen function f(6)
in (4.18). Ideally, the full field stress solution is to be obtained as a linear combination of
the infinite many eigen stresses; by requiring the full field stresses to satisfy the remote
plate boundary conditions, including the applied load, it would yield the coefficients in the
linear combinations of the stresses. In particular, the coefficient associated with A=-1/2
would be the desired crack-tip stress intensity factor K. Bur, this proposition is difficult to
carry out for practical problems and is often avoided.
2.1. STRESS SINGULARITY.
As it can be seen from (4.18), the eigen stress associated with ')..=-1/2 is unbounded at the
crack-tip. Take for instance 0'2 or O'y along the x-axis (6=0); f2(8) = 1 due to symmetry
about the x-axis. Hence
O'y (r) <X 1I...Jr (4.21)
Compare (4.21) to O'y(r) in (4.12); they differ only by a factor of KJ...J(21t). If the full
stress field had been found for the Griffith problem, the dominant term in O'y(r)
corresponding to ')..=-112 would oe exactly the same as in (4.12).
The significance of the singular solution in the general form (4.18) is that all the crack-
tip stresses are dominated by the eigen s. JS corresponding to ')..=-112. In fact, the near-
field stresses and displacements for all the three possible crack extension modes are
expressible in the general form associated only with ')..=-112:
(0'1,0'2,0'6) = [KjI...J(27tr)] fj (6) j = I, II, III

(uj, u2) = [(KjI211)...J(r/21t)] gj (8) j = I, II
U3 = (2KIII/ll)...J(r/21t)] (4.22)
where 11 is the shear modulus of the isotropic solid.
The general forms in (4.22) are independent of the far-field boundary conditions or the
size of the crack; only the K's do. Since the K's are crack-tip quantities, they are related to
the strain energy release rate at the crack-tip:
(4.23)
2.2. THE J-INTEGRAL
The J-integral of Rice [33] considers a finite plate that contains a crack of size a, the crack
being aligned in the x-axis as shown in FigA.S. Suppose that the full field stresses, strains
and displacements are solved; the total potential energy of the plate is given by:
(4.24)
114
Where A is area of the plate (minus the crack); e is the strain energy per unit area of plate, r
is the boundary contour of the plate and t and u are the traction and displacements vectors
on the plate boundary r.
plate boundary r
I---+-~x
contour I
Figure 4.5. The I-integral of Rice
The partial derivation of Yea) with respect to the crack size a corresponds exactly to the
energy release rate defined in (4.10):
G(a) = -dV(a)/da = J (en.1 - t 1• ui , x )ds (4.25)
where ni are direction cosines at a point on the plate boundary r.

Rice shows that the value of the contour integral in (4.25) remains unchanged if r is
replaced by any contour I drawn inside the plate, as long as the contour I contains the
crack-tip; see Fig.4.5. The contour integral in (4.25) evaluated along a chosen contour I is
known as the J-integral, whose value is the same as the energy release rate G at the crack-
tip.
To evaluate the J-integral, the full field solution of the plate must be first obtained, say
by a numerical method (e.g. finite element), without pursuing the singularity at the crack-
tip; the numerical solutions must be sufficiently accurate everywhere, except near the crack-
tip. A contour enclosing the crack-tip is then chosen away from the crack-tip in order to
facilitate the integration of (4.25).
One draw back of the J-integral is that it gives only the total strain energy release rate G
which may contain more than one K; see (4.23); it can not determine all the K,s if the crack
is in mixed modes. Another draw back is that the numerical solutions of the stresses and
displacements on the chosen contour I are approximate to some degree; the accuracy of the
J-integral so evaluated cannot be ascertained automatically.
2.3. SINGULAR-FIELD AND FAR-FIELD MATCHING
An alternative method is to solve the cracked plate problem in three steps: the first step is to
obtain the singular stresses near the crack-tip in the forms of (4.22), with the K's yet to be
detennined. Note that the singular solutions represents the full field solutions only near the
crack-tip; beyond a certain distance r* from the crack-tip, the singular solutions lose their
115
dominance. FigA.6 is a schematic illustration for a stress (J which distributes along the
radial axis r at a chosen e.
far-field
solution
Figure 4.6. Singular field and far-field stress matching for K

The second step is to compute the full neld solutions numerically without the inclusion
of the crack-tip singularities. Assume that the numerical solutions are sufficiently accurate
everywhere, except that it loses accuracy when within r<r*. The third step is to match the
two-step solutions along the ring contour of r=r* for all -1t<8<1t as shown in FigA.6. A
near perfect match of the two solutions furnishes the values for the K's. Details in finding
the desired r* in examples for the matching technique are found in [34].
The matching technique can be combined with the J-integral to compute G at the crack-
tip, along with all the K's in mixed-mode cracking problems.
2.4. CRACK-CLOSURE BY THE FINITE ELEMENT METHOD

The crack-closure technique of Irwin [31], see also (4.14), can be efficiently executed
by the finite element procedure [35]. The basic idea here is illustrated in FigA.? Let the
pre-existing crack of initial length a be aligned along the x-axis; under the applied far-field
stresses, the crack propagates stably a finite incremental length Lla. In this process, the
initial crack-tip opens up, represented by the nodes 1 and 2; the new crack-tip is at node 3.
A finite element analysis of this configuration provides the relative displacements between
nodes 1 and 2:
i = 1,2,3. (4.26)
116
L\a
a
Figure 4.7 Crack closure by the [mite element method

The next step is to close up nodes 1 and 2 so that the crack returns to its initial length a;
the nodal forces Fi necessary to bring nodes 1 and 2 together are computed through a new
round of finite element analysis. Thus, according to Irwin's crack-closure reasoning, the
total crack-tip strain energy release rate G(a) is approximated by
(4.27)
where the terms corresponding to subscripts 1,2 and 3 are, respectively, the sliding (GIl),
opening (GI), and anti-plane shearing (Gill) modes of crack extension.
The computed GI, GIl and Gill can each be related to the respective stress intensity
factors KII, KI and KIll. In particular, if the the crack is inside an isotropic material,
(4.27) corresponds directly to (4.23);
The finite element crack-closure technique can be routinely executed for complicated
cracking cases, such as in composite laminates, where one or more cracks may be present
in the laminate microstructure. But most routine finite element codes do not contain a
singular element near the crack-tip; in that case, the result expressed in (4.27) is only an
approximation.
2.5. INTERFACIAL CRACKS.
In composites, crack-like failures often occur in bi-material interfaces. The picture on the
left of FigA.8 shows the local details of the interfacial crack. This problem can be solved
by all the techniques discussed above. In particular, the near-field singular solution for a
crack in the interface between two isotropic materials, see e.g. Rao [36], admits the stress
and displacements in forms similar to (4.22):
0i =KrA. fi(e); i =1,2,6

Ui = (KIE) rA. gice); i = 1,2,3. (4.28)
117
where the dominant eigen value A may be complex-valued:
11,= -1/2 + ik (4.29)
where i is the complex number ~ (-I) and k is a real constant involving the material
constants E and ~ of material-I and material-2.
Figure 4.8. A crack in the hi-material interface
The complex term in (4.29) causes the singular stress solutions near the crack-tip to be
oscillatory. Within this oscillatory region, the stress intensity factors KI, KII or KIll
become undefined. Moreover, the energy release rates GI, GIl or GIll computed by the
finite extension crack-closure method also become extension-size (~a) dependent inside the
oscillatory stress region, although the totlal strain energy release rate G is indepenndent of
~a; see Sun and Qian [37]. The region of the oscillatory stress is, however, much sarnller
than the region influenced of /..;=-1/2. To show this fact, let a thin layer of material-3 be
introduced in between the bi-material interface and let the crack tip be inside the layer of
thickness t, see the picture on the right of Fig.4.8; then, the dominant A at the crack-tip is
equal to -112. The stress intensity factors are then evaluated in terms of the thickness t of
material-3; in the limit of t""-70, the stress intensity factors would be the ones for the case
shown on the left. Details of this reduction are found in [38].
Finally, it is remarked that the computational approaches discussed above can be
applied readily to materials of anisotropy. Invariably, the mathematical formulation and
solution development become more complicated. In particular, crack modes are generally
coupled; the one-to-one correspondence between GI and KI, GIl and Kn, etc. no longer
exists. Furthermore, in multi-phase materials such as composites, thermal residual stresses
are often present; one cannot proceed to apply the theory of elasticity until the reference
state of the material (zero stress corresponds to zero strain) is determined.
These and other concerns arise from the unique characteristics of composites; often they
impose limitations on the applicability of some of the crack analysis techniques discussed in
118
this paper. Some of these points will be elaborated more in Part V where specific problems
are studied.
Part V. Simulations of Matrix Cracking in Composite Laminates

A class of matrix-dominated cracking problems in composite laminates is discussed in Part
V. Specifically, the developmental processes of intralaminar and interlaminar cracks in
epoxy-based graphite fiber reinforced composite laminates are simulated by a combined
probabilistic and mechanistic approach. Background discussions presented in Parts I-IV of
this series are essential in the crack simulation scheme.
1. Physical Characteristics of Intralaminar Cracking
In laminate design, the 00 , 900 and ±9 ply orientations are amongst the most common
choices (usually, 150~9~50). The mechanical behaviors oflaminates with quasi-isotropic
(0/±45/90), cross-ply (0/90) or angle-ply (±9) stackings are studied extensively in the
laboratory and field settings. Under most design loads, plies aligned in the OO-direction
carry the primary loading, while the off-axis plies (the 900 plies in particular) are subjected
to secondary stressing. Often, these secondary stresses in the off-axis plies may become
tensile and normal to the fibers, thus causing intralaminar cracks in the plies (Part 2, section
3).
The physical characteristics of intralanunar cracking in laminates are well documented
in epoxy-based laminate systems (see, e.g. [39]); transverse cracks are observed to form
inside the 900 plies at multiple locations; the formation of each individual crack is rather
sudden; and it at once severs the cross-section of the 900 plies; see Fig.5 .1.
Actually, the cracking sequence is a unique load-dependent process; it begins with the
first crack, say at the laminate strain Exo and more cracks continue to form at increasing
laminate strains. In some cases, the number of cracks that can form reaches a saturation
value [40]; in other cases, a host of different modes of failure (delamination, longitudinal
splitting, etc.) could occur before the saturation state is reached [20].
A plot of the crack density (number of cracks per unit length of laminate) versus the
laminate strain Ex is shown on the right of Fig.5.l; the plot reveals three key characteristics
in the crack development behavior: the crack onset strain Exo, the rate at which cracks are
formed (the curvature of the crack-density versus laminate strain plot) and the saturation
density.
119
crack
density
saturation
Plane
view
Intralaminar
crack
~ Lan.uante
stram lOx
Figure 5.1. The Physical Characteristics of1ntralaminar Cracking
Clearly, these characteristics are intrinsically influenced by factors of geometric and

material origins in the laminate microstru"tnre. This section discusses briefly some of the
most influential factors and their effects on mtralarninar cracking behaviors.
1.1. THE 900 -PL Y THICKNESS FACTOR
The thickness of the 90 0 plies in which transverse cracks form has a profound effect on all
three characteristics of the crack development curve; the effect on the onset strain lOxO is
fully documented in [39l and the effect on the entire curve is documented in [20].
Briefly, consider the laminate series [±8/90n ls with n =1, 2, 3; here, the value of n
represents the variation of the 90 0 ply thickness. Fig.5.2(a) are qualitative plots of the
crack development curve for n= I, 2 and 3; it shows how the 900 layer thickness affects the
crack development behavior: the thicker the 900 layer the lower the onset strain and the
lower the saturation crack density. The mechanisms responsible for the thickness effect are
discussed later in this section.
1.2. THE STIFFNESS OF THE OUTER PLIES
Consider the laminates [±8/90ls , with 8 = 0 0 ,25 0 and 45 0 . Here, the 900 layer thickness
if kept constant, while the variation of El from 0 0 to 45 0 represents changes of the axial
stiffness of the outer layers. The effect of the outer layer stiffness on the 900 -layer crack
development is shown qualitatively in Fig.5.2(b). It is seen that a stiffer outer layer has
more prohibiting effect on the formation of the transverse cracks. The crack prohibiting
effect is keenly associated with the 900 layer thickness effect, as documented in [39l.
120
crack crack
density density
(a) (b)
Figure 5.2. (a) 9()o ply thickness effect; (b) outer ply stiffness effect
1.3. LAMINATE CURING RESIDUAL STRESSES
The laminate series [±8/90]s studied in [20] are cured at 177 o C; the "residual stress-free"
temperature To is determined at 1500 C. Thus, at temperatures lower than To, thermal
residual stresses are present in the laminatto".
crack
density
Figure 5.3. Thermal residual stress effect on transverse cracking
The effect of temperature, or the thermal residual stresses, on transverse cracking in the
laminates is qualitatively displayed in Fig.5.3, where the test temperature T varies from
-400 to +80 0 C. Clearly, at lower T, transverse tensile stress exists in the 90 0 layer which
in turn enhances transverse cracking at lower laminate strains. Generally speaaking, the
saturation crack density is not affected by the test temperature T.
2. Simulation of Intralaminar Cracking
The key characteristics of intralaminar cracking in laminates discussed above may all be
121
captured in a probabilistic and mechanistic simulation scheme. The scheme involves the
probabilistic representation of the effective flaws in the 90 0 plies that suffer transverse
cracks; and it also involves the judicious use of the fracture mechanics methodology which
captures the unique load-transfer mechanisms in the laminate microstructure. The combined
approach provides a stochastic mimicking of the physical behaviors of transverse cracking.
Some preliminary discussions on the cracking mechanisms and their relations with the 90 0
layer thickness, the outer layer stiffness and the test temperature effects are presented next.
2.1 FROM A FLAW TO A TRANSVERSE CRACK
For purpose of discussion, let the laminate series [±8/90 nl s , with 8=00 and n=l, 2 and 3
be used as examples. In the context of the effective flaws discussed in Part 3, section 2, it
is assumed that the basic 90 0 ply (of thickness t and length unity) is endowed with a
distribution of M effective flaws, characterized by the size distribution (ai) and location
distribution (xi), where O<ai<t; and O<xi<l for all i=l,M. The location distribution {xil
may be of uniform probability, while the size distribution (ail may follow the Weibull
function:
f(ai) = (aJ~) (ai/~)a-l exp[-(ai/~)al (5.1)
where a (>0) is the shape parameter and ~ (O<~<t) is the scale parameter.
The thickness of the 90 0 layer in the [+ Cl/90 nl s laminates is 2nt; the effective flaw size
distribution in the layer is estimated by the Weibull volumetric rule (see Eq.3.4, Part III):
a2n,i = ai (2n)2/a i=l, M (5.2)
--I 1,"' 0,
, ... x ·1
1-- a
2nt
Figure 5.4. Flaw propagation and the associated strain energy release rate
When the laminate is loaded in tension, see Fig.5.4, the flaw of size a (1 <a<2nt) may
be propagated (in pure mode-I) into a transverse crack. The flaw tip strain energy release
rate GI is calculated and expressed in the form:
(5.3)
where b.T=To-T.
122
The functional form of gda) is shown qualitatively in Fig.S.4; the functional form of
gT(a) is similar (for computational details, see appendix in [18]). When the flaw is small,
the flaw tip GI(a), as expressed in (5.3), is rising linearly with size a; when the flaw is
large, the flaw tip stress field is shielded by the more rigid outer [02l-layers. Hence, GI(a)
grows linearly at small a, reaches a maximum and then reduces sharply as a approaches the
outer layer interface. This behavior suggests that the effective flaw will begin propagation
when GI(a) exceeds GIc of the 900 ply material; the propagation is stable and is arrested at
the outer layer interface, assuming the interface is perfectly bonded.
2.1.1. The Outer Layer Stiffness Effect.

Fig.5.5(a) shows how the gda) or the gT(a) curve drops toward the outer layer interface,
depending on the outer layer stiffness relative to that of the 900 layer. The curves drop
sharply if the outer layer stiffness is high; much less, if the outer layer stiffness is low.
Compare the gE(a) curves for 8=00 and 25 0 for instance; the associated crack tip driving
force GI(a) of the former is much reduced due to the higher stiffness of the [02l-layer than
the ±25 0 layer. The extent of the outer layer stiffness effect on GI(a) is more pronounced
for larger flaws in the 900 layer; this can be seen in the difference of the onset laminate
strains between [02/90 nls and [±25/90 n]s shown in Fig.5.2(b).
2.1.2. The 9oo Layer Thickness Effect.

The 90 0 layer thickness effect can be inferred from the geCa) shown in Fig.5.5(b).
According to the volumetric rule in (5.2), HU'ger effective flaws are present in thicker 900
layer. Hence, statistically, a larger flaw in thicker 90 0 layer will propagate at a lower
applied laminate strain. This effect is seen repeatedly in experiment, as displayed in
Fig.5.2(a).
a a
2nt 21 (b) 4t 6t
(a)
Figure 5.5. Effects of outer layer stiffness and 9oo layer thickness.
2.1.3. The Thermal Residual Stress Effect. For a given flaw of size a, the flaw tip energy
release rate is expressible in (5.3) as GJ(a,Ex,LlT); for the flaw to propagate, the Griffth
criterion is applied: GI(a,Ex,LlT)=Glc. Thus, one obtains the laminate strain Ex at the
propagation of the flaw:
(5.4)
123
Clearly, if ~T is high, the critical laminate strain is low, as displayed in Fig.S.3.
2.1.4. The Shear-lag Effect.

The shear-lag zone refers to the extent a transverse crack influences the stress field around
it. Fig.S.6 shows a transverse crack which is formed in the 90 0 layer; along the outer layer
interface, an interfacial shear stress 'tzx exists which is singular at the crack-tip but decays
away from the tip within a small distance. This distance is known as the shear-lag zone.
~J_Z~~=-
1
__I l 21
T
nl
000
00
~
't
zx
L-----------===~--I~ x/nt
shear-lag zone
--I
Figure 5.6. The Shear-lag effect on interface stresses
Within this zone, the tensile normal stress O'x in the 90 0 layer is zero at the transverse
crack face; but it increases away from the crack face and approaches asymptotically towards
the far-field O'x value. Thus, within the shear-lag zone, the in situ tensile O'x is smaller than
the far-field O'x; so the flaws inside the zone is not likely to propagate unless the applied
laminate strain is substantially increased; or, at the higher laminate strain, a flaw of lesser
size outside the zone may be propagated.
The size of the shear-lag zone is proportional to the thickness of the 90 0 layer (=2nt),
see Fig.S.6. The actual zone size can be influenced by a number of factors, including the
axial stiffness of the outer layer and the quality of the interfacial bond. When coupled with
the 90 0 layer thickness and the outer layer stiffness effects, the shear-lag controls the crack
density growth rate in the laminates.
2.2 THE STOCHASTIC SIMULATION PROCEDURE
The transverse cracking processes in the [±S/90n l s series are experimentally documented
124
and numerically simulated in [18], [20] and [41]; the laminates with 9=0 0 are made of
AS4-350 1-06 graphite-epoxy ply system and the laminates with 9=25 0 are made of
T3001934 system. The test and data acquisition details are found in [20]; relevant ply
properties of these ply systems, including the value of GIc, are given in [18,41].
In what follows, the [02/90n]s series is considered and the steps in the simulation are
briefly discussed.
2.2.1. Effective Flaw Distributions.

At the outset, a random distribution of effective flaws, represented by (xi} and {ai} is
assigned to the basic 900 ply of thickness t and of length unity; {ail fits the Weibull
function in the form of (I) for all i=I,M. At this stage, raj} involves three undefined
parameters: M, ex and pand their values will be defined later in the discussion.
The 900 layer thickness in the [02l90n]s series is 2nt; the effective flaws in the layer are
given by the set {a2n,i}, i=l, M, computed using (5.2).
2.2.2. Strain Energy Release Rate Coefficients.

For a given flaw of size a in the laminate, the flaw tip strain energy release rate coefficient
functions ge(a) and gT(a) are calculated via the finite element routine. The flaw tip strain
energy release rate GI(a,Ex,LlT) is expressed in the form of (5.3), where LlT=126 0C is
already known; so GI depends on the flaw size a and the laminate strain ex only.
2.2.3. The First Transverse Crack.

A search of the largest flaw in the {a2n,il set is commenced; say, it is of size ao. The flaw
tip GI{ao) is calculated from (5.3), with a=ao and LlT=126 oC; the laminate strain Ex at the
propagation of the flaw is given by (5.4). This defines the onset of the transverse cracking
process.
The presence of the first transverse crack in the 900 layer casts a localized shear-lag
zone. For this laminate series, the half-zone size is approximately 4 times the 900 layer, see
Fig.5. 7. If a flaw of size ak « ao) is situated inside the shear-lag zone with the distance S
from the transverse crack, the flaw tip GI{ak} is reduced by the shear-lag effect, depending
on the distance S:
(5.5)
where R(S) is the strain energy release rate retention factor depending on S and GI(ak,oc) is
the energy release rate if ak is outside the shear-lag zone. The retention factor R(S) shown
in Fig.5.7 is generated by the finite element routine (see [18] or [41]).
If the flaw of size ak is situated in between two neighboring transverse cracks, the flaw
tip GI(ak) is now doubly reduced by the shear-lag effect from each of the cracks:
(5.6)
where SL and SR are the distances of the flaw to the left and the right crack, respectively.
125
%
100 .---------------------~~---------,
80
crack flaw
60
40
t_88-->0,-;-;_/ _C_O------1H,
j4- S ..... I
20
6 8 10 12 s/nt
Figure 5.7. The energy release rate retention factor due to shear-lag
2.2.4. Multiple Cracks.

After the formation of the first transverse crack from the largest flaw an, a search from the
rest of {a2n,d to find the flaw likely to propagate next is commenced. The search criterion
is to find the flaw that yields the most energetic GI according to (5.5) or (5,6); once found,
the associated laminate strain at the its propagation is given by (4). Note that the laminate
strain at the second crack is higher than the; first; but the flaw size may not be the second
largest in the {a2n,d set, due to possible shear-lag effect.
Subsequent cracks are determined by successive searches and simulations; the process
goes on until it is no longer energetically possible to find any more cracks, or until some
other failure modes (delamination, fiber break, etc.) are set in the simulation process. In the
end, the simulation provides a load-dependent sequence of the crack development process
in the laminate; the plot of crack-density versus laminate strain is just a by-product.
Fig.8 shows the simulated crack-density versus laminate strain plot from four replicate
[02/902]s laminates, the simulation being done with assumed values of M, ex and ~. In the
figure, each small darkened dot represents a simulated transverse crack event. Experimental
data range from four real specimens tested for transverse cracking is also shown along with
the simulated data. The fact that the simulated data agree fairly closely with the real data is
because the values of M, ex and ~ are so chosen as to force an agreement by iterations. The
latter is explained in more detail next.
2.2.5. Determination of M, ex and {3.

As discussed previously, a fully developed crack-density versus the laminate strain plot is
characterized by three parameters: the onset strain, the curvature of the plot (or the crack
development rate) and the saturation crack density. If everything about the laminate is kept
constant, the value of ~ which scales the effective flaw size can shift the plot along the
laminate strain axis; the values of ex which reflects the scatter in the flaw size distribution
can influence the curvature of the plot; and M which is the number of flaws in the {ai} set,
controls the saturation crack density.
126
A best fit between the simulated data and the tested data such as shown in Fig.5.8
yields 0.=4.0, p = 0.6t and M=3.2 flaws/mm for the basic 90 0 ply used in the [±9/90 n]s
series; see [41]. With the values of M, a. and Pdetermined, the transverse cracking process
in all of the [±9/90n]s series can be obtained by the simulation without further adjustment.
2.0 , - - - - - - - - - - - - - - - - - - - - ,
e 1.5
.E
experiment
data range
~::>
..5 1.0
.~
~
~
-tl 0.5
u
e
0.0 l--l~--..l------L.--__,.!.:_I~
0.4 0.6 .8 1.0
Applied laminate strain, %
Figure 5.8. CracK development plot/or [02/902]s
1.5 , - - - - - - - - - - - - - - - , 2.0
aJlD Data Range
aJlD Data Range
I§
1.0- .-
••
1.5
•
-=.. 1.0
a
.u;
~"
-"= 0.5_
o
u
e 0.5
0.0 L....:..!..:.L:.!....-~I_ _--;:;~I_--' O.OL--_...J-JI..--=.;;.;;_ _.....L-_ _ _...J

250 500 500 750 1000
Applied laminate stress (MPa) Applied laminate stress (MPa)
Figure 5.9. Crack development plot/or [02!904Js,lejt; and [02l90Js, right.

127
Fig.5.9 shows a comparison of the simulated crack density plots with experiment for
the [02/90]s and the [02/904]s laminates. In each case, four replicates are tested and the
same four are simulated. Similar comparisons for all the other laminates in the [±8/90 n]s
series are reported in [41].
3. Free-edge Delamination
Delamination is another mode of matrix and/or interface controlled failure in laminates. At
the ply level, delamination may be viewed as a plane crack separating the interface of two
adjacent plies. As most laminates are designed to cany in-plane loads, interlaminar stresses
that promote delamination are generally absent throughout the laminate except near free
edges, cut-outs or small holes.
Fig.S.lO is a display for the free edge induced delamination in the [±2S/90n]s laminate
series subjected to the uniform axial extension Ex. This problem is solved in two stages: the
3D free edge stress field is first treated by means of the 3D elasticity lamination theory (Part
I); the onset and propagation of the delamination is then simulated based on the assumption
of effective flaws (Part 3).
Let a be the initial size of the edge flaw as shown in Fig.S.lO; here, a is randomly
valued represented by the probability function f(a). The flaw tip energy release rate
coefficients ge(a) and gT(a) are generated via the finite element routine [18] and the energy
release rate G(a,e~,LlT) at the flaw tip is calculated by (5.3). The ge(a) curve shows that it
reaches a plateau value at a",,2nt, the 900 layer thickness; the plateau values increases with
the 900 layer thickness. This suggest that delamination growth is a neutral process and has
the 900 layer thickness effect; both are confirmed by experimental observations; see
discussions in [27].
a
2t 4t 6t
Fig.5.l0. Free edge delamination and the crack tip energy release rate.
Now, given the delamination resistance Gc of the laminate material, the laminate strain
Ex for the edge flaw to propagate, or the onset of delamination, is calculated by (5.4). Since
the initial flaw size a is randomly valued, Ex so determined is also probabilistic in nature.
Fig.5.11 illustrates the relationship between the predicted Ex and the flaw size distribution
function f(a). While the exact f(a) is not known, the growth behavior of G(a,Ex,Ll T)
128
coupled with the condition G(a,ex,~T)=Gc yields a lower bound for the onset delamination
strain Ex. For a laminate of sufficient length, this lower bound can be regarded as the
minimum possible onset strain.
Due to space limitation here, several other aspects of delamination cracking could not be
discussed; more are found in [27], however.
E
X
G(a,£x) =Gc
- - -~---,----
'-----..-:::~-----.;;::..-'----~a
Figure 5.11. Determining the range of the onset laminate strain
Concluding Remarks.
In this article, an attempt is made to delineate the distinct roles played by "micromechanics"
in modeling fibrous composite materials. In particular, the "forward" and the "backward"
approaches in utilizing micromechanics are discussed here in some details.
In the forward approach, micromechanics is applied in conjunction with material
homogenization, a process which smears the composite's microstructure. This approach,
however necessary and justified in describing the composite's global behaviors, is often
inadequate in predicting local material failure and/or failure modes. Thus, a backward
approach is needed.
In the backward approach, one attempts to restore at least the essential features of the
composite microstructure that is homogenized in the forward approach. Restoration of the
microstructure in the composite is necessary in order to capture the essence of the
composite failure mechanisms and failure modes. A number of examples and case study
results are referenced here to substantiate the need for the backward approach.
The latter approach, however important, has been little toiled; it is the purpose of this
article to call attention of the interested readers to this fertile field.
References
1. Lekhnitsky, S.G. (1963) Theory of Elasticity of an Anisotropic ELastic Body,
Holden-Day, San Francisco, CA..
2. Tsai, S.W. and Hahn, H.T. (1980) Introduction to composite Materials, Technomic
Pub. Lancaster P A.
129
3. Reissner, E. and Stavsky, Y. (1961) Bending and Stretching of Certain Types of

Heterogeneous Aeolotropic Elastic Plates, Journal Applied Mechanics, 28,402-.
4. Whitney, 1.M. and Sun, C.T. (1973) A Higher Order Theory for Extension Motion
of Laminated Composites, Journal Sound and Vibrations, 30, 85-.
5. Chamis, c.c. (1984) Simplified Composite Micromechanics Equations for Hygral,
Thermal and Mechanical Properties, SAMPE Quarterly, April, 14-
6. Hashin, Z. and Rosen, B.W. (1964) The Elastic Moduli of Fiber-Reinforced
Materials, Journal Applied Mechanics, June, 223-232.
7. Hill, R. (1965) A Self-Consistent Mechanics of Composite Materials, Journal of
Mechanics and Physics of Solids, 13, 213-225.
8. Pagano, N.l. and Brown III, H.W. (1993) The Full-Cell Cracking Mode in
Unidirectional Brittle-Matrix Composites, Composites, 24, 69-83.
9. Whitney, I.M., Daniel, I.M. and Pipes, R. B. (1982) Experimental Mechanics of
fiber-Reinforced Composites, Society for Experimental Stress Analysis, Prentice Hall, NJ.
lO. Askeland, D. (1994) The Science and Engineering of Materials, 3rd Edition, PWS
Publishing, Boston.
11. Daniel, I.M. and Ishai, O. (1994) Engineering Mechanics of Composite Materials,
Oxford University Press, NY.
12. Hill, R. (1948) A Theory of the Yielding and Plastic Flow of Anisotropic Metals,
Proceedings of the Royal Society, A, 193.
13. Azzi, V.O. and Tsai, S.W. (1965) Anisotropic Strength of Composites,
Experimental Mechanics, 5,283-288.
14. Tsai, S.W. and Wu, E.M. (1971) A General Theory of Strength for Orthotropic
Materials, Journal of Composite MateriaL,S, 58-80.
15. Chiu, K.D. (1969) Ultimate Strength of Laminated Composites, Journal of
Composite Materials, 3, 578-582.
16. Petit, P.H. and Waddoups, M.E. (1969) A Method of Predicting the Non-linear
Behavior of Laminated Composites, Journal of Composite Materials, 3,2-19.
17. Hahn, H.T. and Tsai, S.W. (1974) On the Behavior of Composite Laminates After
Initial Failure, Journal of Composite Materials, 8,280-305.
, 18. Wang, A.S.D. (1987) Strength, Failure and Fatigue Analysis of Laminates,
Engineering Materials Handbook, 1, 236-251. ASM International, Metals Park, Ohio.
19. Wang, A.S.D., Kishore, N.N. and Li, c.A. (1985) Crack Development in
Graphite-Epoxy Cross-Ply Laminates under Uniaxial Tension, Composite Science and
Technology, 24, 11-31.
20. Crossman, F.W. and Wang, A.S.D. (1982) The Dependence of Transverse
Cracking and Delamination on Ply Thickness in Graphite-Epoxy Laminates, Damage in
Composite Materials, ASTM STP 775, 118-139.
21. Wang, A.S.D. (1984) Fracture Mechanics of Sublaminate Cracks in Composite
Materials, Composite Technology Review, ASTM, Philadelphia, 45-62.
22. Griffith, A.A. (1921) The Phenomenon of Rupture and Flaw in Solids, Phil.
Trans. Royal Society of London, A. 221, 163-198.
23. Griffith, A.A. (1924) The Theory of Rupture, in Proceedings of 1st Int. Congress
of Applied Mechanics, Biezeno and Burgers, Ed. Waltman (1925), 55-63.
24. Rosen, W.B. (11964) Tensile Failure of Fibrous composites, A1AA Journal, 2,
1985-
25. Wang, A.S.D. and Crossman, F.W. (1980) Initiation and Growth of Transverse
Cracks and Edge Delamination in Composite Laminates: Part 1. An Energy Method,
Journal of Composite Materials, 14, 71-87
26. Wang, A.S.D. in Microstructural Design of Fiber Composites, by Chou, T.W.,
Cambridge University Press (1992), Chapter 3,150-165
130
27. Wang, A.S.D. in Interlaminar Response of Composite Materials, Pagano, N.J.,

Ed. Elsevier (1989), Chapter 2, 69-108
28. Schaffer, J.P, et. al. (1995) The Science and Design of Engineering materials,
Richard D. Irwin, Inc. Chicago. Part II, 372.
29. Westergaard, H.M. (1939) Bearing Pressures and Cracks, Journ. Applied
Mechanics, 61, A, 49-53.
30. Orowan, E. (1955) Energy Criteria of Fracture, Welding Journal, 34, 1575-1605.
31. Irwin, G.R. (1948) Fracture Dynamics, Fracturing of Metals, ASM publication,
147-166.
32. Muskhelishvili, N.J. (1953) Some Basic Problems of Mathematical Theory of
Elasticity, Noordhoff, Neth.
33. Rice, J.R. (1968) A Path-independent Integral and the Approximate Analysis of
Strain Concentration by Notches and Cracks, Trans. ASME, Jour. Applied Mechanics,
35, 379-386
34. Kishore, P.V., Lau, AC.W. and Wang, A.S.D. (1991) A Local-Global Matching
Method for the Single fiber Pull-out Problem, Int. Jour. Computational Mechanics, 14,
84-99.
35. Rybicki, E.F. and Kenninen, M.F. (1977) A Finite Element Calculation of Stress
Intensity Factors by A Modified Crack Closure Integral, Engineering Fracture Mechanics,
9, 931.
36. Rao, A.K. (1971) Strc- Concentrations and Singularities at Interface Corners,
ZAMM, 51, 3395-406.
37. Sun, C.T. and Qian, W. (1997) The Use of Finite Extension Strain Energy release
Rates for Fracture in Interface Cracks, In ·our. Solids and Structures, 34, 2595-2609.
38. Delale, F., Wang, AS.D. and Binienda, W. (1984) On the Strain Energy release
Rate for A Crack at the Interface of Two Bonded Materials, Adv. Aerospace Sciences and
Engineering, ASME AD-08. 113-119.
39. Bader, M.G., Bailey, J.E., Curtis, P.E. and Parvizi, A (1979) The Mechanisms
of Initiation of Development of Damage in Multi-Axial Fiber Reinforced Plastic Laminates,
in Mechanical Behaviors of Materials, ICM-3, vol. 3, p.227.
40. Reifsnider, K.L., and Masters, J.L. (1978) Investigation of Characteristic Damage
States in Composite Laminates, ASME Publications, AERO-4.
41. Lei, S.C. (1986) A Stochastic Model for the Damage Growth During the
Transverse Cracking Processes in Composite Laminates, Ph.D. thesis, Drexel University.
PREDICTION OF THE EFFECTIVE DAMAGE PROPERTIES AND FAILURE
PROPERTIES OF NON LINEAR ANISOTROPIC DISCONTINUOUS REINFORCED
COMPOSITES
K. Derrien, J. Fitoussi, G. Guo, D. Baptiste*,
Laboratory LM3, ENSAM, CNRS URA 1219, 151 Bd de I'Hopital75013 Paris France
* Email: baptiste@paris.ensam.fr
Abstract
We studied the tensile behavior and damage of two different composites. An Al2080
reinforced with different volume fractions of silicon carbide particles and an SMC R (polyester
matrix with randomly oriented glass fibres). The main damage mechanism is particle failure (for
the MMC) or fibre-matrix interface debonding (for the SMC).
Moreover, for the AlSiCp composite, regions of the matrix adjacent to broken particles
are sites with high hydrostatic tension and hence the nucleation of cavities is expected. Using
the J integral and HRR modified solution it is possible to calculate the growth of these voids.
Macroscopic failure is governed by a critical volume fraction of voids.
In this paper, we propose a micromechanical approach to model the elastoplastic
damageable composite behavior of the composites. Moreover, in the case of the MMC, the
proposed model is able to predict the failure strain of the composite as a function of the
microstructure. For the SMC, the elastic damageable behavior law is introduced in a finite
element code in order to predict the damage of a structure. In all cases, the simulated results
are compared to experimental ones and sltow a fairly good agreement.
Introduction
Metal Matrix Composites (MMC) and Organic Matrix Composites (OMC) have been
developed in order to value the combination of the following phases: rigidity of reinforcement,
ductility and fracture toughness of the matrix. They constitute a very attractive range of
materials particularly for the conception of aeronautical or automobile structures looking for
performance improvement as well as structure alleviation.
Our objective is to predict the tensile behaviour of the composites using a micro-macro
relationship taking into account damage and plasticity. In the case of an Aluminium X2080
reinforced by different volume fractions of silicon carbide particles, the model is able to predict
the failure strain. As MMCs and SMCs are very heterogeneous materials, their mechanical
properties are highly dependent on their composition, matrix, type and volume fraction of
reinforcement.
The first stage of this study has been experimental. The results are used as input and
validation data for the model. We use a micromechanical approach based on Mori Tanaka's
model in order to connect the microstructure and the macroscopic properties of the material.
The damage mechanisms are identified at the reinforcement scale, by in-situ tensile tests inside
a scanning electronic microscope, are modelled and integrated in our model, at the micro scale.
This approach leads to a new constitutive law taking into account damage evolution. The final
failure of the MMC is then predicted using the same approach by introducing in the matrix a
cavity growth law of Gurson's type.
131
132
I Materials
11) Microstructures
The first composite material is a Sheet Moulding Compound containing a quasi-random

oriented short-fibre reinforcement (E glass) distributed in a polyester matrix (SMC-R). The
characteristics of the components are listed in Table I. The resin is charged by Calcium
carbonate particles. The matrix is isotropic. The Young's modulus of the matrix is 6.65 GPa
and its Poisson's ratio is 0.28. The glass fibre is about 10 /lm in diameter and 25mm in length.
COMPONENTS Volume E(GPa), v FUNCTION

Percent
E-glass fibres 3 1. 64%±O. 16 72,0.28 Reinforcement
Polyester 44.41%±O.22 3.5,0.35 Resin
Calcium 2 1. 26%±O.11 85,0.28 Filler
carbonate
Others 2.69%±O.49 - Additives
Table I : Properties of Short-Fibre Reinforced SMC-R composite
The second composite material is au iiluminium alloy (X2080) reinforced with SiC
particles.
The materials were made by a powder blending and extruded route. The particles are
almost aligned in the extrusion direction, which leads to an anisotropy of the material. The
average particle's diameter is lOIlffi, their average aspect ratio is 1,5. The material is extruded
like bars with circular section. The specimen are taken from the bar within two directions:
longitudinal and transversal. The material anisotropy is characterised by a macroscopic
behaviour which varies with the way the specimens are taken (longitudinal or transversal).
The mechanical properties of the matrix and MMC's have been measured from tensile
tests. Results are reported in Table 2 :
Material Direction E(Gpa) Rm(MPa) A(%)

AlcoaX2080 Radial 75±O,21 483±44 20±O.42
00/0 Longitudinal 74,6±2.2 536±2.64 17±O.55
AlcoaX2080 Radial 94±13.2 497±14 6±O.73
15% Longitudinal 97.5±3 521+5 5.5±O.85
AlcoaX2080 Radial 94±10 491±l1.8 3.4±O.7
20% Longitudinal 103+7 538+14 3.4±O.4
Table 2 : Mechanical properties of the A1SiCp (indice p means particules)
12) Macroscopic and microscopic experimental analysis
In order to understand the mechanisms governing the deformation in an SMC

composite or in an AVSiCp composite, several types of characterisation are undertaken.
133
a) SMC composite:
At the macroscopic scale, tensile tests are performed and can be coupled with
ultrasonic measurement in order to quantify the evolution of all elastic moduli [I). For the two
studied composites, stress-strain curves can be divided in three parts (see figure 1).
The linear part (AB) corresponding to the elastic behaviour is followed by a non linear
stage (BC). Finally, another linear zone (CD) takes place until final fracture. It can be shown
that the first non-linearity observed in the stress-strain curves corresponds to the first decrease
of the elastic moduli [2]. This means that the macroscopic non-linearity is caused by damage
mechanisms occurring inside the material. In order to identify the damage mechanisms, tensile
tests are performed inside a S.E.M. by using a small tensile test machine. The damage
evolution is observed during the application of the tensile force. In the case of the SMC, these
tests lead to the following conclusions :
1) No significant damage occurs in the first stage (Elastic part (AB»,
2) The stage (BC) corresponds to the occurence of interfacial cracks localised on fibres
oriented from 50° to 90° relative to the tensile direction,
3) Matrix microcraking can also occur during this stage but it is rare. This kind of damage
mechanism is localised in the matrix dominant area and these microcracks are quickly
stabilised,
4) During the stage (CD), the existing interfacial cracks propagate around the fibres and the
crack opening grows. Moreover, new inter.__ .al cracks are created on the less disorientated
fibres,
5) At the end of stage (CD) fibres oriented in the tensile direction break.
This experimental study shows that interfacial damage is predominant for this type of
SMC composite. The other damage mechanisms can be neglected. Depending on the fibre
orientation, interfacial debonding occurs either by shear or by normal stress. The interfacial
cracking is statistically distributed over the whole composite.
In order to quantify the evolution of interfacial failure during the loading, we chose to
follow the interface damage evolution of three bundles of fibre oriented at 90° relative to the
tensile stress. Then we determine by image analysis the evolution of the number of broken
interfaces as a function of the applied stress. Figure 2 shows the evolution of the microcraks
density for the 90° oriented bundles (3 differents bundles are analysed).
The damage kinetics is approximately the same for the three bundles. Thus, we can say
that the average crack density evolution is statistically representative. Note that for the 90°
oriented bundles, the interface failure is due to the normal stress. So in order to characterise
the influence of the shear stress on the interface damage kinetics, we carry out the same
procedure for other fibre orientations. Figure 3 shows that the stress level corresponding to the
damage incipience increases when the angle of fibre orientation decreases.
b) AVSiC composite :
Fractographic investigation has revealed that fracture surface .consists of microvoids of

sizes ranging from the micron to ten microns. The origin of these voids has been attributed to
fracture of either the SiC particles or the inclusions and precipitates. One can see on each side
134
of the failure area the two half of the broken particles, this leads us to think that the damage
was produced by the particle's failure and that the failure way of the composite goes by the
broken particles. The dimples of a 2 microns size are associated to the debonding of the
precipitates in the matrix and are characterizing the ductile failure of the matrix. In order to
confirm the microstructural damage mechanisms at the origin of the MMC static behaviour,
and determine when they occur, tensile tests in-situ in a scanning electron microscope have
been carried out. In the MMCs studied, the main damage mechanism is particle failure which
appears in the largest particles or in the most elongated ones just before the macroscopic yield
stress of the material.
Regions of the matrix adjacent to broken particles are sites with high hydrostatic
tension and hence the nucleation of cavities is expected. The failure between adjacent broken
particles appears to occur by intense growth of these cavities in the matrix. The origin of these
micro-voids is attributed to failure of inclusions and precipitates.
We conducted a statistical survey on the intact specimen and on the broken specimen in
order to quantify the probability of the particle failure depending on their size and their aspect
ratio (Fig 4 and 5). This study was done on X2080 alloy reinforced by 15% of SiC particles.
The specimens have been taken from the bar within the longitudinal direction. Larger particles
or particles with a big aspect ratio have failed first.
II Micromecbanical model
A micro-mechanical model based on' e stiffuess prediction by the Mori and Tanaka
approach is proposed for modelling the elastoplastic and damaged composite behaviour.
Among several well-known micro-mechanical approaches, the Mori and Tanaka's method is an
alternative to find the estimates of elastic moduli and local fields in multiphase composite
materials. The stress tensor is calculated in each phase as a function of the overall macroscopic
stress tensor L, the geometrical parameters as orientation and aspect ratio, and the mechanical
properties of each phase. The combination of the Eshelby equivalent inclusion method and the
Mori and Tanaka's back stress analysis, leads us to the equations used in the prediction of the
composite behaviour.
Ill) Elastoplastic behaviour
The main assumption is contained in the strain localisation relation which defines the
load sharing between the different constituents. This localisation relation is expressed by Mori
and Tanaka [3] :
Er =TrEO Tr =[Lr +Lo(s~I-I)r\oSr (1)

Eo,Lo and Er,Lr are the average strain and stiffness tensor of the matrix and the r-th
reinforcement respectively. Sf is the Eshelby's tensor.
If a uniform stress L is applied to the material, it was shown that the average local stress over
the composite is equal to L, the corresponding strain in the material is noted E and we have:
(2)
( )i means the volume average value in the volume of the phase i
As a result, the estimation of the composite stiffness tensor can be written as:
135
(3 )
The elastic model is extended to the elastoplastic behaviour using the concept of secant
moduli. The theory makes use of a linear comparison material, whose elastic moduli at every
instant are chosen to coincide with the average secant moduli of the matrix to reflect its
elastoplastic state. The composite is subsequently replaced by the comparison material with
equivalent transformation strains. The matrix behaviour is identified from the tensile stress-
strain curve of the non-reinforced alloy and the composite stress-strain curve is determined
step by step by varying progressively the matrix secant modulus.
This approach differs from the original one proposed by Tandom and Weng [4] by the
new definition of the matrix effective stress. The effective stress is not defined in terms of the
averaged stress in the matrix alone but in terms of the average elastic distortional energy in the
matrix, see Qiu and Weng [5] and Hu [6].
The energy can be evaluated from the variation of the effective compliance with respect
to the variation of the local shear modulus such as:
(4)
where M is the composite compliance tensor and 110 the local shear modulus.
IL2 Behaviour oftlte damaged material:
a) Damaged behaviour for the SMC composite
Damage criterion
The failure of the fibre/matrix interface is the principal source of damage. This failure is
determined by a local criterion which combines the normal and the shear stresses by a linear
relation (see next equation).
After debonding of the fibre, its contribution to the composite effective stiffness is
changed. In order to model the damaged composite behaviour we can choose to replace the
proportion of debonded fibres by an equivalent volume of matrix [I], [9] (lower bound) or to
replace it by an equivalent anisotropic undamaged fibre [2]. In order to reproduce numerically
the evolution of the interfacial cracks density as a function of the average interfacial stress, we
introduce in the Mori and Tanaka model a local interface failure criterion. Because the
interfacial damage is distributed statistically as a function of the spatial distribution of the
microstructure, the local interface failure criterion must be written in a statistical form [2] :
Pr ( ")
L-
(J+~l)n
= I-exp(-~ (5)
In this equation, Pr denotes the interface failure probability relative to a given interfacial
stress state (J and 1. (J and l are the normal and the shear stress at the interface which are a
function of the macroscopic stress, L, and of the fibre orientation. ~ is a coupling parameter
136
and Ri denotes the interfacial strength. Finally, n is the statistical parameter. The knowledge of
p, Ri and n defines completely ollr statistical interface failure criterion.
Identification oj the statistical parameters:
The three parameters of the interface failure criterion defined above are numerically
identified by using the micro mechanical model to fit the experimental results. Thus, the model
predicts the evolution of the interface crack density for different fibre orientations as a function
of the interface failure criterion parameters. After several iterations, we identity the values
which lead to a good fit of the experimental results of Figures 2 and 3. Figure 6 shows the
comparison between experimental and simulated data for P=1.73, Ri'O'19 and 4<n<6. (Note
that p and Ri are identified by fitting the curves of Figure 3).
b) Damaged behaviour for the metal matrix composite
Damage criterion:
For modelling the damaged behaviour, we have to define a damage criterion relative to
the experimental results. The particle breaking is the principal source of damage. We use a
Weibull law in order to define the particle fracture. This law is in agreement with our
experimental observations, the fracture probw_ ..ity increases with the particle size and applied
stress.
The fracture probability of each particle is a function of its volume V, and of the
maximum principal stress 0,.
The Weibull's law can be written, see Mochida and al[7] :
Pr{cr r, Vr) = l-exp( - ~: (:: r) (6)
m,ou, are called shape and scale parameters.

Vr is the volume of the r-th group and and Vu is the volume of the reference group for which
the Weibull's parameters are determined.
A broken reinforcement still contributes to the global stiffness of the composite but the
rigidity of a broken reinforcement is lower than the rigidity for an intact reinforcement. We
choose to replace the broken particles by penny shaped cracks located perpendicular to the
loading direction and which are oblate spheroids with the major axis being equal to the radius
of particles and the minor axis being a function of the external load. The analytical value of the
crack opening displacement h is :
h = 2bE33 wheree' is the eigenstrain in the crack and 2b is the crack diameter (Mura [8]).
Using Mori Tanaka's method, we have:
(7)
137
It is now possible to predict the tensile behaviour of composite material including the damage
effect and the plasticity. At each step of the applied macroscopic load, the stress induced in the
SiC particles is calculated followed by their fracture probability. At the following step the new
broken particles are replaced by penny-shaped cracks.
Identification of the statistical parameters
The behaviour will be entirely determined by the knowledge of the statistical

parameters of the Weibulllaw. We determine these parameters using our experimental results:
for a given macroscopic load (here the failure stress) we have determined the fracture
probability of the particles depending on their aspect ratio and their size (Fig 4 and 5). We
adjust the statistical parameters in order to match the results with the theoretical model. We
find the following parameters (determined in a statistical survey on a composite reinforced by
15% ofSiCp).
cr.= 1500MPa, m=4, Vo is the volume for a particle of diameter IOl1m
m Failure criterion (only for the Metal Matrix Composite):

Our assumption is that the failure of the specimen takes place by the coalescence of the
micro-cracks initiated in the matrix from the broken particles. The presence of SiC particles
increases the number of potential sites at th~ v~ginning of the damage and the critical growth
rate of cavities might be limited if the number of broken particles is sufficient and if these sites
are sufficiently close (relative to the distance between two broken particles). From the
theoretical and experimental studies , it seems necessary to have a precise knowledge of the
stress and strain fields close to the broken particles in order to elaborate a failure criterion. Up
to now, we have integrated in our law the porosity due to the particle failure. In order to have
a more accurate description of the damage and to establish failure criteria, we will take into
account the porosity associated to the debonding precipitates. The first stage of a failure
criteria definition is to determine the range of stress and strain fields close to the SiC broken
particles, in accordance with the composition of the representative volume element and the
macroscopic plastic strain. At the second stage we determine the growth rate of cavities
initiated on the precipitates.
III 1) Stress and strain fields near the penny-shaped cracks
The strain distribution near the crack is obtained from the Hutchinson Rice and
Rosengren [10] solution. This solution is not suitable for porous materials because the yield
criterion used is a Von Mises criterion. In our case we want to take into account the local
porosity due to the debonding of the precipitates. So we have modified the original theory,
following the same procedure used by Li and Pan [11] but using a quadratic yield criterion.
The yield criterion is calculated using Mori Tanaka's approach and the plasticity of porous
material developed by Qiu and Weng [5].
(8)
9(1- fO) 15 - 6fo

3u=--- v = - - - (9)
4f02 9f0 2
138
fo=l-fp, fp is the volume fraction of porosity inside the matrix where the stress and strain fields
are calculated.
This expression comes to the Von Mises criterion as u=Q and v= 1.
Thus the HRR modified singularity in conjunction with the value of J specifies the near crack
tip fields.
J represents the amplitude of the singular fields. We calculate J using Mori Tanaka's model:
The energy release rate of a penny-shaped crack can be defined as Mura [8]
J- 1 I~
he ~ (10)
1 •
P=Po + Eint =PO +-(Jf; Vc
2
P is the total potential energy
Po is the total potential energy without any heterogeneity and Ejnt is the interaction energy
between the applied stress and the heterogeneity,
Vo is the volume of the penny-shaped crack and e· is the eigenstrain in the crack.
&~ =(Lo(I-Sc)tl: (11)
J which is calculated on a crack and is in connection with the composition of the composite
surrounding this crack using the e· value.
III2) Failure criterion
The ligament failure is assumed to take place when the growth rate becomes critical at
a distance at least equal to half of the distance between two particles. The volume fraction
matches the average half inter-particle distance, it depends on the radius, R, ofthe particles and
fthe total volume fraction of reinforcement.
2J.critical = R· ( ~ - Jf) (12)
In order to compare the failure strain of the composites reinforced by different fractions
of SiC, we will calculate the growth rate reached at the critical distance Aontic:al for a J value
matching the failure strain ofV.E.R, for one composite (here X2080 + 15% SiCp). One will
use this growth rate as a reference for each composite we calculate the distance I for which this
growth rate is reached. Therefore we impose that the Representative Volume Element (RVE)
failure occurs when AlAcritical =\.
Departing from the failure strain of a given composite, this study allows us to predict
the failure strain of composites based on their particles content.
IV Results at the RYE scale ( tensile test)
IV 1) Results for SMC composites
Figure 7 shows the comparison between experimental and simulated stress-strain

curves (obtained by using the simulation method described in §II.2.b» in the case of a tensile
test for two identical SMC composites but presenting different coating of the fibres (strong or
139
weak interfaces). This figure shows that the model is able to take into account a variation of
the interface resistance between two SMC presenting the same composition. Moreover, the
evolution, with damage, of the Young's modulus of the SMC with a weak fibre-matrix
interface was evaluated by performing unloading and compared to simulated results in Figure
8. The comparison shows a quite good agreement. For SMC with a strong interface, ultrasonic
measurement was performed during the tensile test [I]. Thus, it leads to the evolution of the
different elastic stiffuess coefficients with deformation. The comparison with simulated curves
shows a quite good agreement too (see Figure 8). These results show that the evolution of the
macroscopic anisotropy with micro damage is well predicted by this kind of model.
IV 2) Results for Metal Matrix Composites :
In Figure 9 , we compare experimental and theoretical results for two kinds of materials
: X2080 alloy reinforced by 15 or 20 % of particles. Our simulation gives a good agreement.
Using the same method, we can calculate the failure strain of a composite as a function
of the volume fraction of reinforcement, results are reported in Figure 10.
We obtained the previous results for longitudinal specimen and for SiC particles of
diameter equal to 10 microns. Now we study the influence of particle's size or direction ofload
on the macroscopic failure strain.
For a given volume fraction of reinforcement (here 15%) the failure strain of a
composite material depends little on the particle's size because of two antagonistic phenomena
(Figure 11) .
If the diameter of particles is increased, the distance between two broken particles is
increased but in the same time the energy release rate of cracks is increased. On the other hand,
the failure stress is weaker where the particles are larger because the number of broken
particles is higher and as a consequence there are more cracks in composite.
For a given volume fraction of reinforcement, failure strain of a composite material is

approximately the same (Figure 12) whether the particles are oriented in the loading direction
(longitudinal specimen) or in the perpendicular direction (radial specimen). The distance
between two broken particles is the same. The energy release is higher because the cracks
resulting from the particles oriented perpendicularly with respect of the loading direction are
larger. However, when the particles are oriented perpendicularly to the direction of
sollicitation, they contribute less to the total composite's rigidity, stress in the matrix is lower
and the growth rate is smaller too. Here again, the simulation gives a good agreement.
V Finite element analysis (for the Organic Matrix Composite)
For cases where the geometry of the composite structure is complex, the finite element
calculations are necessary to predict the stress and strain fields as well as the distribution of
damage in the composite structure under an external loading.
Moreover, because on the non-linear behaviour due to damage, we can not consider
the composites as purely elastic. We have to use a behaviour law taking into account damage
effects. We propose here to use the developed multi-scale model in a finite element analysis in
the case of complex loading such as torsion-bending. The tests are realised experimentally and
also simulated using the elastic damageable law in order to validate the approach.
140
Using our multi-scale model, the effective stiffuess of the composite can be estimated
when the local damage occurs. The local stresses are searched as a function of the micro-
structural parameters and the applied « macro» strain. The damageable elastic behaviour law is
then introduced into a finite element code, ABAQUS, by using the procedure UMAT. The
evolution of damage for each element of the analysed structure is thus visualised.
Numerical procedure
The stress and strain tensors as well as the increment of strain at the beginning of the
increment are given, for each integration point of each element, by the code ABAQUS. The
Jacobian tensor and the increment of stress tensor at the integration points are calculated by
the model developed in the subroutine, UMAT, at the end of the loading increment.
The difficulty is to calculate the Jacobian tensor. It is defined as:

J =iJiJ Au
A& (13)
This tensor J(i, j) defines the change in the i-th stress component at the end of the time
increment caused by an infinitesimal perturbation of the j-th component of the strain increment
array.
The effective stiffhess of the composite, L comp , is degraded during the loading, so it is a
function of the applied deformation, Eo:
Lcomp.=Lcomp.(Et) i=l,6 (14)
Using Hooke's relation and the equations (13, 14), we have the Jacobian tensor:
iJL
Jif(t + At) = iJE:E+J(t)=Lik.jEk+Lq(t) i,j,k=1,6(VoigtNotation) (15)
The finite element calculations with this definition of Jacobian tensor are difficult to
converge because of numerical problem during the calculation of J. Using a central difference
operator [7], the Jacobian tensor is defined as:
J(/+LJt)= ]12LJ Lcamp. +Lcomp.(t) (16)
The damage state for every fibre family, for every point of integration, is defined as a
state variable. These state variables are calculated at each step of calculation and allow us to
insure the condition of irreversibility of the damage.
The results of these calculations are presented here by comparing the simulated results
to the simulation in the case of a combined torsionlbending.
The tests are realised on a tensile test machine, INSTRON (50 kN), with a setting of 4
arms schematised by Figure 13.
Figure 14 shows the comparison between the simulation and experimental results. The
applied force, F, is given in function of the longitudinal strain. The experiment - simulation
comparison shows a good correlation.
Figure 15 shows the moment-longitudinal strain curves comparison. The difference
between the calculation and the test is about 12% at the end of the loading.
This procedure allows us to calculate the composite structures submitted to a non

homogeneous loading. The evolution of the effective stiffness of the composite relative to the
local damage occurring during the loading, can be estimated by the calculation (see Figure 16
141
in the case of a bending test), Test simulation comparisons on a SMC type composite have
given satisfactory results [9].
The installation of this numerical tool into a finite element code constitutes an
important stage for the use of composite materials, The engineers will be able to strengthen the
part of the composite structures where the damage is more sensitive. It allows indeed to
calculate the global composite structure by taking into account the local damage mechanisms
occurring at the reinforcement scale (some microns).
Conclusion
It is well known that the non linearity observed in the experimental stress-strain curves
of composite materials is the consequence of physical phenomena occurring in both matrix,
fibres and interface. Plasticity and damage are the principal phenomena shown by in-situ tensile
test.
Thus, a multi-scale analysis is proposed by introducing experimental data obtained at
the microscopic scale (damage or plasticity) in a multi-scale model in order to predict the
macroscopic response of the composite.
This paper is an overviem l)f some models which can be developed on the basis of
multi-scale approach. We have shown the potentiality of this kind of models in the case of
elastic damageable laws (SMC), elastoplastic damageable law (CMM), prediction of failure
(CMM) and finally, analysis of structure usin~ 'us approach(SMC).
142
Figures
y
180
.-.
CIa 1~0
~ 10'"
~ 120 /
......
'/
"-'
......~ 90 C
GI
60 B/
==...
~
= 30 /
0 ~ I
o 0,006 0,012 0,018
Tensile strain
Figure 1 : Description of an SMC tensile test.
::-
'1ij
c ....
'" n'
-0
.;.:
~1
...00:1 '"
0
~.
0; 03
'u 0.2
~
..
-
~l
c:'" II
Stress (MPa)
Figure 2 : Evolution of the crack density for the 90° oriented bundles of fibers.
143
0·'1-·
0.1
06
0'
04
UJ
,?
lO
"
Stress (MPa)
Figure 3 : Influence of the orientation on the evolution of the interfacial failure kinetics.
v oIum~ fraction (eft,)
~.O
4,5 Dlntact specimen
4,0 : broken specimen

3,5
3,0
2,5
2,0
r
1,5
1,0
0,5
0,0 ....
~I
6 8 10 12
Diameter (!un)
Figure 4 : The statistical distribution of broken particles: a function of their diameter.

Volume fraction (0/.)
10 a intact specimen
9 : broktn specimen
1,5 2,5
A.spect Ratio
Figure 5 : The statistical distribution of broken particles: a function of their aspect ratio.
144
Stress (MPa)
Figure 6: Numerical identification

145
Stress (MPH)
180 Strong interface
160
140
120
Weak interface
100
80
60
40
20
o ~---;---~
o 0,002 0,004 0,006 0,008 0,01 0,012 0,014 0,016 0,018

Deformation
Figure 7 : Comparison between experimental and simulated tensile tests, effect of interface
0,5 0,5
Evolution of C 11 EvolutionofC23 :
°° 0,02 0,Q2
0,5 0,5
Evolution ofC22 Evolution of C55 :
° °
° 0,02
SMC strong interface

° 0,01 0,02
0,8
0,6
SMC weak interface
0,4 -l-----+------t----+------;---t----i
20 40 60 80 100 120
Streu(MPa)
-Experimental
-Simulated
Figure 8 : Comparison between experimental and simulated evolutions of elastic moduli with
interfacial damage
146
fv=20%
model: dash line
600 experimental : solid line
=- 500
~ 400
~ 300
fv=15%
"
"is. model : dash line
8 experimental: solid line
200
.."E
~ 100
0
0 2 6
Macroscopic strain ("!o)
Figure 9 : Behaviour law for materials reinforced by two different volume fractions of SiCp
8
~
>'! 7
~
~ 6
SCo
S 5
.
....8
c
4
OJ 3
!a
e 2
.:";
"" 0
0 10 20 30
Volume fraction of SiCp in composite
Figure 10 : Failure strain of a composite as a function of the volume fraction of SiCp contained
in the matrix
147
1So;. SiCp longitudin.1 .pecimen ISo;. SiCp,lonllitudinol

600
600 dash line: • orr.rliel., ~ G I'm pcdmt:n
\
-;- 500 ~ .. ' . ' .
~
~"
C.
400
JOO
I
IS% SiCp longitudinal .pecimen
+
J S% SiCp, radial .petimen
'Olid line: experimental
ee... 200
solid line : of particle. - ill'I" 1S% SiCp, radial .pedmen
dash Un. : model
i 100
O+----t-------II---~
o o
Motro.coplc .traln (%) Mocroscopic .traln (%)
Figure 11 Effect of the particle's Figure 12 Effect of the load's direction

diameter on the macroscopic failure strain on the macroscopic failure strain
Bending
Bendlna ... TOrtiOn
Figure 13 : Bending + torsion test sample

148
Forcc(N)
."- L
'\\ """ 71
,[J{'_
~, -
"-'\ .,. 1'.1
£l
- F. E. Face in traction
-Test. Face in traction
,
~ j - - F.E. Face in compreslion
'r~
-Tesl, Facc in oomprcssion
u II
'''\If
u
1,50% -1,00% ~50% 0,00% 0,50% 1,00% 1,50% 2,00%
Longitudinal strain &:('/0)
Figure 14 : Bending + torsion test

Total force -longitudinal strain curves comparison
Moment(Nm)
k wi
"\.\
~
.. //
/. jf
'\"I, . f'
# ---F.E. Faec in tnttiOll.
-Tat. Face in tAction
\D .I --F.E. Fac in compn:uion.
-Tat. Face iacampre_ioo.
~
..... ,I
-1.50% -1,00% ~50% 0,00% 0,50% 1,00% 1,50% 2,00%
Longitudinal strain &" (0/0)
Figure 15 : Bending + torsion test

Moment - longitudinal strain curves comparison
149
width (mm)
4 4.75
Figure 16 : Four points bending test

(Distribution of longitudinal Young' s modulus on the cross section, Iso-damage Surface)
150
References
1. Keyvani, M., Contribution a la caracterisation micromecanique et ullrasOllore du

comportement et de l'endommagement des composites de type SMC (sheet molding
compound), Thesis of Ecole Centrale Paris, France, 1992.
2. Fitoussi, 1., Etude micromecanique de /'influence de I'endommagement a l'interface

a
fibre/matrice sur Ie comportement de composites renforts discontinus, Thesis of ENS AM
Paris, France, 1995.
3. Mori T and Tanaka K, Average stress in matrix and average elastic energy ofmaterials
with misfiring inclusions, Acta Metall Mater, vo121, 1973: 571-574.
4. Tandon G.P., Weng GJ., A theory ofparticle reinforced plasticity, Journal of Applied
Mechanics, March 1988, vol. 55, pp 126-135.
5. Qiu Y.P.,Weng GJ., A Theory ofplasticity for porous materials and particle-reinforced
composites, Journal of Applied Mechanics, June 1992:59:261-268.
6. Hu G., A method ofplasticity for general aligned spheroidal void or fiber reinforced
composites, International Journal of Plasticity, 1996, vol 12, n04, pp 439-449.
7. Mochida T, Taya M, Lloyd DJ, Fracture ofparticles in a particle/metal matrix composite

under plastic straining and its effect on the Young's modulus of the composite, Mat.
Tansactions, 1991:32(10) :931-942.
8. Mura T., Micromechanics of defects in solids, second revised edition, Martinus Nijhoff
Publishers, 1987.
9. Guo G., Fitoussi, 1., and Baptiste, D., Modelling of damage behaviour of a short fiber
reinforced composite structure by the finite element analysis using a micro-macro law,
International Journal of Damage Mechanics, 1997 :6(3) :278-299.
10. Hutchinson J.W., Plastic stress and strain fields at a crack tip. J.Mech Phys Solids,
1968: 16:337-347.
11. Li F.Z, Pan J, Plane-Strain Crack Tip fields for pressure sensitive dilatant materials,
Journal of Applied Mechanics, 1990: 57:40-52.

MECHANICS OF THREE-DIMENSIONAL TEXTILE STRUCTURAL
COMPOSITES: PROCESSING
TSU-WEI CHOU
Center for Composite Materials and Department of Mechanical Engineering
University of Delaware, Newark, DE 19716 USA
TIMOTHY D. KOSTAR
Foster-Miller,1nc.
303 Bear Hill Road, Waltham, MA 02154 USA
1. Introduction
The design and fabrication of preforms for advanced composites have gained considerable
attention in light of the recently developed textile preforming techniques. It is within this
realm of preforming technology that the full advantage of the knowledge of process-
structure-property relations may be rea___ ~d. The fabrication history of these preforms
directly determines composite micro-structure and resulting mechanical properties.
Textile preforms may be loosely classified into two-dimensional (2-D) and three-
dimensional (3-D) structures, depending on the degree of reinforcement between layers
[1]. Figure 1 summarizes the major types and fabrication schemes for textile preforms.
1.1 THREE-DIMENSIONAL TEXTILE PREFORMS
Perhaps the simplest way of adding through-the-thickness reinforcement is the stitching

process. An industrial size sewing machine is usually employed whereby a needle is used
to penetrate the layers of fabric and pull the stitching yarn through the preform. Though
cost-effective, considerable fiber damage occurs due to needle penetration. The resulting
reduction in composite strength can be appreciable, making stitching an unattractive
option. In recent years, novel stitching techniques have been developed where the fibers
are effectively spaced to greatly reduce breakage during needle penetration.
Three-dimensional knitted fabrics are akin to their two-dimensional brothers. They
may be produced by either weft knitting or warp knitting process. Additional
strengthening is accomplished by the use of laid-in yarns in the mutually orthogonal
direction. The knitted preform which deserves the most attention is the multi-axial warp
knit. The knit consists of longitudinal, latitudinal, and bias (+1-8) yarns held together by a
through-the-thickness tricot stitch. These three-dimensional knits possess the
151
152
TEXTILE PREFORMS
I
I
TWO-DIMENSIONAL THREE-DIMENSIONAL
WEAVING
I
KNITTING
I
I
BRAIDING STITCHING WEAVING KNITTING BRAIDING
,l, r-l. J-, r---l ~

biaxial triaxial weft knit warp knit circular flat angle-interlock orthogonal 2-step 4-step solid
Figure 1: The classification oftextile preforms (after ref. 1).
Layer
3 weft yam
a) A sample angle-interlock weave b) A sample orthogonal weave
Figure 2: Examples of angle-interlock and orthogonal 3-D weaves (after ref [1]).
characteristics of unidirectional laminates while enjoying enhanced stiffness and strength

in the thickness direction.
Three-dimensional weaving is achieved through a modification of the traditional 2-D
weaving process. The two main types of 3-D woven fabrics are angle-interlock and
orthogonal structure. Angle-interlock weaving is carried out by utilizing multiple
harnesses on a conventional loom. The shifting sequence of the harnesses determines the
undulation of the warp yarns (Fig. 2a). Many geometric variations are possible due to the
unlimited combinations of loom configuration and harness sequencing. These multi-layer
interlocked structures are ideal for thick-section composites. The reinforcement in the
thickness direction may be tailor designed to enhance composite impact resistance.
Additionally, laid-in yarns may be used to increase stiffness in a desired direction.
Orthogonal woven fabrics possess three sets of mutually perpendicular yarns (Fig.
2b). Inherent in such a structure are matrix rich regions between the intersections of the
three sets of yarns. Fabrication of these preforms is accomplished by inserting
153
a) Horngear arrangement b) Solid braid architectures
Figure 3: Solid braid fabrication and geometry (after ref. [1]).
alternating, in-plane yarns between the stationary thickness direction yarns. In this
fashion, both Cartesian and cylindrical geometries are possible.
Horngear machines with sq e or circular arrangement are employed in the
fabrication of solid braids (Fig. 3a). Present-day machines are limited in the number of
yarn carriers and therefore limit the sizp 'l.nd shape of preform. The micro-geometry of
braid is also restricted and is shown in Figure 3b. As can be seen, the braider yarns form
inter-twined helical paths throughout the structure.
To allow for more flexibility in preform size, shape, and micro-structure, new
braiding processes have been introduced. These include AYPEX, interlock twiner, 2-step
[2], and Rowand Column [3] which is more commonly referred to as 4-step in the
literature. A brief introduction of the 2-step and 4-step processes follows.
1.2 TWO-STEP AND FOUR-STEP BRAIDING PROCESSES
The two-step braiding process is so termed because it involves two, distinct motions
of each yarn carrier. The braid consists of an array of longitudinal (axial) yarns arranged
in a prescribed configuration such as rectangular, circular, box, etc., and braider yarns
positioned at select locations on the perimeter of the axial array. The shape of the axial
yarn configuration will determine the final shape of preform. The braider yarns, which
move along alternating diagonals of the axial array, interlink the axials and hold them in
the desired shape. Figure 4a shows a rectangular arrangement of axial yarns with the
locations and paths of the braider yarns. Figure 4b shows the cross-section and surface of
a composite so produced by such an arrangement. The two-step braided composite not
only offers high stiffness and strength in the axial direction, but also yields enhanced in-
plane reinforcement due to the presence of the braider yarns. Furthermore, the addition or
subtraction of axial yarns during fabrication allows for on-line change of preform cross-
section.
The four-step braiding process involves four distinct Cartesian motions of groups of
yarns termed rows and columns. For a given step, alternate rows (orcolumns) are shifted
a prescribed distance relative to each other. The next step involves the alternate shifting
154
AXIAL YARN • BRAIDER YARN •
step 1 step 2
corner axial 17 side axial V
interior axial. braider ' "
Cross-Section Surface
Figure 4: The two-step process and braided composite.
of the columns (or rows) a prescribed distance. The third and fourth steps are simply the
reverse shifting sequence of the first and second steps, respectively. A complete set of
four steps is called a machine cycle (Fig. 5a). It should be noted that after one machine
cycle the rows and columns have returned to their original positions. The braid pattern
shown is of the 1 x 1 variety, so called because the relation between the shifting distance
of rows and columns is one-to-one. Track and column braiders of the type depicted in
Figure 5 may be used to fabricate preforms of rectangular cross-section such as T-beam,
I-beam, and box beam if each column and row may be independently displaced. Figure
5b shows the cross-section and surface of the square-shaped composite formed from the
configuration shown in Figure 5a. Four-step braided composites offer excellent shear
resistance and quasi-isotropic elastic behavior due to their symmetric, intertwined
structure. However, the lack of unidirectional reinforcement results in low stiffness and
strength, and high Poisson effect. To help eliminate this, some advanced machines allow
for axial yarns to be fed into the structure during fabrication.
155
•
step zero step one step two
• • ••
:
I; :. •
step three step four path of carrier "a"
cross-section surface
••
(a) (b)
Figure 5: The four-step process (a) and braided composite (b).
2. Multi-Step Braiding
Conventional track and column braiders usually only allow for all tracks or columns to be
moved simultaneously. Additionally, they are only able to produce track to column
shifting ratios of a set 1 to 1, 3 to I, etc ... Given the possibility of individual control of
each track and column and a displacement of each of up to three units, much may be
gained by the added flexibility. This extended track and column braiding has been termed
"multi-step braiding" [4]. To better understand the advantages of this extended process, it
has been simulated and the results are explained below. Also, with the added ability to
insert axial yarns into a four-step braid, it will be seen that two-step braids may be
automatically, and efficiently fabricated on a track and column type braider. A
mathematical approach to the treatment of braided structures is presented. An analysis of
the braid architecture and the corresponding process which produces it will form a
relationship between the two and, subsequently, form a foundation for the fabrication of a
variety of structures.
2.1 CARTESIAN BRAIDING SIMULATION
By simulating the sequential Cartesian motion of braider yarn elements, the idealized
spatial orientations of the yarns may be identified. This knowledge, coupled with a model
of yarn packing within the braid, can become a useful tool in the design process.
The "multi-step braiding simulation" is an easy to operate, versatile tool designed to
be used in conjunction with the automated braiding equipment. The essence of this tool is
explained below. A grid consisting of 14xlO squares representing possible carrier
locations is graphically displayed. The user is free to choose any number of squares in
defining a rectangular machine base array and is then prompted to define a single carrier
(square) to be traced. Next, the desired machine cycle is entered. This may consist of any
156
COLUMN
1 2 3 4 5 6 7 8 9 10 11 12 13 14
1
2
3
4
5 ROW
6
7
8
9
10
1] COLUMN MOTION SPECIFICATION

2] ROW MOTION SPECIFICATION
3] END INPUT
ENTER YOUR CHOICE (1,2,3)?
Figure 6: Sample screen output for multi-step braiding simulation
number of steps in a machine cycle (multiple steps) and any desired displacement of
individual tracks/columns. Once finished, the program checks the entered machine cycle
to see if all tracks!columns finish at their initial locations. Figure 6 shows a typical screen
which is prompting for the machine cycle. That is, the program asks for the column/row
shifting sequence.
Once the machine cycle is programmed, the user has the following options: (1) an
animated run through of consecutive machine cycles (step by step) may be carried out
automatically or step-wise controlled by the user. This will continue until the defined
tracer carrier returns to its original position after a set number of machine cycles; (2) the
user may redefine the tracer carrier; (3) the user may view the path of the tracer carrier
on the machine bed; (4) the planar projection of the tracer yarnfs spatial orientation onto
the sides of the simulated braid may be seen; (5) the program can determine the number
of yarn groups, the number of yarns per group, and their location on the machine bed; (6)
the unit cells, or the repeat geometry, of the simulated braid may be determined; and (7) a
new bed arrangement may be set.
2.2 YARN GROUPS AND REPEAT GEOMETRY
Yarn groups are sets of yarn tows that travel the same path. A multi-step braiding process
may have multiple yarn groups and a varying number of yarns per group. The process
simulation software records the location of a given yarn carrier at the end of each machine
cycle until the carrier returns to its original position. The data is stored in a two-
dimensional array where by the same numerical value of an array element signifies a yarn
157
in the same group. This process is continued until all the array elements are linked with
numerical values. The simulation then displays the results by showing a single letter for
the yarns in a given group.
The existence of yarn groups implies that sets of yarns trace the same path on the
machine bed. After one complete machine cycle, each yarn in a group has moved to its
leading yarnfs location. This in turn implies that the braid geometry produced during one
machine cycle (repeat) is the repeating geometry for the entire structure. That is to say, a
cross-sectional slab of preform with the length produced during one repeat may be
"stacked -up" on top of one another to reproduce the entire preform. It may now be seen
that knowledge of this repeat braid geometry is essential for future prediction of braided
composite properties.
The process simulation records the location of each yarn carrier at discrete time
increments; initial location, location every two steps (column I track shift) of the complete
cycle, and location at one complete machine cycle. The data are contained in separate
two-dimensional arrays which are linked by the numerical values of the array's elements.
A three-dimensional view of the repeat braid geometry may be graphically displayed
where differently colored line seb ___ ..mts represent the individual yarns of the structure.
These line segments are an idealization since actual yarn tows within the braid will
deform and translate due to interactior ,etween yarns. Examples of the repeat braid
geometry produced by the above mentioned bed arrangement and process (here shown in
black and white) will be shown later. This information about braid geometry may be
incorporated into an analytical model for prediction of composite properties.
2.3 EXAMPLE BRAID CYCLES
Along with the possibility of having independent control of track/column motion and
displacement comes the possibility of designing new braid cycles and, as a result,
innovative braid structures. It may readily be seen that quite a large number of potential
braid cycles exist. With this in mind, we introduce a few such braid cycles.
To exemplify the applicability of individual track/column displacement, consider the
braid cycle shown in Figure 7. Here, a base array of 4x4 yarns is used. This process is
labeled 4-step 211 x2lI because there are both 2 and I units of displacement for the rows
and columns. The cycle yields a total of six yarn groups as is also shown in the figure.
The groups "e" and "f' tend to remain in the center of the braid thereby serving to
interlace the core of the structure with the other yarns. This allows for a distinction
between yarns in different yarn groups leaving more room for design of the braid
structure. Figure 8 shows the repeat braid geometry of the resulting architecture.
The process simulation is termed "multi-step" because any number of steps may be
specified in a given machine cycle. As an example, consider the machine cycle depicted
in Figure 9. The cycle consists of eight steps with a one unit displacement for each.
Again, a 4x4 base array was used. Notice the number and location of the yarn groups.
Groups "a" and "d" tend to occupy the corner locations while groups "b" and "c" the sides
and interior. This suggests an application to hybrid composites where high performance
fibers may be incorporated into the braid structure where added stiffness is needed while
158
--....
r-
I
step 1 step 2
--
.... Group "a" has 6 yarns
Group "b" has 6 yarns
Group "c" has 6 yarns
Group "d" has 6 yarns
Group "e" has 2 yarns
Group "f' has 2 yarns
step 3 step 4
(a) (b)
Figure 7: Four-step 211 x 2/1 braid cycle and yarn groups. (a) A four-step 211 x 211
braid cycle. The heavy arrows indicate two units of displacement. (b) location of
yarn groups
Figure 8: Sample simulation output of repeat braid geometry

159
a b
step 1 step 2 step 3 step 4
Group "a" has 5 yarns

step 5 step 6 step 7 step 8 Group "d" has 5 yarns
(a) (b)
Figure 9: Eight-step 1 x 1 braid cycle and yarn groups. (a) Eight-step 1 x 1 braid
cycle. (b) Location of the yarn groups.
~~II!:J (ste~:l & 2)
J hl4
(steps 3 & 4)
J hl4
(steps 5 & 6)
J(ste~:7 & 8)
~
Figure 10: Simulation output ofrepeat braid geometry for eight-step 1 x 1 braid cycle
160
....- ® Yam Carrier
0
~
Step Two Empty Carrier
....-
~
• Axial Yam
+t +t +t Step One
Figure 11: Schematic showing selection and path of yarn carrier of four-step yarn array
for the two-step process. The yarn which occupies the initial position of "0" starts on a
side having 3 axial yarns and takes 2+ 1-3 machine cycles to complete its "two steps."
the structure still benefits from the structural and delamination resistant properties of the
3-D braid. Figure 10 shows the repeat geometry of the structure.
3. Comparison of Two-Step and Four-Step Braiding
A four-step Ixl pattern was run on the simulation and the paths of selected tracer yarns
examined. It was found that a certain sub-set of the outer yarns for a given array
sequentially traced the paths of a group of braider yarns in a two-step process. This
suggests, and was later verified, that two-step braiding is, in fact, four-step braiding with
axial yarn insertion and employment of selected outer yarn carriers. Figure 11 shows a
four-step 1 x 1 bed pattern which utilizes only selected outer yarn carriers. The black dots
represent the axial array, here, a 3 x 2 array is shown. The path of one such braider yarn
is seen to trace the familiar two-step pattern. In general, for a two-step axial array of m x
n, a four-step yarn array of 2m x 2n is needed. This assures an even number of tracks and
columns, which is required due to the symmetry of the process and resulting structure.
After further investigation, it was also found that a braider yarn which starts on a side
having m axial yarns will require n+ 1 machine cycles to complete its two steps where m
and n are interchangeable.
4. Concluding Remarks
This paper has examined the recent developments in multi-step braiding processes for
manufacturing of preforms for composites. The Cartesian braiding simulation has
161
resulted in the unification of 3-D braiding. No longer may we speak of two-step or four-
step braiding as unique processes. They now become special cases of general "multi-
step" braiding where axial yarns may be inserted at will and repeat geometry, along with
yarn groups, tailor designed. That is to say, the door to the design of composite
microstructure, hybrid structures, and composite cross-sectional shape has been opened.
The methods employed to reach these goals will be discussed in detail in the following
paper.
References
1. Chou, T-W. (1991) Microstructural Design of Fiber Composites, Cambridge

University Press, Cambridge, England, chapters 6 and 7.
2. McConnell, R. and Popper, P. (1988) Complex Shaped Braided Structures, U.S.

Patent 4,719,837.
3. Florentine, R. (1982) Apparatus for Weaving a Three-Dimensional Article, U.S.

Patent 4,312,261.
4. Kostar, T.D. and Chou, T-W. (1994) Process Simulation and Fabrication of
Advanced Multi-Step Three-Dimensional Braided Preforms, Journal of Material
Science, 29, 2159-2167.
MECHANICS OF THREE-DIMENSIONAL TEXTILE STRUCTURAL
COMPOSITES: ANALYSIS
TSU-WEI CHOU
University of Delaware, Newark, DE 19716 USA
TIMOTHY D. KOSTAR
Foster-Miller, Inc.
303 Bear Hill Road, Waltham, MA 02154 USA
1. Introduction
This paper focuses on the analysis of the process-structure relationship of three-

dimensional braided preforms. The final architecture and structure of the braided preform
depends on many factors. These include the density and packing of the yarns used, the
length of braid formed during one complite shifting sequence (pitch length), the overall
fiber volume fraction, and the final preform dimensions. It is desirous to relate the
control parameters such as yarn density, shifting sequence, and pitch length to the
dependent parameters such as yarn orientation, preform dimensions, and volume fraction.
With this knowledge, design of desired preform architecture may be realized [1].
2. Yarn Packing Within Preform
From an observation of the transverse cross-section of four-step preforms, yarn packing

may be idealized as shown in Figure la. The arrows in Fig. la represent the directions
that the base of the yarns are moving within the braiding plane. The idealized
arrangement is due to the shifting sequence of individual yarns and the movement of
yarns within the preform during fabrication. This may be attributed to yarn tension and
interaction. An individual yarn's elliptical cross-section may be partially inscribed inside
a rectangle as suggested in Figure 1b. Since the four-step process is a specific example of
the more general multi-step braiding, it is safe to assume that similar yarn packing will
occur in any 3-D braid formed by a row/column process. In other words, the transverse
cross-section of any multi-step braid may be idealized as elliptical yarn cross-sections
arranged in the fashion shown in Fig. la. This follows from an assumed circular yarn
cross-section.
163
164
fly
a) Assumed yam packing b) Inscribing rectangle
Figure 1: Idealized yarn packing in preform transverse cross-section. (a) Assumed

yarn packing in three-dimensional braid. (b) Model of inscribing rectangle.
i=1;!"
k hI(N121~~~~=--L:..
x~
j=2
Figure 2: Definitions of effective yarn diameter (d), pitch length (h), rotation angle
(a), and inclination angle (0) for a given yarn.
The dimensions of the inscribing rectangle (2u x 2v) may be related to the in-plane
rotation angle, inclination angle, and pitch length as defined in Figure 2. From Fig. 2, the
values of i and j may be directly related to row/column shifting distance of individual
yarns. The inscribing rectangle half-width u is found by imposing a tangent condition
between the line L1 and the ellipse with minor and major axes a and b, respectively
(Figure Ib). Since the ellipse represents a circular yarn cross-section cut at angle withe
respect to yarn axis, the ellipse minor axis a may be assumed to be half the effective yarn
diameter d. To this end, we may start with the equation of an ellipse in the x'-y'
coordinate system
(1)
165
The relation between the minor and major axes a and b, respectively, of an ellipse
which is formed by cutting a circular rod of radius, a, at an angle, e, with respect to the
rod axis is given simply from
b=_a_ (2)
sin e
Also, from Figure 2, the following important relation may be obtained
tan9= h(N/2) (3)

~(j2u y + (i2v y
where N is the total number of steps in the braid cycle, and i and j are the non-
dimensional, in-plane displacements of individual yarns. Equation (3) represents a
complex relation between the control parameters h, d, i, j and the yarn orientation
parameters ex. and e. It should be tovident that the value of the rotation angle ex. may be
estimated from
i
tana=- (4)
j
After defining the non-dimensional pitch hd as
h
h =-
d d
(5)
and carrying out the necessary manipulations, a more useful equation expressing hd as a
function of e, i, and j may be obtained
(6)
where
f = (i 2 Ysin 2 e+ l Xi 2 + y)
i 2 Ysin 4 e+ l sin 2 e+ i 4 sin 2 e+ i 2 Y _i 2 j2 cos 4 e
and
166
25
EI I x I aJpha=45 deg.
20
• I x2 alpha=26.2 deg .
• Ix3 alpha=18.4 deg.

~ 2x3 aJpha=33.7 deg.
15
--=
..c: 10
~
5
o~ ______ ~ __ ~ __ ~ __ ~ __ ~ __ ~ __ ~
30 35 40 45 50 55 60 65 70
theta (deg.)
Figure 3: Non-dimensional pitch (hid) vs. inclination angle (9) for various rotation
angles (a) where N=4.
The quantity in brackets is subject to the minimum value condition [1].
The NI2 in Eq. (6) represents the fraction of the pitch length displayed in Fig. 2.
e
Figure 3 shows the relation between hd and for different values of a. (i I j). Here, N=4
which corresponds to a four-step process. In other words, the relation displayed in Fig. 3
is valid for one-half of the pitch length or two out of the total of four steps. Alternatively,
given a higher number of total steps such as 6 or 8, Fig. 3 would be scaled according to
Eq. (6). As a result, the greater pitch length would make room for the increase in the
number of slabs formed during each of the two steps. It should be noted that Eq. (6) is
true for each yarn in the preform. Therefore, it must be applied on a yarn-to-yarn basis
and will not account for yarn interaction. For example, if each yarn has an effective
diameter d, and given a pitch length h, an individual yarn in the preform which has been
e
displaced i by j must be inclined at an angle as dictated by Eq. (6). Viewed in this way,
Eq. (6) is seen to be a design constraint, which is valid for every two steps of a multi-step
process. Note that yarn angles are calculated every two steps of a multi-step process. The
rationale for this is that change in an individual yarns orientation may occur at two step
intervals.
3. Yarn Volume Fraction
When determining final composite properties, it is important not only to have knowledge
of fiber orientation, but fiber volume fraction as well. To start, we may define yarn
167
volume fraction Vy as total yarn volume in a preform cross-sectional slab of height

h/(NI2) divided by the volume of the slab. The total volume of the yarns will be given by
V =1td 2 "'1 (7)

yarn 4 £.. lj
where 1.. is the length of a yarn displaced by i and j in the column and row direction,
IJ
respectively, and is given by
(8)
The summation in Eq. (7) is carried out for all the yarns in the slab and it is assumed that
all yarns have the same effective diameter d. The volume of the slab is expressed as
v = (_h_k A (9)
,I,b ,,{N/2}) ij
where Aij is the area of the inscribing rectangle of a yarn displaced by i and j. This area
is found simply from
Ai; =4uv (10)
where u and v are determined on a yarn-to-yarn basis and will vary for yarns with
different orientations. Again, the summation is applicable to all the yarns in the structure.
Combining Eqs. (7) through (10) we arrive at an expression for yarn volume fraction
(11)
Note that u and v in the above expression are the inscribing rectangle half dimensions
corresponding to the yarn of length lij. To determine the actual fiber volume fraction, we
need only to multiply yarn volume fraction by the appropriate packing fraction for the
given yarn tow where packing fraction is defined as ratio of total fiber cross-sectional area
to total cross-sectional area of a yarn tow. Note that fiber volume fraction defined in this
way is a piecewise calculation for each fractional segment of the total pitch length. A
simple average over the entire pitch length may be used for estimating total fiber volume
fraction.
Due to the deformability of preform cross-sectional shape, yarn volume fraction of
preform has little meaning. A more useful and measurable quantity that does not depend
on cross-sectional area may be introduced. This quantity is the volume of yarn in a given
168
0.26 -r----------------.....,
0.24
i:::!,
-= 0.22
~
..=
""...
(U
0.20
0.18
3.2 5.2 7.2 9.2 11.2 13.2 15.2 17.2 19.2
hid
Figure 4 Yarn volume fraction per pith length vs. non-dimensional pitch.
fraction of pitch length divided by that fraction of pitch length. Again, this quantity must
be calculated for each slab formed every two steps of a multi-step process since it need
not be continuous. To obtain this relation, we need only to combine Eqs. (7) and (8) and
divide Vyarn by the appropriate fraction of pitch length. By utilizing this, the final
composite fiber volume fraction may be predicted by simply dividing Vyarn /[hI(N/2)] by
the composite cross-sectional area. Figure 4 shows how Vyarn /h varies with non-
dimensional pitch for a standard four-step lxl braiding process and a yarn diameter of
0.7391 mm (0.0291 inches). Here, Vyarn /h is used since the fiber volume fraction of a
four-step lxl braid is continuous in the length direction.
4. Design of Braid Structures
To fully realize the usefulness of the above simulation and analysis capabilities, computer
code was developed to tie together the analytical findings in an intelligent fashion. The
user starts by specifying desired preform size (i.e. wand t) and effective yarn diameter (d)
to be used. Next, in-plane rotation angle (a.) and inclination (9) angle are chosen and the
corresponding percentage of yarns at this orientation specified. This continues until 100%
of the yarns have a specified orientation. The program will give the orientation with the
highest percent of yarns priority and will remain as close to this orientation as is allowed.
For this orientation, an hid value is calculated and used in the following two cases.
169
25~----------------' 25~----------------,
" Ix! alpha=4S " lxl alpha=45

20 20
• lx2 alpha=26.2 • lx2 alpha=26.2
• lx3 alpha=18.4 • lx3 alpha=18.4
• 2x3 alpha=33.7 • 2x3 alpha=33.7
15 15
old alpha
10 10
esired new alpha
alpha calculated
hid =;rt:;;::;::;;.....~=z....;...---:::;;:rf'....
o+-~~~~~~~~~
30
prime theta old theta new theta prime theta desired theta
theta (deg.) theta (deg.)
a) Priority put on rotation angle b) Priority put on inclination angle
Figure 5: Illustration of optimization procedures used in design program. (a) User

priority put on rotation i:lIlgle. (b) User priority put on inclination angle.
First, the user may place importa~~~ on rotation angle a. In this scenario, new q
values are found for the specified a values and calculated hid value. For example,
consider Fig. 5a. Here, the user has specified the highest percentage of yarns at a=43
degrees and a=26.2 degrees. For this ala. combination, an hid value of approximately 5.5
is calculated. The user then specifies a ala. combination of 53 degrees and 45 degrees,
respectively, for the remaining percentage of yarns. However, the calculated hid value
will not allow this second ala. combination. With user priority being placed on rotation
angle a., a new a value is found which corresponds to the desired a value and calculated
hid value.
Second, the user may place importance on inclination angle a. For this case, new
a. values are found which minimize the total error in desired a values and calculated hid
value. This procedure is exemplified in Figure 5b. Here, the user has specified the
highest percentage of yarns at a and a. values of 41 degrees and 26.2 degrees,
respectively. The remaining percentage of yarns is originally specified at a and a. values
of 62 degrees and 26.2 degrees, respectively. Since the calculated hid value will not allow
this second ala. combination, a new a. value must be found which minimizes any changes
to the high priority ala. combination. Once completed, final orientation, pitch length, and
machine set-up are displayed in either tabular or graphical form.
Overall, simulation of the 3-D Cartesian braiding process has given insight to the
placement offibers within a braid for a given track I column shifting sequence.
S. Complex 3-D Braids
The relationship between the Cartesian braiding process and the resulting braid
architecture establishes a method of tailored fiber placement. The importance of this to
170
the design of fibrous composites is revealed when one considers hybrid composites,
composites with complex cross-sectional shapes, and textile composites with surrogate
material added or removed. A tried-and-true approach to the fabrication of these unique
materials is required to fully exploit the advantages of 3-D "multi-step" braiding [2].
The recent development of three-dimensional, multi-step braided composites has
given birth to other exciting design possibilities. One of these design approaches is the
grouping of yarns at a desired location in order to form a hybrid composite. Hybrid
design was studied via. the aforementioned simulation of the multi-step braiding process.
It was found that an iteration procedure is first required to insure the desired yarn
grouping is achieved. After further study, some general rules governing the braiding of
hybrid composites were discovered. Both the iteration procedure and the fundamental
rules are discussed below.
Yarn groups are sets of yarn tows which travel the same path. A four-step braiding
process will yield either one, two, or more yarn groups depending on factors such as the
even/odd number of tracks and columns. In addition, as previously shown, these groups
will always form in a symmetrically distributed pattern and contain the same number of
yarns. However, a multi-step braiding process may have multiple yarn groups and a
varying number of yarns per group. Different braiding schemes will yield different
distributions of yarn groups. For exarr:-'~, certain track and column shifting sequences
will group certain yarns near the sides or interior. This phenomenon has a direct
application to hybrid composites where high performance yarns such as carbon may be
placed only where needed.
Suppose that individual control of each track or column displacement is possible (up
to three units) and any number of steps may be specified in a given braid cycle. To
exemplify this idea, consider the braid cycle depicted in Figure 6. The cycle consists of
eight steps with a one unit displacement for each step. For simplicity, a square base array
of 4x4 is used. Notice the number and location of the yarn groups. Groups "a" and "d"
tend to occupy the corner locations while groups "b" and "c" the sides and interior. It is
readily apparent that knowledge of the relationship between the braid cycle and this
grouping phenomenon would open the door to the design of hybrid composites.
Preforms of a variety of cross-sections may be fabricated through a suitable use of
rectangles. Because of this, we will focus our attention on the design of hybrids (or
grouping of yarns) within a rectangular domain. In addition, due to the extremely large
number of braiding sequences which will yield the defined grouping of yarns, an
analytical solution to this problem is not plausible. Instead, we are forced to limit our
approach to one of simulation.
The simulation approach has its disadvantages. However, it is a pragmatic means to
determine unique braid cycles that yield the desired grouping of yarns. The idea is simply
to iterate through all possible track and column shifting sequences, in an intelligent
fashion, until the appropriate braid cycle is found. Additionally, through this approach,
knowledge is acquired which allows for the development of fundamental rules.
First, the user inputs the effective yarn diameter (d), the braid pitch length (h), and
the macro-dimensions of the preform. Using the established geometric models for the
size of the inscribing rectangle, the size of the base array (m x n) is calculated. Sides,
corners, and interior are then defined and the specific desired grouping locations are
171
Group "a" has 5 yarns

Group "d" has 5 yams
a) Eight-step 1 by 1 braid cycle b) Location of yam groups
Figure 6: Eight-step 1 x 1 braid cycle and yarn groups.
r-
fo-
f0- r-
I I
I
I
I I
- ..... ~
fo-
'-
.....
Define Specify Locations Generate Shifting

Locations and Percentages Sequence
Figure 7: Approach to the design of a hybrid composite.
identified. The percentage of yarns that start in these identified locations is also specified
and the iterations are started (Fig. 7). A shifting sequence is then found which yields a
minimum of these percentages at the desired locations. If the resulting braid architecture,
as seen through the simulation software (i.e. color coordinated to display distribution of
yarn groups), is not satisfactory, the iterations may be continued. Ideally, this iterative
procedure may be expanded upon. By coupling the idealized architecture of the
simulation with the idealized yarn packing arrangement, the effective elastic properties
may be quickly calculated. If these are not suitable for the intended application, the
iterations may continue. Reference [2] gives an extensive coverage on the analysis and
design of complex 3-D braids.
172
This paper has outlined the methodology for analyzing Cartesian braiding to fully
understand the limitations and potentials of the process. This methodology enables the
design of unique and complex braid structures.
References
1. Kostar, T.D. and Chou, T-W. (1994) Microstructural design of advanced three-
dimensional braided preforms, Journal o/Composite Materials, 28, No. 13,
1180-1201.
2. Kostar, T.D. (1998) Analysis, Design, Fabrication and Performance of Three-

Dimensional Braided Composites, Ph.D. dissertation, University of Delaware,
Newark, Delaware, USA.
MECHANICS OF THREE·DIMENSIONAL TEXTILE STRUCTURAL
COMPOSITES: PERFORMANCE MODELING
TSU-WEI CHOU
University of Delaware. Newark. DE 19716 USA
KISHORE POCHIRAJU
Stevens Institute of Technology
Hoboken. NJ 07030 USA
1. Introduction
This paper presents a model for the prediction of anisotropic elastic stiffness of 3-D
textile structural composites based on the preform architecture and the preforming process
parameters. The methodology employed in this paper uses preform modeling to represent
the fiber microstructure and uses micro-____chanical analysis to determine the macroscopic
anisotropic stiffness. Representative volume elements (macro-cells) are identified for
three-dimensionally woven preforms and geometric descriptions of the path and
interactions between the tows within the cells are generated. This geometric modeling
enables the estimation of the fiber volume fraction and the directional distribution of the
fiber in the composite. The macro-cell is further decomposed into simpler elements
whose stiffness is determined from tow geometry and the stiffness of the fiber/matrix
constituents. The macroscopic stiffness of the textile composite is obtained from the
macro-cell definition using an Effective Response Comparison (ERC) technique.
Numerical results are compared with data from experimental characterization of several
woven composites.
2. Geometric Modeling of Fiber Preform
The elastic stiffness of the 3-D textile structural composite is determined based on the
fiber and matrix material properties and a model of the fiber microstructure. The
development of the geometric model for the fiber microstructure is based on the following
assumptions:
173
174
Prefonning
Method and
~;S,," -v,'W
Geometric , Micro-mechanical
Modeling .--"'"""''''"---. Modeling
M:· "
'11">-/
Assembly Scheme
Macroscopic
Elastic Stiffness
of Textile Composite
Figure 1: Methodology for elastic stiffness determination
1. The microstructure of the composite is deterministic and spatially periodic.

This enables the identification of a representative volume element (macro-
cell). The properties and performance of the macro-cell represent the
macroscopic composite properties.
2. The textile composite microstructures can be categorized into

classes/families based on the fiber preform manufacturing processes.
Macro-cell representations are constructed such that they are valid for
several possible variations of fiber architectures in the identified preform
class.
3. The mechanical behavior of a textile composite is dependent upon the

preform consolidation process. But, in composites with high strength and
stiffness fibers and lower modulus resin, the preform construction and fiber
orientation within the preform are expected to be the major factors in
determining the behavior.
Figure 1 illustrates the proposed approach. The model may be divided into geometric
modeling phase and micro-mechanical modeling phase. The geometric modeling phase
entails a) the fiber volume fraction and directional distribution of the fiber volume using a
macro-cell representation for the preform, b) modeling individual yarn paths using a unit-
175
cell decomposition, and c) establishing the geometrical relationships among the yarns
using an "assembly model". Micro-mechanical modeling phase of the proposed model
determines a) the stiffness of the individual unit-cells, b) load sharing relationships among
the unit-cells with the aid of assembly model, and c) the macroscopic stiffness of the
macro-cell (hence the composite) using an averaging method called Effective Response
Comparison (ERC).
In the geometric modeling phase of the stiffness prediction, a parametric
representation of the microstructure of the 3-D textile composite is constructed. The
geometric models consider the fiber preforming process and the possible resulting
variations in the composite microstructure. The geometric modeling phase can be further
sub-divided into four specific tasks: (1) determination of the size and shape of the macro-
cell for each class of fiber preforms (for example: layer-to-layer angle interlock woven,
through-the-thickness angle interlock woven, and 3-D triaxially braided) (2) scheme for
estimation of the fiber volume fraction and distribution, (3) units cells and their
parameterization, and (4) assembly model that identifies the load sharing relationships
between the unit-cells. A complete list of preform parameters, their descriptions and
example values are given in Table 1.
2.1 ANGLE-INTERLOCK WOVEN P...... FORMS
Figure 2 shows the geometric model for layer-to-layer angle interlock weaves. The
architecture of these angle-interlock woven preforms can be represented in the Cartesian
coordinate system. The x-axis is considered along the warp-weaver yarn direction or the
length-wise textile forming direction (machine direction). The y-axis is along the fill
direction or the weft direction of the preform. The z-axis is oriented in the thickness
direction of the preform. Straight warp-stuffer yarns are employed to increase the
longitudinal stiffness in these preforms. The macro-cell of the layer-to-layer structure
consists of half-a-cycle of warp-weaver yarns in the longitudinal direction and entire
thickness of the preform. The width of the macro-cell is determined from the warp-
stuffer and warp-weaver spacing along the weft-yarn direction. The larger one of the two
spacings, lIS, and llW, is considered as the length of the macro-cell along the weft
direction.
The macro-cell of the through-the-thickness angle interlock is topologically similar
and also has a length equal to half-a-cycle (half-pitch) of the warp-weaver yarn. However,
the warp-weaver yarns interlock all the weft yarn stacks present along the thickness and
therefore form a larger macro-cell than the layer-to-layer angle interlock composites.
Given the linear spacing of the yarns and the number of layers per thickness and the
fiber diameter (d s for warp-stuffer, df for weft, and dw for warp-weaver) for both the
classes of the woven composites, the fiber volume fraction within the yarns and
distribution of the fiber for warp-stuffer, weft and warp-weaver yarns can be computed
from the preform parameters and the tow size.
Adding the contribution to fiber volume from each of the yarns in the macro-cell and
dividing by the volume of the macro-cell determines the total fiber volume fraction within
the composite. For layer-to-layer and through-the-thickness woven composites, the fiber
volume fraction, can be determined. The layer-to-layer angle interlock preform described
176
Actual Microstructure
p
Weaver Yarn
j
TIS :
t
warp-Stuller Yarns
Macro-Cell
Idealized Microstructure
r
TfW !.-.
TIFrt . - . : . ,
tJ
.W'~Srutf"
f"::~
Wuv., ~ T
• • •
T • iet
=..::-I.I --.. L.
"51
Macro-Cell and Geometric Parametrization

Figure 2: Geometric modeling for layer-to-Iayer angle interlock weaves.
177
in Table 1 is utilized as an example to illustrate the present methodology. All the yarns
are made of Hercules-AS4 carbon fibers with mentioned tow sizes. The fiber properties
are considered isotropic and are given by the Young's modulus E= 235 GPa (34.11 MSI)
and the Poisson's ratio Vr = 0.3. The matrix is 3M-PR 500 epoxy resin with E = 3.45 GPa
(0.5 MSI) and Vm = 0.3. The geometric modeling phase provides the necessary outputs to
the micro-mechanical models. Then, the total fiber volume fraction and the relative
distribution of fiber volume within each of the yarns are given in Table 2.
TABLE 1: Preform Parameters for woven preform architectures
Architecture Parameter Description Example Values i

P Pitch 3.0 mm (0.12 in)
W Width 1.47 mm (0.06 in)
SI Stuffers/inch 17.5
WI Weavers/inch 7
Layer-to- FI Wefts/inch 15.5
Layer St StufferslThickness 5
Woven Wt WeaverslThickness 4
Ft WeftslThickness 6
Ts Stuffer Fiber Count 6000
Tw Weaver Fiber Count 3000
Tf Weft Fiber Count 6000
P Pitch 3.0 mm (0.12 in)
W Width 1.47 mm (0.06 in)
SI Stuffers/inch 17.5
WI Weavers/inch 7
Through-the- FI Weftslinch 15.5
Thickness St StufferslThickness 5 I I
II
Woven Wt WeaverslThickness 1
Ft WeftslThickness 15.5 !
Ts Stuffer Fiber Count 6000
Tw Weaver Fiber Count 3000
Tf Weft Fiber Count 6000
TABLE 2: Preform Parameters for woven preform architectures
Volume Relative Volumf

Fraction Fraction
Warp-stuffer fiber volume fraction 25.6 % 44.37 %
Weft (fill) fiber volume fraction 27.2 % 47.16 %
Warp-weaver (z) fiber volume fraction 4.9% 8.47 %
Total fiber volume fraction (V f) 57.6 %
178
3. Anisotropic Elastic Stiffness Modeling
The preform geometric modeling determines the fiber volume fraction and fiber
orientation distribution within the preform. Also, an approximation for the yarn path
within the macro-cell -- straight, inclined, and undulating -- can be identified. The micro-
mechanical modeling is conducted with this information as the starting point. The
assumptions made in the elastic stiffness modeling are listed as follows:
• The macro-cell is further decomposed into mechanically simpler unit-cells

containing individual yarns in the macro-cell. Each yarn present in the
macro-cell is typically considered as a unit-cell.
• The unit-cell stiffness is considered anisotropic but spatially invariant. A

constant stiffness and stress and strain components are assumed through-out
the yarn. Hence, only the macroscopic yarn behavior is considered.
• The load and strain distribution relationships among the fiber yarns and the
interstitial matrix material in a macro-cell are identified using the geometric
relationships between the unit-cells. A mechanical assembly model is
utilized to describe these relationships.
3.1 MACRO-CELL AND UNIT-CELL DECOMPOSITION
The subdivision of the macro-cell into unit-cells for stiffness modeling is accomplished as
follows. The yarns with a characteristic path within the macro-cell, such as warp-stuffer
(straight path or slightly undulating path), weft (straight path or slightly undulating path),
and warp-weaver yarn (undulating path) are identified along with the period of repetition
for that yarn. A unit-cell is defined to consist of the simplest repeatable geometry of a
yarn and each yarn segment present in the macro-cell is associated with one unit-cell.
3.2 STIFFNESS OF UNIT-CELLS
The stiffness of a straight yarn is determined from the fiber and matrix properties and a
local fiber volume fraction called the packing fraction (K.). This local volume fraction in
a 3-D textile composite varies throughout the yarn path and location. The local packing is
also influenced by the processing variations. Using the Concentric Cylinder Assemblage
(CCA) model [1], the elastic properties of a straight tow are computed from the fiber and
matrix properties using the defined packing fraction. The transformation of the stiffness
required for unit-cells with inclined yarns is conducted using tensor transformations and
Tsai-Pagano invariants.
179
Effective Phase with A ve,aged

Warp-slUffe, & Warp-weave,
Unit-Cell Propenies ...---------~
Woven Archirectures
Woven Architectures
Figure 3: Assembly model for woven composites
3.3 LOAD SHARING RELATIONS
Based on the assembly model (Figure 3), the relationships between the stress and strain
components of the unit-cells and those of the macro-cell (and hence the composite) are
defined. These relationships can be interpreted as a multi-dimensional network of springs
that relate the unit-cell strain and stress components to that of the effective response of the
macro-cell.
3.4 EFFECTIVE ELASTIC PROPERTIES -- ERC APPROACH
Several existing models for the elastic property determination of textile composite
materials follow stiffness and/or compliance averaging schemes to obtain the effective
properties of a representative volume element (RVE). Modeling the fiber reinforcement
geometry within the RVE, elastic modulus prediction methods determine the stiffness and
compliance properties of constituent yarns. The elastic properties of the RVE are then
determined through a stiffness or a compliance averaging scheme. The present method
differs significantly from some of the existing models in that we use a stress/strain
component level averaging rather than a stiffness/compliance matrix level averaging. The
effective stiffness of a woven composite in terms of the stiffness matrices of the unit-cells
and their respective volume fractions has been obtained using the Effective Response
Comparison (ERC) approach.
4. Model Prediction-Experimental Data Correlation
The results obtained from the tensile testing have been compared with the predictions
from the model. Table 3 shows the elastic modulus, Poisson's ratio and shear modulus
predictions obtained from the models. Figures 4 and 5 show the correlation between the
predicted Young's moduli, Ell and E22 , and the experimental measurements. The
predictions are seen to be within 10% of the experimental values. Hartranf et. at. [2]
utilized the present model for the prediction of the elastic modulus of thick textile
composites with similar results.
180
TABLE 3: Predicted elastic properties
Architecture Ell E22 E33 VI2 V23 V32 G23 G31 G12
GPa GPa GPa GPa GPa GPa
(MSI) (MSI) (MSI) (MSI) (MSI) (MSI)
LTL-I 73.44 65.80 10.27 0.08 0.46 0.38 8.41 11.36 7.03
(10.66) (9.55) (1.49) (1.22) (1.65 ) (1.02)
LTL-II 74.69 66.97 10.34 0.086 0.44 0.386 8.96 10.54 7.1
(10.84) (9.72) (1.50) (1.3) ( 1.53) (1.03)
LTL-III 74.07 63.59 9.69 0.092 0.40 0.394 8.19 8.61 6.59
(10.75) (9.23) (1.407) (1.19) ( 1.25) (0.95
7)
TIT-I 68.48 66.00 11.30 0.097 0.346 0.315 8.34 10.27 6.88
(9.94) (9.58) (1.64) (1.21) (1.49) (0.99
9)
ITT-II 68.55 66.90 10.68 0.093 0.377 0.34 8.27 10.13 6.8 1
(9.95) (9.71) (1.55) (1.20) ( 1.47) (0.98
8)
TIT-III 67.38 63 .66 9.92 0.089 0.425 0.36 8.06 10.89 6.46
(9.78) (9.24) (1.44) ( 1.17) (1 .58) (0.937)
15.00
Experimental Measurement
~
o Theoretical Prediction
C/)
~
:: 10.00
UJ
(/)
:::I
"S
-0
0
:2 5.00
.~
iii
CD
iIi
0.00
L TL-I LTL-II TTT-I BRO-I BRO-II
Preform Architecture
Figure 4: Correlation between experimental and predicted modulus. Ell" BRD - I and
BRD - II indicate two types of 3-D braided composites; these data are given for
comparison purpose.
181
15.00
(SI Experimental Measurement
o Theoretical Prediction
({)
~
W !:l 10.00
5.00
0.00
LTL-I L TL-II TTT-I BRD-I BRD-II
Preform Architecture
Figure 5: Correlation between experiJ ntal and predicted modulus, E22 . BRD - I and
BRD - II indicate two types of 3-D braided composites; these data are given for
comparIson purpose.
References
1. Hashin, Z. (1983) Analysis of composite materials --- A survey, Journal of

Applied Mechanics, 50:481.
2. Hartranft, D, Parvizi-Majidi, A, and Chou, T-W. (1995) Modeling and

Characterization of through-the-thickness properties of 3-D woven composites.
In Mechanics of textile composites conference, NASA Conference Publication
3311, page 251. NASA.
TESTING AND MODELING OF INELASTIC REHAVIOR OF
FIBROUS COMPOSITES
C. T. SUN
School ofAeronautics & Astronautics
Purdue University
West Lafayette, Indiana 47907-1282, U.S.A.
Abstract
An orthotropic plasticity model was developed for unidirectionally reinforced fibrous

composites. A simplified nl'lStic potential function with a single parameter was
proposed for the flow rule. The parameters of the plasticity model were determined by
simple tension tests on off-axis composite specimens. It was shown that this plasticity
model was able to characterize the nonlinear behavior of polymeric and metal-matrix
composites.
1. Introduction
Fiber composites exhibit physically nonlinear behavior. This nonlinearity is especially

pronounced in metal-matrix and thermoplastic composites. A number of models have
been used to describe the nonlinear stress-strain relationship. Hahn and Tsai [1]
employed a complementary elastic energy density function that contained a bi-quadratic
term for in-plane shear stress. As a result, the strain-stress relation is linear in simple
longitudinal and transverse extensions. Moreover, the shear strain is a quadratic
polynomial function in the shear stress, which is unsuitable for describing the nonlinear
behavior exhibited by many composites. Assuming that fibers are linearly elastic, Sun
et al. [2] modeled the composite by employing nonlinear matrix layers alternating with
effective linearly elastic fibrous layers. This approach is limited by the fact that the
nonlinearity was obtained by including third order terms in strains in the stress-strain
relations.
The physical nonlinearity in fiber composites can also be considered a plastic behavior.
Dvorak and Bahei-EI-Din [3-4] used a micromechanical model consisting of elastic
filaments of vanishing diameters and an elastic-plastic matrix. An elastic-plastic
continuum model was developed for which the constitutive relations depend on the
constituent properties, their volume fractions, and mutual constraints between the phases
associated with the geometry of the microstructure. A few authors [5-8] used fmite
element method to analyze a representative volume of a unidirectional composite to
understand the elastic-plastic behavior of the composite.
183
184
Inherent in these micromechanical models is the need to know the constituent material
properties, fiber array, and fiber matrix interfacial condition. As noted in Bahei-el-Din
and Dvorak [9], the in-situ matrix property can be quite different from that of the bulk
matrix material. Consequently, adjustment of the matrix property in the
micromechanical model is often necessary to yield good results. Another difficulty
encountered in formulating micromechanical models lies in the 3-D elastic constants of
the fibers, which are extremely difficult to measure because of their small size.
Recently, Johnson et al. [10] indicated that the fiber matrix interfacial bond in some
metal-matrix composites is very weak as opposed to the perfect bond assumed in
developing these micromechanical models.
To overcome the aforementioned uncertainties present in the micromechanical level of
the composite, an alternative approach is to regard the composite as a homogeneous
elastic-plastic continuum and measure its macromechanical properties directly. This
approach enables one to employ well documented classical plasticity theories and many
existing fmite element codes for structural analysis.
For unidirectional composites, orthotropy exists in their plastic behavior under
monotonic loading. Griffin et al. [11] proposed a three-dimensional flow theory based
on Hill's [12] orthotropic yield criterion to describe the nonlinear behavior of a
boron/epoxy composite. The present author and co-workers have recently employed a
more general plastic potential functi~- which is a quadratic form of the stress
components, to characterize the orthotropic plasticity in a boron-aluminum metal-matrix
composite [13-14]. A simplified version of the plastic potential function with a single
parameter proposed by Sun and Chen [15] was successfully employed by Sun and Yoon
[16] to characterize the elastic-plastic behavior of AS4IPEEK thermoplastic composite
at various temperatures. This plasticity model is simple in form and can easily be
obtained experimentally using off-axis specimens. Because of its simplicity, this testing
procedure can easily be performed at elevated temperature environments.
2. A 3-D Plasticity Model
A plastic potential function is assumed for the general 3-D fiber composite as
2 f (Oij) = al 101 I + a22 0 22 + a33 0 33
2 2 2
+ 2a12011022 + 2a\3011033 +
(1)
+ 2a23022033 + 2a440~3 + 2assOf3 + 2a660f2 = k
where k is a state variable and the stresses O"ij refer to the principal material directions.
The coefficients a;j describe the amount of anisotropy in the initial plasticity. This
expression of f O"ij satisfies the orthotropy condition. The values of a;j can be determined
from the experimental data. This criterion reduces to the Von Mises yield criterion for
isotropic solids when the a;j have the values
all = a22 = a33 = 2/3

al2 =a13 =a23 =-113 (2)
a44 =ass =a66 =1
185
The Hill-type yield criterion for orthotropic materials [12] is a special case of Eq. (1)
when the function f aij is assumed to be independent of the hydrostatic stress. Hill's
yield function is obtained if
2aI2 = a33 -all-a22

2a23 =all-a22 -a33 (3)
2al3 = a22 - all - a33
By using the associated flow rule, the yield function is taken as the plastic potential
function from which the incremental plastic strains can be derived as
(4)
where the superscript p denotes plasticity, and dA. is a proportionality factor. In general,
the plastic potential given by Eq. (1) implies neither the incompressibility of plastic
strains nor the assumption that hydrostatic stresses result in no plastic deformation.
The increment of plastic work per unit volume is given by
dWP = aijdE~ = 2fdA. (5)
Let the effective stress (j be dermed as
(6)
The effective plastic strain increment dE"P can be defined that
dWP = a··dEP.
1J 1J
= crdE"P (7)
Substitution of Eqs. (5) and (6) into (7) yields
dE"P =lcrdA. (8)

3
and
(9)
Since the incremental strain dEij is assumed to be small, we are able to decompose it
linearly into the elastic part dE ~ and the plastic part dE P. as
u y
(10)
Consider a state of plane stress parallel to the Xl - X2 plane. The plastic potential
function reduces to
(11)
Without loss of generality, we further set a22 = 1.

The effective plastic strain increment can be derived using Eqs. (5-7) and the inversion
ofEq. (3). It is explicitly written as
186
,\deP)2 =-(al1-
2 2 t['\dEn
at2f p)2 +al1\dE
' p )2 -2at2dEl1E22+
P P (al1-at2){\dYt2
p )2] (12)
22
3 ~~
From Eqs. (4) and (11), the plastic strain increments are obtained as
(13)
3. Off-Axis Tension Test
The complete orthotropic plastic flow rule is defmed if the parameters al1, a12, 1i66, and
dA. are determined. To determine dA., the effective stress-effective plastic strain relation
must be established. This can be accomplished from the results of tension tests on off-
axis specimens.
Let the x-axis be the uniaxial loading direction which makes an angle e with the fiber
direction Xl-axis. The stresses referring to the material principal axes (Xl and X 2 ) are
related to the applied uniaxial stress O'x p"
Cftt = cos 29 Cfx
Cf22 = sin 2 9 Cf x (14)
Cf12 = -sin9cos9Cf x
Substitution ofEqs. (13) into Eqs. (lO) and (5) yields
u=h(9)crx (15)
where
(16)
Using the coordinate transformation for strains, the plastic strain increment in the
loading direction is given by
(17)
Substituting Eq. (13) into Eq. (17) and then using the relations (14), we obtain
dE~ =~h2(9)CfxdA. (18)

3
In view ofEq. (15), we thus obtain
dE~ =~h(9)UdA. (19)
Comparison ofEqs. (19) with (8) leads to
de P = dE~ Ih(9) (20)

187
For monotonic loading, the above equation is integrable. Thus,
E"P = E~ Ih(9) (21)
The desired relation between

cr and E"P can now be obtained from the experimentally obtained relation between ax
and E~. From Eqs. (15) and (21), it follows that
dcr = h 2(9) dO"x (22)

dE"P dE~
and
(23)
Since the cr - E"P relation should be unique in monotonic loading for the given material,
the parameters a1l , al2 and %6 mllst be chosen so that the resulting cr - E"P relation is
independent of e.
The above procedure can only determine the sum al2 + %6, as is evident from Eq. (16).
To separate the two parameters, the pIa!> .." Poisson's ratio must also be used.
Using the coordinate transformation law on strain components and Eqs. (13) and (14),
we obtain the plastic strain increments in the x-direction and the y-direction as
(24)
and
dE~ = [(I + all - 2a66 )sin 2 9cos 2 9 + aI2 (sin 4 9 + cos 4 9)]dAo"x (25)
respectively. The plastic Poisson's ratio is defined as
(l + all - 2a66)sin2 9cos 2 9+ al2(sin 4 9+ cos 4 9)

(26)
all cos 4 9+ sin 49 + 2(a12 + a66)sin 2 9cos 2 e
For 9 = 00 , 90 0 , and 45 0 the plastic Poisson's ratios reduce to
vg=-~
all
v~o = -al2 (27)
P l+all-2(al2 -a66)
v45 =
1+ all + 2(al2 + a66)
In the off-axis tension test, the transverse strain is measured, from which the transverse
plastic strain E~ can be obtained, and, thus, the plastic Poisson's ratio. The additional
data allow us to separate al2 and %6.
188
4. One-Parameter Plasticity Model
Numerous experimental data show that fiber composites behave linearly when loaded in
the fiber direction. Thus, it is reasonable to assume that
(28)
which leads to the condition

(29)
As a result, we obtain a one-parameter plastic potential
2f = 0"~2 + 2a660"l2 (30)

From this plastic potential, the plastic strain increments are derived
(31)
where y 12 is an engineering shear st . n. The corresponding effective stress and

effective plastic strain increment reduce to
(32)
and
(33)
respectively, and h(9) becomes
2 2 J'2
h(9) = [%(Sin 4 9+2a66 sin 9cos 9 (34)
This simplification results in a single parameter 146 to be determined. The procedure for
determining the one-parameter plasticity model becomes very simple. In theory, only
two off-axis specimens with different 9's are required to determine 146. Therefore, there
is no need to measure the plastic Poisson's ratio to aid in determining of the parameters
as in the 3-parameter model. Note that for 9 = 90°, h(9) = Jill, and thus, the stress-
strain relation for the 90°-specimen is not affected by the choice of 146 and can be
conveniently used to construct the master effective stress-effective plastic strain curve
for the composite. However, the main drawback in the 90° specimen is that its strength
is low, and the corresponding stress-strain curve is relatively short.
The plastic Poisson's ratio corresponding to the one-parameter plastic potential is given
by
189
(35)
As e~ 0°, the plastic Poisson's ratio approaches
(36)
5. AS4IPEEK Thermoplastic Composite
The one-parameter plasticity model was employed by Sun and Yoon (1989) to
characterize the plastic behavior of the AS4IPEEK (APC-2) thermoplastic composite.
Specimens with e = 0°, 15°. 30°, 45°, and 90° were tested. The stress-strain curves
obtained at 24°C and 177°C are presented in Figs. land 2, respectively.
Trans. - - Long.
400
0 300
a..
~
.S 200
III
III
'"
~
Vi 100
1] lS 2.0 l5 3]
Strain in %
Figure 1. Stress-strain curves for AS4/PEEK off-axis specimens at 24° C.

(-----: strain gage debonded but strength obtained)
Trans. - - Long.
400r-~--Tr-r~--~~--~--~-.
0 300
a..
~
o 0.5 1.0 1.5 2.0 2.5 3.0

Strain in %
Figure 2. Stress-strain curves for AS4/PEEK off-axis specimensat 121 ° C.

(-----: strain gage debonded but strength obtained)
190
Experimental results indicate that for the thermoplastic fiber composite there is no well
defmed yield point. In view of this, the master effective stress-effective plastic strain
curve can be fitted as a power law,
eP = A(cr)D (cr in MPa) (37)

The parameter ~, coefficients A and n may be functions of temperature, strain rate and
other environmental factors.
The effective stress-effective plastic strain relations of all off-axis specimens at
24°C, 66°C, 121 0, and 177°C are shown in Fig. 3. It is seen that the effective stress-
effective plastic strain data for all off-axis angles more or less collapse into one curve
with 1166 = 1.5 for all temperatures. This indicates that the orthotropy of plasticity
AS4IPEEK can be regarded independent of temperature.
300~--------------------------~
•• ,.30-
'.15·
• ,. '5·
• •• goo
~ _ _ _ _ 15O"F "'·CI
=--_-------.
_ ......_ ....._••~_._ 250·F ,'2"CI
350"F "??OCI
0.5 1.0 1.5 2.0 2.5 3.0

Ettoctl¥. plastic .train kI %
Figure 3. Effective stress and effective plastic strain curves for AS4IPEEK
off-axis specimens.
The coefficients A and n can be determined from fitting the data into the power law of
Eq. (37). The results are presented in Table 1, which indicate that n = 7.0 is adequate
for temperatures up to 121°C. It is also found that coefficient A is related to
temperature as
In(A) = {0.036T - 40.7 (Tin dc)

(38)
0.02T - 41.3 (Tin OF)
for temperatures below the glass transition temperature. At 177°C, which is above the
T.G. (glass transition temperature), n = 2.9, and the value of A drops significantly.
191
Table 1. Coefficients A and n for AS4IPEEK
Temperature 24°C 66°C 121°C 177°C
(7S0F) (ISO OF) (2S00F) (3S00F)

Coefficient
n 7.0 7.0 7.0 2.9

A S.l4xlO-18 2.18 X 10-17 1.70xl0-16 7.06xl0-li
In(A) -39.8 -38.4 -36.3 -13.S
The total incremental strain in the loading direction can be decomposed into the elastic
part dE ~ and the plastic part dE ~ as
(39)
In the monotonic loading case Eq. (39) can be integrated to yield
(40)
In Eq. (40), dE ~ is the elastic strain in the loading direction given by
(41)
where Ex is the apparent elastic modulus of the off-axis specimen which can be obtained
from the transform equation
_1_= ...!..cos4 8+(_1__ 2v12 )Sin2 8cos28+_1_sin 4 8 (42)

Ex E) G 12 E) E2
Using Eqs. (15), (21) and (37), dE~ can be expressed as
E~ = [h(8)]n+) AO'~ (43)

Therefore, the stress-strain curve in monotonic loading can be predicted using the
following equation.
(44)
Figure 4 shows the total stress-strain curves for the off-axis specimens. Agreement
between the experimental data and predictions according to the model is excellent.
The value of Il66 can be checked by the comparison of the plastic Poisson's ratios
obtained from theory and experiment. Figure 5 shows the predicted plastic Poisson's
ratio (with 1166 = 1.5) and the experimental data up to 121°C. The present plasticity
model seems to yield reasonably accurate plastic Poisson's ratios.
192
400r------------------------,
0 300
a..
::::E
c
·;;;200
III 30°
~
Vi 100
0.5 1.0 1.5 2.0 2.5 3.0

Strain in %
Figure 4. Measured and predicted stress-strain curves for AS4IPEEK

off-axis specimens.
1.0
- - Predicted
o Experimental
~0.8
III
• • 75°F
• 150°F
·gO.6 • 250°F
III
III
'0
a.. 0.4
.!:1
iii
~0.2
a..
15.0 30.0 45.0 60.0 90.0

8 in degrees
Figure 5. Measured and predicted plastic Poisson's ratios for AS4IPEEK
6. Conclusion
Polymeric and metal matrix composites displayed significant nonlinear inelastic

behaviors. These orthotropic nonlinear stress-strain relations can be modeled very well
by an orthotropic plasticity model which contains a single orthotropy parameter. The
simplification of the model is achieved by assuming that plasticity in the longitudinal
(fiber) direction is negligible.
The one-parameter plasticity model was verified experimentally to be suitable for the
description of plastic behavior in polymeric metal matrix composites. For AS4IPEEK
composite, the orthotropy parameter 1166 and the power index n can be regarded as
constants below the glass transition temperature.
193
7. References
1. Hahn, H. T., Tsai, S. W. (1973) Nonlinear elastic behavior of unidirectional

composite Laminae, J. Composite Materials, 7 102-118.
2. Sun, C.T., Feng, W.H., Koh, S.L. (1974) A theory for physically nonlinear elastic
fiber-reinforced composites, International J. Engineering Science, 12919-935.
3. Dvorak, G. J., Bahei-EI-Din, Y. A. (1979) Elastic-plastic behavior of fibrous
composites, J. ofMechanical and. Physical Solids, 27 51-72.
4. Dvorak, G. 1., Bahei-EI-Din, Y. A. (1982) Plasticity analysis of fibrous composites,
J. Applied Mechanics, 49327-335.
5. Chamis, C. C., Sullivan, T. L. (1973) Theoretical and experimental investigation of
the nonlinear behavior of boron aluminum composites, NASA TM X-68-205,
NASA-Lewis Research Center, Cleveland, OH.
6. Adams, D. F. (1970) Inelastic analy~is of a unidirectional composite subjected to
transverse normal loading, J. Composite. Materials, 4, 310-328.
7. Foye, R. L. (1973) Theoretical post-yielsding behavior of composite laminates, Part
I - Inelastic micromechanics, J. Composite Materials, 7, 178-193.
8. Lin, T. H., Salinas, D., Ito, Y. M. (1972), Initial yield surface of a unidirectionally
reinforced composite, J. ofApplied Mechanics., 39 321-326.
9. Bahei-EI-Din, Y.A., Dvorak, GJ. :"180) Plastic deformation of a laminated plate
with a hole, J. ofApplied Mechanics, 42, 385-389.
10. Johnson, W. S., Lubowinski, S. J., Highsmith, A. L., Brewer, W. D., Hoogstraten,
C. A. (1988) Mechanical characterization of SCS6/Ti-15-3 metal mamx composites
at room temperature, Mechanical Characterization of SCS6/Ti-J5-3 MetalMatrix
Composites at Room Temperature, NASA-Langley Research Center, Hampton,
VA:NASP Technical Memorandum 1014.
11. Griffm, O. H., Kamat, M. P., Herakovich, C. T. (1981), Three-dimensional inelastic
finite element analysis of laminated composites, J. Composite Materials,S, 543-
560.
12 Hill, R. (1948), A theory of the yielding and plastic flow of anisotropic metals,
Proceedings of the Royal Society, Mathematical and Physical Sciences, 193, 1033,
281-297.
13. Kenaga, D., Doyle, J. F., Sun, C. T. (1987), The characterization of
boron/aluminum composite in the nonlinear range as an orthotropic elastic-plastic
material, J. Composite Materials, 21516-531
14. Sun, C. T., Chen, J. K. (1987) Effect of plasticity on free edge stresses in boron-
aluminum composite laminates, J. Composite Materials, 21969-985.
15. Sun, C. T., Chen, J. L. (1989) A simple flow rule for characterizing nonlinear
behavior of fiber composites, J. Composite Materials, 36 321-337
16.. Sun, C. T., Yoon, K. 1. (1989) Characterization of elastic-plastic behavior of
AS4IPEEK thermoplastic composite for temperature variation, J. Composite
Materials, 25 1297-1313
CHARACTERIZATION OF STRAIN RATE-DEPENDENT
BEHAVIOR IN POLYMERIC COMPOSITES
C.T.SUN
School of Aeronautics & Astronautics, Purdue University,
West Lafayette, Indiana 47907-1282 USA
Abstract
Viscoplasticity models for describing strain rate-dependent behavior in fibrous

composites are reviewed. The focus is on the use of a one-parameter plastic potential
function in characterizing the nonlinear stress-strain relations and on the use of off-axis
composite specimens to generate data that capture the anisotropic properties.
Experimental results of an S2 glass composite are used to verify the constitutive
models.
1. Introduction
High strain rate characterization of composites has recentlY been a subject of interest
primarily because of many new application~ "f composites that are subjected to blast
and impact loading. In general, the constitutive model necessary to completely and
accurately describe the rate-dependent behavior of the composite can be extremely
complex and mathematically intractable because of the large number of components of
stress and strain that are involved in the formulation. Experimental investigations tend
to primarily focus on a one-dimensional state of stress or strain. Various techniques
have been employed over the years to generate different regimes of strain rate. Material
testing machines have been employed to generate strain rates in the order of O-IO/s;
Hopkinson bar impact experiments have been used to generate strain rates in the order
from 10-1 OOO/s, while plate impact experiments have been used to generate even
higher strain rates.
Because of the anisotropic properties of composites, a successful constitutive model is
one which can collapse the entire data into a single equation and have the capability of
extrapolation and interpolation. Various plasticity models for composites have been
developed in the last two decades. In general, two approaches were used in the
constitutive modeling of composites. One is the micromechanics approach in which
the composite is considered to be heterogeneous and is composed of fiber and matrix
phases and the properties of the composite are determined from the properties of the
constituent phases [1,2]. The other is the macromechanics approach in which the
composite is considered as a homogeneous and orthotropic solid of which the
properties are determined from experimental characterizations. Many factors influence
the predictions based on micromechanics, such as fiber arrangement, fiber-matrix
interface, matrix damage, etc. As a result, it appears more realistic to use
macromechanics based models for engineering applications.
195
196
Among the macromechanical models for the nonlinear behavior of composites, the
plasticity theory by Hill [3] which was a direct extension of the classical von Mises
flow rule to orthotropic materials, has been quite popular. Hill's theory [3] was
developed based on the classical assumption that hydrostatic stress does not influence
plasticity. Although this assumption is true for isotropic solids, it was shown by Chen
and Sun [4] that for anisotropic solids such as fibrous composites, the hydrostatic stress
produces significant plasticity if the matrix is elastic-plastic. Instead, they found that a
state of uniform dilatation in composites produces little plasticity. By using the fact
that fibrous composites usually exhibit little plasticity in the fiber direction, Sun and
Chen [5] proposed a one parameter plastic potential function to model nonlinear
behavior in composites. Using the one parameter plastic potential function in
conjunction with the overstress concept, Gates and Sun [6] and Yoon and Sun [7]
proposed an elastic/viscoplastic constitutive model for describing the rate dependent
behavior of AS4IPEEK thermoplastic composite. Later, this model was extended by
Gates [8] to characterize the viscoplastic behavior of polymeric composites at both
room and elevated temperatures for IM7/5260 and IM7/8320. Weeks and Sun [9]
conducted experiments using an MTS machine and the Split Hopkinson Pressure Bar
for AS4IPEEK composite, and produced strain rates ranging from O.OOOOI/s to 1000/s.
They also proposed a number t-• .;onstitutive models to describe the rate-dependent
behavior. Recently, Thiruppukuzhi and Sun [10] have investigated the rate effects in
unidirectional and woven glass composites.
In this paper, the viscoplasticity models tor loading as well as for unloading are
reviewed. The experimental methods suitable for characterizing the models are
discussed. Data for actual composite materials are presented.
2. Viscopiasticity Behavior
Many authors have proposed various constitutive equations in order to describe the
mechanical properties of materials that exhibit rate effects. In the following, two models
suitable for composites are presented.
2.1 OVERSTRESS MODEL
In the overstress viscoplasticity model, the plastic strain rate is assumed to be a

function of only the dynamic overstress. This model was proposed by Malvern [11] for
wave propagation in work hardening materials. Based on an additive decomposition of
the strain rate into an elastic and plastic component the total strain rate can be
expressed as
. ·e·p (I)
&ij = &ij +£ij
where the elastic strain rates are given by
·e S . (2)
"ij = ijkl<Jkl
in which Sijkl is the elastic compliance tensor for the composite. The plastic strain
rates are obtained from the flow rule requiring the normality condition, i.e.,
197
. P _. Of
E·· - A.-- (3)
IJ acrij
where
1 2 2 (4)
f = "2(0"22 + 2a660"12)
is the plastic potential, and the proportionality factor i is given by

3 -:- P
i=-!. (5)
20"
In equation (5), the effective stress;;: is defined by
;;:=.!3f (6)
and the corresponding effective plastic strain rate is
P )2 _1_C )2] p
7p _
E - "32[CE22 + 2a66 )'12 (7)
By introducing the viscoplastic modulus
H=~ (8)
P iP
the proportionality factor i can be expresseo In the form
(9)
1 9 . 9 .
= --::-(- 0"22 0"22 +-a66 0"12 0"12)
HpO" 4 2
A power law functional form can be chosen to express effective plastic strain rate as a
function of dynamic overstress as
iP =x(o-o*)mH(o-o*) (10)
where X and m are the two material parameters determined from experiments, 0 and
0* are the effective dynamic stress and the effective equilibrium (quasi-static) stress,
respectively, and H denotes the Heaviside step function.
2.2 TOTAL STRESS MODEL
For the total stress model, the effective plastic strain rate is assumed to be a function of
the effective stress and effective plastic strain, i.e.,
iP = B(£"P)u ( ;;: )13 (II)
Unlike the overstress viscoplasticity model, this model does not employ the equilibrium
(back) stress. For determining the viscoplasticity parameters as well as numerical
implementation, the above relation is rewritten in the form
£"p =x(iP)m (;;.)" (12)
where
198
a =~. j}=-~ (13)

m m
The parameters X. m, and n are the three parameters to be determined from experiments.
3. Material Characterization by Off-axis Testing
To establish a viscoplasticity model for a fibrous composite, we need to defme the

plastic potential function f by determining the orthotropy coefficient ~. and to
determine the coefficients in equations (10) and (12). The coefficient ~ can be
determined by quasi-static testing of off-axis specimens with a vanishing loading rate.
Determinations of the other coefficients require the results of different strain rates.
3.1 DETERMINATION OF ~
Figure 1 shows a schematic of the off-axis specimen configuration used.

Thiruppukuzhi and Sun [10] conducted uniaxial tension tests on various off-axis
specimens of S2 gJass/8553 composite with e = 0°, 15°, 30°, 45°, 60°, and 90° for three
different strain rates of 0.000 lis, O.Olls and lis. Figure 2 shows typical experimental
stress-strain curves obtained at O.OOOlls which was regarded as the quasi-static
(equilibrium) state.
y
Strain Gage
Applied Displacement
Figure I. Schematic ofthe experimental setup for uniaxial testing.

199
300
250
200 -+-0
-;s --+- 15
___ 30
e
~
ISO ....... 45
i
!::
+60
--.- 90
'" 100
50
0.004 0.008 0.012 0.016

Strain (mm/mm)
Figure 2. Stress-strain curves for the unidirectional S2 glass at O.OOOl/s.

The effective stress-effective plastic strain relation should be unique in monotonic
loading for a given material system and strain rate. Therefore, the parameter ~ in the
plastic potential function is chosen such that the resulting effective stress-effective
plastic strain relation is independent of the off axis angle e. By selecting ~ = 1.40, the
effective stress-plastic strain curves for all off-axis specimens collapse into a single
master curve. Figure 3 shows collapsing curves for various off-axis orientations with
~ ~ 1.40.
120
100
,
-;s 80
, ,
e.,
~ -------~----------------
, --.- 15
, ____ 30
~
.. , -.-45
.
60 --------, -*-60
iZi
-+90
t... , - - Master Curve
40 ______ ..! ________ ,. _______ _ '.L-____ ,.-----'
....
~
'"
Effective Plastic Strain (mm/mm)
Figure 3. Effective stress-effective plastic strain curves for the

unidirectional S2 glass at O.OOOl/s.
200
The value of a66 depends on the composite system. For AS4/PEEK composite, Weeks
and Sun [9) found a66 = 2.0.
3.2 OVERSTRESS MODEL
Using the quasi-static master curve given by Fig. 3 and the results of the off-axis tests
at different strain rates, the dynamic overstress a-a· in equation (10) can be
calculated for each off-axis specimen over the entire loading history. This result can be
plotted as shown in Fig. 4 in a log-log scale from which the coefficients x and m in
equation (10) can be obtained. For S2 glass/8553 composite we have m = 5.64 and
x = 8.4 x 10" (MPa)"m. .
100 ,...----,...--_ _,....--_ _ _,....--_ _ _,....--_ _---,
VI
!I!!
~ 10
u
E
III _ 15
~
--: I •
A 45
30
- - Curve fit
Effective plastic strain rate

Figure 4. Log-log plot for determining the overstress viscoplasticity
parameters for 82 glass composite.
3.3 THE TOTAL STRESS MODEL
The total stress viscoplasticity model given by equation (II) can be expressed in the
form
(14)
where
(15)
Thus, this model can be viewed as a plasticity model with strain rate-dependent
amplitude A. For the S2 glass/8553 composite, the experimental data indicate that
exponent n is constant for all strain rates tested. Theoretically, two constant effective
plastic strain rate experiments are sufficient to determine the two viscoplasticity
201
parameters a and P in this model. It is noted that constant total strain rate experiments
do not result in constant effective plastic strain rates. However, experimental data
obtained from constant total strain rate experiments indicate that the effective plastic
strain rate reached a steady state value in a narrow bandwidth after initial loading.
Thus, the data corresponding to the initial part of loading is truncated in determining
the viscoplasticity parameters. This does not influence the model predictions as the
response of the material is dominated by elastic behavior during the initial part of
loading.
Figure 5 shows amplitude A as a function of the effective plastic strain rate for
different off axis orientations on a log-log scale for the unidirectional S2 glass
composite. The viscoplasticity parameters a and Pare determined from this plot as the
intercept and the slope, respectively. We have m = -0.14 and X = 1.05 X 10- 11 (MPa)"n.
Having obtained the viscoplasticity parameters, the off axis material response for the
uniaxial tension tests can then be predicted using the following nonlinear differential
equation
(\6)
where
(17)
in which
(\8)
~
CI 1E-011
.3
1E-005 0.0001 0.001 0.01 0.1 10
Log(Effective Plastic Strain Rate)

Figure 5. Material parameters ofthe total stress modelfor the
unidirectional 82 glass material system.
Figure 6 shows the model predictions using the total stress model for the unidirectional
S2 glass 30° off-axis specimen for the three strain rates of 0.000 lis, 0.0 lis and lis.
202
300
I I A ..~ ....
____ __ 1_ ____ __ 1__ ___ __1 _ _ _ _ _ _ J _____ _ l,k"~"'~ __
250 I I I I ... r
I I I I ........ I
-J.'" ..~~
: : : "l,r' ..............
200 _ _ _ _ _ _ '- _ _ _ _ _ _ 1_ _ _ _ _ _ .J __ !"'II'~ _ _ _ ....... *"!... t __ ___ _
I I I ~... I...... I
Ii I I ~' ......,. I
Il.
~
I ..
I........'. I
I
I
I
150 -:i,.t/tC.. - - - - - .... - - - - -- + - - - - - -
I/)
I/)
- -- -- -1-------1- - -
I *
•• I I
I!! I .-;;/t"
in .";.ft- * O.0001IS (Expt)

100 • O.01IS (Expt)
.. 1IS (Expt)
I
_._.- o.o001lS (Model
50 I
1- - - - - - -1- - - - - - -.
I - O.01IS (Model)
I ----- 1IS (Model)
I ' - -_ _~---...J
0
0 0.002 0.004 0.006 0.008 0.01 0.012
Strain (mmlmm)
Figure 6. Model predictions using the total stress modelfor the

unidirectional S2 glG;- r 7r the 300 offaxis orientation.
The model predictions are found to be generally in good agreement with the
experiments.
Due to their extension-shear coupling property, off-axis composite specimens cannot
be tested directly by Split Hopkinson Bar to generate data beyond the rates capable of
usual material testing machines. However, as shown by Weeks and Sun [9] using
Hopkinson Bar and [±e] angle ply AS4IPEEK laminate specimens, the viscoplasticity
model obtained based on the loading rate below lis is valid for strain rates exceeding
1000/s.
4. Conclusion
The nonlinear and strain rate-dependent constitutive behavior of fiber-reinforced

polymeric composites can be accurately modeled with a one-parameter plastic potential
function. This parameter, a.., is independent of strain rates and can be determined
easily from testing off-axis composite specimens. Since the constitutive models are
expressed in terms of effective stress and effective plastic strain, they can be used for
general loading conditions.
The most attractive features of the present viscoplasticity models are their simple
functional forms, the ease in performing the off-axis test, and the small number of
coefficients involved. Moreover, the single scalar constitutive equation via the use of
effective stress and effective plastic strain is able to handle general loading.
203
5. References
1. Dvorak, GJ. and Bahei-EI-Din, Y.A. (1982) Plasticity analysis of fibrous

composites. 1. ofApplied Mechanics, 49,327-335.
2. Adams, D.F., and Crane, D.A. (1984) Combined loading micromechanical
analysis ofa unidirectional composite, Composites, 15, 181-92.
3. Hill, R. (1950). The Mathematical Theory of Plasticity, Oxford University Press,
London.
4. Chen, J. L., and Sun, C. T. (1993) A plastic potential function suitable for
anisotropic fiber composites, 1. of Composite Materia/s, 27, 14, 1379-1390.
5. Sun, C.T. and Chen, J.L. (1989) A simple flow rule for characterizing nonlinear
behavior of fiber composites, 1. of Composite Materials, 23, 1009-1020.
6. Gates, T.S. and Sun, C.T. (1991) Elastic/viscoplastic constitutive model for fiber
reinforced thermoplastic composites, AIAA 1., 29, 457-463.
7. Yoon, K.J. and Sun, C.T. (1991) Characterization of elastic-viscoplastic properties
of an AS4IPEEK thermoplastic composite, 1. of Composite Materials, 25, 1277-
1298.
8. Gates, T.S. (1991) Effects of elevated tl. ...tlerature on the viscoplastic modeling of
graphite/polymeric composites. NASA TM 104160.
9. Weeks, C.A. and Sun, C.T. (1995) Nonlinear rate dependent response of thick-
section composite laminates, Proceedings of International Mechanical Congress
and Exposition, San Francisco, November 12-17.
10. Thiruppukuzhi, S., and Sun, C.T. (1996) High strain rate characterization of
unidirectional fiber and woven glass composites, Proceedings of the 11 th
Technical Conference, American Society for Composites, 201-210.
II. Malvern, L.E. (1951) The propagation of longitudinal waves of plastic
deformation in a bar of material exhibiting a strain rate effect. 1. of Applied
Mechanics, 18,203-208.
A TREATMENT OF INTERFACIAL CRACKS IN THE
PRESENCE OF FRICTION
C. T. SUN and W. QIAN

School ofAeronautics and Astronautics
Purdue University
West Lafayette, Indiana 47907-1282, USA
Abstract
Frictional sliding on interface crack surfaces results in weak crack tip stress singularity
and zero strain energy release rate. A fracture criterion based on finite extension strain
energy release rate, is proPOSI;:U to capture the intrinsic fracture toughness. The fmite
extension strain energy release rate is shown to represent the magnitude of the singular
stress field. Numerical simulations of a center crack in a bimaterial infmite media under
remote shear as well as fiber pull-out and push-out are presented to illustrate the
frictional effect in both small and large scale contacts near the crack tip.
1. Introduction
The development of classical fracture mechanics is based on the open crack

assumption, i.e., the crack surfaces are not in contact. That assumption leads to, in
linear elastic solids, stress singularity of inverse square root that enables one to employ
either strain energy release rate or stress intensity factor to characterize fracture
toughness and predict crack movement.
When crack surfaces are under combined compression and shear, the effect of
sliding frictional stresses on the crack surfaces cannot always be ignored. Only in the
case in which the neartip contact zone is extremely small, the K-dominated field
basically agrees with that of the open crack solution. This was discussed in detail by
Sun and Qian [1] for bimaterial interfacial cracks under "Mode I" loading. On the other
hand, if the contact zone is fmite, frictional contact of the crack surfaces can no longer
be neglected. Most importantly, the frictional contact alters the stress singularity to
weak singularities [2]. Consequently, the strain energy release rate as conventionally
defined vanishes [3] and cannot be used as a fracture parameter.
Many researchers have attempted to extend the fracture mechanics approach to treat
various engineering problems involving interfacial cracks with frictional contact. Fiber
pull-out and push-out problems have attracted the most attention (e.g., [4-7]).
Delamination in thin film/substrate micro-electronics systems [8] and composite
205
206
laminates [9] also have been investigated. Various methods were proposed by these
authors to derive the strain energy release rate or to employ different parameters for
characterizing the interfacial crack toughness. Many of these approaches suffer
deficiencies mainly due to the weak singularity and the energy dissipation associated
with the frictional sliding of crack surfaces.
In this study, the concept of strain energy release rate for interfacial cracks in the
presence of friction is reexamined. A finite element based numerical procedure is
introduced to calculate the strain energy release rate and energy dissipation due to
friction for a finite crack extension. An explicit relation between the generalized stress
intensity factor and the finite extension strain energy release rate is derived based on the
near tip field. Thus, the finite extension strain energy release rate with a fixed crack
extension can be used to represent the magnitude of the singular stress field and,
therefore, to quantitatively characterize the intrinsic fracture toughness. For numerical
examples, a center crack in a plane strain infinite bimaterial medium under shear
loading (see Figure 1) and fiber pull-out and push-out tests (see Figure 2) are simulated.
't
l istiffer
l i
l ~,Vl
i
l ~, ,v,
i .
open tip closed tip
l i
l +-+-+-+-+-+- i
softer
Figure J A cracked bimaterial panel, 20 m x20 m, under remote loading
Figure 2. Bimaterial crack and its coordinate system

207
2. Near Tip Field and Strain Energy Release Rate
For a crack in homogenous media with friction or an interface crack without friction,
there is always square root singularity associated with the neartip field. However, it
was originally shown by Comninou [2] for an interface crack between two dissimilar
isotropic media with friction and later confirmed and extended by Deng [3] for an
interface crack between two dissimilar anisotropic media that the stress singularity r- A
is always weak (A < 0.5) for a stationary crack under monotonic loading. The
singularity index A is related to the coefficient of friction Il as [2]
coth = Il~ (1)
where
~ = 1l1(K2 -1)- 1l2(Kl -I)
(2)
1l1(K2 +1)+1l2(Kl +1)
is a Dundurs parameter, and J.lI and J.l2 are shear moduli. In equation (2), subscripts I
and 2 stand for upper and lower materials, respectively (See Figure I), and
K = 13-
3-v
--
I+v
A.. for plane strain
for plane stress
(3)
in which v is the Poisson ratio. It is easy to see that A is 0.5 if ~ or J.l is zero, which
corresponds to homogenous media or frictionless condition, respectively.
The near tip stress field ahead of and behind the crack tip along the bimaterial interface
and the relative crack surface sliding displacement in plane strain condition are
O'xy(r.O) =Kn(2ltr)-A
O'xy(r.±It) = KU cosA1t(21tr)-A (4)

O'yy(r,±1t) = -K ul3sin A1t(21tr)-A
and
~ux (r) = Ux (r, It) - Ux (r,-It) = [ K IIYsin Alt / 2(1 - A)(21t)A }l-A (5)
respectively, where the generalized stress intensity factor Ku is defined as
Kn = lim (2rrr)AO'xy(r,O) (6)

r~O
and
The solutions above are obtained with the assumptions that the crack surfaces slide and
208
the nonnal and shear stresses behind the crack tip follow the Coulomb frictional law,
i.e.,
O"yy(r,±1t) < 0 (8)
(9)
(10)
The assumption of equation (8) implies that Kn13 > 0, while equation (9) gives JlKn > O.
Thus, we have the condition Jl13 > 0 which in conjunction with equation (1) leads to the
conclusion that A < 0.5.
It has been pointed out by Dundurs and Comninou [10] that the problem of an
interface crack with friction is a linear process in the context of monotonic loading. In
this context, the strain energy released for a crack extension Aa can be obtained using
Irwin's crack closure integral. We have
G(&) =~Jga~x1r,0)-(Jx1&-r, 7t)~x(&-r)dr
KfIysinA7t A 1-;" r(2-A)r(1-A) cos A7t ]( 11 )

ua L~--~~--~
4(1- A)(27t)2A r(3 - 2A) 2(1- A)
in which r is the gamma function. The conventional strain energy release rate is
defmed as G = lim G(Aa). It is seen from equation (11) that the strain energy release
Aa~O
rate G vanishes as Aa ~ 0 due to the fact that A < 0.5 in the presence of friction.
Consequently, G cannot be used as a parameter in the fracture criterion. On the other
hand, for Aa = Aa.. "* 0, G(Aa..) is uniquely related to the stress intensity factor Kn and,
thus, to the near tip stress field. By selecting a proper characteristic crack closure
distance Aa.., it seems possible to use G(Aa..) as a fracture parameter when friction is
present.
3. Finite Element Procedure for Energy Calculation

Analytical solutions for interfacial cracks with frictional sliding are extremely difficult
to obtain. Numerical methods such as the fmite element method are necessary for
practical applications. In view of the foregoing, we have developed a fmite element
procedure to calculate the strain energy release G(Aa) and the energy dissipation due to
friction during crack extension. This procedure is based on crack tip nodal forces and
displacements. To separate the energy release from the frictional energy dissipation at
the crack tip, the elastic nodal force must be separated from the frictional nodal force
that results from the frictional traction on the crack surface.For illustration purposes,
assume that the crack tip is modeled using four-noded plane strain (or plane stress)
elements as shown in Figure 2. To facilitate friction as well as nodal release, interface
elements INTER2 in the commercial code ABACUS are used along the interface ahead
209
and behind the crack tip. Let F~ and F~ be the horizontal and vertical nodal forces,
respectively, at the crack tip node in the loaded medium just before crack extension.
Crack extension is simulated by releasing the crack tip node into two separate nodes
allowing a frictional sliding. The relative sliding of the nodes a and b (see Figure 3) is
denoted by ~u~. Then the strain energy release rate associated with a fmite extension
~a is
G( ~a) = _1_( F: - Ii Fya )~u~ (12)

2~a
The term F~ - 11 F~ is recognized as the elastic nodal force.

0.07 ....-...-.........................---.----.........-.--.,.......,
0.06
0.05
-~
III
>.
0.04
0.03 I ~=O.21
.... ,,=O.S
0.02
0.01
o
.(J.Ol !-;;"';;-':!'::--::!7"~~-::':~~':;-::~
-1 .(J.B .(J.B .(J.4 .(J.2 0 0.2 0.4 0.6 O.B 1
xJa
Figure 3. Relative crack surface normal displacement

During the assumed crack extension, the total dissipation energy rate associated with
crack surface friction is given by
(13)
where G~(..1.a)is the portion of the dissipation energy rate produced by the newly
formed crack surface, and Gd(~a) is the portion produced by the existing crack
surfaces. We have
where (IlF;(I) and IlF;(2) ) denote the vertical tip nodal force before and after crack
extension, respectively. The calculation of G ~ (~a) is carried out in a similar manner.

However, it should include all the nodes in the contact region before crack extension.
A center crack lying between two dissimilar isotropic media under remote shear loading
210
t (see Figure 1) is used to examine the accuracy of the aforementioned technique. The
commercial code ABAQUS was used to perform the analysis. The size of the panel
was 20m x 20m with a crack of 2m. A similar problem of an infmite plate was also
investigated by Dundurs and Comninou [10] using the elastic dislocation approach.
The stress intensity factor obtained by calculating the strain energy release rate
G(da) and using the relation in equation (11) for the closed crack tip (right tip in Figure
I) is given together with those from Dundurs and Comninou [10] in Table 1. The
agreement between these two methods is excellent. The slight difference is believed to
be due to the fmite size effect of the panel in the present study, in contrast to the infmite
medium assumed by Dunders and Comninou [10].
Table 1. Comparison ofstress intensity factor Knfrom present analysis and K~
obtained by Comninou, et al. {lO}. Krr = (1+ 11 2p2)K rr /t(2al, 11 = 0.5, P = 0.5
Aa/a KU (present) d d
I(KU -KU)/KU IxIOO
3.I8xl02 " "723 1.28
1.59 x 10 2 0.9720 1.25
7.96 x 10 3 0.9724 1.29
3.98 x 10 3 0.9737 1.42
4. A Fracture Criterion
The fmite element procedure presented in the previous section was used to study the
interfacial crack under shear loading as shown in Figure 1. The dimensions of the
bimaterial medium were 20 m x 20 m with a half crack length a = 1 m. The applied
shear stress t was assumed to be 0.12 MPa. The material constants were J.l.1= 35 MPa,
J.l.2 = 0.35 MPa, and VI = V2 = O. These material constants yield ~ = 0.49.
The relative crack surface normal displacements for J.I. = 0.2 and 0.5, respectively, are
shown in Figure 3. It is seen that there is a sizable contact zone near the right tip while
the crack is open at the left tip. Theoretically, there is an extremely small contact zone
at the left tip. However, this contact zone is so small that the left crack tip can be
treated as an open crack with the classical oscillatory stress field [1].
The fmite extension strain energy release rate G(~a), dissipation energy rate G d (~) as
well as total energy release rate G(~a)+Gd(~) for the problem of Figure 1 with J.I. = 0.5
corresponding to various crack extensions are shown in Figure 4 and Figure 5 for the
open (left) and closed (right) crack tips, respectively. At the left crack tip, the contact
zone rc/a is usually around 10-5 to 10-7 and can be virtually treated as an open crack
tip. The modified crack closure technique can therefore be applied to calculate the
strain energy release rates. It is seen in Figure 4 that individual finite extension strain
energy release rates GI and Gn are Aa dependent, while the total strain energy release
rate G = GI + Gn as well as the dissipation energy rate Gd remain constant for different
crack extensions. It is noted that the dissipation energy rate is fairly small compared
with strain energy release rate. This indicates that the frictional effect on the open crack
211
tip is fairly small. It can also be shown that the stress distribution ahead of the open
crack tip is very close to those of the oscillatory solution obtained based on traction free
crack surface condition. It is concluded that frictional effect at the open crack tip is
negligible. All existing approaches for the calculation of fracture parameters for
bimaterial interfacial cracks are thus applicable (e.g., see Ref. 1).
--
N- 30 y-------------------,
E
~ ~I*I~I~==*I========~I
>-
2>
(])
an 0 •• .1
0.00 0.02 0.04
l1ala
Figure 4. Finite extension energy release rates at the open crack tip for the panel
shown in Figure I
•• • • •
!(]) 15
II)
CO
(])
Q)
0::
>-
2>
::=: =e
--d
- - Gd.
(])
--G+Gd
c:::
W
o
0.00 0.02 0.04
l1ala
Figure 5. Finite extension energy release rates at the closed crack tip
for the panel shown in Figure 1
As for the fracture behavior at the closed crack tip (the right tip), it is clearly seen in
212
Figure 5 that the fmite extension strain energy release rate G decreases, and dissipation
energy rate Gd increases, while the total energy release rate G + G d remains constant
when crack extension ~a decreases. Recall the relationship between strain energy
release rate G and crack extension ~a in equation (11) that strain energy release rate
indeed decreases and eventually vanishes when ~a approaches zero due to the weak
stress singularity. The strain energy release rate is therefore ~a -dependent, while the
increase of dissipation energy rate makes up the decrease of strain energy release rate
for total energy balance when ~a decreases. The implication of decreasing strain
energy release rate (as ~a ~ 0) is that the classical strain energy release rate concept is
no longer valid because of its vanishing value. Consequently, a strain energy release
rate associated with a finite crack extension is needed to quantify the weak singular
stress field ahead ofthe crack tip.
The decreasing behavior of strain energy release rate was also noticed by Stringfellow
and Freund [8] who computed the J-integral for the frictional sliding fracture in a thin
film on a substrate. It was pointed out later by Deng [3] that the path independence of J-
integral no longer exists due to the crack surface traction resulting from friction, and the
J-integral for a vanishingly small contour becomes zero indicating strain energy release
rate also vanishes.
In view of the foregoing, a fracture crit~Hon using the strain energy release rate Gof a
characteristic crack extension ~ao as a fracture toughness parameter is proposed, i.e.,
,~We ~u '
=
G(~aO) ---Gd(~ao)-- =G c (15)
~a ~a
where ~We, ~U, and G d (~a 0) are the external work done, strain energy change and
the dissipation energy rate for a crack extension of ~ao, respectively. From equation
(11), it is seen that, for a fixed ~ao the finite extension strain energy release rate G has
a unique relation with the generalized stress intensity factor K II. The fracture criterion
given by equation (15) states that the interfacial crack would grow if the near tip stress
field reaches a certain critical state.
Note that the simple relations
dU 1 dWe G = dU (16)
da""=2da' da
between strain energy release rate, strain energy and work done in linear elastic fracture
mechanics are no longer valid in the presence of friction. Table 2 shows the result
obtained from the finite element analysis for the center crack problem (Figure 1) with ~
= 0.49 and different coefficients of friction. It is evident that the above relations are no
more valid in the presence of friction.
213
Table 2: Energy ratios/or f3 = 0.49 and different coefficients o/friction ~
f.l l1u/ l1We (11Wd + G(l1a) .l1a)/ 11We

0.02 0.4917 0.5084
0.2 0.4883 0.5117
0.5 0.4705 0.5295
5. Effect of Compressive Loading
An open crack may be forced to close if compressive loads normal to the crack surfaces
are applied in addition to the shear load. An example is the residual stresses between
the fiber and matrix resulting from the mismatch of coefficients of thermal expansion.
Theoretically, the open region of the crack cannot be completely closed by externally
applied compression. However, if the open crack is almost closed dure to compressive
loads, then the associated near tip stress field is dominated by a singularity stronger than
A. = 0.5. This has been discussed by Qian and Sun [II]. For this type of "closed"
interfacial crack, the strain energy rek~ne rate for a finite extension l1a is given by
equation (11). It is obvious that if A. > .05, then Gbecomes unbounded as l1a ~O.
Consequently, the intrinsic interfacial fracture toughness must be measured in terms of
the finite extension strain energy release rate Gfor a characteristic distance <la.
However, the toughnesses of the two closed interfacial cracks with stronger and weaker
singularities, respectively, should not be interchanged.
6. Conclusions
The presence of friction on interfacial cracks lying between two dissimilar media may
cause the crack surfaces to close and crack tip stress singularity to become less singular
than the inverse square root singularity. Depending on the loading direction, the
contact size may be extremely small or finite. For the case of small contact zones, the
crack tip can be treated effectively as an open crack and the fracture criterion in terms
of total strain energy release rate and mode mixity can be employed. For the case of
finite contact zones, the classical strain energy release rate vanishes and thus cannot be
used as a fracture parameter as in linear fracture mechanics. A strain energy release
rate based on a finite crack extension is shown to be uniquely related to the near tip
stress field and is capable of capturing the intrinsic fracture toughness of the interfacial
crack.
References
1. Sun, C.T. and Qian, W. (1997) The use of finite extension strain energy release
214
rates for fracture in interface cracks, International J. of Solids and Structures 34,
20,2595-2609.
2. Comninou, M. (1977) Interface crack with friction in the contact zone, J. of
Applied Mechanics 44, 780-781.
3. Deng, x., (1994) An asymptotic analysis of stationary and moving cracks with
frictional contact along bimaterial interfaces and in homogeneous solids,
International J. ofSolids and Structures 31, 17,2407-2429.
4. Gao, Y. C, Mai, Y.W. and Cotterell, B. (1988) Fracture of fiber reinforced
materials, J. of Applied Mathematics and PhysiCS. (ZAMP) 39.550-572.
5. Sigl, L.S. and Evans, A.G. (1989) Effects of residual stress and frictional sliding
on cracking and pullout in brittle matrix composites, Mechanics of Materials, 8, 1-
12.
6. Kerans, R.I. and Parthasarathy, T.A (1991) Theoretical analysis of the fiber pullout
and pushout tests, J. of the American Ceramics Society, 74, 1585-1596.
7. Hutchinson, lW. and Jensen, H.M. (1990) Models of fiber debonding and pullout
in brittle composites with friction, Mechanics of Materials, 9,139-163.
8. Stringfellow, R.G. and Fl __nd, L.B. (1993) The effect of interfacial friction on the
buckle-driven spontaneous delamination of a compressed thin film, International J.
ofSolids and Structures, 30, 10, 13'70,1395.
9. Buchholz, F.G., Wang, H., Ding, S. and Rikards, R. (1995) Delamination analysis
for cross-ply laminates under bending with consideration of crack face contact and
friction, Proceedings of the lath International Conference on Composite Materials
(ICCM-IO) 1, Whistler, BC, Canada, August 14-18, pp. 141-148.
10. Comninou, M. and Dundurs, 1 (1980) Effect of friction on the interface crack
loaded in shear," J. ofElasticity 10, 2, pp. 203-212.
11. Qian, W. and Sun, C.T, Frictional interfacial crack under combined shear and
compressive loading, to appear in Composites Science and Technology.
IMPACT AND HIGH RATE TESTING OF COMPOSITES
An Overview
B.R.K. BLACKMAN AND J.G. WILLIAMS

Imperial College, Exhibition Road
London SW7 2BX. UK.
1. Introduction
1.1 OUTLINE
This paper details the first of two lectures presented at the NATO Advanced Study
Institute on the Mechanics of Composite Materials and Structures held from 12-24th
July 1998. The aim was to present, in a single lecture, an overview of the field of
impact and high rate testing of composite materials. Limits imposed on the length of
published contributions required that the treatment here be succinct however, the reader
should find sufficient additional inform_.. on in the cited references to follow up any
particular test method in more detail.
1.2 BACKGROUND
The use of carbon fibre reinforced polymers (CFRPs) in demanding engineering

applications has increased significantly in recent years. These composite materials are
increasingly replacing metals due to their superior specific strengths and stiffness and
their excellent corrosion resistance. Also, recent advances in manufacturing techniques
have extended the application of composite materials. However, a number of review
articles [1-3] have clearly demonstrated that CFRPs are very susceptible to impact
damage. Unlike most metals which will deform plastically during impact to absorb
large quantities of energy, CFRPs will initially deform in an elastic manner, after which
damage will occur. The problem is particularly acute in CFRPs due to the low failure
strain of carbon fibres. Glass fibre reinforced polymers (GFRPs) have significantly
better impact properties due to the higher failure strain of glass fibre, but are inferior to
CFRP in terms of strength and stiffness.
The development of tougher, thermoplastic matrices e.g. poly ether-ether ketone (PEEK)
has led to a new generation of tougher CFRPs with better impact properties. However,
the high cost of these materials has limited their use and so there is still great interest in
improving the energy absorption properties of epoxy based CFRPs. The toughness of
215
216
epoxy-matrix resins are being significantly improved via the inclusion of toughening
agents and the use of other techniques to enhance composite toughness e.g. fibre
hybridisation, the insertion of interleaf layers and by through-thickness stitching are all
active research areas which were recently reviewed in [4].
1.3 THE NEED FOR IMPACT AND HIGH RATE DATA
The terms impact testing and high rate testing are often used interchangeably however, a
broad distinction can be made. In an impact test there must be an impact event and we
are interested in how the material responds globally, how and where damage occurs and
how much energy is absorbed by these processes. In a high rate test we are interested in
measuring or characterising material properties at high rates. The high rate loading of
the test piece mayor may not involve an impact event.
Impact tests are usually undertaken to simulate an impact loading which may
occasionally be encountered ir.-",;rvice, e.g. a tool dropped on a CFRP aircraft wing or
road debris impacting a racing car travelling at a high speed. In certain applications,
even higher impact velocities may occ--- Satellites for example, have to withstand
impacts with space debris at ballistic velocities. High rate tests are often performed to
measure material properties at the rate of loading caused by the impact. The
measurement of accurate material properties at these higher test rates is important for
structural design and impact modelling, as properties measured at slow rates may be
significantly different and consequently their use can lead to misleading results being
obtained. High rate tests have been and are being developed to measure the tensile,
compressive and shear strengths and moduli of composites as reviewed recently in [5].
Further, high rate delamination tests are now frequently performed to characterise the
dependence of fracture toughness on strain rate. This review does not consider the
measurement of residual strengths after impact.
1.4 TYPES OF IMPACT DAMAGE
The major types of damage which occur in composite materials as a result of impact are
delamination, matrix cracking, fibre failure and penetration. Of these, delamination is
the most common. Delamination involves the initiation and propagation of cracks
running in the interlaminar region between two adjacent laminae in the composite.
Although the presence of delaminations will have a negative effect on most material
properties, the effects on the compressive and interlaminar shear strengths are
particularly deleterious. Delamination damage, even when severe, often cannot be
detected visually and so non destructive test methods are routinely carried out.
Delamination is more likely to occur with short spans, in thick laminates or in
composites with lower interlaminar shear strengths [2]. Fracture mechanics has been
widely used for the measurement of resistance to delamination in polymers and
217
composites at both low rates e.g. [6] and high rates e.g. [7]. Other damage can take the
form of matrix cracking or debonding between the fibre and the matrix and is often the
first failure induced by transverse, low-velocity impact [3]. Fibre fracture will usually
only occur after delamination and matrix cracking and will often precede catastrophic
penetration. Penetration may occur when an impacting object strikes a composite
structure if the structure does not have the time to respond globally to the loading
conditions. The impact of a small projectile on a composite structure will cause high
local stresses and penetration may occur. The penetration of an impacting object may
additionally cause flexural failure on the rear surface and multiple delaminations
throughout the structure.
2. Review of test methods
2.1 BACKGROUND
A review of the most popular •.:st methods for impact and high rate testing of
composites is attempted here however, the review is not exhaustive. The main
advantages and disadvantages of each me .• Jd are discussed. Some relevant current ISO
standards are tabulated later in Section 4.
2.2 TEST METHODS
2.2.1 Pendulum impact tests

The Izod and Charpy tests are perhaps the most well known pendulum impact tests and
the test specimens are shown in Figure 1. In the Izod test a cantilever beam is clamped
in the vertical plane. The bar is notched close to the support clamp and a pendulum
then strikes the specimen. In the Charpy test, the specimen is simply supported in a
horizontal plane and loaded in three point bending with the pendulum striking the centre
of the specimen opposite the notch. Both tests were originally developed for metals and
their popUlarity stemmed from their simplicity. In both tests, the initial and final
potential energies of the pendulum are measured and the potential energy lost by the
pendulum is attributed to fracturing the specimen. More accurate data can be obtained
by the use of an instrumented striker, where a force transducer or calibrated strain gauge
is incorporated onto the striker. A transient recorder may then be used to capture the
force versus time data during the test. ISO standards are available for the testing of
plastics using the Izod or Charpy methods. For the Charpy test, both non-instrumented
and instrumented standard methods are available, as listed in Table 2 later.
The Izod test is not a very suitable test for composites. Not only is no consideration
given to the kinetic energy of the test piece after the impact, but also the radius of the
notch tip prescribed at 250 microns in the standard is too blunt to give conservative
data. Attempts to compute the toughness from Izod data can be very misleading, as was
highlighted in [8]. The Charpy test is better, particularly when instrumented, and quite
218
accurate fracture energies can be measured. However, the Charpy test is prone to
dynamic effects, as described later in Section 3, and the short thick beam specimens are
untypical of engineering CFRP components.
(a) (b)
impactor
specimen
specimen
impactor
clamping jaws
support
Figure 1 The (a) Izod and (b) Charpy tests
2.2.2 Falling weight drop tests

In the falling weight drop test (FWDT), the test specimen is usually supported in a
horizontal plane and an impactor or striker (of known mass) is allowed to fall from a
known height to strike the specimen. The specimen may be simply-supported, as
described in [9] or clamped as in [10]. The energy absorbed may be deduced from the
initial potential energy of the striker, taking into account any rebound of the striker, or
may be deduced from the kinetic energy of the striker just prior to impact.
Displacement gauges are often used to measure the velocity of the striker just prior to
impact. Again, the striker is often instrumented to allow the force-time record of the
impact event to be captured. This is the most popular impact test for composites and
has the advantage that realistic specimen geometries can be tested. Also, the design of
the striker can accurately simulate the geometry of any anticipated in-service impact. A
disadvantage of the test however, is that when penetration occurs, considerable energy is
consumed by friction. Increasingly, servo-hydraulic test machines are being used to
perform impact tests. An instrumented striker may be attached to the hydraulic ram of
the test machine to replace the falling weight. ISO standards for FWDTs applied to
composite materials have recently been drafted for both non-instrumented and
instrumented versions of the test, as shown later in Table 2.
219
2.2.3 Gas Guns: Projectile impact

Experimental gas gun set-ups are used to generate impact velocities from a few tens of
metres per second up to several km per second. Essentially, a mass (often a ball
bearing) is fired along a length of tube and impacts the specimen, which is usually
clamped vertically. The specimen may be in a stressed or unstressed condition, and hence
the effect of any pre-load can be determined [11]. The mass is inserted into one end of
the tube, often within a carrying device. To accelerate the mass, compressed air or
nitrogen gas is first allowed to fill a reservoir at the far end of the apparatus. The gas is
then released into the tube behind the mass via either an electro-pneumatic value, or via
the bursting of a plastic diaphragm. A wide range of projectile masses may be
employed, depending upon the application. The velocity of the mass may be calibrated
against reservoir pressure, or measured optically just prior to impact. The nature of the
damage induced in the test specimen depends upon the mass and velocity of the
projectile and also on the properties of the composite material. Impacted samples are
routinely inspected by C-scan or X-ray radiography to measure the damaged area, which
is then correlated to the incident projectile velocity or energy .
2.2.4 Hopkinson bar method

The split Hopkinson pressure bar may be used to measure tensile, compressive and shear
properties at strain rates of up to about IOOOS·I. These strain rates are much higher than
can be achieved using servo-hydraulic testing machines. The test methods have been
reviewed by Harding [12] and here only the tensile Hopkinson bar method will be
described. A schematic of the tensile version of the split Hopkinson bar is shown in
Figure 2. All Hopkinson bar specimen dimensions are made small to minimise radial
inertia and wave propagation within the specimens.
High pressure reservoir
Strain gauge stations Loading block

Projectile
..
Output bar
Specimen Loading bar
Figure 2 Tensile version of the split Hopkinson bar
In the tensile version, a specimen (which is often waisted) is gripped between an input
and an output bar. Both bars are strain gauged. The high pressure reservoir propels a
cylindrical projectile which then strikes the loading block and loads the specimen. An
analysis of the two strain gauge signals, one on the input bar and one on the output bar,
allow the full dynamic stress strain curve to be derived. Due to the low strains to failure
220
in composite materials, the specimen is sometimes also strain gauged to improve the
accuracy.
2.2.5 High rate tensile testing (and tensile impact)

High rate tensile testing can be performed on servo-hydraulic testing machines, e.g. as
described in [13], to determine the tensile strength and tensile modulus of a composite
material. The samples may either be strain gauged or an optical extensometer used to
monitor the strain of the test piece. Loads can be measured using piezo-electric crystal
load cells. Beguelin et al [13] used a specially developed pick-up unit to smoothly
accelerate the test specimen and reduce impact and associated dynamic effects. Through
the careful design of the specimen and striker, high rate tensile impact tests may be
performed using pendulum machines. In the Ceast apparatus [14], the specimen is
clamped to a calibrated tensile grip, which the pendulum then strikes, loading the
specimen.
2.2.6 High rate delamination testing

Slow speed delamination tel" techniques have been developed for unidirectionally
reinforced fibre composites in mode I (tensile opening), mode II (in-plane shear) and
mode III (out-of-plane shear) and also t~~'miques to combine modes I and II are being
developed. In mode I, the double cantilever beam (DCB) test geometry has been the
most popular. In mode II, the end-notched flexure test (ENF) and the end-loaded split
(ELS) tests have both been extensively used. Mixed-mode 1111 test geometries have
included the fixed-ratio mixed mode (FRMM) test and the mixed-mode bend (MMB) test.
These geometries are shown in Figure 3.
Model Mode II Mixed-mode
DeB
I
• ENF
•• •I • FRMM::1
:2 •
*
•
iF ELS MMB
•I
Figure 3 Some typical interlaminar test geometries for composites.
In recent years there has been a great deal of research interest directed towards extending
the various delamination tests to high rates. Much use has been made of servo-hydraulic
test machines which are now capable of displacement rates of up to 30m/s. Motion is
often transferred to the test specimen via a pick-up unit or lost-motion device. The
221
specimens initially contain a moulded in defect to act as an initial delamination and then
this delamination is extended during the test in one of the test modes. Measurements are
made of the forces, the beam displacements and the delamination lengths during the
progress of the test. From these data, the delamination resistance or toughness, G c, can
be deduced. However, it is particularly important that adequate consideration is paid to
the dynamic effects that are encountered during the tests, if accurate Gc values are to be
obtained as was highlighted in [15-17]. The high rate delamination testing of
composites was recently reviewed by Cantwell and Blyton [18].
2.3 SUMMARY OF TEST RATES AND STRAIN RATES
In the preceding section, a variety of impact and high rate tests have been described. The
choice of the test will depend upon equipment availability and the anticipated in-service
impact event. These tests offer a wide range of test rates and specimen strain rates.
These are summarised in Table 1.
TABLE 1. Comparison of approximate test and strain

rates produced in the various tests.
Test Test rate Strain rate

(m/s) (s'\)
Izod 2 60
Charpy 3 10
FWDT 1-4 10'\_10
Gas Gun 20-240 10.2_10"
Hopkinson bar 103
High rate delamination 10. 1-10
3. Dynamic effects
3.1 THE NATURE OF DYNAMIC EFFECTS
Perhaps the most significant problem associated with impact or high rate testing is the
occurrence of dynamic effects during the test. , These effects originate from the stress
waves which propagate through a test piece on rapid loading. During a slow speed or
static test, there is sufficient time for the test piece to deform and force equilibrium is
maintained throughout. As the test rate is increased, the response of typical composite
components becomes dominated by flexural waves. This response leads to violent
oscillations in the recorded values of the load measured and may also lead to complex
crack growth behaviour where the stress waves interact with a growing crack leading to
accelerations, deceleration and sometimes to stick-slip crack growth where the crack
grows in short bursts followed by a period of crack arrest [15-16]. If the rate of loading
is increased even further and thus the impact times become very short, then the response
of the composite sample becomes dominated by dilatational waves [19]. These waves
travel very much faster than flexural waves and lead to very localised damage e.g.
222
penetration of the striker into the test piece without global deformation of the specimen
occurring. The kinetic energy imparted to the test piece will also result in inertial loads
being recorded.
3.2 SCHEMES TO MITIGATE DYNAMIC EFFECTS
If any useful information is to be gained from performing an impact or high rate test,
then careful consideration needs to be given to the dynamic effects encountered. If
dynamic effects are not adequately accounted for, then very misleading and erroneous
conclusions can be reached. The first and perhaps most common approach followed by
researchers is to experimentally minimise the dynamic effects encountered. This can be
achieved by either limiting the permissible impact velocity or damping the contact
between the striker and the specimen. A second approach frequently followed involves
digitally filtering the force signals recorded during the test, so as to 'remove' or separate
the structural response from the desired material response. Such filtering however,
requires a very detailed knowledge of structural dynamics and great care needs to be taken
when filtering signals if the true material response is to remain unaltered. A third
approach which is becoming increasinrlu popular is to modify the test procedure, so
that rather than measure the load (which is severely impeded by dynamic effects) the
specimen displacement or time to fracture is measured. These methods have the
advantage that provided an appropriate value of the material modulus is known, then the
fracture energy can be measured without needing to measure the load. This technique
has been applied to high rate delamination testing [15-17] and is also being applied to
the impact testing of polymers at rates greater than I mls by the European Structural
Integrity Society (ESIS), Technical Committee (TC4) on Polymers and Composites. A
final approach which has been followed is to model the impact event either analytically
or numerically, and so attempt to understand the dynamic character of the test, and
separate this from the material property. For example, a mass-spring model was
proposed [20] for the Charpy impact test which modelled the contact stiffness between
the striker and the specimen. For this simple geometry, useful results were obtained.
However, for more complex tests, analytical solutions are rare and much use is now
made of finite element code to numerically model the dynamics of the test.
4. Standardisation of impact and high rate tests
Satisfactory standards for impact and high rate testing of composites are rare, although
much progress is currently being made, as evidenced by a number of new draft
International Standards (DIS) now under consideration as shown in Table 2. An
instrumented impact test based on the Charpy method is now available for plastics as an
ISO standard. The falling weight drop test is being developed and both non-
instrumented and instrumented versions of the standard are currently in draft (ISO DIS)
form. The draft standard allows the use of either a drop weight tower or a servo-
223
hydraulic test machine. A new protocol is now under consideration as a New Work item
for the determination of G1C or K1C for plastics at a test rate of hnls. Currently, no
standards exist for the high rate delamination testing of composites.
TABLE 2. Availability of current impact and high rate standards and draft standards
Type Number Name Year
Izod ISO 180 Plastics-Determination of Izod impact strength 1993

Charpy ISO DIS 179-1 Part 1: Non instrumented impact test *
Charpy ISO 179-2 Part 2: Instrumented impact test 1997
FWDT ISOIDIS 6603-1 Determination of puncture impact behaviour of
rigid plastics: Non instrumented impact test
*
FWDT ISOIDIS 6603-2 Determination of puncture impact behaviour of *
rigid plastics: Instrumented imc:ct test
KIC orGlc ISOINWIP 694 Determination of fracture toug ess KIC or G IC +
of plastics at high loading rates
(Note: (*) Draft international standard (DIS), e) New work item at time of writing.
5. Conclusions and outlook
Composites, particularly when reinforced with carbon-fibres, are prone to impact

damage. Improvements in matrix toughness and the use of interleaf layers, hybrid fibre
and through thickness stitching have all contributed to improved impact resistance.
However, impact resistance remains a key issue. Due to the complexities involved in
impact and high rate testing, few satisfactory standards currently exist, although progress
is being made and much effort is being directed towards devising better test methods.
However, until these exist, a wide variety of tests will continue to be used.
6. References
1. Cantwell, WJ., and Morton, J., (1991), The impact resistance of composite
materials - a review. Composites 22 (5), 347-362.
2. Matthews, FL, and Rawlings, R.D., (1994), Composite Materials: Engineering
and Science. Chapman Hall, London.
3. Richardson, M.O.W., and Wiseheart, MJ., (1996), Review oflow-velocity
impact properties of composite materials, Composites Part A, 27A, 1123-1131.
4. Zhong, W., and Jang, B.Z., (1998), Material design approaches for improving
impact resistance of composites. Key Engineering Materials (141·143), 169-
186.
5. Harding, J., and Ruiz, C., (1998), The mechanical behaviour of composite
materials under impact loading. Key Engineering Materials (141.143), 403-426.
6. Hashemi, S., Kinloch, A.J., and Williams, J.G., (1990), The analysis of
interlaminar fracture in uniaxial fibre-polymer composites, Proc. R. Soc.
Lond. A427, 173-199.
224
7. Williams, J.G., (1994), High speed delamination of composites, Eur. 1. Mech.,

AlSolids, 13 (4) -suppl., 227-241.
8. Lowe, AC., Moore, D.R., and Rutter, P.M., (1995), An investigation of the
measurement of the toughness in phenolic resins and their short fibre reinforced
composites. Impact and Dynamic Fracture of Polymers and Composites, ESIS
19, Mechanical Engineering Publications, 383-400.
9. Winkel, J.D., and Adams, D.F., (1985), Instrumented drop weight impact testing
of cross-ply and fabric composites. Composites 16 (4), 268-278.
10. Found, M.S., and Howard, I.C., (1995). Single and multiple impact behaviour of
a CFRP laminate. Composite Structures 32, 159-163.
11. Hancox, N.L., (1996) Making an impact: Composites in the automotive
industry. Materials World (AprU), 197-199.
12. Harding, J., (1992), Shock-wave and high strain rate phenomena in materials,
Marcel Dekker Inc., New York.
13. Beguelin, Ph., Barbezat, M. and Kausch, H.H., (1991) Mechanical
characterization of polymers and composites with a servohydraulic high-speed
tensile tester, 1. Phys. III France 1, 1867-1880.
14. CEAST (1996), Introduction to impact. Technical Report CE-TR-Ol (96).
15. Blackman, B.R.K., Dear, J.P., K!-'och AJ., MacGillivray, H., Wang, Y.,
Williams, J.G and Yayla, P. (1995), The failure of fibre composites and
adhesively bonded fibre composites under high rates of test. Part I: Mode I
loading- experimental studies, 1. Materials Science 30, 5885-5900.
16. Blackman, B.R.K., Kinloch, AJ., Wang, Y., and Williams, J.G., (1996), The
failure of fibre composites and adhesively bonded fibre composites under high
rates of test. Part II: Mode I loading- Dynamic effects. 1. Materials Science 31,
4451-4466.
17. Blackman, B.R.K., Dear, J.P., Kinloch A.J., MacGillivray, H., Wang, Y.,
Williams, J.G and Yayla, P. (1996) The failure of fibre composites and
adhesively bonded fibre composites under high rates of test. Part III. Mixed-
mode I/II and mode II loadings. 1. Materials Science 31, 4467-4477.
18. Cantwell, W.J., and Blyton, M., (1998), The interlaminar fracture properties of
composite materials at high rates of strain. Key Engineering Materials (141-
143), 463-476.
19. Olsson, R., (1998). Impact response and delamination of composite plates.
Report 98-12, ISSN 0280-4646.
20. Williams, J.G., (1984), Fracture Mechanics of Polymers, Ellis Horwood Ltd.,
Chichester.
IMPACT AND HIGH RATE TESTING OF COMPOSITES
High rate delamination testing
B.R.K. BLACKMAN AND J.G. WILLIAMS

Imperial College, Exhibition Road
London SW7 2BX. UK.
1. Introduction
1.1 OUTLINE
This paper details the second of two lectures presented at the NATO Advanced Study
Institute on the Mechanics of Composite Materials and Structures held from 12-24th
July 1998. The aim of the second lecture was to describe in detail a study which was
undertaken to investigate the high rate delamination resistance of unidirectionally
reinforced carbon fibre polymer composites. Limits imposed on the length of the
published contributions necessitate that_Jch of the detail be omitted here. However, a
full description and presentation of the results and analyses may be found in the cited
references.
1.2 BACKGROUND
Many studies, e.g. as reviewed in [1], on the impact resistance of CFRPs have
concluded that this class of material is particularly susceptible to impact damage.
Delamination represents a major form of impact damage in CFRPs and is particularly
dangerous because it is often not visible from the surface. Research efforts to measure
of the resistance of composite materials to delamination have made extensive use of
fracture mechanics. The property measured is the critical energy release rate, or fracture
toughness, Ge, and is a measure of the energy consumed during the creation of unit area
of fracture surface during delamination. Three modes of crack loading can occur, namely
mode I (tensile opening), mode II (in-plane shear) and mode III (out-of-plane shear). In
practice, modes I and II and combinations of I and II are the most important.
Test methods for measuring the interlaminar fracture toughness at slow rates in mode I,
II and mixed-mode 1111 are well established and several standards exist for mode I [2-4]
and various test methods are currently being pursued for the other modes. Due to the
popularity of the test methods there has been much interest recently in applying them at
higher rates.
225
226
However, when extending the slow speed test methods to high rates, a number of
significant obstacles need to be overcome. Firstly, experimental test equipment capable
of rapidly accelerating the test specimen and then accurately recording the forces applied
and the deformation behaviour of the beams are required. Secondly, dynamic effects are
invariably induced when performing these high rate tests and it is critical that these
effects be carefully considered and accurately accounted for if accurate and valid Gc values
are to be measured. Indeed, this probably accounts for the conflicting nature of some of
the test results reported in the literature. For example, Smiley and Pipes [5] reported
results on two unidirectionally reinforced CFRPs, the first based upon a brittle epoxy
and the second based upon a tough thermoplastic matrix, poly ether-ether ketone
(PEEK). Their results pointed to very large reductions in the values of G,c and Gnc for
both materials as the test rate was increased from a few mm/min to about Im/s.
Beguelin et al. [6] reported mode I results on a unidirectional CFRP based upon a PEEK
matrix and their results indicated that only a small reduction in the value of G,c occmred
as the test rate was similarly increased. In a third study by Aliyu and Daniel [7] on
similar materials, increasing followed by decreasing values of G,c were reported as the
test rate was increased.
The differences in the experimental resl'lt~ reported were further highlighted in a recent
review by Cantwell and Blyton [8]. Their review indicated that the rate sensitivity of
unidirectional CFRP was dominated by the toughness of the matrix, with brittle matrix
composites exhibiting much less of a rate effect than tough matrix composites.
However, they did not suggest any reasons why such different results had been obtained
by different workers on very similar materials. The detailed study which forms the basis
of this work [9-11] concluded that the differing results were caused by the different
experimental approaches used, and the different analysis methods employed.
2. Experimental
2.1 MA1ERIALS EMPLOYED
Two different unidirectional CFRP composites were employed in the study. These
were:
2.1.1 Epoxy/carbon-jibre composite

A unidirectional carbon-fibre epoxy-resin composite (Fiberdux 6376C supplied by
Hexcel Composites, UK). The volume fraction of the thermosetting matrix was
nominally 35% and the carbon fibres were T-400 type (from Toray Inc., Japan).
2.1.2 PEEKlcarbon-fibre composite

A unidirectional carbon-fibre poly (ether-ether) ketone (PEEK) composite (APC-2
supplied by ICI PIc.) The volume fraction of the thermoplastic PEEK matrix was
nominally 35% and the carbon fibres were AS4 type (from Hercules Inc., USA).
227
2.2 TEST SPECIMENS AND TEST RIG
Test coupons were prepared from manufactured sheets of composite containing a release
film material at one end. The geometries used in this study are shown in Figure 1.
MODEIDCB MODE II ELS MIXED MODE IIII
Figure 1 The test geometries used in this study
For mode I, the double cantilever beam (DCB) test specimen was used. For mode II, the
end-loaded split (ELS) test geometry was used and for mixed-mode, the fixed-ratio
mixed-mode (FRMM) geometry was employed where the ratio of GlG n was 4/3.
Samples contained film inserts (approximately 12 microns thick) to act as crack starters.
Samples were 25mm wide and were pre-cracked in mode II prior to testing. This ensured
that the starter crack had grown away from any matrix-rich region in the vicinity of the
film insert and that fibre bridging did not develop prior to mode I loading. The fixture
used for the ELS and FRMM testing incorporated rollers, which allowed the specimens
to slide freely in the horizontal plane. The data acquisition for the high rate test rig
employed is shown schematically in Figure 2.
l Instron Control I Signal from LVDT

-----
Linear Actuator
t
Module oram, oram
DCB
D
I
~
P, t
20MHz Load-cell
Oscilloscope
I Amplifier
-----
Piezo-electric
Load-cell
IEEE
~,a,a,t Projector:
~
Comruter Analysis or the H~h-Speed
(Del 486) films amera
Figure 2 Schematic of the data acquisition used in the high rate tests.
228
A servo-hydraulic testing machine (lnstron model 1343) was used which was capable of
displacement rates of up to 3Om/s. Static tests, below l00mmlmin, were performed on
a screw driven tensile testing machine. DeB, ELS and FRMM fixtures were attached to
the test machine and the specimens were coupled to the hydraulic ram via a lost motion
coupling device, which ensured a constant loading rate was attained during the tests.
Full details of the test rigs can be found in [9] and [11].
2.3 LOAD MEASUREMENT
The high speed test rig incorporated a piezo-electric load cell connected to a signal
amplifier. Data was captured on a 20 MHz digital oscilloscope and the signals were
transferred to computer where signal analysis could be conducted. The load cell was
positioned very close to the specimen in all tests. At rates above about Im1s, the load
traces were severely impeded by dynamic effects, and any attempt to extract load values
from these signals for the calculation of Gc values would be very suspect. It was
strongly recommended in [9· •• ] that these load values should not be used in the
analysis.
2.4 DISPLACEMENT AND CRACK LENGTH MEASUREMENT
High speed photography was used to measure the crack length and the beam deformation
behaviour. The camera was a 16mm, rotating prism design, capable of framing rates of
up to 40000s· l • Each test was filmed, providing accurate values of the beam
displacement and the crack length as a function of time. The crack velocities could
therefore be deduced from the film. The films clearly revealed the dynamic character of
the beam response at test rates above Im1s and it was evident that the DCB tests became
partially asymmetric at the highest test rates. The position of the hydraulic ram was
also recorded using an L VDT on the test machine. It was found that very misleading
results would be obtained if the displacement of the specimen was deduced from the
position of the hydraulic ram. This would underestimate the value of the beam
displacement at high test rates when a loss of contact occurred between the hydraulic ram
and the pick-up unit. This could result in very low values of Gc being deduced.
3. Analysis
3.1 STATIC ANALYSIS
Analyses for determining the interlaminar fracture toughness for composites tested in
mode I. mode IT and mixed-mode I/II have been reported for static, slow speed testing
e.g. [12,13]. Such analyses ignore the contribution of kinetic energy to the energy
balance and are unable to predict the onset of any dynamic effects. Static expressions for
Gc usually require the measurement of the load, displacement and crack length in
229
addition to the width and thickness of the beam. At slow test rates, the measurement of
all of these variables can be undertaken with good accuracy. At higher test rates
however, dynamic effects render the measurement of load, displacement and crack length
difficult. However, by the use of high speed photography, the displacement and crack
length were accurately monitored and thus the only variable with significant uncertainty
was the load, which oscillated violently at test rates greater than about Im/s. An
alternative form of the analysis, which is more suited to high rate testing, is independent
of the load and requires the axial modulus of the specimen to be known. Values of Gc
may therefore be determined without requiring accurate load values to be known. The
modulus value used in this analysis should be that measured at an equivalent test rate
i.e. a high rate or dynamic value is required. These values were deduced by an ultra-
sonic method using Lamb waves [14]. Mechanical tensile modulus testing by Harding
and Welsch [15] indicated that the tensile dominated properties of unidirectional CFRPs
were not strain rate dependent and this agreed well with the ultra-sonic results.
3.2 DYNAMIC ANALYSIS
A dynamic analysis should include the kinetic energy contribution in the energy balance.
Exact analytical solutions including kit . 'IC energy are uncommon for beam problems
due to the nature of the equation of motion [16]. However, a useful approximation was
obtained using the ''Berry Method" where the static displacement profile was assumed
[17], hence the velocities were derived and the kinetic energies computed. In the DeB
test, there are two cases of interest. Firstly, the instant just prior to crack initiation
when the crack velocity was zero and secondly the steady-state propagation of the crack,
when there was a contribution to the kinetic energy from the crack motion. These
kinetic energy terms were included in the calculations of Gc .
3.3 TRANSIENT EFFECTS
During steady-state crack propagation, the crack is 'fully aware' of the boundary
conditions. The boundary condition assumed is {a=O at t=O}, where a and t are the crack
length and time respectively [18]. However, for the interlaminar test geometries, there
is always an initial crack length and so the crack initiates after a time delay with a finite
velocity. In addition, due to the different kinetic energy values for crack initiation and
for steady-state propagation, the energy requirements to initiate the crack are greater than
are required to propagate it. There is therefore a transition growth region after initiation
and before the crack reaches the steady-state. This transient crack growth was detected
experimentally in the high speed films [9] and has also been seen in finite element
modelling results [19]. An expression for the kinetic energy when the crack is in the
transition region between the stationary and the steady-state can be written, but the form
of this differential equation is somewhat intractable. However, approximate solutions
have been obtained by the use of a perturbation analysis [10].
230
4. Results
4.1 INTERLAMINAR FRACTURE TOUGHNESS
The values of Gle as a function of test rate are shown in Figures 3(a) and 3(b) for the
epoxy/carbon-fibre composite and the PEEK/carbon-fibre composite respectively.
Values shown are for crack initiation. However, for the PEEK/carbon-fibre composite,
where stick-slip crack growth was observed, initiation and arrest values are shown. The
results show that the values of Gle for the epoxy/carbon-fibre composite exhibited no
rate dependence when the test rate was increased up to 15m/s. However, the
PEEK/carbon-fibre composite showed a somewhat larger rate dependence, with the
values of Gle at crack initiation decreasing by up to 20% of the static value. However,
the reduction was much less severe than had been reported elsewhere in the literature.
The arrest values were largely insensitive to test rate.
(a) (b)
·······r·······l········r·······l········j········r···...
I:: :ir~-rm_:r:·::tt:~:~Li.fn::
o.~
.r-
2.0
, . .
=nWt ........;......... !....... .
•....•• : .••..... !........: ....... !
• !~~.m"j
o
~
=TFJlil o
10 5
:JPi!
1at 10. 1
~
10.2
:
10" let Id 102

!'~-ldl~,~~
0.0
1~ 1~ 1~ I~ 1~1 ,~ 1~ ,~
Displacement rato (mIs) Displacement rato (mil)
Figure 3. Gle versus specimen displacement rate for (a) the epoxy/carbon-fibre
composite and (b) the PEEK/carbon-fibre composite. (Circles: initiation values, squares:
arrest values).
Figures 4(a) and 4(b) show the mode II ELS test results for the two composites and
Figures 5(a) and (b) show the mixed-mode FRMM test results. These results show a
similar trend to the mode I results, namely that there was negligible reduction in Glle or
Glem and Gllem when testing the epoxy/carbon-fibre composite, but small reductions
were measured in these values when testing the PEEK/carbon-fibre composite at test
rates above about 3m/s.
231
(a) (b)
, , ,
~,:d!~ijl~ ~ ': L1IjrtL~

~g 1.5 --------j----- ---r-·--.. ··-r--------:--- ---.. -~.-.
0.3 ______ ., _______ , _______, ______ : ____ _
~O~.~,~~.~,O~.~·~~~,O~'~,~~~'~,~~~
Displacement rate (mls)
Figure 4. GIIC versus specimen displacement rate for: (a) the epoxy/carbon-fibre
composite and (b) the PEEK/carbon-fibre composite.
(a) (b)
1.4 .............,.~rr"."...,--~""
: : : :
~~~~41~ ~: ~ rtitl~;11
0.10
0.050
-------~---.- •••;.-----.-~--. -_ •. j- --------j. ... _.••
0.20 -------r -.----~.---. -+----- -.-{ ------+--
0.0 L.,.....L~""'___"~L....."""_~-'--'-.......'---'-'-''"'" 0,0 L...........L~""--'-~L.....,....L~.........~L...........J

10-' lc}s 10. 103 10-2 10-' 100 10' 10" 10" 10.4 10.3 10- 2 10' 10° 10'
Displacement rate (mls) Displacement rate (m/s)
Figure 5. G IC m and G ncm versus specimen displacement rate for (a) the epoxy/carbon-
fibre composite and (b) the PEEK/carbon-fibre composite. (Filled points: G IC m values,
open points: GIIC m values).
4.2 TRANSIENT CRACK GROWTH
The transient analysis was applied to the DCB test results. The crack growth behaviour
measured using high speed photography was compared to that predicted by steady-state
theory. Values of crack length and crack velocity were examined. Figures 6(a) and 6(b)
show the results obtained from a DCB test at 8m1s using the epoxy/carbon-fibre
composite. Steady-state theory predicted that the crack should initiate at t=O, with
infinite velocity. However, the effective initial crack length, ao • in this test was 40mm
and crack initiation did not occur until t=O.5 milliseconds, with a velocity of about
92m1s. Thus, the real crack initially lagged behind the steady-state, then overshot it.
During the remainder of the test, the crack 'hunts' the steady-state value, as can be seen
in Figure 6.
232
(a) (b)
140 r-~"":~..,:,..............,...,:~.,.......,:~.........:~,......,
E 120 ·'~.:~:"'i""'··"j"·""'·y"·"··:"··· ... 80
...J
5
100
80
········T········T········1"········~ ....... ,. ....... .
! ! . :"'! !
···ru·····T··..····r·······
-·······T···-·~·~ ..·· f
!
80
40
.5 "~:lll:!
20~~~~~~~~~~~~
l
~ .0
o 0.5 1.5 2 2.5 0.5 1 1.5 2 2.5

Time (milliseconds) Time (milliseconds)
Figure 6 (a) Crack length values and (b) crack velocity values versus time for the
epoxy/carbon-fibre composite during a DeB test at a rate of 8m/s. Data points represent
the experimental values and the solid lines represent the values predicted by steady-state
theory.
The transient analysis yielded approximate solutions to the crack length and velocity
perturbations i.e. the disturbance from UI~ steady-state. The experimentally measured
perturbations (shown as the points in Figure 7) were compared to the theoretical values,
the bold lines in Figure 7. The form and magnitude of the predicted normalised
perturbations agreed well with the experimentally measured values.
(a) (b)
0.6
0.'
0.' D.'
.......
0.1 D.'
.!!,.
{= 0.0 0.0
-0.1 -0.'
-0.' .Q.'
·0.6
0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.0 0.5 1.0 1.5 '.0 '.5 3.0
Time (mlUlseconds) Time (milliseconds)
Figure 7. Values of the normalised perturbations versus time the epoxy/carbon-fibre

composite at a test rate of 8m/s. (a) Normalised crack length perturbation (P/~), (b)
Normalised crack velocity perturbation (Pt.,l~). Points are experimental data, solid lines
are the theoretically predicted values.
233
5. Conclusions
Values of G)C were measured for two unidirectionally reinforced carbon-fibre polymer
composites at test rates from 2mm/min up to about 15m/s. Values of Gnc and rnixed-
mode G)Cm and Gnc m were also measured for the two composites up to test rates of about
5m/s. For the epoxy/carbon-fibre composite, no significant reductions in Gc were
recorded in any mode as the test rate was increased. For the PEEK/carbon-fibre
composite, modest reductions were observed in the values of Gc, at test rates above
about 3m/s. However, the maximum loss of toughness measured at these high rates did
not exceed 20% of the static value. This reduction was much less than had been reported
elsewhere in the literature. A dynamic transient analysis was able to predict quite
accurately the nature of the crack growth at high rates. In particular, the disturbance
from the steady-state, or perturbation, was examined and the transient analysis agreed
well with the experimental results.
6. Acknowledgements
The authors gratefully acknowledge th. ~ontributions of their co-workers at Imperial

College, London, during the course of this study: J.P Dear, AJ. Kinloch, H.
MacGillivray, Y. Wang and P. Yay la, and the EPSRC for research funding.
7. References
1. Cantwell, W.J., and Morton, J., (1991), The impact resistance of composite
materials - a review. Composites 22 (5), 347-362.
2. ISO CD 15024 (Version 97-02-24) Standard Test Methodfor Mode I
interlaminar Fracture Toughness, Gic, of unidirectional Fibre-Reinforced
Polymer Matrix Composites, International Organisation for Standardisation.
3. AS1M D 5528-94a: Standard Test Methodfor Mode I Interlaminar Fracture
Toughness of Unidirectional Fiber-Reinforced Polymer Matrix
Composites, AS1M Annual Book of Standards, Vol. 15.03,272-281.
4. 1IS K 7086 (1993) Japanese Industrial Standard testing methods for interlaminar
fracture toughness of carbon fibre reinforced plastics, Japanese Standards
Association.
5. Smiley, AJ. and Pipes, R.B., (1987), Rate effects on mode I interlaminar
fracture toughness in composite materials, J. Composite Materials, 21 (7)
670-687.
6. Beguelin, Ph., Barbezat, M. and Kausch, H.H., (1991) Mechanical
characterization of polymers and composites with a servohydraulic high-speed
tensile tester, J. Phys. III France 1, 1867-1880.
7. Aliyu, AA and Daniel, I.M., (1985) Effects of strain rate on delamination
fracture toughness of graphite/epoxy. Delamination and Debonding of Materials,
234
ASTM STP 876, American Society for Testing and Materials, Philadelphia
336-348.
8. Cantwell, W.J., and Blyton, M., (1998), The interlaminar fracture properties of
composite materials at high rates of strain. Key Engineering Materials (141-
143), 463-476.
9. Blackman, B.R.K., Dear, J.P., Kinloch AJ., MacGillivray, H., Wang, Y.,
adhesively bonded fibre composites under high rates of test. Part I: Mode 1
loading- experimental studies, J. Materials Science 30, 5885-5900.
10. Blackman, B.R.K., Kinloch, A.J., Wang, Y. and Williams, J.G. (1996)
The failure of fibre composites and adhesively bonded fibre composites under
high rates of test. Part II: Mode I loading-dynamic effects, J. Materials
Science 31 4451-4466.
11. Blackman, B.R.K., Dear, J.P., Kinloch, A.J., MacGillivray, H., Wang, Y.,
adhesively bonded fibre composites under high rates of test.
Part III: Mixed-mode IIII and mode II loading, J. Materials Science 31,
4467-4477.
12 Hashemi, S., Kinloch, A.J. and UTUliams, J.G., (1990) The analysis of
interlaminar fracture in uni-axial fibre-polymer composites, Proc. R. Soc. Lond.
A 427, 173-199.
13. Kinloch, A.J., Wang, Y., Williams, J.G., and Yayla, P. (1993) The mixed-
mode delamination of fibre composite materials, Composite Science and
Technology 47, 225-237.
14. Alleyne, D.N. and Cawley, P., (1992), Optimisation of lamb wave inspection
techniques, NDT & E International 25, 11-22.
15. Harding, J. and Welsh, L.M., (1983), A tensile testing technique for fibre-
reinforced composites at impact rates of strain, 1. Materials Science 18, 1810-
1826.
16. Berry, J.P., (1960), Some kinetic considerations of the Griffith criterion for
fracture -I: Equations of motion at constant force, 1. Mech. Phys. Solids 8 (3),
194-206.
17. Berry, J.P., (1960), Some kinetic considerations of the Griffith criterion for
fracture -I: Equations of motion at constant deformation, J. Mech. Phys. Solids
8 (3), 207-216.
18. Williams, J.G., (1994), High speed delamination of composites, Eur. J.
Mech., AlSolids, 13 (4-suppl.), 227-241.
19. Wang, Y. and Williams, J.G., (1994), A numerical study of dynamic crack
growth in isotropic DCB specimens, Composites 2S (5) 323-331.
COMPOSITES FOR MARINE APPLICA TIONS
Part 1. Testing of Materials and Structures for Surface Vessels
P. DAVIES
Marine Materials Laboratory
IFREMER Centre de Brest
BP 70, 29280 Plouzane
France
1. Introduction
The use of fibre reinforced polymer composites in marine structures is widespread, and
has been reviewed in several recent publications [1-4]. These structures are designed for
service in a hostile environment, and two conditions must be met if satisfactory
performance is to be assured:
i) the behaviour of the materials in this environment must be known,
ii) the tools used for structural design must be validated
Laboratory testing is an important part of the design process, as shown in Figure I.
Valid atio n
Laboratory
Simulation < ) Modelling
T",
Improvement
~
Se a Trials
& In-S e rvice
experience
Figure I. The role of laboratory testing in design of marine structures
The tests employed to simulate the marine environment are rarely standardized, but
evolve continuously based on the lessons learned from in-service experience. This paper
will present some of the tests used at IFREMER to simulate the mechanical loading of
marine materials and structures. Hydrostatic pressure tests are not included, they will be
presented in the following chapter on underwater applications [5].
235
236
The paper consists of three parts. First, the materials usually considered for marine
applications will be briefly described. These materials differ from those commonly
employed in the aerospace industry, which has several implications for test procedures.
Tests performed on coupons to characterize short and long term material behaviour will
then be described. In particular, scatter and scale effects will be discussed. In the final
part of the paper, tests on structures will be presented, and examples involving both
static and impact loading will be given.
2. Materials
The materials most commonly used are glass/polyester laminates made by hand lay-up,
and sandwich combinations of laminates with foam or balsa cores. Filament wound
materials are also used in some higher performance applications. Table I presents the
materials most widely used in marine structures.
TABLE 1. Marine composite constituents

Reinforcement Matrix resin Comments
E glass Hand lay-up
Woven, Chopped Mat, Polyester. (Vinyl ester, Low fibre fraction
'Rovimat', Epoxy, J:'henolic) (Vc 30%)
(woven rovings + mat) Thick composites, often
Stitched fabrics >IOmm
E, S or R-glass, Carbon Epoxy Filament winding
V f (60%)
3. Tests to characterize materials
There are a large number of standard test methods available for laminates. Although
these are generally satisfactory for in-plane elastic properties there remains considerable
controversy over tests to determine strengths. Indeed, for compression and shear there
are several different standards, none of which is completely satisfactory. These tests
were mostly developed for aerospace materials, so the first question to address is :
To what extellt call standard tests be applied to marille composites?
The difficulties in applying these tests originate from the large weave size, the low
volume fractions and high potential scatter in properties due to hand lay-up fabrication.
These aspects will now be addressed. Fracture tests are not yet completely standardized,
and will be treated separately. Finally, sandwich materials pose particular problems
which will also be discussed below.
3.1. SIZE EFFECTS

Weave patterns may be 5 to lOmm wide, so the dimensions of specimens may not be
large enough to be representative of bigger structures. In order to study this, and also to
237
examine scale effects, a series of tests was performed on glass rovimat reinforced
specimens scaled in dimensions from I to 10, i.e. from one rovimat layer (lmm) thick
and 62.5 by 6.25mm dimensions, up to ten layers (10 mrn thick) and 625 by 62.5mm
dimensions and la ers [6]. Some results are shown in Fi ures 2 and 3.
._.__..___..__.._.M..._._.._..__..._____.___.__.__.__
250 18
x-x
+-/ "x-x--x 17
CI:S
tl.. 200 CI:S
tl..
::E .....x 16 tJ
..ci'
.....
bI)
~
150
x
----. en
-
15 ;:s
;:s
"d
f::!
..... 14 0
::E
-..... .-
til
0 100 13 ....
0
CJl CJl
~ ~
~ -x- Strength 12 0
50 E-<
-+--- Modulus 11
0 10
0 5 10
Scale (Number oflayers)
Fi ure 2. Tensile results from scaled s ecimens
75
CI:S 70
Il..
::E 65
.sbI) 60
~
0
tl
CJl 55
la0 50 -+--- Increasing scale
..c: - - Increasing width
til
45
40
0 100 200 300 400 500
Cross section, ~
Figure 3. Apparent in-plane shear strength versus specimen section area.

Within experimental scatter the only effects of specimen scale on results were at the
very small scale, for which it was hard to obtain the same fibre fraction as for thicker
specimens, and for in-plane shear strength values. The latter increased both with
238
specimen scale and width, as larger specimens could accomodate more damage before
final failure, and a plateau value appeared only for the largest specimens. Nevertheless,
overall these results are encouraging as they suggest that tests on standard specimens
may be used to predict the behaviour of large structures.
3.2 SCATIER IN PROPERTIES

The dominant fabrication technique for marine applications is hand lay-up. This is true
for both pleasure boats and large military craft, although recent legislation and
technological advances are resulting in increased interest in closed mould techniques
(RTM, 'SCRIMP'). The hand lay-up route involves the placement of reinforcement
layers and their impregnation by rollers, and there is a risk of variability in properties.
An exercise was performed to evaluate the variation in properties of composites
produced in 14 different boatyards, who were each asked to produce one square meter of
a 5 ply thick rovimatlpolyester laminate [7]. Figure 4 shows some results.
cu 250
~!1~
~ 0°
~ 200
....
.c:~
rIott:!
.
c: 150
OJ)
....
~
45°
-.... 100
(Il
50
(Il
c:~
Eo-;
0
0 5 10 15
Boatyard
Figure 4. Tensile strengths, specimens from 14 boatyards.
At first sight these appeared to suggest a very large variation in properties. However,
when the results are considered more carefully and the details of the fibres and resins
used by each boatyard are examined it becomes apparent that there are reasons for many
of the anomalies. In some cases resin-rich surface layers were applied, as this is the
practice in some boatyards, in others the use of separate mat and woven rovings rather
than rovimat resulted in lower fibre contents. In one case very high void content (10%)
was also noted. Taking 12 of the 14 panels the coefficients of variation for tensile
strength and modulus on specimens tested in the fibre direction and at 45° vary from 9
to 14%. This suggests that high safety factors to account for property variations in
marine composites are not justified provided reasonable quality procedures are
followed.
3.3 FRACTURE OF MARINE COMPOSITES

There has been considerable activity directed towards the standardization of
interlaminar fracture tests on composites in recent years [8,9], but these tests have been
mainly applied to unidirectional aerospace materials. Interlaminar fracture also occurs in
239
marine composites, but the characterization of delamination resistance in these materials

is complicated by off-axis fibres, low stiffness specimens and the influence of large
weave size which causes considerable heterogeneity in the interlaminar region.
Nevertheless valid tests can be performed and Figure 5 shows an example of the use of
mixed mode fracture tests to examine the influence of a delay overnight during
lamination on the mixed mode toughness of a glass/polyester composite [10]. The mode
I response is only slightly affected but the resistance to shear dominated loading is
seriously reduced by this delay.
/'
:§
......
1000
B
....CIS 800
Q)
rJ) .......
CIS
,.
Q)
] 600 ~
»
bI) /'
/'
....
Q)
400 /
t::
Q) ,. /'
t::
·ca -+- DELAY 16 h
.b
rJ)
200
~ CONTINUOUS
....
'"ta
0
E-< 0
0 0,2 0,4 0,6 0,8 1
Percentage Mode II
Figure 5. Mixed mode bending (MMB) test results for glass QXlpolyester composite
fracture. Influence of a delay during fabrication.
3.4 TESTS ON SANDWICH SPECIMENS

The tests commonly performed on sandwich materials are of two types: quality control
tests, designed to check the core/facing integrity, and tests to determine properties for
calculations. The former generally use a short beam loaded in flexure, while the latter
involve more complex arrangements designed to yield shear, tension and compression
data, Figure 6. However, by careful test design it is possible to use a single test to
characterize sandwich beams. This involves four point loading of instrumented beams,
as recommended in a German standard DIN 53293, Figure 6d. By measurement of
upper load point displacement, lower facing central displacement and strains in the
facings it is possible to determine: D, the effective flexural stiffness, N, the effective
shear stiffness, the facing moduli and to estimate the shear modulus G. This test has also
been used in creep tests to estimate long term behaviour of sandwich structures [11].
r----r
240
----
t ~
B ~ B
~ t ~
a) Tension b) Shear c) Compression
dl ~ IOh
~
I Ih
i 20h
d2
t
d) Four point flexure
Figure 6. Sandwich tests.
4. Tests to simulate loading of structures
Testing of structures is usually very expensive and the range of possible loading
conditions is very large. It is never possible to run an exhaustive test programme, so the
aim of this type of test is usually to validate models for a loading which is typical of that
dominating the design loads to be considered. Some examples are shown in Table 2
below. They include tests on composite panels, sandwich panels and cylindrical
structures. Impact loadings have also been simulated and these will be briefly described.
4.1 TRANSVERSE PRESSURE LOADING OF THICK COMPOSITE PANELS.
The advantage of this set-up is that it enables panels to be subjected to loadings which
are more representative of in-service loads than those generally applied to test coupons.
Other authors have presented similar tests [19,20]. A 100-ton rated test frame has been
built which allows panels up to 2 meters by 1 meter to be loaded transversely by uniform
pressures up to 5 bars. Air cushions are inflated below the panel to apply the pressure
and the loads generated are reacted by mobile loading points attached to the test frame,
Figure 7.
241
TABLE 2. Examples of tests performed on composite structures at IFREMER

Application Mechanical Simulation test Ref. Material
loadin
Surface
transport,
Buoys
SIalic + creep
t
• • t
11 PVC foam
core
sandwich
4 point flexure
Static + cree
12
Bulkheads In-plane shear Phenolic

with sandwich
fire resistance
(]
Panels 1.8m x 1.8m x 78mm
Hull Uniform 13 Thick

wave pressure 14 composite
&
t t t t tt Sandwich
Pressure
Panels 2m x 1m x SOmm
Cooling water Internal IS, Filament

circuits pressure 16 wound
Glass!
Boats, Offshore p epoxy
±5So
/
Tubes 1m x IS0mm diameter
Instrumentation External 17, Filament

housings pressure 18 wound
Carbon
AUV and Glass!
(Autonomous epoxy
Underwater
Vehicle) /'
242
Two examples of tests performed on structural elements will be described. In both cases
the loading is uniform pressure, intended to simulate the wave loading of a boat hull.
Certification society rules use this type of loading in defining the minimum thickness of
hull structures. The first example is a wide beam, 2 meters long and 1 metre wide, with
1.1 meters between supports. This is a simplified case. The second is a square panel with
fixed edges simulated by adhesively bonding steel profiles to the edges. In this case the
free area is 1 meter square.
The main aim of these tests is to validate design tools, particularly finite element codes.
To do this three aspects must be carefully studied:
(i) the material characteristics must be well known,

(ii) the boundary conditions must be well defined, and
(iii) the test must be carefully instrumented to provide reliable measurements for
correlations.
Figure 7. Test set up showing 2 metre long 12 mm thick composite panel

under transverse pressure loading.
Elastic properties were measured for each panel, as shown in Table 3
. I e IasHc propertIes 0 f 8 an d 121 ayer th'ICk. pane s

TABLE3 . Tlyplca
Thickness Ex Ey G XY v xy
mm GPa GPa GPa
8 layer 7.6 15.42 12.65 3.32 0.26
12lay_er 11.9 15.42 14.11 3.50 0.26
A first series of tests was performed on 6 panels, three of each type, up to 0.2 bar
pressure. A second series was then performed up to failure. Extensive instrumentation
was used, both LVDT's to measure out-of-plane displacements along the panel centre
lines, and strain gauges to record in-plane strains on both upper and lower panel faces .
Figures 8 and 9 show examples of results.
243
80
aa 70
60
..:
= 50
a!U 40
!U
-
U
11:1 30
c::l.
....M 20
~ 10
0
0 0,2 0,4 0,6 0,8 1
Pressure, bars
Figure 8. Measured displacements of 12-layer thick panel at centre.

showing non-linear behaviour up to failure.
25
e
e
..: 20
e= 15
Qj
-...
Qj
~
~
~
10
-=
rIl
-=r.
Qj
5
Qj
U
0
0 100 200 300 400
Applied pressure, mbar
Figure 9. Measured displacements along centre line, 8-layer panel.

results from three tests on different panels.
The correlation with finite element codes was a major exercise involving several
partners and different codes. More details of this project are given elsewhere [21]. Here
some results from one set of calculations made at IFREMER using ADINA (large
displacements) are shown in Table 4.
244
TABLE 4. Correlation between prediction and measurements.

Results for 0.2 12 ply 12 ply 8 ply 8 ply
bar pressure Centre Centre upper Centre Centre upper
displacement microstrain Ex displacement micros train Ex
mrn mm
FE-ADINA 14.6-22.2* 1500-1820* 14.6 1030
Measured 19.4 1260 15.5 490
*The range corresponds to different boundary conditions, from perfectly clamped to the
use of torsion springs
It is apparent that the correlations with FE predictions are not particularly good. While
displacements can be approximated, and this is generally sufficient for structures
designed for stiffness, the correlation with strains suggests that the deformed shape of
the panels is not simulated correctly.
This test set-up has also been for longer tests, in order to validate the use of coupon
creep test data in the prediction of panel deflection with time.
4.2. PRESSURE TESTS ON SANDWICH PANELS

Two other series of tests have been performed using this test set-up, on PVC foam core
sandwich panels. The first involved light cores, 80 kg/m3, and has been described in
detail [13,14]. The second series was completed recently on four sandwich panels with
heavier cores (200 kglm 3). Here it was shown that by using non-linear material
behaviour in the model reasonable correlation with test results could be obtained [22].
4.3 IMPACT TESTS ON PANELS

There is a vast literature on impact, with results from both experimental and theoretical
studies available [e.g.23-25]. Much of this work is concerned with carbon fibre
reinforced aerospace composites and very high (ballistic) strain rates. Most marine
applications we are concerned with involve glass fibre composites and are limited to
certain types of low velocity impact loading. Figure 10 shows some panel impact tests
performed by IFREMER in recent years. The low energy drop test is intended to
simulate the damage to structures such as boat decks or hulls, induced by falling or
floating objects. A number of tests of this type have been performed in different recent
studies, both at IFREMER and elsewhere [e.g. 31-34], and these will not be described in
detail here. However, one aspect of the impact of sandwich panels which has received
little attention is the effect of scale: A question which is of great interest is:
Is it possible to scale impact tests, to predict the behaviour of large structures based all
tests on smaller scaled-down specimens?
This is of considerable practical importance, as large scale tests are expensive and
require special test equipment, whereas impact towers for small specimens are widely
available. The energies involved in the impacts of interest to the offshore industry
illustrate this. An example is the case of a container being accidentally dropped on the
deck of an offshore platform. This impact case was studied in a recent exercise to
demonstrate that the composite flooring concepts proposed to replace steel floors could
withstand the accidental drop of a 4 ton container, and more important could prevent it
245
from passing through to the deck below. Both steel and composite floors, 4 meters by 5
meters, were manufactured and tested. Figure 11 shows the experimental set-up.
A lication Load in Ref.
Surface Impact Low 26

transport energy 27
up to 500 J 28
Offshore
top ide deck
Impact
Medium
energy
to 3 kJ
• 29
Topside deck Impact 29

30
High Mass
(4Tons)
Energy up to
120 kJ
Hull Wave 29
slamming
Figure 10. Recent impact studies performed by IFREMER

246
Figure 11 . Steel container in place above composite floor before 3 metre drop
The containers and floors were instrumented, the former with accelerometers, the latter
with force and displacement transducers and strain gauges. High speed video (1000
images per second) was also used to record the impact events.
These tests have been described in more detail elsewhere [29,30), but it is interesting to
note that one of the composite flooring concepts, based on pultruded beams,
successfully passed the test by stopping the container. The steel flooring also passed the
test but suffered considerable damage and would have needed to be replaced.
247
This illustrates one of the high energy impact situations which new composite structural
somlutions must be able to resist. By developing validated impact models for composite
and sandwich structures it is hoped that scaling laws for impact behaviour can be
developed allowing new concepts for large marine structures to be tested at a reasonable
size, before final design choices are made.
5. Conclusions
Coupon tests to determine the properties of marine materials are widely used but there
are significant problems in obtaining valid values, particularly for sandwich materials.
Tests on structures are therefore invaluable, both in order to check design tools such as
finite element codes and also to study the behaviour of composite structures under more
realistic loading conditions. The uniform pressure loading test bed is a versatile set-up
which allows data to be generated on a range of materials at a reasonable cost. It may
also be adapted to both long term and cyclic loading. Given the high cost of full scale
structural testing, scaling laws are an essential part of the test philosophy. More work is
required to validate these lawf +~r both static and dynamic loading conditions.
Acknowledgements
The contriburtions of colleagues in the Ocean Engineering Department at IFREMER,
and particularly in the Marine Materials Laboreatory, to the work presented here is
gratefully acknowledged.
6. References
[1] Smith CS, 'Design of Marine Structures in Composite Materials', Elsevier, 1990.
[2] Davies p. in 'Durability analysis of structural composite systems', ed Cardon. 1996.
p33.Balkema Publishers
[3] Shenoi RA. Wellicome JF. 'Composite Materials in Maritime Structures'.
Cambridge Ocean Technology series, 1993.
[4] Proceedings Int. conference on Nautical Construction with Composite Materials. ed.
P. Davies. L. Lemoine. Paris, IFREMER Actes No. 15. 1992.
[5] Davies P, Chauchot P, Composites for Marine Structures, Part 2 Underwater
Structures, in this volume.
[6] Davies p. Petton D. to appear in Composites Part A, 1998.
[7] Baudin M, Proc Int Conf. on Nautical Construction with Composite Materials. 1992,
Paris. IFREMER Actes No. 15, p301.
[8] Davies P, Blackman BRK, Brunner AJ, to appear in Applied Compo Materials, 1998
[9] Brunner AJ, 'Interlaminar Fracture tesing of unidirectional composites', in this
volume.
[10] Davies P, in Fracture of Composites, ed. Armanios, Transtec publications. 1996.
p583.
[II] Davies P, Craveur L, Proc. ECCM7-CTS-3, 1996 Woodhead Publishers, p367.
[12] Davies p. Choqueuse D, Bigourdan B. Proc. 3rd Int. Conf. on Sandwich
Construction, Southampton. EMAS publishers, 1989. p647.
[13] Davies P, Choqueuse D. Bigourdan B. Marine Structures. 7. 1994 p345.
248
[14] Choqueuse D, Davies P, Bigourdan B, in Ref. 7, 1992, p276.

[15] Croquette J, Parquic JC, Forestier JM, Dufour X, in Ref. 7, 1992, p412.
[16] Baizeau R, Davies P, Choqueuse, LeBras J, Proc DURACOSYS 1995, ed Cardon,
Fukuda & Reifsnider, Balkema, p225.
[17] Bigourdan B, Chauchot P, Hassim A, Lene F, Mechanics and Mechanisms of
Damage in Composites and Multi-Materials, ed Baptiste, ESIS 11, MEP, 1991, p203.
[18] Davies P, Choqueuse D et a!., Proc. JNClO, (in French), 1996 p525.
[19] Reichard RP, Proc. Sandwich Construction I, 1989, EMAS.
[20] Bertelsen WD, Proc. Sandwich Construction 2, Gainsville, Florida, 1992.
[21] Bureau Veritas, Projet National 'Composites Navals', Final Report, Sept. 1995.
[22] Hayman B, Wiese M, Davies P, Choqueuse D, Hoyning B, Mitusch P, Proceedings
4th Int. Conf. on Sandwich Construction, June 1998.
[23] Abrate S, Applied Mech. Review, 1991,44, p155.
[24] Cantwell WJ, Morton J, Composites 22, 1991, p347.
[25] Greszuck LB, in Impact Dynamics, John Wiley, 1982 p55.
[26] Davies P, Choqueuse D, Verniolle P, Prevosto M, Genin D, Hamelin P, ESIS 19,
ed. Williams & Pavan, 1995, MEP, p341.
[27] Davies P, Choqueuse D, Riou L, Wahab A, IFP review Vol 50, 1, Jan. 1995, p75.
[28] Collombet F et a!., Proc conf. 'Mechanics of Sandwich Structures' , Euromech 360,
St. Etienne, May 1987 p255.
[29] Choqueuse D, Davies P, Baizeau R, s __ .nitted to Marine structures, 1997.
[30] Croquette J, Proceedings Composite Materials for Offshore Operations conference
(CMOO-2), Houston October 1997.
[31] Pecault S, PhD thesis, Bordeaux, December 1996, (in French)
[32] Wahab A, PhD thesis, Bordeaux, February 1997, (in French)
[33] Nemes JA, Simmonds KE, J. Compo Mats., 26, 1992, p500.
[34] Nilsen PE, Moan T, Gustafson C-G, Proc. Sandwich Constructions 2, Florida,
1992.
COMPOSITES FOR MARINE APPLICATIONS
Part 2. Underwater structures
P. DAVIES & P. CHAUCHOT

Ocean Engineering Department
IFREMER Centre de Brest
BP 70, 29280 Plouzani
France
1. Introduction
The advantages of composite materials for marine structures, notably their high specific
properties, resistance to degradation in water, and flexible fabrication to produce special
shapes, become even more valUable when underwater applications are considered.
Nevertheless, apart from a few exceptions there are not many structural applications of
composites underwater at present. The reasons will be discussed here. Three types of
application will be considered:
(i) Oceanographic
(ii) Submarine
(iii) Offshore
These will be described first, then some VI the design considerations to be taken into
account for underwater vessels will be detailed. Finally, some areas requiring further
\lIork will be discussed.
1.1 . OCEANOGRAPHIC APPLICATIONS
Potential applications of interest to oceanographic institutes such as IFREMER may be

classed as:
- Frames, support structures and benthic landers,
- Instrumentation housings,
- Underwater vehicles, submersibles.
The former are often simply space frames designed to provide a stable support for
instrumentation. A recent review published in 1995 indicated twenty-seven known
benthic landers, the majority in aluminium or steel [1]. Recently a composite lander has
been developed at IFREMER, the MAP ('Module Autonome Pluridisciplinaire') in the
MAST project 'Alipor'. This is a glass/epoxy structure, 1.5 metres high, which is
designed to support a range of probes and instrumentation on the ocean floor, Figure I.
It can operate down to 6000 metres depth and replaces an aluminium alloy structure. A
major reason for the replacement was the corrosion products from the aluminium, which
interfered with some of the measuring devices. This type of support frame can be made
by hand lay-up. It must be able to withstand accidental impact, but the loading during
service is not very severe.
249
250
Figure 1. 'MAP' instrumentation support structure in composite (white) for 6000

metres. (The capsules above are foam buoyancy modules).
For instrumentation housings and underwater vehicles the loadings are more severe and
these are generally filament wound cylinders designed to resist external pressure.
Potential applications in underwater vehicles range from the protective outer casings of
deep diving vehicles (eg the French manned submersibles 'Nautile' and 'Cyana', and the
American 'Alvin' and the US Navy's DSRV (Deep Submergence Rescue Vehicle), to the
hull structure of Autonomous Underwater Vehicles (AUV's) such as 'Autosub'. Several
recent papers have described the materials challenges in designing deep sea exploration
submersibles (eg 2,3], and the design of pressure hulls will be discussed in more detail
below.
251
1.2. SUBMARINE APPLICATIONS

An extension of these underwater vehicle applications leads to military submarines,
which already employ considerable amounts of composites. Lemiere has described some
of these components in French submarines, on which these materials have been used for
many years [4]. Large applications include sonar domes, and deck structures. The
former are thick monolithic structures, while the latter are sandwich structures with
surface areas around 1200 m2 . Graner has described some large submarine fairwaters
and sonar domes used in the USA for over 30 years [5], while the UK Navy has also
used composite fins and casings for submarines for many years [6]. There has been
much military research on composites for other submarine applications, with large
programmes in the USA and ex-USSR. Although most of the results from such research
are confidential, some details of a study financed by the US agency ARPA ( Advanced
Research Projects Agency) of a man-rated demonstration article for a dry submarine
deck shelter showed that considerable development work has been performed on
thermoplastic composites [7]. In the ex-USSR some very large composite cylinders have
been pressure tested, and considerable experience is also available in this area. A
European military programme just finishing, EUCLID RTP 3.8, also includes work on
composites for submarine applications.
1.3 OFFSHORE APPLICATIONS UNDERWATER

Finally, the offshore industry is studying cumposites closely for underwater installations.
Gibson has given an overview of offshore applications of composites [8]. An underwater
example is the sandwich wellhead protection structures described by Brevik [9] for the
North Sea. The major design requirement in this case was the need to protect wellhead
equipment from high energy impacts. Other large underwater structures are composite
risers, which were demonstrated by IFP [10]. The latter are attracting considerable
interest in the USA and elsewhere, as lighter materials are needed to explore and exploit
deep sea reserves [11].
2. Factors in the design of underwater vehicles

The remainder of the paper is devoted to consideration of the hull structure of
underwater vehicles. The emphasis is on AUV's, but many of the factors involved
(material properties, buckling, crushing) are also critical for smaller structures (eg
instrument housings) or larger submarine applications. The design of underwater
vehicles such as AUV's involves careful balancing of weight, payload, and energy
requirements. Lightening the hull structure by using composites is very attractive, as
discussed by Perry et al. [12], but adopting these orthotropic materials requires far more
property data than the conventional metal options if thorough buckling and failure
analyses are to be performed. These aspects will be discussed below, but first the
underwater environment will be briefly described.
2.1 THE UNDERWATER ENVIRONMENT

The main factor peculiar to composite structures operating underwater is the presence
of hydrostatic pressures, but other factors may need to be taken into account:
- Temperature: Sea temperature decreases with increasing depth, varying in the range
from 30°C at the surface to SOC at 1000 meters depth.
252
- Medium: Sea water parameters include salinity (varying from 27 gil to 37 gil and
usually decreasing with depth), and pH, which varies from 7.5 to 8.5. Biological
parameters may also play a role in material aging processes, but few studies have
addressed these.
- Pressure: Hydrostatic pressure is directly proportional to depth. While many
applications are limited to a few hundred meters some oceanographic equipment must
spend months or years at up to 6000 meters where a hydrostatic pressure of 625 bars
(62.5 MPa) is acting.
2.2 MATERIALS CONSIDERATIONS

The competing materials for deep sea applications are shown in Table 1. Polymer matrix
glass and carbon composites look attractive, as they are readily available and their
technology is far more developed than that of MMC's and ceramics. Carbon composites
look significantly better than the metallic solutions in terms of stiffness to weight ratio,
and hence in buckling performance, while both glass and carbon composites should be
significantly better than the metals in terms of strength. However, the values in Table 1
are nominal values and three aspects to be considered are:
i) The dominant loading is com,-_ Jssion, a mode of loading which is notoriously hard to
characterize in composites,
ii) The thickness of the composites involved, often 50 mrn or more, requires knowledge
of through-thickness properties ; in-plane u<;naviour is not sufficient. Through thickness
strength is not the strong point of layered composites.
iii) The cylindrical form of most of the structures described above implies that the
preferred fabrication method is filament winding. Characterization of curved specimens
is therefore preferred, as the fibre content and internal stress distributions in flat plates
may differ significantly from those in cylinders.
TABLE 1. Materials for deeE sea vehicles

Material Density Modulus Stiffness Compo. Strength
kg/m3 GPa to strength, to
Weight MPa weight
Steel HY80-HY130 7.8 207 27 550-890 70-115
Aluminium 7075 2.8 70 25 500 180
Ti6A14V 4.5 110 24 830 185
Glass/epoxy 0/90° 2.1 50 24 1000 475
Carbon/epoxy 0/90° 1.7 100 59 1000 590
SiC/AI (MMC), UD 2.7 140 52 3000 111O
Ceramic Al 20 3 3.6 380 105 2060 570
The determination of material properties can be based on calculations or on tests, but in

general a combination of the two is employed as both have their limitations. Considering
first the determination of material properties by calculation. Homogenization methods
based on the theory of periodic media have been developed, such as the MODULEF
programmes [13]. These enable elastic properties in the three dimensions to be obtained,
requiring as input:
- Fibre properties
- Matrix properties
- Fibre content and distribution in the basic unit cell.
253
These methods are very useful as they enable parametric studies to be performed, to
assess the influence of factors such as fibre orientation and void content, but they must
be validated by tests. Such validations have been performed in some cases, but the range
of tests available for the characterization of thick composites is incomplete. Within a
European MAST (Marine Science & Technology) programme [14,15] both tests and
homogenization calculations were carried out on specimens from carbon/epoxy
cylinders wound at ±S5° (see Figure 2). Table 2 below shows some results.
---------------T-he~t-. ~'\
(-
__
~o-{l~z
R
Figure 2. C J under angle and nomenclature used.
TABLE 2. Material elastic properties fnPa) for carbon/epoxy composites, from

homogenization and tests.
Property Test Homogenization
Ez 10.95 10.2
Ee 32.6 31.5
21.8 30.8
3.0 3.8
3.3 3.9
Shear (Ze and ZR) and compression tests were performed on specimens cut from 14mm
thick carbon/epoxy cylinders in the axial direction, It should be noted that the stress-
strain plots are both non-linear, Both shear and compression behaviour are dominated by
the matrix and values are very low. Compression data in the circumferential direction
(i.e. with fibres at ±35° to the loading direction) are more difficult to obtain. Moduli
values (Eee) may be obtained from diametral loading of rings, or vibration analysis but
strength values are very hard to determine. External pressure testing of rings is one
method which has been used [16], but this requires pressure vessels. If pressure vessels
are available it may be more efficient to work directly on cylinders, and this is the
approach used at IFREMER where a range of large pressure vessels is available (up to
2400 bars). This has the advantage of producing a biaxial compression stress, rather
than the uniaxial coupon loading. Given the uncertainty in failure criteria, particularly
applied to biaxial compression, this is obviously attractive, but scale effects must then be
considered: Can results from a test on a 150 mm diameter cylinder be extrapolated to
predict the behaviour of structures of diameters 1 metre or more ? There are several
parameters influencing this extrapolation and the absolute value of radius is important.
Pressure tests will be considered below.
254
2.3 STRUCTURAL CONSIDERA nONS
2.3.1 Buckling
Underwater vessels are designed to withstand external hydrostatic pressure. Under such
loading the optimum shape of a housing is a sphere. Nevertheless, operational
constraints lead to cylindrical geometries. According to the immersion depth and the
material used, the ultimate strength of a vessel may depend on either its buckling
resistance or on its material failure characteristics. There are simple analytical
expresions available which give a reasonable first estimate of buckling pressures in
some cases [17,18]. For example Mistry et al [18] propose;
P cr = 0.807 Ee (R Y~)2.5
3/~1-vxevex LA R
with Pcr the critical buckling pressure, expressed in terms of inner radius R, length L,
thickness h and material elastic properties.
3500
,-..,
..ef
.....
'-'
3000
...c
0/)
'0 2500
~
:a- 2000
'-'
...c::
..... 1500
0-
Il)
"0
Il) 1000
:i2
u
;:s
p:) 500
°
° 0,05 0,1
Thickness/Radius ratio
0,15 0,2
----- Alwniniwn .. ·x ... Ti6-4 ----- Glass/Epoxy 55

- -+ - . C/epoxy 55 ---0-- G/Epoxy 0/90 - - C/epoxy 0/90
Figure 3. Comparison of potential huH materials, based on simple buckling expression

[18] for buckling pressure, normalized by weight.
Figure 3 shows results from simple analytical buckling expressions for glass and
carbon/epoxy cylinders, with values for aluminium and titanium al\oys for comparison.
Using the same thickness of cylinder the metal options are superior, but if the buckling
255
pressures are normalized by weight the advantage of using carbon/epoxy becomes

apparent. Unfortunately, the absolute values predicted by these simple expressions often
greatly overestimate measured buckling pressures. Thus, for example, a half-scale test
on a candidate AUV hull in carbon/epoxy was performed at IFREMER within a
European MAST project (partners lOS, ORA, NTUA) , [I5] ona cylinder of 450 mm
inner diameter and 40 mm nominal thickness. The analytical expression above would
predict a buckling pressure of 134 MPa, but in fact the cylinder failed at a pressure of 61
MPa.
The development of finite element calculations has enabled more detailed modelling of
complex structures and material behaviour. Some special finite elements take into
account composite material ply characteristics. A linear buckling analysis (LBA) can
provide the first buckling mode, but it usually overestimates the buckling resistance. To
get a correct approximation of the buckling load, the best finite element procedure is
based on a non linear analysis performed with an iterative load displacement controlled
(LOC) calculation, starting with an initial geometrical imperfection having the shape of
the first buckling mode obtained by the linear analysis.
Nevertheless, even with such a procedure, results are very sensItive to boundary
conditions and imperfections. The latter may be geometrical (thickness variation for
thick hulls) or material (internal defects such as voids or delarninations, mechanical
characteristics variations) imperfections. For the case of the MAST candidate AUV hull
the following finite element calculations have been performed with the ADINA code
(figure 4) :
- LBA with nominal characteristics

- non linear LOC calculation taking into account the non linear contact conditions with
initial gap between the cylinder and its titanium end closure, with:
· nominal characteristics
· real mean thickness
· real mechanical properties of titanium ends and rings
· non linear properties of the adhesive used between the composite cylinder and its
titanium end rings.
With the last calculation, the adjustment of the measured strains with the computed ones
has provided the mechanical characteristics of the composite hull which failed at 61 MPa.
Figure 5 shows that the buckling mode is initiated at a pressure below 500 bars.
A parameter of interest in designing such cylinders is the fibre orientation. Some authors
have indicated that in ribbed composite shells the highest buckling pressure will be
found for [02n5/-75/0]s lay-up [19]. For fabrication simplification, it is interesting to
optimise cylinders without stiffeners. According to the size of the hull to be designed (t,
Rlt), optimised winding sequences can be found with a strong reinforcement at a 90°
angle since buckling modes are circumferential.
256
Figure 4. Finite element model, AUV hull.
Central section srrain gauge response
o 60 120 180 240 300

0 --100bar
t:<
.~ I
b -1000 -_ .... ...... -" I------- 200 bar
CIl - .. -- .. .. _- ..... - .. ---
.... -2000
0
--300 bar
·s
u
-3000
0.. -4000 , I --400 bar
0
0
::r: -5000 "'-'-.. .// ' '" '-.. , / --500 bar
-6000 - 6 0 0 bar
Angle (degrees)
Figure 5. Buckling of carbon/epoxy cylinder
2.3.2 Material failure
At higher thickness-to-radius ratios buckling failure is no longer observed as material

failure occurs first. Figure 6 shows data from a number of published references, in which
crushing (compression failures) occurred. This plot shows some interesting features.
257
First, for carbon/epoxy the lower hoop strengths, around 500 to 600 MPa, came from
early studies [20,21], while a more recent value was for a cylinder of similar dimensions
which did not fail at a pressure 50% higher than the earlier tests [22]. As the fibres were
of the same type (IM6) this suggests that significant improvements have been made in
fabrication technology, as difficulties in avoiding fibre waviness were blamed for the
low early results.
1000 ~-----------,
800
ell
••••
ell
.~
. . . ro 600
ell p..
§"::8
400 r.t E- glass ±55 °
::r:o
A S-gIass 0/90°
200
• Carbon 0/90°
o +-------~----_+------_r----~
o 0,1 0,2 0,3 0,4

thickness/inner radius ratio
Figure 6. Failure hoop stresses for glass and carbon/epoxy cylinders
Results from S-glass and R-glass fibre reinforced cylinders are less widespread. One
might expect the slightly higher modulus of these fibres (86 GPa compared with 73 GPa
for E-glass fibres) to give improved buckling resistance, and a small effect was noted in
buckling tests with E and R-glass at IFREMER. The improvement noted in the sets of
data shown in Figure 6, compared with the E-glass at similar rlt ratios, may be the result
of either the improved fibre or the different fibre orientation (0190). This requires further
evaluation. The E-glass/epoxy results shown here are all from cylinders wound at ±55°.
Kaddour et al. [23] suggest a plateau value of around 800 MPa for this material and
layup, but this plateau was obtained on small diameter (51 mm) thick cylinders. Larger
diameter glass/epoxy cylinders do not reach this level even at similar tIR ratios where no
evidence of buckling is detected. This raises the question mentioned above, of whether
there are scale effects in pressure tests on cylinders. Fibre orientation is also a critical
parameter. Most work in the USA has been on 0190 0 cylinders, while in Europe ±55°
has been studied extensively. It is not easy to run theoretical studies to find whether
these or other layups offer the optimum resistance to external pressure as there is so
much uncertainty in the material strength data. This requires more work, and several
studies are underway in this area. The angle chosen depends on the depth considered,
but fabrication parameters must also be considered and may dominate the final choice.
258
2.3.3 Defects and NDT
Filament winding of thick structures, 50 mm or more, is not straightforward.

Temperature rise due to cure exotherm must be controlled by appropriate cure cycles
and there is a risk of extensive delamination. Even if large delaminations are avoided
small defects are always present, and the non-destructive control of thick cylinders is an
essential part of their fabrication process. However, even if NDT is successful and a
defect is detected the question of whether or not to scrap the cylinder must be addressed.
This aspect is currently being studied in a joint IFREMER-F1orida Atlantic University
project, financed by the ONR (Office of Naval Research). Analyses of crack
propagation between dissimilar plies in curved specimens are being developed.
2.3.4 Durability and long term behaviour
The long term behaviour of composite cylinders is the subject of other programmes,
including a MAST project on benthic housings. There is little data available on the
cyclic and creep behaviour of these structures. These points must be addressed, as
excessive design coefficients must be avoided if composite materials are not to be
penalized with respect to alternative materials.
2.3. 5 Impact damage resistance
It is of interest to know how these cylinders resist impact damage and a number of
studies have been published in which the influence of boundary conditions, impact
geometry and impact energy have been described [24-27]. Dynamic analyses have been
presented, and comparisons with quasistatic indentation response have been suggested to
predict damage. There has also been a recent attempt to use fracture mechanics·
parameters, notably GUe , to correlate impact energy with delamination [27]. That latter
work was performed on pipes placed on flat plates and impacted at relatively low impact
energies (up to 20 J).
A study is underway at IFREMER to examine this approach further, with respect to

tubes impacted under conditions reasonably close to those that might be expected in
service: i.e.
- Tubes supported over spans up to 3.2 metres

- Impact energies in the range from 55 to 275 Joules
Tubes are all of 150 mm internal diameter, with three different wall thicknesses of 3.2,
5.5 and 15 mm. Different impacter diameters are used, and the damage introduced into
the tubes at different impact energies has been assessed in different ways. The simplest
description is that of projected surface area, which can be determined using strong back
lighting. An example for thin cylinders is shown in Figure 7 below. This suggests a
strong influence of impacter diameter and boundary conditions, but is of limited use as it
gives no indication of the crack surfaces created through the thickness. Sectioning and
measuring the total delaminated area are necessary in order to assess whether an energy
balance can be used.
259
10000
5000
o 50 100 150 200 250 300

Impact energy, J
-Diam.= 20rmn, L=1.2m

-=- Diam.=lOOmm, L=3.2m
.....c:>-Diam.=lOOmm, L=1.2m
Figure 7. Damage in cylinders, projected areas
2.3.6 Connections
A final point which must not be overlooked is the connections between composite
cylinders and extremities, inserts, and other components. It is clear to any experimental
researcher who has tried to test a cylinder under pressure that the end fittings are critical
to a successful test with controlled boundary conditions [eg 28]. A composite
underwater vehicle will require cut-outs, passages, heat sinks, dismountable fittings,
internal supports and a host of other appendages which will transform it from a simple
axisymmetric structure into an extremely complex multi material assembly. Interactions
between materials in sea water under pressure must be considered to ensure satisfactory
long term integrity.
260
3. Conclusions
Composite materials show great potential for underwater applications. They are already
applied in many secondary structures and protective housings and there is much work
underway to examine their introduction into primary hull structures. This paper has
described some of the points still needing clarification before such structures can be
designed with confidence. Of particular importance are:
- scale effects
- the influence of defects and damage and their detection.
- long term and cyclic behaviour.
All of these are being addressed by current programmes, and as experience is gained
from the development of small unmanned vehicles larger structures will follow.
References
[1] Tengberg A et aI., Proc. Oceanog., 35, 1995, p253.
[2] Garvey RE, Marine Sci. & Tech., 24, 2, June 1990, p49.
[3] Hawkes GS, Ballou PJ, Marine Sci. & Tech., 24, 2, June 1990, p79.
[4] Lemiere Y, Proc. IFREMER Conf. on Nautical applications of composite materials,
Paris 1992, p 441
[5] Graner WR, ch. 27, Handbook of Composites, ed. Lubin, 1982, van Nostrand.
[6] Chalmers DW, Marine Structures, 7, 11,~ "p441
[7] Journal of Thermoplastic Composite Materials, Volume 8, Jan. 1995.
[8] Gibson AG, Chapter 11 in 'Composite Materials in Marine Structures', ed Shenoi &
Wellicome, Cambridge Ocean Technology, 1993.
[9] Brevik AF, Proc. Sandwich Constr. 3, 1996, EMAS, p27
[10] Tamarelle PJC, Sparks CP, Proc. OTC 5384, 1987, Houston.
[11] Fischer EJ, IFP Review, 50, I, Jan 1995, p35
[12] Perry TG, Douglas CD, Gorman 11, SNAME 1992, p3-1.
[13] Bigourdan B, Chauchot P, Hassim A, Lene F, ESISll, 1991, MEP, p203.
[14] Collar P, Babb RJ et aI., Proc. OCEANS 94, Sept. 19941-158.
[15] Davies P, Choqueuse D et al. Proc 10th French National Composite conf., JNClO,
Paris 1996 (in French), Vol. 3, p525.
[16] Kim RY, Tsai SW, Proc 33rd SAMPE Symp., March 1988, p1159.
[17 Smith CS, 'Design of Marine Structures in Composite Materials', Elsevier, 1990.
[18] Mistry J, Gibson AG, Wu YS, Compo Structures, 22,1992, p193.
[19] Chin HB, Prevorsek DC, Proc 4th Japan-US conf on Composite Materials, June
1988, Technomic p953
[20] Stachiw JD, Frame B, NOSC Report 1245, August 1988.
[21] Garala HJ, Chaudhuri RA, AMD-VoI162, Mechanics of Thick Composites, ASME
1993,p227
[22] Starbuck MJ, Blake HW, ASTM STP 1185, 1994, p159.
[23] Kaddour AS, Soden PD, Hinton MJ, Proc ICCMll, 1997, V-786.
[24] Alderson KL, Evans KE, Compo Structures, 20, 1992, p37
[25] Evans KE, Alderson KL, Compo Structures, 20, 1992, p47
[26] Swanson SR, Smith NL, Qian Y, Compo Structures 18, 1991, p95
[27] Frost SR, Proc. FRP 94, Newcastle 1994, Inst. Of Materials, paper 3/1.
[28] Blake HW, Starbuck MJ, ASTM STP 1185, 1994, p113.
EXAMPLES OF NON·DESTRUCTIVE TESTING OF GLASS· AND CARBON·
FIBER·REINFORCED POLYMER·MATRIX COMPOSITES
AJ. BRUNNER, 1. NEUENSCHWANDER

EMPA, Swiss Federal Laboratories for Materials Testing and Research
Oberlandstrasse 129, CH·8600 Dubendorf, Switzerland
ABSTRACT
This paper discusses examples of non-destructive test (NDT) methods applied to glass-
fiber- (GF) and carbon-fiber- (CF) reinforced polymer-matrix (RP) composites. In the
introduction, a brief overview of NDT-methods for GFRP- and CFRP-composites is
presented. Visual Inspection, X-ray projection imaging, Ultrasonic C- and B-scans, and
Acoustic Emission serve as examples of NDT applied to GFRP and CFRP. Certain
advantages and limitations of the selected NDT-methods applied to FRP-composites
are discussed using examples taken from experiments at EMPA.
1. Introduction
1.1 DEFINITIONS AND LITERATUIL.
In a standard [1] NDT is defined as the development and application of technical

methods for (1) the detection, location, measurement and evaluation of discontinuities,
defects and other imperfections, (2) the assessment of integrity, (3) the assessment of
properties and composition, or (4) the measurement of geometrical characters of
materials and components without impairinl: their intended use or <llWlication. Sources
for literature on NDT-methods are listed in Table 1. A composite is a solid with at least
two macroscopically distinct phases which retain different properties. Information on
glass-fiber (GF) and carbon-fiber (CF) reinforced polymer-matrix (RP) composites and
their properties can be found in, e.g., [2].
1.2 OVERVIEW OF NDT-METHODS
The determination of certain properties of materials or structures with NDT-methods

contrasts destructive testing (e.g., determination of ultimate tensile strength, cutting or
polishing specimens for microscopy, etc.). Application of NDT-methods can affect the
test object and may change some properties. Unless limits for acceptable changes are
included in the test standard or procedure, they have to be defined by the user.
Therefore, the evaluation of NDT-results requires a sufficient knowledge of material
and structural behavior. The classification of NDT-methods in Table 2 is based on the
overview in [3, Vol. 10]. The "probe-type", i.e., the energy source serves as definition
261
262
for the different categories. Table 3 lists examples of NDT-methods applicable to
GFRP and CFRP and the quantities used for the measurement.
Detection of "defects", i.e., locating and/or sizing features that are not acceptable with
NDT-methods according to [1] consists of three stages: (1) looking for "indications",
(2) interpretation of indications as false, relevant or non-relevant, and (3) evaluation of
relevant indications (meet specified criteria =accept, else reject).
TABLE 1. Literature Sources on NDT-methods (N.B. The listing of references is not exhaustive)
Source Type Type of Literature Remarks
Standards Test procedures (with or without evaluation criteria), e.g. ASTM, ISO *
guidelines, tenninology and defmitions
Handbooks Review articles (with references), introduction to test e.g., [3]
methods, commercial applications, special applications
Journals Reviewlresearch articles, new developments
Proceedings Research papers, special applications e.g., Conferences
Monographs Specific test methods, sp . II applications e.g., [4]
Technical Description of instrument and accessories, instrument use, **
Documentations special applications
Lecture notes Background on specific methods, applications (e.g., for ***
certification of operators)
* ASTM - American Society for Testing and Materials, ISO = International organisation for
Standardisation
** See, e.g., documentations provided by NDT-instrument manufacturers.
*** See, e.g., NDT-courses offered by national NOT-societies.
2. Non-Destructive Test Methods for FRP-Composites
2.1 INTRODUCTION
Each NDT-method can be characterised by type of (1) probe (usually energy source),
(2) interaction with test object, (3) signal detection, (4) signal recording, and (5) signal
interpretation. The NDT-methods discussed in this paper use two different probes, (a)
electromagnetic radiation (Visual Inspection and X-ray Imaging) and (b) elastic waves
(Ultrasonics and Acoustic Emission). Visual Inspection and X-ray Imaging again use
different regions of the electromagnetic spectrum, the wavelength of visible light is
between about 300 and 800 om, that of X-rays around 0.1 nm. Frequencies of ultrasonic
wave pulses lie in the range between I and 50 MHz. Higher frequency yields higher
spatial resolution but also higher attenuation. The choice of the frequency, therefore, is
a compromise between resolution and attenuation. Acoustic Emission sensors typically
cover the range between 50 kHz and about 1 MHz. There are two types of sensors, (a)
resonant sensors (with a rather narrow resonance within the above frequency range) and
(b) broadband sensors which cover part or all of the above range with similar
sensitivity.
263
TABLE 2. Categories ofNDT-methods and examples (N.B. The listing of examples is not exhaustive)
Category [3, Vol. 10] Examples of NDT-Methods [3]
Mechanical Tensile and bending tests

Vibration analysis
Geometrical Measurement (dimensions, coordinate measurement)
Scanning Probe Microscopy" (Scanning Tunnelling Microscopy = STM, Atomic
Force Microscopy =AFM)
Optical Visual Inspection (by eye, photography, video, high-speed film, dye penetrants)
Photoelasticity (by eye, photography)
Microscopy" (light microscopy, fluorescence microscopy, scanning electron
microscopy)
Structured Light (moire, fringe projection)
Interferometric Methods (holography, shearography, electronic speclde
interferometry =ESPI)
Penetrating radiation X-ray (radioscopy/radiography, laminography, computer tomography, Compton
backscattering, diffraction)
Gamma-radiation (radio---,hy)
Neutron-radiation (radiography)
Electromagnetic/ Eddy current measurement
electronic Dielectric measurement
Magnetic flux, magnetic particle measurement
Soniclultrasonic Tapping test
Acoustic emission (burst emission, continuous emission)
Ultrasonics (A-scan, B-scan, C·scan): pulse-echo or transmission
Acousto-Ultrasonic
Thermal/infrared Thermography: passive, impulse, lock-in
ChemicaII Infrared Spectrometry .. (Fourier Transform-Infrared spectrometry =FTIR; chemical
analytical analysis offunctional groups of polymers)
Raman Spectrometry .. (chemical analysis)
Ultraviolet spectrometry .. (photoelectron spectrometry, electronic properties)
X-ray" (fluorescence, photoelectron spectrometry, diffraction)
Magnetic Resonance .. (magnetic resonance imaging =MRI)
.. Specimen preparation may be destructive (often because of size limitation).
Visual Inspection, X-ray Projection Imaging, and Ultrasonics (B- and C-scan) are
examples for "imaging" NDT, Acoustic Emission an example of a "non-imaging"
technique. Acoustic Emission can also be classified as "global" technique since the
sensors register all signals within a certain radius (determined by signal attenuation),
while the other techniques are "local", i.e. limited to the size of the respective viewing
window (e.g., film size in X-ray imaging).
For X-ray projection imaging an imaging and analysis system for computer-
tomography (CT) techniques has been implemented at EMPA. However, CT-X-ray
results will not be discussed in this paper.
2.2 DESCRIPTION OF TEST PLATE FROM CFRP
For a comparison of different NDT-methods a CFRP-test plate with simulated defects

has been used. This test plate (CF-epoxy) is 356 mm long, 255 mm wide and 1 mm
264
TABLE 3. Selection of NDT-methods for CFRP-composites [3]
NOT-Method Detection of indication for * Signal detection' remarks **

(selected examples)
Tensile or bending test Fiber volume fraction Compliance' GL

(linear elastic range) Fiber lay-up
Porosity
Vibration analysis Delamination, Crack Eigenfrequency , GL, 1M
Inclusion
Porosity
Visual inspection Surface finish Reflected light' CL, 1M, LO
Delamination (near surface)
(Surface) Cracks
Projection Moire Geometrical properties Moirt interference pattern' CL, 1M,
(Large) Surface cracks LO
Radioscopy/radiography Inclusion Difference in absorption of X-rays'
(Conventional X-Ray source) Delamination, Crack IM,LO
Projection Radiography Inclusion Difference in absorption of X-rays I
(Microfocal X-Ray source) Delamination, Crack CL,IM,LO
X-ray Computer-Tomography Dimensions Difference in absorption of X-rays'
(Conventional or Microfocal X- Density variation CL,IM,LO
Ray source) Voids, Inclusions
X-ray Compton Backscattering Laminate orientation Backscattered X-ray photons' CL,
Delamination IM,LO
X-ray Chemical Analysis (e.g., Qualitative and/or quantitative e.g., X-ray or electron energy' CL,
X-rays from Synchrotron) composition IM,LO
Tapping Delamination Eigenfrequency I GL
Cracks
Voids
Porosity
Ultrasonic B-Scan Interface, Delamination, Crack Reflected ultrasonic wave' LO
(time-profile of signal along line (Large) Inclusion
of test object) Voids, Porosity
Ultrasonic C-Scan (parallel line Interface, Delamination, Crack Reflected ultrasonic wave' 1M, LO
scans of reflected or transmitted (Large) Inclusion
signal amplitude on area of test Voids, Porosity
object) Fiber orientation
Acoustic Emission ("burst" Fiber breaks Elastic wave induced by stress
emission) Delamination growth concentration change I GL
Crack growth ("continuous" emission appears
Friction between surfaces e.g., in process monitoring)
Thermography (impulse, lock-in) InterfacelDelamination Difference in thermal conductivity'
(Large) Inclusion CL,IM,LO
(Large) VoidslPores
Eddy Current Fiber orientation Resistivity' CL, 1M, LO
Fiber fracture
* NOT for delamination growth is discussed in more detail below.

** CL = contact-less' 1M = imaging' LO = local' GL = global
265
thick. It consists of three layers of roving (0°/45°/0°). Folded PTFE foils are used to
simulate delaminations. There is a total of 29 square or rectangular simulated
delaminations, ranging from 3 mm x 3 mm to 25 mm x 25 mm in size. The PTFE-foils
have been inserted either between the first and second (type A) or between the second
and third layer of roving (type B). The position and size of the simulated defects have
roughly been marked on the surface of the test-plate.
2.3 VISUAL INSPECTION
CFRP-composites are usually opaque and visual inspection of the test plate is hence
limited to surface features. Except for the intentional marking of their position and size,
no indications of the simulated defects has been found with visual inspection.
2.4 X-RAY PROJECTION IMAGING
X-ray projection imaging is performed by placing a X-ray sensitive medium (e.g., X-

ray film or imaging system) at a certain distance behind the test object (as viewed from
the X-ray source). The closer the test object to the X-ray source and the larger the
distance between source and imaging system, the larger the magnification. Due to the
projection of the X-ray source volume images become blurred if conventional X-ray
sources with typical source diameters o~ _ - 2 mm are used (Figure 1).
Distance between source

and test object (focus-object
distance)
Unsharpness
(due to source
volume)
Distance between source and imaging

system (film-focus distance)
Figure 1. Schematic representation of X-ray projection imaging.
Special "microfocal" X-ray sources with electron beam diameters around 5 J.1m limit
this unsharpness, see, e.g., [5]. The magnification obtainable with microfocal X-ray
sources is limited by (1) the source volume and (2) the absorption of the X-ray between
source and imaging system. The maximum X-ray energy of microfocal X-ray sources is
around 300 keV, since the electron beam that generates X-rays penetrates a target
volume of> 10 J.1ID diameter if accelerated to higher energies [6]. Magnifications of up
to 100 x have been achieved at EMPA with suitable test objects.
266
X-ray projection images of details of the test-plate have been taken at maximum film-
focus distances of tOOO mm, focus-object distances of 150 mm, i.e., magnifications of
about 7x. Typical parameter settings were X-ray energy 17 - 18 keY, anode current 0.8
rnA and exposure times of the X-ray film around 600 seconds. Due to the cone-angle of
the emitted X-rays (about 40°), it was not possible to take magnified images of the
whole test-plate in one single exposure.
CFRP- and GFRP-composites are low-density materials and hence do not absorb X-
rays efficiently. In order to obtain absorption differences that yield contrast sufficient
for imaging, low energy X-rays (typically 10 - 40 keY) are recommended for CFRP [7].
For film-focus-distances larger than tOOO mm absorption of the low-energy X-rays (17
- 18 keY) in air became the limiting factor. X-ray intensity was too low to yield film
densities with sufficient contrast even for exposure times significantly longer than 600
seconds. In order to reduce absorption of low energy X-rays by the air between the X-
ray source and the imaging system, use of low-density helium has been recommended
as medium instead of air [8]. This requires the construction of a special enclosure for
the X-ray system (source and imaging system) and has not been attempted at EMPA.
The location of all simulated delaminations in the test plate could be identified in the X-
ray images. Magnified X-ray projection images (magnification around 7x) showed
details of the folded PTFE-films usee. •.) simulate delaminations. In one case, the
location of the PTFE film was different from its intended position (marked on the
surface of the test plate). Several PTFE-films were crimped along the fold line. The
position in depth of the PTFE-films (between first and second; type A, or between
second and third layer, type B) could not be determined with the X-ray projection
images. The detection of the simulated delaminations with X-ray was feasible because
of the absorption by the PTFE-material. Delaminations in laminates, however, do not
change the amount of material and produce no detectable change in X-ray absorption
and hence do not yield image contrast.
High density materials, e.g., liquid halogen compounds such as ZnI (so-called contrast
agents) can be used to improve contrast. If delaminations and cracks extend to the
surface contrast agents can be injected or the test object be immersed in the contrast
agent. The question whether contrast agents affect the matrix or other components of
the FRP-composite deserves attention. CF-PEEK immersed in methylene-di-iodide for
> 24 hours did not show visible changes [9] but microscopic investigation of CF-
polyimide immersed in ZnI did yield indications for dye penetrant induced
microcracking [10].
Good wetability of contrast agents is important for penetrating to the tip of the crack.
Depending on the type of signal recording (e.g., X-ray film, scintillating screen,
electronic imaging system) and analysis (e.g., electronic image processing), a certain
amount of contrast agent is needed to obtain measurablelvisible contrast [9].
2.5 ULTRASONIC A-, B- AND C-SCANS
The ultrasonic waves are excited by an emitter (e.g., a piezocrystal excited by a voltage
pulse) coupled to the test object, propagate, and are scattered and attenuated. The
detection of the signal is either in transmission or pulse-echo mode. At each sensor
267
position, the wave as a function of time (A-scan) or the maximum amplitude of the
(transmitted or reflected) signal can be recorded. If the waveform versus time is
recorded along a line of the test object, the resulting position-time plot is called B-scan.
The C-scan usually shows the maximum amplitude of the signal as a function of
position on a two-dimensional surface of the test object. All results on the test plate
have been obtained with the immersion technique (water) in pulse-echo mode.
The attenuation of elastic waves in FRP-composites depends on the frequency of the

wave. Higher frequencies are more strongly attenuated than lower frequencies. In the
test plate, ultrasonic B-scans at 20 MHz show the reflection from the opposite surface
(thickness 1 mm) while B-scans at 50 MHz do not (due to attenuation). On the other
hand, the spatial resolution is increasing with increasing frequency. B-scans at 50 MHz
with a step-size around 25 ,.1m clearly show the corrugation (texture) of the top layer of
roving.
As with X-ray imaging, the location of all simulated delaminations in the test plate
could be detected with ultrasonics (both B- and C-scans). B-scans also indicated the
depth position (type A or B). By -'loosing a suitable setting of the gate-window (pulse-
echo mode), this information could also be obtained from the C-scans. Even at high
resolution (scan step size around 100 /.lm) the folding and crimping of the PTFE-films
could not be detected with ultrasonics.
Another problem is posed by the shape of the test object. Ideally, the excitation wave
should be incident normal to the surface. Even a slight curvature can change the
amplitude of the reflected signal in pulse-echo mode. An example of this are CFRP-
plates where changes in amplitude of the reflected signal have been attributed to a
slight curvature, i.e., a deflection of about 10 mm over a length of 300 mm [11].
Selectively tilting the sensor with respect to the plane surface of a selected area of the
test-plate (by + 7.5° about the x- or y-axis) yielded C-scans with line features
predominantly parallel to the respective tilt axis. Structures or parts with shapes with
small radii of curvature are also difficult to scan [12].
Edge effects (e.g., reflection at edges) also affect the recorded signal. An example of
this has been seen in C-scans of CFRP-coupons with center holes (unpublished results
by the authors). Due to these effects, the detection of delaminations originating at edges
is difficult, unless they have grown a certain distance into the laminate.
Contact-less ultrasonic testing is feasible. Generation of ultrasonic waves with laser-

pulses and an application are discussed, e.g., in [13]. Ultrasonic waves can also be
coupled by air [14] and are the topic of current research activities. For the test-plate,
results from air-coupled ultrasonics are not available.
2.6 ACOUSTIC EMISSION
The elastic waves in Acoustic Emission (AE) are caused by a variety of processes,
often stress energy release inside the test object. AE is hence based on an irreversible
process. The transient waves, the so-called "burst" emission, are detected by sensors
(usually piezoelectric transducers) on the surface. The voltage signals of the sensors are
electronically amplified and analysed. Signal recording consists either of electronic
268
records of a set of AE-waveform parameters (e.g., amplitude, duration, number of
threshold crossings) or of the complete transient AE-waveforms.
Broad-band sensors are preferred for recording transient waveforms, since they distort
the AE-signalless than the resonant sensors. Resonant sensors usually provide higher
sensitivity than broad-band sensors. The choice of the resonance frequency depends on
noise sources (e.g., low frequency mechanical vibrations or high frequency electrical
interference) and attenuation (dependent on material properties). A compromise
suitable for many applications is a resonance frequency around 150 kHz.
Resonant AE-sensors typically detect signals within an area with a diameter between 30
- 50 cm in FRP (determined by wave attenuation in the material). A limited number of
sensors suffices to completely cover FRP-parts or structures. Criteria for determining
the number of sensors and their location are available [15].
AE was used on a different test-plate from CFRP-composites. This CFRP-plate also

contained simulated delaminations made from thin foil inserts. The plate was loaded
under three- and four-point bending and AE was used to detect indications for
delamination growth from the simulated defects. Some macroscopically visible cracks
were detected on the surface of the plate after testing. Later, ultrasonic C-scans did not
yield indications of delamination grow ___ . The relatively low number of detected AE-
signals seems to be consistent with this. The AE-signals were hence tentatively
attributed to microscopic and macroscopic crack formation.
Time
Figure 2. Schematic "stair-step" load pattern used for structural integrity assessment with
Acoustic Emission. The Felicity-ratio is the ratio of the load level at which
significant AE activity is observed upon reloading to the previous maximum load.
This ratio can be > 1, = I, or < 1 [15].
Empirical criteria (AE activity, AE intensity and the so-called "Felicity"-ratio [3, Vol.
5]) are used to evaluate the stress energy release e.g., due to crack formation or
delamination growth. These effects relate to the integrity of structures. If a defined
"stair-step" load pattern (Figure 2) is applied to the test object, the empirical criteria can
269
be quantitatively evaluated. The criteria had originally been developed for AE from
pressure vessels made of glass-fiber reinforced composites but can be applied to other
types of structures and to CFRP-structures as well [15].
Source location in AE analysis uses triangulation analysing arrival time differences at

the sensors but the difference in signal propagation speed between the components of
the composites and complex wave propagation effects (reflection, wave-mode
conversion, etc. [16]) often limit the location to zonal location (area around "first-hit"
sensor). This is the reason why AE is usually less accurate in locating cracks or
delaminations compared with imaging NDT-methods (e.g., ultrasonic C-scan).
The identification of AE source mechanisms is mostly based on empirical criteria (e.g.,

amplitude ranges of the AE signals). Recently sophisticated waveform analyses (using
broadband sensors) have been developed. AE signals in FRP-composites are caused by
several mechanisms, e.g., delamination growth, microcrack formation, fiber debonding,
fiber breaks, and friction between surfaces.
Designing tests with a single d( 'nant damage mechanism (e.g., delamination, matrix
splitting) where the other damage mechanisms contribute a negligible amount of
signals, are useful to create a data base for signal identification with AE signal
parameters and/or waveforms from knOWlI sources [17].
3. Delamination Growth in FRP-Composites Detected by NDT
3.1 INTRODUCTION
Delamination growth is an important failure mode in both glass-fiber (GF) and carbon-
fiber (CF) reinforced composites. Delaminations can be caused, e.g., by impact, e.g.,
[18] or by fatigue, e.g., [19]. Fracture mechanics is used to predict quantitatively how a
crack or delamination will behave under given loading conditions (e.g., service loads).
Therefore, detection of the location of the delamination and the accurate measurement
of delamination lengths are important for making predictions about the service life of
FRP-structures. Another application is the determination of fracture toughness values
for FRP-materials that relies on measurements of critical load, of displacement and of
delamination length.
Table 4 lists NDT-methods for qualitative and/or quantitative delamination growth

detection. The discussion below concentrates on NDT-methods that can be applied in-
situ for delamination sizing and have potential for "real-time" data analysis. They are:
(1) Visual inspection, (2) real-time X-ray projection imaging, and (3) Acoustic
Emission. Ultrasonic C-scan is the NDT-method usually chosen for ex-situ detection of
delaminations.
3.2 VISUAL INSPECTION
In opaque composites, such as CFRP, delamination growth can be detected by visual

inspection if the delamination extends to the surface (visible as crack). Dye penetrants
can be applied to highlight microcracks on the surface. Delaminations inside
270
TABLE 4. Categories and examples of NOT-methods for delamination length measurements (N.B. the
listing of the examples is not exhaustive).
Category [3, Vol. 10] Method Quantity used for measurement
Mechanical Tensile or bending test Compliance change

Optical Visual inspection Crack length (with respect to
specimen marking)
Video or high-speed film Crack length (dito)
Projection Moire, e.g., [20] Moire pattern
Penetrating radiation X-ray imaging (including Crack length (with respect to
projection imaging) reference marker)
Real-time X-ray imaging Crack length (dito)
Electromagneticl Resistive network (e.g., "silver Resistance change of network
electronic paint") across delamination
Electrical emission, e.g., [21] Induced voltage
Soniclultrasonic Ultrasonic C-scan Reflected ultrasonic wave (pulse-echo
(Ultrasonic B lIl) mode)
Transmitted ultrasonic wave
(transmission mode)
Acoustic Emission Stress release due to crack growth
transparent or semitransparent GFRP-composites can often be located and sized with

visual inspection.
In fracture toughness tests, visual inspection by eye or with a travelling microscope is

the main method for measuring delamination lengths. It is implemented in all standards
and protocols for fracture toughness testing to date. The position of the tip of the
delamination is determined on the edge of the specimen as a function of time. Markers
(e.g., every mm) are put on the edge of the specimen before the test.
3.3 REAL-TIME X-RAY PROJECTION IMAGING
Special X-ray image converters allow X-ray imaging in "real-time", the time resolution
is usually the video recording frequency (25 Hz). Because of the short exposure time
per image, contrast is lower than in still images with longer exposure times taken with
image converters or on X-ray film.
Delamination growth can be followed in "real-time" with X-ray projection imaging if

(1) an X-ray image converter with video recording is used, and (2) the delamination is
highlighted by a contrast agent. "Real-time" means that the video images can be
watched and/or recorded at typically 25 images per second. Quantitative analysis with
image processing and analysis is usually performed after the test [9].
271
Contrast agents are essential for detecting delaminations or delamination growth. The
simulated delaminations in the test plate consist of additional layers of material that
absorb X-rays and hence yield contrast. Delamination growth in CFRP-composites does
not change the amount of material and thus does not yield show in the X-ray images.
Delamination growth in composites may, e.g., affect the fiber orientation, and changes
in fiber orientation may be seen in X-ray projection images. The interpretation and
evaluation of such indications is nevertheless difficult and often ambiguous.
In fracture toughness tests, real-time X-ray projection imaging was used to clarify the
time of the onset of delamination growth from the tip of the starter crack (so-called
"initiation"). The question was whether the change of the slope observed in the load-
displacement plot could be attributed to the onset of delamination growth and thus be
used as alternative to the operator-dependent visual observation on the edge of the
specimens. Contrast agents are necessary to highlight the shape of the delamination.
Contrast agents have been shown to induce microcracking in a polymer-matrix [10], but
the test time of a few minutes is considered short enough to exclude significant effects
on the fracture toughness mesurement from microcracking due to contrast agents.
Video recording of the X-ray images and image processing (frame averaging) yielded
additional insight into the delamination process. Delamination initiation typically
started inside the specimen, in the form vi protrusions slowly growing from the insert
tip. The protrusions then coalesced into the typical curved shape of the delamination tip
[9]. The delamination hence extends further inside the specimen than along its edges.
Visual inspection and structured light techniques [3, Vol. 8] determine delamination
length on the edges. This raises the question of whether optical NDT-methods tend to
underestimate the "real" delamination length. Another question in the case of curved
crack tips is which delamination length should be used in the data analysis. Measuring
grey level differences, the position of the crack tip across the specimen width can be
determined to within a few image-pixels [9].
Once a curved tip had formed from the straight tip of the insert film, delamination
growth continued in a similar fashion, i.e., by the formation of small protrusions in the
center of the specimen, followed by coalescence into another curved tip [9]. This
indicates that the assumed continuous, stable delamination growth is an idealisation.
Stable delamination growth in fiber-reinforced composites seems to be the result of
spatially localised microscopic events (e.g., micro-crack formation yielding damage-
zones, i.e., the small protrusions highlighted by the contrast agent). Acoustic emission
monitoring (see section 3.4) supports this interpretation.
In most tests there is also a variation in the speed of delamination growth that correlates
with the variation in load (decreasing load after the first maximum with higher speed
and vice versa). In a few cases delamination tip shapes have been observed that were
asymmetrical with respect to the median of the specimens. These may indicate
problems with the load introduction (asymmetrical loading) or inhomogeneities in the
material. As noted in [9] the equipment cost and time spent for the analysis make real-
time X-ray imaging too expensive for routine testing.
272
3.4 ACOUSTIC EMISSION
Acoustic Emission (AE) monitoring has been used simultaneously with real-time X-ray
projection imaging for investigating delamination growth in fracture toughness tests
under Mode I loading [9]. The AE-activity measured in number of "hits" per 5 seconds
from such a test is shown in Figure 3. The main goal of the tests was to investigate
delamination initiation. Real-time X-ray images show that peaks in the AE-activity
coincide with fast delamination growth. In periods with low AE-activity, delamination
growth is slower or stops. The amplitudes of the AE-signals are mostly between 60 - 80
dB. This range has been empirically correlated with delamination growth in FRP-
composites [17].
1000
20
I
U; 800 , M
~
- J
~~
=
::a:
600 ~..!!
~-o .. ~
,
~
~ 0 100 200 300 400 500 I
~
w
400
1
""~
-. ,.,.,J
c(
200
\ !~
\
o .J
o 200 400 600 800 1000 1200
Time[s]
Figure 3. Acoustic Emission activity from a fracture toughness test (delamination growth under Mode I
opening load).
The non-linear point in the corresponding load-displacement plot (not shown) that is
used as indication of delamination initiation occurs about 480 seconds after the start of
the test. Real-time X-ray projection imaging shows the formation of the first
protrusions from the tip of the insert (starter crack) about 10 - 20 seconds earlier. Visual
observation of delamination growth (on the edge of the specimen) follows even later
and the maximum load point, the last of the three "initiation" points (non-linear, visual,
and maximum load), coincides with the ftrst peak in the AE-activity.
The inset in Figure 3 indicates that AE activity appears ftrst about 50 seconds after
starting the test. AE-activity and amplitude of the AE signals increase steadily after
about 200 seconds. The AE-signals indicate changes in the specimens long before X-
ray and visual inspection yield indications for delamination initiation. This raises the
question whether AE could be used to identify delamination initiation with higher
sensitivity than the other NDT-methods. The source mechanisms of the AE signals can,
at present, not be identifted with certainty. Empirically, the amplitude range (between
40 and 60 dB) points to microcrack formation [17]. It is possible that microcracks serve
273
as precursors for delamination growth. The formation of protrusions seen in X-ray

images has tentatively been attributed to microcracks growing into a local damage zone
(see 3.3 above).
Preliminary Acoustic Emission measurements on fracture toughness tests in Mode II

were difficult to interpret. Friction between the two surfaces of the delamination and
between the specimen and the fixture is probably a dominant source of AE-signals.
4. Discussion of NOT-Methods for FRP-Composites
4.1 INTRODUCTION
The advantages and disadvantages of the four NOT-methods can not be fully discussed
here, even limiting the application to FRP-composites. The CFRP-test plate used at
EMPA for a comparison of Visual inspection, Ultrasonic B- and C-scan, X-ray
Projection Imaging and Lock-in Thermography (unpublished results by the authors and
P. Wyss) shows that, with thp exception of Visual Inspection, the methods yield
comparable indications of the s;~'Jlated delaminations. Of course, the test parameters
have to be suitably chosen for each NOT-method.
The approach taken in this section is to ull;hlight selected advantages and disadvantages
of the NDT -methods that are used to detect and size delaminations in FRP-parts or
structures. The next section will then discuss criteria for the selection of NDT-methods
for application on FRP-composites.
4.2 LOCATION OF DELAMINATIONS IN FRP-COMPOSlTES
In composite structures or parts the NDT-methods first have to locate the delamination.
The example of the test plate shows that several NDT-methods yield comparable
location results. Sometimes, areas with "critical" stress concentrations or "critical"
deformations under load can be identified from Finite Element modelling (stress
distributions) or from Acoustic Emission monitoring during loading (spatial clustering
of AE events).
Frequently, delaminations from impact or fatigue loading occur between laminae inside
the structure or part and do not extend to the surface. This limits the applicability of
visual inspection and of dye penetrants to enhance detection by visual inspection.
As discussed above, X-ray imaging depends on the use of contrast agents for detecting
the location of real delaminations. X-ray Compton backscattering can indicate the
location of delaminations without contrast agents [22]. The method is using a special
source-detector geometry. It also provides depth information, i.e., indicates the
delaminated layers. Portable Compton backscattering equipment for field testing of
FRP-structures (e.g., airplane wings) is available.
The method of choice for assessing delaminations in FRP-structures and -parts is the
ultrasonic C-scan [18]. It is routinely used to test CFRP-panels after fabrication (ex-
situ) and to periodically inspect CFRP-structures. Ultrasonic B-scan [3, Vol. 7] yields
274
information on the layer in which the delamination occurs and supports the interpretion
of the C-scan images.
One limitation of Ultrasonics are edge effects (see above), another is the so-called
"shadowing" effect. B- and C-scans of CFRP-coupons with center holes showed that
delaminations that strongly reflect the ultrasonic waves tend to hide delaminations in
layers below. X-ray imaging with contrast agents also showed that debonding of fibers
and microcrack formation can occur beyond the tip of the delamination seen in
ultrasonic C-scans. Both of the above effects can lead to an underestimation of the
damage in multi-layer FRP-composites.
The "shadowing" effect can be reduced by scanning a part from both sides. However,
this will increase test time (and cost) and the geometry of the part (e.g., stringer
reinforced thin shells) or the lay-up (e.g., sandwich structures) often does not allow
scanning from both sides.
4.3 DELAMINATION LENGTH MEASUREMENTS DURING DELAMINATION

GROWTH
The required time resolution in standardised fracture toughness tests, and to some
extent, the total number of measuremer._ affect the selection of the NDT-method. The
measurement frequency has to be chosen such that the delamination length increment is
obtained with sufficient accuracy. At typical quasi-static load-rates of a few mm1min
stable delamination growth in most GFRP- or CFRP-composites is on the order of a
few mmls at most. Furthermore, length readings are required every 1 or 5 mm which
yields a measurement frequency of about 1 Hz or less. This allows to use visual
detection of the position of the crack tip.
The limitations of visual detection and measurement of delamination growth are: (1)
observation of delamination growth depends on the operator (eye sight, alertness,
experience), (2) if delamination growth is not recorded (e.g., on video) the data can not
be rechecked afterwards, and (3) for some loading modes (Mode II or Mixed Mode 1111
tests, e.g., [23]) it is difficult to detect the delamination tip. Analysis of round robin
fracture toughness test data has shown large in-laboratory and inter-laboratory
variations for visual detection of delamination growth initiation, e.g., [23].
As discussed (see 3.3 above), X-ray projection imaging with video recording of images
from an X-ray image converter yields data that can be analysed visually. In case of
doubt, the records can be rechecked. Electronic image processing and analysis yields
quantitative data (e.g., delaminations lengths or delamination shapes) but is more time-
consuming.
Acoustic emission (AE) opens the interesting possibility of microscopic investigation

of fracture phenomena. However, it is difficult to unambiguously define macroscopic
delamination initiation based on microscopic events recorded by AE. Any choice of a
limit of either AE-activity or of AE-signal amplitude is, at least to some extent,
arbitrary, since, e.g., lowering the instrument threshold might yield an even earlier
onset of AE-activity. AE signal detection also depends, e.g., on sensor position and
coupling.
275
At present, the location accuracy for AE signal sources in FRP-composites is limited.

The location of transverse matrix cracks in FRP, for example, is described in [16] but
the direction of crack growth has not been resolved in this case. At slower speeds of
crack growth (e.g., stable delamination growth in fracture toughness tests), the position
of the crack tip as a function time can be detected with AE [24].
s. Selection Criteria for NDT-Methods Applied to FRP-Composites

5.1 INTRODUCTION
Ultrasonics for ex-situ use [3, Vol. 7], X-ray [3, Vol. 3] and LaserlLight techniques [3,
Vol. 8] for both in-situ and ex-situ use have been and still are the main choices for
detecting indications or measuring other properties of FRP-composites. Acoustic
Emission [3, Vol. 5] is less well known but ideal for many NDT-problems in FRP-
composites. Some advantages and disadvantages of the selected NDT-methods have
been discussed in sections 3 and 4 above. Besides these technical considerations (e.g.,
resolution, in particular for imr .=',g, penetration, and signal-to noise ratio [3]), other
aspects, such as shape and size of the test objects and the cost for testing have to be
taken into account for selecting a NDT-method for FRP-composites.
5.2 SHAPE AND SIZE CONSIDERATION
Size and shape of the test object pose an important limitation for NDT. Small parts
(several mm to cm) can be tested with commercial equipment. Large parts (several
meter or more) often require portable equipment or robots. Large size implies (1) time-
consuming measurements, and (2) large amount of data, requiring (3) time-consuming
analysis. For imaging, the test-time frequently depends on the spatial resolution (scan
step size). Complex shapes may hide parts of the test object and make imaging and/or
interpretation of the images difficult.
5.3 COST CONSIDERATION
The selection of the NDT-method may involve aspects other than its technical merit,
e.g., availability of a method and/or operator, certification for method or operator, cost
of test (including specimen preparation) and analysis, and compatibility with previous
results (for comparison). Cost is likely to be the most important factor. Significant cost
factors are: (1) Test duration including set-up and analysis time, (2) equipment, and (3)
certified operators and their training. Rules of thumb are, e.g., that time spent on
analysing AE data can be 2 - 3 times that spent on testing; and that Computer
Tomography image reconstruction/analysis is equivalent to measurement time.
If available, a combination of NDT -methods is recommended for solving a particular

problem. A guiding principle for this is to apply NDT -methods that yield
complementary information. Such a combination of NDT-methods is, e.g., "global"
Acoustic Emission to identify the location of possible "critical" areas followed by
276
"local" investigation (e.g., imaging) with Ultrasonics (C-scan), X-ray (film, image
converter), or LaserILight techniques in these critical areas. Combining the sensitivity
of Acoustic Emission for detecting damage accumulation and the spatial resolution and
contrast of imaging NDT has proven to be a simple but powerful approach. Modelling
and simulation (e.g., with Finite Element Methods) can support the NDT-investigation.
Defects, e.g., introduced during manufacturing, however, may cause critical stress
concentrations in areas other than those theroretically predicted to yield large stresses.
There are five trends in NDT that are likely to continue: (1) Increasing the sensitivity,
(2) increasing spatial resolution, (3) decreasing test- and analysis-time, (4) increasing
automation, and (5) development of real-time NDT-methods. Improving time-
resolution sometimes results in a trade-off with less contrast or spatial resolution. All
but two NDT-methods (visual inspection and tapping) used today have been introduced
within the last 100 years (discovery of X-rays in 1896), and about 50% within the last
30 years. Therefore, it is likely that the development of new NDT-methods will
continue. Existing methods will be pushed closer to the limits allowed by the respective
physical principle(s). Electronic signal processing will play an important role, both for
imaging methods and for classification of indications by expert systems. Automated, in-
situ NDT-methods will provide means to implement total quality assurance (100%
testing) and statistical process control in production lines; as well as continuous or
periodic in-service monitoring of the prL_cts throughout their lifetime.
7. Acknowledgement
Contributions of P. Flueler, M. Heusser, R. Jaggi, B. Jahne, R.A. Nordstrom, St.

Tanner, Ch. Walder (Polymer/Composites section), E. Blaser, Th. Luthi and P. Wyss
(NDTIRopeway Technology section), and E. Hack (Electronics/Measuring Techniques
section) are gratefully acknowledged.
8. References
1. (1997) ASTM E 1316, Standard Terminology for Nondestructive

Examinations, in Annual Book of ASTM Standards, Volume 03.03
Nondestructive Testing, American Society for Testing and Materials, West
Conshohocken, USA.
2. Engineered Materials Handbook (1987), Volume 1, Composites, American
Society for Metals (ASM), Metals Park, USA.
3. McMaster R.C., McIntire P., Moore P.O., eds. (1982-1996) Nondestructive
Testing Handbook, 2nd Edition, Vol. 1 Leak Testing, Vol. 2 Liquid Penetrant
Tests, Vol. 3 Radiography & Radiation Testing, Vol. 4 Electromagnetic
Testing, Vol. 5 Acoustic Emission Testing, Vol. 6 Magnetic Particle Testing,
Vol. 7 Ultrasonic Testing, Vol. 8 Visual and Optical Testing, Vol. 9 Special
Nondestructive Testing Methods, Vol. 10 Nondestructive Testing Overview.
American Society for Nondestructive Testing (ASNT), Columbus, USA.
4. Summers cales 1., ed. (198711990) Non-Destructive Testing of Fibre-Reinforced
Plastics Composites, Vol. 1, Vol. 2, Elsevier, Oxford, UK.
5. Ely R.V., ed. (1980) Microfocal Radiography, Academic Press, London, UK.
277
6. Goldstein J.I., Newbury D.E., Echlin P., Joy D.C., Fiori C, Lifshin E., eds.
(1981) Scanning Electron Microscopy and X-Ray Microanalysis, Plenum
Press, New York, USA, p. 107f. (Kanaya and Okayama model).
7. Barry R.C., Betz R.A. (1991) Ultralow-Energy Real-Time Radiographic
Techniques for Composites, Materials Evaluation, 49, 474- 477.
8. Rudich I., Lesensky L., Kapelewski F. (1986) High-X-Ray Flux, High-
Contrast Radiography of Carbon Fiber-Reinforced Epoxy Aircraft Structures,
Materials Evaluation 44, 1158-1184.
9. Fltieler P., Brunner A.J. (1992) Crack Propagation in Fiber Reinforced
Composite Materials Analysed with In-situ Microfocal X-Ray Radiography
and Simultaneous Acoustic Emission Monitoring, Proceedings European
Conference on Composites Testing and Standardisation CTS, 385-394,
European Association for Composite Materials (EACM), Bordeaux, France.
10. Paige Clifton Furrow A., Dillard D.A., St. Clair T.L., Hinkley J. (1998) Dye
Penetrant Induced Microcracking in High Performance Thermoplastic
Polyimide Composites, Journal of Composite Materials, 32, 31-48.
11. Neuenschwander J., Luthi Th., )lordstrom R., Brunner A.J. (1993)
Kohlenstoff-faserverstarkter Kunststoff: Charakterisierung von Bauteilen aus
CF-PEEK mit bildgebenden Ultraschalltechniken und Schallemission (in
German), Proceedings German Society for Non-destructive Testing, Vol. 37,
Part 2, 1022-1029.
12. Kreier P., Gribi M., Durocher J.M., Hay D.R., Pelletier A., Edelmann X.
(1993) Computer Intergrated Testing, European Journal of NDT, 2,94-99.
13. Chang F.H., Drake T.E., Osterkamp M.A., Prowant R.S., Monchalin J.P., Heon
R., Bouchard P., Padioleau C., Froom D.A., Frazier W., Barton J.P. (1993)
Laser-Ultrasonic Inspection of Honeycomb Aircraft Structures, Review of
Progress in Quantitative Nondestructive Evaluation, Vol. 12A, 611-616,
Plenum Press, New York.
14. Schindel D.W., Hutchins D.A. (1995) Through-thickness characterization of
solids by wideband air-coupled ultrasound, Ultrasonics, 33, 11-17.
15. Committee on Acoustic Emission from Reinforced Plastics, CARP (1987)
Recommended Practice For Acoustic Emission Testing of Fiberglass
Reinforced Plastic Resin (PR) TankslVessles, Society of the Plastics Industries
(SPI), New York, USA.
16. Prosser W.H., Jackson K.E., Kellas S., Smith B.T. McKeon J., Friedman A.
(1995) Advanced Waveform-based Acoustic Emission Detection of Matrix
Cracking in Composites, Materials Evaluation, 53, 1052-1058.
17. Brunner A.J., Nordstrom R.A, Fltieler P. (1997) Fracture Phenomena
Characterization in FRP-Composites by Acoustic Emission, Proceedings
European Conference on Macromolecular Physics: Surfaces and Interfaces in
Polymers and Composites, European Physical Society, Vol. 21B, 83-84.
18. Potel C., Chotard T., de Belleval J.-F., Benzeggagh M. (1998) Characterization
of composite materials by ultrasonic methods: modelization and application to
impact damage, Composites Part B, 29B, 159-169.
19. Roy C., EIGhorba M. (1988) Monitoring Progression of Mode II Delamination
During Fatigue Loading Through Acoustic Emission in Laminated Glass Fiber
Composite, Polymer Composites, 9, 345-351.
278
20. Arakawa K., Ishiguma M., Takahashi K. (1994) Study of mode I interlaminar
fracture in CFRP laminates by moire interferometry, International Journal of
Fracture,66,205-212.
21. Sklarczyk C., Winkler S., Thielicke B. (1996) Die elektrische Emission beim
Versagen von Faserverbundwerkstoffen und ihren Komponenten (in German),
Mat.-wiss. u. Werkstofftech., 27, 559-566.
22. Kosanetzky J., Harding G., Fischer K.H., Meyer A. (1988) Compton
Backscattering Tomography of Low Atomic Number Materials with the
Surpass System, Proceedings 4th European Conference on Nondestructive
Testing, 4th ECNDT, 2118-2132, Pergamon Press, New York, USA.
23. Davies P., Blackman B.R.K., Brunner AJ. (1998) Towards Standard Fracture
and Fatigue Test Methods for Composite Materials, Applied Composite
Materials (to be published).
24. Bohse J., Krietsch T. (1998) Interpretation of Fracture Toughness in
Unidirectional Glas-FiberIPolypropylene Composites by Acoustic Emission
Analysis of Damage Mechanisms, Progress in Acoustic Emission IX:
Transitions in AE for the 21st Century, (K. Ono, ed.) Acoustic Emission
Working Group and Acoustic Emission Group, V55- V62.
TOWARDS STANDARDISED INTERLAMINAR FRACTURE TESTS FOR
UNIDIRECTIONAL FIBER·REINFORCED POLYMER·MATRIX
COMPOSITES
AJ. BRUNNER
EMPA, Swiss Federal Laboratories/or Materials Testing and Research
Uberlandstrasse 129, CH-8600 DUbendorf, Switzerland
ABSTRACT
This paper deals with interlaminar fracture testing of unidirectional (UD) fiber-
reinforced polymer-matrix fFRP) composites. The emphasis will be on the
development of standardised test procedures. After defining the terminology, an
overview is given of interlaminar fracture toughness testing under different loading
conditions. A selection of experimental aspects from the test development is presented
in detail. Discussion and conclusions then focus on results from recent research.
1. Introduction
1.1 DEFINITIONS
Unidirectional (UD) fiber-reinforced polymer-matrix (FRP) composites consists of

continuous fibers. The fibers are all aligned parallel and embedded in a polmyer matrix.
Interlaminar fracture follows from delamination growth between two adjacent laminae
(stacked layers with UD-fibers). The critical fracture toughness, Kc, or the critical
energy release rate, Gc, are considered to be material properties and to describe the
resistance of UD-FRP to crack growth or delamination. Most of the work to date is on
carbon- and glass-fibers.
Three basic modes of stress or loading, labelled Mode I (opening load), Mode II (shear
load), and Mode III (tearing load) yield corresponding values of critical fracture
toughness or critical energy release rate. Delamination growth will occur if the energy
release rate at the crack tip, G, exceeds the critical value, Gc. Mixed mode loading
conditions, e.g., 1111, are also evaluated for standardisation.
1.2 LITERATURE
The development of interlaminar fracture toughness tests for UD FRP-composites,

under both quasi-static and fatigue loading, has recently been described in [1]. That
paper contains references to selected review and research papers, as well as to
published standards and draft procedures submittted for standardisation.
279
280
2. Overview of Interlaminar Fracture Toughness Testing
2.1 STANDARDISATION
Standardisation of interlaminar fracture toughness tests has concentrated upon UD

FRP-composites for two reasons: (1) Research indicated that (at least for Mode I) UD-
laminates provide conservative, i.e., the lowest, values compared to other lay-ups, and
(2) the main industrial application of standards is quality control at manufacturing sites
where UD-laminates are easily prepared.
Table 1 summarizes the development of interlaminar fracture toughness tests for UD

FRP-composites. Standardisation of interlaminar fracture tests for FRP has, among
others, been pursed by: (a) American Society for Testing and Materials (ASTM),
Subcommittee D30.06, (b) European Structural Integrity Society (ESIS), Technical
Committee 4, and (c) Japanese Industrial Standardisation (JIS) group, High Polymer
Center. JIS (through the Japanese Standards Association) and ASTM issue national
standards, while ESIS submits test protocols to the International Organisation for
Standardisation (ISO). ASTM and JIS often submit their national standards also to ISO.
TABLE 1: Selected specimens and test protocols for interlaminar fracture touglmess testing of UD FRP-
composites
Mode Specimen type * Standard or test protocol
I tensile DCB ISO CD 150241 ASTM 055281

ns K7086 1 pr EN 6033
WOO [2]
U shear ENF ns K 70861 ASTM draft
ELS 1 SENF 14ENF [3] VAMAS draft **
CDO[4]
mtearing Ecr ASTMdraft
Notched ACPB [1]
IIII mixed tensile/shear MMB *** ASTMdraft
ADCB *** ESIS draft
I1IU mixed tensile/tearing
lIlUI mixed shearltearing
* DeB - Double Cantilever Beam, WOO - Wedge Driven Delamination, ENF = End Notch Flexure,
ELS =End Loaded Split, SENF =Stabilized End Notched Flexure, 4ENF =Four Point End Notched
Flexure, COO =Curvature Driven Delamination, ECT =Edge Crack Torsion, ACPB =Anticlastic
= =
Plate Bending, MMB Mixed Mode Bending, ADCB Asymmetric Double Cantilever Beam
** VAMAS =Versailles Project on Advanced Materials and Standards; this draft compiles four test types
from the three organisations involved in test development [5].
*** MMB: Covers a range of Mode Ito U ratios, ADCB: Fixed Mode I to II ratio of 4:3.
Most progress has been made towards a Mode I test and there exist national standards
(JSA, ASTM) [1] and an ISO committee draft [6], all using the Double Cantilever
Beam (DCB) specimen. For Mode II, there exists a Japanese standard [5] but recent
research shows that the problems are far from being solved [7]. An international round
robin on Mode II between 1996-1998 initiated by V AMAS (Versailles Project on
Advanced Materials and Standards) compares four different tests (see Table 1). Mode
281
III testing is at an early stage [8]. Mixed Mode IIII tests have been investigated for
some time but standardisation will likely have to wait for the resolution of the Mode II
problems [1]. Mixed Modes I1IIl and IlIIII have not been pursued in national or
international standardisation for UD FRP-composites so far.
2.2 CURRENT DEVELOPMENTS
A selection of current research activities is shown in Table 2. Test procedures for

climates other than 23°C and 50% relative humidity or specimens conditioned in
different climates have not been submitted for standardisation yet. These developments
may become important for testing "real" engineering materials under actual service
load conditions and would serve two areas: (1) Determination of design values and (2)
failure analysis. Failure analysis requires an alternative method to produce starter
cracks as the material will have been taken from a real component and will thus not
have been manufactured containing an insert film. Starter cracks may be introduced in
these "real materials" by wedge precracking, i.e., driving a wedge into the material [9].
The limits of the wedge precracking technique with respect to laminate lay-up and type
of reinforcement (e.g., woven rovings) have not been explored yet.
TABLE 2: Selection of current developments in interlaminar fracture toughness testing of FRP-composites
Category Specific property Activity Refs.
Material Additional matrix materials Tests by manufacturers

(e.g., tougher resins) (covered by scope of standards)
Glass-fiber reinforcements * ESIS TC41 VAMAS RR ** e.g., [9]
Cross-ply lay-up ESIS TC4 RR ** e.g., [10]
Adhesives ESIS TC4 RR ** e.g., [II]
Loading Fatigue Mode I ASTMRR **, ASTM standard e.g., [12]
Fatigue Mode II ASTM RR **, ASTM draft e.g., [13]
High-rate (> 5 mm/min) Research e.g., [14]
Specimen type Curved beams (from pipes) Applied Research e.g., [15]
Sandwich beams Applied Research e.g., [16]
* Originally, standards were developed for carbon-fiber reinforced composites, research efforts mostly
by ESIS TC4 lead to incorporation of glass-fiber reinforced materials into the scope of the standards.
** VAMAS =Versailles Project on Advanced Materials and Standards; RR =Round Robin (national or
international inter-laboratory comparison with identical FRP-material or FRP-materials).
3. Issues in Fracture Toughness Testing
3.1 INTRODUCTION
The analysis of measurements of the critical interlaminar fracture toughness (Gc) is

based on the Irwin-Kies equation (I):
282
P2dC
G =-- (1)
c 2Bda
where "P" denotes the critical load, "B" the specimen width, "c" the compliance
(displacement divided by load) and "a" the delamination length. The behavior of the
compliance as a function of delamination length (dC/da) can be derived from beam
theory or experimental compliance calibration [I]. Equation (1) can hence be evaluated
using corresponding values of critical load, displacement, and delamination length.
The crucial issue of determining the delamination length with non-destructive test
(NDT) methods is presented in an accompanying paper. The discussion of issues in this
paper (type of starter crack, data scatter, selection of data points and analysis, and
validation of results) will use examples from the development of the Mode I Double
Cantilever Beam (DCB) test. Most of the issues apply to tests for other modes as well.
3.2 TYPE OF STARTER CRACK
The standard type of crack Slillter is a thin « 13 11m), non-adherent polymer film
(Aluminum films are considered unsuitable due to folding and crimping) embedded in
the laminate during fabrication [6]. T1';" provides a more or less straight crack tip
across the specimen width while precracking, i.e., a brief Mode I loading yielding a
crack extension of 3 - 5 rom or wedge precracking will produce a curved crack tip
(close to the shape observed during Mode I delamination growth [17]).
The main argument against precracking is that initiation values (see section 3.3) may
be affected by fiber-bridging. The bridging of the delamination by fibers extending
between the two specimen beams in the zone behind the crack tip is considered an
artefact of UD-composites that would not occur in "engineering" materials with non-
UD lay-up. The shape of the plot of Gc versus delamination length (the so-called R-
curve; "R" for resistance) is attributed to this fiber-bridging. A second argument against
precracking is the possible creation of a damage zone ahead of the crack tip.
The R-curve consists of the initiation value of GIc (see section 3.3) and of larger
propagation values of GIc that finally tum into a roughly constant plateau-value. There
are cases in both Mode I and Mode II where precracking yielded lower initiation values
than insert films [1, 9]. It is unlikely that the difference in shape of the crack starter tip
(straight versus curved) is causing this behavior. At present, it can not be predicted
whether any given material will yield lower values from insert or from precracking
(clearly an area for future research). Damage tolerance of the matrix affecting the
severity of damage and the size of the damage zone may be an important factor.
Wedge-precracking [6] or the combination of wedge-precracking followed by a brief

Mode I loading (precracking) provides a starter crack for materials without insert foils.
The main application for this technique is failure analysis of FRP-structures. It has
been shown for one type of GF-epoxy that wedge precracking with different precrack
lengths tends to increase the scatter (larger coefficient of variation) but does not
significantly change the GIc-values [9]. Wedge-precracking produces a curved crack-
tip similar to Mode I precracking (unpublished X-ray images taken by the author).
283
3.3 SELECTION OF DATA FOR ANALYSIS
The selection of the so-called "initiation" point of delamination growth (start of

delamination growth) has been and still is a matter of debate. Standards or test
protocols developed for standardisation currently use three different definitions: (1) the
non-linear (NL), or (2) the maximum load or 5% increase in compliance (MAX/5%)
points in the load-displacement plot, and (3) the visual (VIS) point where delamination
growth is visually observed by the operator. Visual observation is operator dependent
and can not be rechecked after the test unless the tests are recorded (e.g., on video).
In-situ X-ray imaging has shown that the detection of delamination initiation in the X-
ray images roughly occurs around the same time as the non-linearity in the load-
displacement plot. It is not clear whether both events are causally connected or not and
a detailed analysis shows that frequently changes in the X-ray images are observed
before the non-linearity. A comparison of real-time X-ray images and simultaneous
acoustic emission monitoring shows that detecting initiation of delamination growth
depends on the sensitivity of the particular NDT-method [17]. A definition of initiation
that depends on sensitivity is clearly arbitrary and hence questionable.
A round robin has shown that the variation in the MAXl5% is smaller than in the NL
point when different operators analySe lhe same load-displacement plot [18]. The
MAXl5% point is not "conservative", Le., yielding the lowest Ole-value, it corresponds
to a delamination growth of about 1 mm. Determining conservative values is assumed
to be the goal of testing, even though this is rarely stated. On the other hand, results are
calculated as an average of (usually) five specimens, typically with a standard deviation
of the order of 5 - 10%. Depending on the intended use, other presentations of results
may be more informative, e.g., taking the lowest valid value, or the spread between the
lowest and the highest valid value (Le., the "band-width") from the test series. The
problem in this case is to determine which values are valid (see section 3.5).
3.4 SCATTER IN THE DATA
The development of computer-spreadsheets for data analysis contributed to reduced

scatter in inter-laboratory comparison by consistently including all corrections. The use
of spreadsheets also pointed out differences in the interpretation of the test protocols.
This resulted in redrafting the ESIS test protocols in order to remove ambiguities [19].
In the analysis of Mode I DCB data, it has been observed that values from linear fits for
calculation of OIc or corrections show rather large in-laboratory and inter-laboratory
variations. By sequentially decreasing the number of data points for each specimen
from the end of the data table to less than about 10 - 15 yielded pronounced variations
in the values calculated from the linear fits as a function of the number of data points
(Figure 1). This variation was significantly smaller for the same number of data points
when leaving the initial and final data sets (e.g., erasing every second data set). This
was explained by statistical fitting effects that resulted in a larger influence of data
from the "end zones" of the fitting range [19]. Prescribing a minimum number (> 10) of
data points for the data analysis reduced scatter in inter- and in-laboratory comparison.
284
40
--- A 0
0 Specimen 7.1
-0(
~
..J
•6
6 0
• Specimen 7.2
r.o:l 20 A Specimen 7.3

~
::..... •
• • A6
6
• Specimen 7.4
C
.....~
<.I
~
0 00
fl.
ot:C
~A
·.~8" ••
·':l~~~~MolB:a~~Q.~
n Specimen 7.5
""0""
U 0
-20
0 10 20 30 40
Number of Data Points
Figure 1. Dependence of ~-values from linear fits on number of data points (~ is a delamination length
correction used in the beam theory anal ysis for Gc)
3.5 VALIDAnON
The standards and test protocols contain a number of validation criteria. Early round
robins varied some geometrical characteristics of the specimens [1]. Explicit criteria
are, e.g., the requirements on specimen shape and size, the most important one is the
thickness of the insert foil « 13 f.lm) that serves as crack starter. Limiting the thickness
variation along the specimen length is also considered to be important and current test
protocols set a limit of ~ 0.1 mm, while the overall thickness and width are less crucial
(as long as the specimens are sufficiently stiff).
For most loading modes there exist alternative test types (see Table 1). Identical
material has rarely been used with different test types and information on whether
different tests yield consistent data (a criterion for validation) is lacking. Probably the
largest effort in that respect is the VAMAS round robin on Mode II that compares four
different test types and two materials [5]. This may lead to the question whether the
critical fracture toughness for any particular loading mode determined in these tests is
really a material property independent of specimen type and size.
Because of the statistical effects, explicit criteria for the number of data points used in
the analysis have been introduced and in order to improve the linear fits, taking
additional data points in the early and late phases (readings taken, e.g., every mm
within the first and last 5 mm of delamination growth) has been prescribed [19].
The occurrence of stable delamination growth is another criterion. Unstable

delamination growth is rapid, almost instantaneous delamination growth that yields
drops in the load-displacement curves with virtually infinite slope. It usually stops in
so-called "arrest" points and those are currently excluded from the analysis. Data
analysis of round robins on several materials had shown that arrest points sometimes
285
yield low fracture toughness values, often lower than the initiation values that are
considered to be conservative, i.e., lower bound values (unpublished round robin data
on CF-epoxy obtained by the author). There are matrix-materials that indicate a more
pronounced trend towards unstable delamination growth [10].
4. Discussion of Current Research
Recently, attention has focused on the question whether Mode II test results represent a
material property or contain contributions from other effects [20]. Preliminary analyses
indicate that friction effects may contribute significantly. Development of corrections
for friction effects in Mode II is under way. Applying the corrections developed so far,
tends to decrease the values of GIIc but analogous corrections for Mode I may also
decrease GIc values. Measured Gllc values, so far, are always larger than GIc values
for the same material [20]. The question whether Mode II values of Gc corrected for
friction effects are larger or the same as Mode I values (Glc) can, at present, not
conclusively be answered.
Microscopic load or stress distributions are likely to be complex in a FRP-composite.

Therefore, it seems unlikely that macroscopic Mode II loading yields Mode II on a
microscopic scale in FRP and this Cotuu affect Mode II measurements. For Mode I
tests, in principle, the same argument would apply but this does not prevent the
determination of consistent and meaningful (macroscopic) GIc-values [I].
From experience, it is often argued that "real" delaminations tend to "follow" Mode I.
Mode II loading of UD FRP-composites may also include significant Mode I
contributions on a microscopic scale [20]. In that respect, procedures for the other
"pure" or "basic" modes (II, and in particular III) are considered to be of more
academic interest, e.g., in determining failure envelopes [1]. On the other hand, failures
of structures or parts from CFRP-composites may involve mixed mode loading
conditions.
Unstable delamination growth has not received much attention yet and if occurring, the
Gc-values affected by unstable delamination growth (e.g., arrest points) are classified
as invalid [6]. Brittle matrix materials may yield a larger percentage of unstable
delamination growth than less brittle materials. How fracture toughness values for
brittle materials can be reliably measured, or how the results calculated from unstable
delamination growth have to be interpreted, remain open questions.
International standardisation of the Mode I test procedure of unidirectional FRP-

composites appears within reach. Technically, a compromise acceptable for all parties
involved in the development has been found [8]. Formal editing is still being discussed.
As mentioned above, UD-composites are of limited use but the standards provide a
starting point for test procedures for non-UD composites used in engineering
applications.
286
The development of mixed mode test methods is expected to follow a similar course to
that of Mode I tests, i.e., initially unidirectional materials are studied, after which the
more realistic cross-ply and multidirectional (quasi-isotropic) laminates are considered.
The most important problem, at present, is the test procedure and the interpretation of
the results for Mode II loading. Stable, reproducible delamination growth under Mode
II loading (necessary for determining the R-curve) has quite likely been achieved with
the SENF [5], the ELS [7] and the 4ENF [3] tests. The corrections in the analysis of the
data are under investigation, since there are indications that GIIc may contain
considerable contributions from friction. It has been suspected that GIIc values may be
close or become equivalent to values for OIc, once friction effects are properly
accounted for. Further research is expected to shed more light on this problem.
6. Acknowledgement
Contributions from the members of the Technical Committee 4 on Polymers &

Composites of the European Structural Integrity Society (ESIS), in particular of B.R.K.
Blackman, P. Davies, J.G. Williams, and experimental work ofTh. DeKalbermatten, R.
Jaggi, R.A. Nordstrom, Ch. Walder at EMPA, and discussions with P. Flueler
(Polymers/Composites section) are gratefully acknowledged.
7. References
1. Davies P., Blackman B.R.K., Brunner A.J. (1998) Towards Standard Fracture
and Fatigue Test Methods for Composite Materials, Applied Composite
Materials (to be published).
2. Glessner A.L., Takemori M.T., Vallance M.A., Gifford S.K. (1989) Mode I
Interlaminar Fracture Toughness of Unidirectional Carbon Fiber Composites
Using a Novel Wedge-Driven Delamination Design, in Composite Materials:
Fatigue and Fracture. ASTM STP 1012, Lagace P.A., ed., American Society
for Testing and Materials (ASTM), Philadelphia, USA.
3. Martin R.H., Davidson B.D. (1997) Mode II Fracture Toughness Evaluation
Using a Four Point Bend End Notch Flexure Test, Proceedings 4th
International Conference on Deformation and Fracture of Composites, 243-
252.
4. Flueler P., Brunner A.J. (1992) Investigation of Mode II Crack Propagation in
Fiber-Reinforced Composite Materials Using a Setup for Curvature Driven
Delamination, Proceedings European Conference on Composites Testing and
Standardisation, CTS, 395-404, European Association for Composite
Materials (EACM), Bordeaux, France.
5. Davies P. (1996) Mode II test procedures for a VAMAS International Round
Robin, IFREMER Technical report: Dm-GO-MM 96-11, Version 1.
6. ISO CD 15024 (Version 97-02-24) Standard Test Method for Mode I
Interlaminar Fracture Toughness. GIc. of Unidirectional Fibre-Reinforced
Polymer Matrix Composites, International Organisation for Standardisation.
7. Davies P., Ducept F., Brunner A.I., Blackman B.R.K., deMorais A.B. (1996)
Development of a standard mode II shear fracture test procedure, Proceedings
287
European Conference on Composite Materials (ECCM-7), Vol. 2, 9-15,

Woodhead Publishing, Cambridge, England.
8. ASTM Committee 030.06 (Version 1214/96) A Protocol for Edge Crack
Torsion Mode llllnteriaminar Fracture Test Method.
9. Brunner AJ., Fliieler P., Davies P., Blackman B.R.K., Williams lG. (1996)
Determination of the Delamination Resistance of Fiber-Reinforced
Composites: Current Scope of Test Protocols and Future Potential,
Proceedings 7th European Conference on Composite Materials, ECCM-7,
Vol. 2, 3-8, Woodhead Publishing, Cambridge, England.
10. Blackman B.R.K., Brunner AJ. (1998) Mode I Fracture Toughness of Fibre-
Reinforced Polymer Composites: Unidirectional Versus Cross-ply Lay-up,
Proceedings 12th European Conference on Fracture. ECF-12: Fracturefrom
Defects. Vol. III" 1471-1476 EMAS Publishing, Cradley Heath, U.K.
11. ESIS TC4 (Version 97-04) Determination of the Mode I Adhesive Fracture
Energy, Glc, of Structural Adhesives using the Double Cantilever Beam
(DCB) and Tapered Double Cantilever Beam (TDCB) Specimens.
12. ASTM D6115-97: Standard Test Method for Mode I Fatigue Delamination
Growth Onset of Unidirectional Fiber-Reinforced Polymer Matrix
Composites, ASTM Vol. 15.03, American Society for Testing and Materials,
West Conshohocken, USA.
13. ASTM Committee D30.06 (Versivu July 1994) A Protocol for Interlaminar
Fatigue Fracture Testing. End-Notched Flexure (ENF).
14. Blackman B.R.K., Dear J.P., Kinloch A.J., MacGillivray H., Wang Y.,
Williams J.G., Yayla P. (1996) The failure of fibre composites and adhesively
bonded fibre composites under high rates of test, J. Mater. Sci. 31,4467-4477.
15. Beckert W., Lauke B., Friedrich K. (1995) Delamination Toughness
Computation for Curved Thermoplastic Composites, Applied Composite
Materials, 1,395-400.
16. Cantwell W.J., Davies P. (1996) A Study of Skin-Core Adhesion in Glass
Fibre Reinforced Sandwich Materials, Applied Composite Materials, 3,407-
420.
17. Floeler P., Brunner A.J. (1992) Crack Propagation in Fiber-Reinforced
Composite Materials Analysed with In-situ Microfocal Radiography and
Simultaneous Acoustic Emission Monitoring, Proceedings European
Conference on Composites Testing and Standardisation. CTS, 385-394,
European Association for Composite Materials (EACM), Bordeaux, France.
18. Davies P. (1996) Uncertainty in the determination of initiation values of OIC
in the mode I interlaminar fracture test, Applied Composite Materials, 3, 135-
140.
19. Brunner A.l, Tanner S., Davies P., Wittich H. (1994) Interlaminar Fracture
Testing of Unidirectional Fibre-Reinforced Composites: Results from ESIS
Round Robins, Proceedings European Conference on Composite Materials,
Composites Testing and Standardisation, ECCM-CTS 2,523-532, Woodhead
Publishing, Cambridge, England.
20. O'Brien T.K. (1997) Composite Interlaminar Shear Fracture Toughness,
Gllc: Shear Measurement or Sheer Myth?, NASA Technical Memorandum
110280. NASA Langley Research Center, Hampton, USA.
ADVANTAGES AND DISADVANTAGES OF THERMOSET AND
THERMOPLASTIC MATRICES FOR CONTINUOUS FIBRE COMPOSITES
A. CERVENKA
Manchester Materials Science Centre,
University of Manchester and UMIST, Grosvenor Street, Manchester
Ml 7HS, Great Britain
1. Introduction
Merits of composite precursors, i.e. thermoplastic and thermoset matrices for

continuous fibre composites, can only be established providing that their chemical,
physical and engineering aspects are considered on three levels required for
converting simple organic molecules into a composite structure that meets design
requirements during its full service life. Following this approach, attention is paid to
events spanning between 10-5 and 108 secs and involving length scales between 10-8
and 103 cm.
The objective will be achieved in tr-~~ steps, i.e. concentrating on thermosets
and thermoplastics as matrices (Section 2), forming the composite lamina by
reinforcing a matrix with a fibre (Section 3) and discussing aspects related to a
satisfactory functioning of a composite structure (Section 4). Whilst the discussion
of matrices addresses issues such as the structural aspects, thermal response,
building blocks available, reactive processing and the importance of crosslinking
density, the composite structure is considered in terms of its manufacturing, desired
property balance and durability.
2. Matrices
2.1. STRUCTURAL ASPECTS
Polymers can be conveniently subdivided according to their response to temperature

associated with their different structures. Thermoplastics can be seen as ensembles
of randomly packed and entangled linear chains. Their mobility depends on the
molecular build up in terms of molecular weight, nature of the repeating unit,
presence of intermolecular forces by virtue of polar groups and tendency to form
higher level, i.e. overmolecular, structures such as domains (block copolymers) or
crystallites (semicrystalline polymers).
Thermosets are covalently bonded networks with all polymer segments present
being attached - under ideal conditions - to one huge macromolecule. Although the
289
290
chemical nature of a thermoset polymer influences the chain mobility, this is

primarily controlled by the crosslinking density proportional to the concentration of
the junction points incorporated into the network and reducing the overall chain
movements. Both the number of junction points and their nature, as given by the
functionality of the crosslinking agent or hardener, are structural aspects dominating
the crosslinking density by controlling the length of polymer chains between
crosslinks (the effective molecular weight). Thermosets are typically prepared by
means of polycondensation or polyaddition of the mixture consisting of a
bifunctional monomer (resin) and a polyfunctional hardener.
2.2. THERMAL RESPONSE
Structural features responsible for different behaviours of the two polymer classes to
the temperature are entanglement of linear chains with thermoplastics and network
structure achieved by covalent bonding for thermosets. Using the polymer stiffness
as a yardstick, Young's modulus of a thermoplastic falls with increasing temperature
in two sudden steps between three plateau regions. The first step corresponds to the
so-called glass transition temperature Tg separating the glassy and rubbery states.
This is the vitrification point of a thermoplastic material of a given type and
corresponding to its particular molecular "'~ight. The glass transition is associated
with a sudden increase in the segmental motion and the tensile modulus is reduced
by a factor of 2000. The second step, at a higher temperature, corresponds to the
viscous flow in which elasticity of the rubbery phase gradually degenerates into the
viscosity of the polymer melt.
Characteristics of thermoplastics such as Tg , the stiffness of the glassy state
and the viscosity quantifying the terminal plateau are closely related to the overall
molecular weight of a polymer and the chemical nature of the repeating unit. A
molecular designer enjoys an appreciable freedom to increase the tensile modulus
and the glass transition temperature. However, molecular design of a high Tg
material (a major theme pursued towards the so-called engineering polymers) will
always result in a material which will be difficult to process, possible at high
temperatures only.
Focusing our attention on thermosets, the temperature dependence of Young's
modulus for this material class is simpler, because there is only one transition from
the plateau corresponding to the glassy state to the rubbery state. Covalent bonds
constituting network do not permit a thermoset to be melted. This implies that if a
thermoset is to be processed in a structural element of a given shape, this must be
achieved during network formation. Thus shaping thermosets is always associated
with processing that involves chemical reactions (usually termed 'reactive
processing'), making thermosets generically resemble ceramics more.
The crosslinking density controls both the glass transition temperature Tg and
the level of the rubbery plateau. An increase in the crosslinking density by
controlling the network structure chemically results in a thermoset of a higher Tg
with the rubbery plateau attaining a higher stiffness value. The state of material
formed is governed not only by chemical means such as functionality and
concentration of a hardener, but also by the manner of controlling network
291
formation by 'driving' chemistry in terms of reaction times and temperature. This is

discussed in Chapter 2.4.
2.3 BUILDING BLOCKS FOR THE MOLECULAR ENGINEERING
In the case of thermoplastics prepared by polymerisation, the nature of the resulting

polymer is given by the monomer type and features of the molecular weight
distribution (MWD). The monomer type is associated with features such as
length/polarity of pending side groups attached to ethylene, i.e. the simplest vinyl
monomer representative. MWD is effectively controlled by the concentration of
initiating species, polymerisation temperature, reaction medium and the type of
reaction used
If polycondensation or polyaddition of bifunctional monomers is used, more
freedom in engineering of a thermoplastic is possible. Firstly reactive groups A and
B can be attached to different fragments Re and Rt, with a wealth of possibilities to
control the nature of R by a proper choice of its length, substitution, etc. see e.g.
[1,2].
For thermosets, the building kit cor·~;.ns identical tools described above with
some additional ones, further extending engineering options. Leaving aside
possibilities of choosing a branching reagent to be copolymerised with a vinyl
monomer, it is the functionalities f. and fb (specifying the number of active groups A
and B attached to the fragments Re and Rt,) which are of the utmost importance.
Although the common practice is to cure a bifunctional resin with a polyfunctional
hardener, any combination off. and fb satisfying the stoichiometry condition f. *[A]
~ fb * [B) is possible.
In addition to forming networks from liquid organic oligomers, thermosets can
also be obtained by curing linear thermoplastic precursors. Three examples are: 1)
Vulcanisation of unsaturated rubbers with sulphur, 2) Curing of unsaturated
polyesters, and 3) Crosslinking of vinyl esters with a vinyl monomer.
Finally, the list of building blocks for thermosets would be incomplete without
stating that a) polyfunctionality of one of the reagents is responsible for the non-
linear increase in the molecular weight with the reaction time and b) curing reaction
is never completed. As the reaction kinetics play an important role here, the network
formation in the time! temperature domain is addressed below.
2.4 REACTIVE PROCESSING OF THERMOSETS
A convenient way of dealing with thernlOset processing is the so-called

'Temperature - Time - Transposition' proposed by Gilham [3] which maps changes
in T g with the reaction time and temperature. As a rule, a polyfunctional hardener is
responsible for the weight average molecular weight to reach infinity at gelation. At
the same point only a few macromolecules are incorporated into the network. The
rest constitutes an extractable sol fraction. The gelation curve - its position is
governed primarily by the functionalities and chemical structure of reagents -
together with the already discussed vitrification curve - are important boundaries in
292
the TIT plot defining coexistence of polymeric products in different structural states
of the matter. These are: liquid and ungelled glass based on linear and slightly
branched species on the left side of the gelation curve, crosslinked polymers in the
form of glasses or rubbers (below and above the vitrification curve, respectively).
The fact that all these can be obtained from a single, simple mixture of oligomers of
a well defined initial value Tg ;: Tg,o requires that a network is fully specified only
and only if the starting formulation is accompanied by a definition of the curing
route.
The TIT diagram, however, allows some more and equally important
conclusions to be drawn:
1) Resin!hardener mixtures are inherently unstable. Thus thermal stability of a
fully formulated mix requires either a low temperature storage or mixing of
chemicals on site just before the start of production of a finite product.
2) As many manufacturing steps combine formation of fully or partially cured
networks with other operations (fibre impregnation, mould filling, winding) the
curing reaction must be conducted in such a way that different, and sometimes
contradicting requirements, are satisfied by following the optimum TIt path.
3) H full cure is the objective, thermoset formation must not be 'quenched' by
reacting at too low a temperature that leads to vitrification of the reacting system
prior to gelling. H, on the other hand, a ~t'<\pable intermediate based on a partially
cured matrix is desired (case of the SO<al.led B-stage pre-pregs for vacuum bagging
of intricate shapes), then the reaction temperature between gel Tg and Tgo isto be
used. 4) Prolonged curing. at high temperatures leads to an inferior network due to Tg
having been reduced by polymer degradation or even charring.
5) As networks are formulated on the basis of very reactive species (high
reaction rates and reaction completion reasons), curing reactions are usually
exothermic, therefore the core temperature might be excessive and the state of cure
heterogeneous.
2.5 CROSSLINKING DENSITY AND MATRIX PROPERTIES
As the crosslinking density has been shown to be practically zero for amorphous
thermoplastics and the structural characteristic that can be well controlled in
thermosets, there is a clear distinction of the two material classes in terms of:
1) Solubility in organic solvents. Thermoset networks may be considered
insoluble, albeit without excluding a possibility for slight swelling. However,
thermoplastics are in principle soluble materials, so their chemical resistance,
particularly when stressed (environmental stress cracking), tends to be inferior to
that of thermosets. Chemical attack on amorphous polymers is stronger than on
semicrystalline types.
2) Toughness. Thermoplastics allow more energy to be absorbed due to chain
slippage. As this is limited with thermosets, these are generally more brittle, the
toughness characteristics being a decreasing function of the crosslinking density.
3) Ultimate strain. Arguing along similar lines as in point 2), typical failure
values are 2% for thermosets and 10% for thermoplastics.
293
4) Dimensional stability and strength. Comparable at ambient temperature,

but thermoplastics decay faster with increasing temperature.
Thermoplastics are more prone to creep.
3. Lamina
An optimum lamina represents a marriage of two equal partners fulfilling different

functions: fibre provides stiffness and strength, matrix keeps fibres apart in fixed
positions, protects them from chemical attack and provides resistance to shear and
transverse stresses. Meeting these objectives, a matrix guarantees composite
integrity and distributes loads effectively.
Proximity of a matrix to a fibre is called an interphase (if expected to have a
finite thickness and properties different to the matrix) or referred to imprecisely as
fibre/matrix 'interfacial' properties. A number of conditions need to be satisfied if
the family (composite) is to prosper: Matrix needs to wet, infiltrate and adhere to
reinforcement, to solidify when required and not to release volatile products.
3.1 THERMOSET LAMINA
Low viscosity of a typical thermoset formulation combined with a surface treatment

of a reinforcement (sizing in the case of glass fibres, oxidation of carbon fibres)
facilitates an easy wetout as soon as a continuous fibre has been brought into contact
with a resin. This physical operation is followed by a partial curing to the so-called
B-stage brought about by a temperature! time regime which leads to vitrification of
branched polymer species. This 'pre-preging' operation results in the elementary
brick of a majority of thermoset-based composites, the so-called 'pre-preg'. Its
flexibility, drapability and tackiness allow relatively easy shaping and stacking to a
desired thickness.
A way unique to a thermoset-based composite is the so-called 'Resin transfer
moulding' (RTM). A very thorough control of rheology and kinetics is needed
during this complex operation because of two requirements: effective mixing during
the reactive injection moulding step and fast and homogeneous infiltration of a
complex (sometimes 3-dimensional pre-knitted), reinforcing pre-form positioned in
the mould cavity.
The actual nature of interfaces in a thermoset-based lamina has been. the main
topic of a long and active controversy about composites. Different views have been
expressed such as:
1) A chemical reaction of the coupling agent which effectively grafts a matrix
(epoxy, polyester mainly) to the reinforcement,
2) A different curing regime of the matrix due to fibre proximity likely to result
in a gradient in matrix properties,
3) Mechanical clamping of the fibre due to a mismatch in coefficients of the
thermal expansion, and
4) Weakening of a matrix by void formation along the reinforcement.
Although an analysis of these opinions does not fall within the scope of this paper,
in our opinion the last alternative might be most relevant to thermoset lamina.
294
3.2 THERMOPLASTIC LAMINA
Concentrating our attention on the thermoplastic lamina, the physical aspects that
need to be considered are:
1) Viscosity of molten thermoplastics is inherently higher than that of
thermoset precursors and steps need to be taken to decrease it to its minimum
without any significant loss in mechanical properties of the matrix such as the
tensile strength and toughness. Three ways of how to achieve this is to use a
relatively low molecular weight polymer, a narrow MWD and a high temperature of
the impregnation process.
The last item is responsible for laminating of thermoplastic matrices to be a very
energy-consurning step. Typically, processing temperatures are recommended to
exceed Tg of amorphous and semicrystaIline matrices by 100 and 200°C,
respectively. Fibre/matrix contact is achieved either in a fluidised bed or in a melt
bath both followed by a consolidation step that involves a high pressure pultrusion.
2) Work of adhesion is generally low because the surface free energy of
thermoplastics is also low.
3) Although the lamina fOuuation does not involve any reactive processing, the
actual processing regime (in T/t terms) must not be underestimated. This is
particularly relevant to solidifying semicr:""''llline matrices.
4) Temperature regime during formation of a thermoplastic lamina has an
appreciable impact on the fibre/matrix interfacing. This is particularly the case,
when the fibre surface initiates a high density heterogeneous nucleation so that
crystal growth in the matrix is restricted to the lateral direction and a columnar
layer develops around the fibre - a phenomenon called 'transcrystallinity [4]. Again
and without labouring its consequences on lamina properties, it is enough to stress
that the presence of the interphase can be clearly seen. Its structure appears to be
different from the morphology of the bulk matrix and, most importantly, represents
a feature not occurring in thermoset laminae.
Whereas the RTM process is specific to the thermoset matrices, there are two
types of precursors to thermoplastic composites typifying tIlis matrix type: matrix
fibres comnlingled with fibres [5], i.e. an approach allowing hybridisation, 'pre-
stitching of anisotropy' and yielding drapability, and Sheet moulding compound
(SMC), obtained by sandwiching a curvilinear (swirl) continuous fibre (glass)
between two thermoplastic films.
4. Composite structure
The two previous Sections dealt with chemical aspects of matrices (Section 2) and
physical/chemical ways of obtaining composite precursors (Section 3). Issues
connected with a 'composite structure' require engineering considerations to be
included as well. These aim at answering questions such as: 1) How can it be made,
2) Has the choice of a composite precursor/lanlina and a processing route given a
structure of adequate properties and 3) Is that structure going to last for the whole
service life?
295
4.1 MANUFACTURING OF A COMPOSITE STRUCTURE
There are several primary criteria not directly related to the title of this paper: actual
geometry of a final product, its production rate and the type of industry. For
laminate plates, processing such as vacuum bagging, autoclave moulding and
compression moulding are in principle usable for both types of matrices. Since for
thermosets further chemical reaction is required, cycle times are generally longer,
the overall energy input needed for making a monolithic structure is relatively low
and a typical Tit profile is more complex and involves different pressures
For thermoplastic matrices, the cycle consists of compacting, consolidating and
demoulding, the last step being responsible for a controlled cooling that leads to the
optimum structure/property balance. Processing is carried out at high temperatures
which might be in excess of operational limits of some autoclaves. As no chemistry
is involved, the time periods are shorter leading to a substantial reduction in overall
cycle time. This is an attractive feature to industries producing a large number of
identical items for which either an investment into specific tools or use of existing
machinery for steel-plate shaping would be an advantage. Another point is that the
existence of a dividing line between chemistry and physics allows clear specification
of the feedstock for shaping (pre-preg), implementation of quality aspects and
definition of responsibilities between the fpj>.cJstock supplier and the manufacturer.
Both matrix types can also be successfully processed by the pultrusion
technique. With thermoplastics, the technique is mainly used for producing the
feedstock for injection moulding of short-fibre composites.
Manufacturing routes which do not require any composite precursor in the
form of the lamina belong entirely to thermosets. One has already been introduced
as RTM (also called reinforced reaction injection moulding RRIM), the other two
are successfully used on a large scale for the commercial production of pipes by
filament winding and complex shapes by the 'wet hand lay-up' approach. For
filament winding, a reinforcement strand is impregnated with a formulated resin,
wound on a rotating mandrel and postcured in an oven. Filament winding for
thermoplastic materials is much less developed and uses the pre-preg in the form of
a tape 'welded' into a cylindrical structure by local heating
4.2 BALANCE IN STRUCTURAL PROPERTIES
Using the lamina that contains 62% (vol.) AS carbon fibre in a thermoset (H3501)
and thermoplastic matrix (polyether ether ketone - PEEK) as an example, lamina
properties important for designing are the stiffness (engineering constants) and
strength. There are differences between the two laminae discussed [6]: El = - 2.9%,
E2 = -0.6%, G12 = - 28.2%, ul2 = -7.1%, X = -0.1%, X' = - 24.0%, Y = + 54.7%,
Y' = - 1.4% and S = + 72.0%. Quoted values represent a relevant deviation at a
given property for the APC-2 thermoplastic material after normalisation on the
same characteristic of the thermoset-based lamina. The meaning of symbols is as
usual: El, E2 and G12 are the longitudinal, transverse and shear moduli, U12 is the
Poisson ratio, X, Y and S are the longitudinal, transverse and shear strengths in
tension; the apostrophe ' refers to strength characteristics in compression. The
296
design properties for the thermoplastic composite are practically identical with the
exception of G12 ' X' being lower and Y, S characteristics being higher for the
thermoplast composite. In the first approximation, the stiffness! strength response of
a composite structure can be said to be chiefly dependent on the number of plies and
their orientation regardless of the matrix used.
Performance aspects of thermoplastic composites are further closely related to
the viscoelastic nature of a matrix responsible for non-linear behaviour:
1) lower X' due to a low modulus at failure,
2) toughness being at a fixed matrix modulus higher than that for thermosets
and thus making thermoplastic composites more robust to impact damage,
3) 'scissoring effect' [1] observed for ±45° laminates due to fibre reorientation
as a response to longitudinal tension and seen as a possible way of enhancing the
'weepage' stress of composite pipes,
4) better fatigue resistance due to existence of an unreinforced thermoplastic
interlayer between plies, and
5) tendency to creep.
These effects increase with increasing temperature for the reasons stated in Section
2.2.
4.3 DURABILITY ASPECfS
Presence of polar groups in thermosets makes composites hygroscopic and matrix

dominated properties are expected to be decreasing functions of the time (due to
increasing moisture content). Thus the operational envelope defining the borderline
between operation and failure will be shrinking. This ageing phenomenon results
from a high solubility of water in a matrix combined with a relatively fast diffusion.
It must not be underestimated, particularly for those composites formulated on very
high Tg thermoset matrices. For thermoplastics, uptake of water is much lower, so
their hygrothermal ageing affects properties to a lesser degree. Although PEEK has
been demonstrated to be solvent resistant, organic vapoursl solvents permeating
other thermoplastic matrices might behave like water in thermosets.
Service life of thermoplastic composite structures does not seem to be
drastically reduced by an impact that can be accommodated by their higher
toughness characteristics. In addition, crack self-healing is occasionally mentioned
as an advantage of thermoplastics.
One aspect of durability is that of repairability. Both composite types can in
principle be rejuvenated by replacing a damaged section. Thermoplastic-based
structures are argued to lend themselves to easier repair on the grounds of
impossibility to crosslink two networks of different populations. Welding is the
method usually advocated for thermoplastics, however, strictly along the same lines
of argumentation, continuous semicrystallinity running through the whole structure
is very difficult to achieve.
Finally, whereas an engineer may desire a structure to last as long as possible,
durability of thermosets is a concern to environmentalists. 'Green' considerations
favour thermoplastics because their recycling could follow the path of 'downgraded
anisotropy', as illustrated by the following example. A structure based on a
thermoplastic reinforced with a continuous fibre (filament wound pipe. pultruded
297
beam) ~ shredding into pellets ~ injection moulding of a composite structure
reinforced with finite length fibres ~ reinforced filler for matrices such as bitumen
~ road surface with an advantageous wet grip/rolling resistance balance.
Although this approach seems to be rational from the material point of view,
the amount of energy required for the multiple processing steps would be large and
might initiate another negative response from the 'green' lobby.
5. Composites of the future
An appropriate way of concluding this treatise is to indicate current efforts aiming at

wider composite acceptance. This can be achieved only by overcoming current
weaknesses of polymeriC matrices as such. Few examples below illustrate possible
avenues currently pursued:
1) Thermosets less vulnerable to impact have been successfully formulated by
either blending a resinlhardener mix with a thermoplastic [8] or by polymerising a
vinyl monomer during thermoset curing [9].
2) Significant improvements have been achieved for matrices, however,
toughness increment of the composite is only marginal. Insertion of thermoplastic
layers between neighbouring thermoset p':. has been found [10] to be an efficient
option for extending the fatigue life of composites.
3) Difficulties due to high melt viscosities of thermoplastics are being
overcome [11] by impregnating reinforcement with low viscosity liquids (cyclic
oligomers) and fast cyclo- polymerisation of the matrix within a shaped structure.
4) Problems arising from an effort to incorporate macroscopic reinforcement
into reacting microscopic matrices are being resolved by formulating the so-called
'molecular composites' [12] using a polymer that serves both as the reinforcement
(liquid crystalline form) and the matrix (coil conformation).
5) Finally, a recent development of rotaxenes and catenanes [13] might be an
efficient way to increase 'crosslinking density' of otherwise linear polymers.
These few examples illustrate that two distinct trajectories followed for the two
generic types of composites gradually merge and the historical rivalry will probably
cease. This is assisted by a number of current chemical, physical and engineering
problems of composites being resolved by researchers learning from the composites'
shortcomings and using their advantages only. Particularly timely is the exploitation
of their synergism. This will prove to be important for formulating matrices that will
fully deserve qualification as 'polymer' matrices. Engineering of the anisotropy with
polymeric matrices will then yield a new material class complementing those that
have already found their place in technology - ceramics and metals.
References
1) Bicerno, 1. (1993) Prediction of Polymer Properties, Marcel Dekker, Inc.,

Basel.
2) Knight, GJ. (1980) High temperature properties of thermally stable resins,
Developments in reinforced plastics -I, ed. G. Pritchard, Elsevier, London.
298
3) Wisanrakkit, G. and Gilham J.K. (1990) The glass transition temperature as an

index of chemical conversion for a high Tg amine-epoxy system: Chemical and
diffusion controlled reaction kinetics, J. Appl. Polymer Sci 41, 2885-2929.
4) Thomason, lL. and van Rooyen, AA (1992) Transcrystallised interphase in
thermoplastic composites, J. Mater. Sci. 27, 889-896 .
5) C01ECR Non Crimp Fabrics, Product brochure of Tech Textiles (Andover,
UK, 1990).
6) Tsai, S.W. (1988) Think Composites, 4th Ed., Paris.
7) Wang, C. and Sun, C.T. (1992) Behaviour of thermoplastic composite during
cooling from moulding temperature, Proc. Am. Soc. for Composites, 7th
Techn. Conf, Technomic Publishing Co., Inc., Basel.
8) Bucknall, C.B.and Partridge, I.K. (1983) Polymer 24, 639-644.
9) van Veelen, Aand Stamhuis, lE. (1991) Fracture behaviour of
unidirectionally glass-fibre reinforced composites based on toughened epoxy
resin systems, 1st Intern. Con! on Deformation and fracture of composites,
Manchester.
10) Partridge, I.K. (1993) Effects of resin-rich interply zones on fatigue of
unidirectional and cross-ply laminates, 2nd Intern. Con! on Deformation and
fracture ofcomposites, Manchester.
11) Salem, AJ. (1991) Thermoplastic- "'\latrix composites from cyclic resins,
(Gordon Research Conference on Composites, Ventura, California, USA
12) Ree, M., Yoon, D.Y. and Volksen, W. (1990) Polyimide molecular composites
via in-situ rod formation, Am. Chem. Soc., Div. Polymer Chem., Polymer
Preprints 199, 319.
13) Shaffer, T.D. and Tsay, L.M. (1990) On the way to polymeric rotaxenes and
catenanes,Am. Chem. Soc., Div. Polymer Chem., Polymer Preprints 199, 210.
HYGROmERMAL AGEING OF GLASS REINFORCED COMPOSITE
PIPES
A. CERVENKA
University o/Manchester and UMIST, Grosvenor Street,
Manchester, Ml 7HS, United Kingdom
1. Introduction
Structural applications of composite materials imply that a given structure is

exposed to a complex loading spectrum over a long period of time. This loading
spectrum can be both of a mechanical nature (dynamic, static or their
combination) and of a non-mechanical origin associated with external
environment such as temperature and concentration of chemical species present.
Fully recognising probable interaction between mechanical and
environmental factors, the current research is conducted along separate lines
with the aim to unveil time dependence of relevant engineering properties for
both: Whilst damage accumulation due· purely mechanical loadings such as
fatigue, static pressurisation of pipes and impacting has been dealt with in [1],
the purpose of this contribution is to outline a predictive algorithm that relates
failure of the most popular type of composite pipes - glass epoxy matrix
reinforced by continuous glass fibres (GRE) - to the operating temperature and
the amount of water or any other low molecular weight species that have
ingressed during the service life.
Modelling of environmental ageing is enabled by setting up a conceptual
framework consisting of 1) hygrothennal behaviour of polymers, 2) lamina
micromechanics, 3) hygrothermal effects on the behaviour of laminates and 4)
diffilsion and relationships among characteristics that govern transport of the
water.
Modelling is accomplished by means of a computer code. Its input is
documented and the output illustrated by monitoring hygrothermal ageing of a
glass reinforced epoxy (GRE) pipe. This is achieved by analysing the behaviour
of ±57° angle-plied laminate exposed to moisture at a high (T= l20°e)
temperature with the average water concentration increasing from zero to 1.6%
(w). Issues such as the relative roles of temperature and moisture during this
hygrothermal process, behaviour of individual plies and the time development of
the failure envelope of the weakest ply are studied.
299
300
2. Modelling of environmental ageing
2.1 HYGROTHERMAL BEHAVIOUR OF POLYMERS
Concentrating fIrst on thennal effects only, stiffness of a dry polymer decrease

with temperature. The dependence is nearly linear in the glassy region when
T<T800 and becomes progressively non-linear as the glassy state is transformed in
that of the rubber, i.e. in the region of the glass transition temperature T8oo. Thus
the difference T800 - T can be used as an approximation of the polymer stiffness.
The subscript "0" signifies the dry state of the polymer.
Ingress of low molecular weight species, particularly those of a high
chemical affinity to polymer chains, e.g. water to epoxies, has been realised to
plasticise a polymer. This causes a reduction in the glass temperature in the 'wet'
state and thus T80w < Tg.o implying that at a given temperature the 'wet' polymer
is a more compliant solid than its 'dry' version. A number of schemes (e.g.
Karasz [2]) have been derived to relate the reduction in Tg to the actual
concentration [w] of species w. When the concentrations are relatively low a
linear depression appears to be adequate:
TglN =Tg,o{1- ~[wD (1)
For linear polymers, the temperatur _ Jependence of the polymer stiffness
can be derived by the algorithm used in the BIOSYM modelling software for
deriving the Qualitative Structure! Property Relationships (QSPR) and
documented by Bicerano [3].
For crosslinked polymer networks such as epoxies, the nature of a repeating
unit is ill defIned and other structural parameters (e.g. the cross-linking density)
control Tg.o in addition to the chain architecture. In absence of a more detailed
information between the matrix formulation (selection of resins and curing
agents, their stoichiometry) which is, perhaps the most guarded secret in the
trade, we simply assume all epoxy resins to be identical in their molecular
characteristics (and thus to exhibit the 'universal' stiffness of 3.3 GPa at the
ambient temperature and when dry) but different in T800 with the crosslinking
density governing the glass transition temperature.
The pragmatic approach described thus allows the polymer stiffness
characterising the reference state to be reduced due to the combined effect of T
and w with the T80W acting as the pivot for the modulation process based on the
Bicerano algorithm.
2.2 LAMINA AND MICROMECHANICS
Orthotropic lamina can be understood to be the building block for all composite
structures. Its behaviour is strongly anisotropic with four engineering constants
characterising its stiffnesses and fIve parameters describing its strength. Using
the notation stiffness/strength, the following table relates the symbols used to the
deformation modes and loading directions:
301
deformation / direction longitudinal transverse

tension
compression
shear
The engineering constants can be derived micromechanically from the

properties of constituents and their concentrations. For the composites based on
isotropic constituents (such as epoxies reinforced with glass fibres) any
engineering constant of the lamina can be derived as
Ec,j = j(E j , Vj,vj) (2)
where j identifies the type of an engineering constants, i stands for a constituent
(matrix and fibre), E is the Young's modulus, v the Poisson ratio and v the
volume fraction of a constituent i. Concerning the function j in the equation
above, the computer code SMC [4] available from the Composite Centre of the
University of Delaware can be recommended for this purpose.
Micromechanical modelling of the engineering constants of a lamina is very
important as we are able to analyse data measured at different fibre
concentrations, to relate behaviour of systems reinforced with different fibres to
the reinforcement type and, from the IX' ~ of composite ageing, to synthesise
properties of a 'hot and wet' composite on the basis of our knowledge Tg,w of the
matrix. This, of course, providing that a given reinforcement is inert during the
ageing process and the so called interfaces are either absent or age as the matrix.
The former is likely to be true for glass fibres, the latter does not appear to be
critical as it follows from Raman studies of interfaces in microcomposites
undergoing in the Manchester Materials Science Centre [5].
2.3 HYGROTHERMAL RESPONSE OF A LAMINATE
Effects of moisture and temperature are accommodated by expanding 'the state

equation of the composites' (see e.g. [6]), i.e. the relationship that relates forces
N and moments M to extensions E and curvatures K by means of the elongational
lA!, coupling IBI and bending IDI matrices. The equation is modified by addition
of non-mechanical forces N"m and non-mechanical moments MTI'l':
INI = IAI*IEI + IBI*IKI - INlnm (3)

IMI = IBI*IEI + IDI*IKI - IMF
with
f f
s s
NF m = Cij(J,w)ajIlTdz + Cij(T,w){Jjwdz
o 0
(4)
f f
s s
MF m = Cij(T,w)ajll Tzdz + Cij(T,M)Pjwzdz
o 0
Here a and f3 are expansion matrices either due to thermal (temperature T)
or swelling (moisture w) effects, s is the laminate thickness and C stands for the
stiffness matrix simply relating stress/strain behaviour of the composite of which
302
properties have been degraded by presence of the moisture and operation at a

temperature different than those characterising the reference composite. As well
as the matrix ICI, also IAI, IBI and IDI are moisture/temperature dependent.
Quantification ofhygrothermal effects is feasible providing that: a)
expansion matrices 1a.1 and I~I are available, b) Engineering constants of an aged
"hot and wet" lamina are known and c) Variation in the lamina strength
characteristics with both the temperature and the moisture is known. As the
point b) has been treated in the Section 2.2, the issues under a) and c) are tackled
as follows
Concerning the 1a.1 and 1131 matrices, these can be derived for a laminate in
the same manner as the stiffness characteristics [7]. The lamina response to
either a temperature AT or a concentration [w] gradient is again anisotropic and
the longitudinal and transverse parts need to be established either experimentally
(dimensional change in an appropriate direction as a function of the driving
force), found in the literature [8] or calculated micromechanically from
constituent properties [9].
Turning our attention no.. to the lamina strength characteristics, these
represent events in the region of higher strains and are an outcome of a
complicated damage accumulation. It i.s for these reasons that their
micro mechanical modelling is currently not possible and the lamina strength
degradation factors s(T,w) - ~ in the contrast to ~ used for dealing with the
matrix stiffness - have to be established by firstly generating the necessary
experimental data and secondly describing them with a general function.
Gadke [10] has made a pioneering contribution in the experimental area by
conducting long-term exposures to water and determining all the strength
characteristics, X, X', Y, Y' and S at different equilibrium water concentrations
[w] and over a wide temperature interval. His results describing behaviour of
carbon reinforced epox)' matrices based on 914C, Code 69 and F550 clearly
indicate that the strength characteristics dominated by the matrix, namely Y and
S are affected most and that the maximum reduction is achieved by the
combined effect of a high concentration of water and the testing temperature.
We have fitted the Gadke's results by means of modifying the equation
originally proposed by Tsai [11]. Our modification is based on the relationships:
P(J,w) = P(Jrer,w = O){s(J,w)V
T -T
s(J, w) = ---",g,c.;..w __
(5)
Tg,o - Trer
where T ref is a reference temperature (usually set at the ambient 295K level) and
p an exponent that signifies deviation from the linear behaviour. The
dimensionless quantity s(T,w) is within the interval zero and one. The extreme
values are attained when T=Tg,w for the "wet" composite (terminal state when a
property is zero) and T=Tref when the composite is "dry" (composite reference
state). The absolute value of the exponent p reflects by which composite
constituent is a given strength dominated: Values p~ indicate the fibre
domination (e.g. longitudinal tensile strength X) and p>O reflects the matrix
303
domination (Y and S). The following values of these exponents have been
determined by analysing the Gadke's data:
p X X' Y Y' S
p 0.08 0.15 0.50 0.15 0.30
Although the type of epoxy matrix and, likely, also the volume fraction of
reinforcing fibres will have an effect on the p-values, we propose the values
above to be used as the first approximation indicators for any epoxy resin at this
moment of time. Further refinement will be easily made by a deeper analysis of
more experimental data generated for resins of known chemical formulations.
2.4 MOISTURE INGRESS IN THE COMPOSITE
Abundant literature information and particularly contributions from the

Springer's group [11,12] facilitate quantification of the ageing process between
the start of exposure and the state corre!>pullding to the full saturation. Fickean
diffusion is assumed to describe the mechanism that governs water in epoxy
based composites and water concentration profiles are generated nsing now well
known equations available from dedicated monographs (e.g. [13]).
The scaling length of the diffusion process is the dimensionless quantity
DtJL2 where t is the time, L the specimen thickness and D is the diffusion
coefficient predetermining how quickly the saturation level w(oo) the so called
equilibrium solubility, is reached. As the D values for typical (60% volume fibre
loading) epoxy composites at the ambient temperature are round 2xlO-9 cm2/sec,
the time for reaching the complete saturation - even for a sheet 2 mm thick - are
in excess of four years. The temperature accelerates the transport and an
Arhenius type of dependence has been demonstrated to apply.
As other composite properties, also the diffusion coefficients are anisotropic
properties and the longitudinal and transverse characteristics correlate with the
volume fraction Vr of reinforcing fibres. Mass transport into glass fibre reinforced
epoxy (GRE) pipes is associated with ingress that is perpendicular to the fibres
and thus D should be inversely proportional to the square root of the volume
fraction. Concerning the D dependence on the matrix, the diffusion coefficient
should be a decreasing function of the cross-link density. A large experimental
scatter, lack of systematic studies and the tendency of chemical companies to
code rather than to declare actual chemical formulations are all the factors
responsible for not providing a more positive generalisation. D values have been
found to decrease with the molecular size of ingressing species and to be
invariant with the level of mechanical stresses.
Concentration profiles that develop in a composite with the time depend on
the boundary conditions. For a 'double-sided' diffusion into a slab of the
thickness L, with w(oo) being constant at x=±LI2 at all times and no water
present in the slab at t=O, water distribution in the slab and with the time is
given by:
304
w(x,t)
---'--C-~ = 1- -4~(-1)1_
L.. --cos
(2}+l)1lX
exp
{-(2)+ l)2;r2Dt}
6
w(oo) ;r j=O 2} + 1 2L 4L2
Thus an average water content in each ply can be derived, hygrothermal
effects on stiffness and strength characteristics evaluated and the failure
envelopes for the composite drawn for a particular time 1.
3. Computer programme
3.1 PROCEDURE, INPUT AND OUTPUT
The concept outlined above has been incorporated into a computer code that
allows each stage of composite ageing between two states at which all plies
behave identically: The reference state that corresponds to the unexposed (dry)
laminate at the ambient temperature at the time t=O and that associated with
the complete water saturation at the exposure temperature t=oo. Typically, eight
ageing stages are evaluated in equal incrt'rn~nts of log 1. The state of each ply is
quantified in terms of the avera.,_ water content and the corresponding decrease
in both the matrix stiffness due to a reduction in the glass transition temperature
and the lamina strength characteristics.
Extensive computational input that is required for each run falls into two
categories: 1) Data bank of constituent properties and 2) conditions for a
particular simulation. The former (l) contains constituent properties (elastic
constants Ei and Vi of both constituents, their expansion coefficients <X.j and ~i'
glass transition of the matrix in dry state and its sensitivity to the water content
(L\w· see eqn. 1), the temperature dependence ofD and the saturation level w(oo).
Furthermore, all the lamina strength characteristics and their dependence on T
and w (defined by exponents p in eqn. 5) need to be known. Concerning the
input under (2), the micromechanical calculations involving the lamina require
the volume fraction VF to be specified, the laminate construction is defined by the
stacking sequence, number of plies and their orientation, the boundary
conditions are given by the exposure temperature and humidity, the loading
regime identified by forces and/or moments together with their direction and,
finally, the type of output required.
The output, a sizeable map of stresses and strains that develop in the axial
and hoop directions of the pipe , is streamlined in the manner allowing three
important areas to be addressed: i) partition of the hygrothermal response with
respect to the temperature and the role played by the moisture, ii) the mutual
position of individual plies contained in the pipe wall at a fixed exposure time
and iii) investigation of the movement of an isolated ply, usually the weakest ply
in the wall laminate as a function of the exposure time. The ply failure concept
has been chosen not only because of its relevance to the design considerations
but also for the recognised advantages: Concise, global representation of the
laminate behaviour taking into account stress interaction in anisotropic solids
and their sensitivity to the lamina (stiffness and strength) characteristics.
3.2 ILLUSTRATION
GRE pipe is modelled as 4 mm thick structure consisting of eight plies stacked

305
symmetrically in the sequence {+57/-57/+57/-57}..,m. The elementary lamina is

defined to consist of an epo>..}' resin (Em=3.4 GPa, vm=O.35, a".=48xlO-6 KI,
Pm=O.064 (%rl, InDm=O.67-6200/[T+273])reinforced by 62% (v) glass fibres
(EF72 GPa, VFO.22, <w51O-6 KI, Pm=O (%r\ Hygrothermal ageing is defined
as a process taking part at 120°C with water vapour ingressing through both
laminate planes. The water content is assumed to increase from the initial zero
value to the saturated level of 1.6% (w). The composite pipe response is studied
in the form of ply failure envelopes in the stress domain (tensile quadrant with
both the axial and hoop stresses positive) relevant to the pipe operation.
3.2.1. The role of temperature and moisture effects

The exposure temperature of 393K can be demonstrated to lead to full saturation
within 6.4 months; approximately 16 years would be required if the water
ingress happened at the ambient temperature. In contrast to the mass transfer,
the heat transport can be stated, on the basis of thermal diffusivities, to be much
faster with the thermal equilibrium attained within 20 sees.
Concerning the position of ply failure envelopes, the equilibrium situation at
the times zero and infinity when all the plies behave identically is illustrated in
Fig. 1. The dry pipe at the ambient temperature (T=295K, w=O) is seen to be the
strongest structure. The increase in tempe ure only brings ·about an appreciable
reduction which is further amplified by saturating the laminate to the level
w= 1.6% at that temperature. The most significant shift is exhibited in the axial
direction where the matrix domination is more pronounced.
3.22 Behaviour ofindividual plies

In contrast to the equilibrium states, transient times are associated with each ply
behaving individually as their properties have been affected (reduced) by varying
levels of moisture. This is illustrated in Fig. 2 constructed for a short (tI=:: 1.5
days) exposure time when a steep water concentration develops across the
laminate. The failure envelope of the external ply containing the highest water
concentration is diminished more than the drier internal plies. As to the
principal stresses, plies exhibit pronounced differences in the axial direction and
practical invariance in the hoop direction.
3.23 Time dependence of the failure envelope of the weakest ply

Full failure envelopes for the weakest, i.e. the external ply of the symmetrical
laminate, generated for different exposure times are shown in Fig. 3 The loci
exhibit a rather complicated pattern, namely initial shrinkage of ellipses at very
low exposure times that is followed by their expansion towards the equilibrium
state as the time progresses. Thus the fully saturated laminate does not seem to
be the most vulnerable entity. Using the area contained in the tensile quadrant as
the benchmark, 80% of the pipe capacity to cope with combined tensile stresses
is lost within 2 days of the exposure. Having passed the minimum, the laminate
recuperates and this process enhances the operational window which eventually
reaches the plateau on 37010 of the original potential.
306
0.30,---------------------------,
+
+
0
x
.. 0
.. X
0
+
a. 0.15
o
o 0 0
+
r. X
+
0
0 X +
0
0 X 0
+
+
;0
0
0.00+-----~-~~~~-r_-~~-------_4
..;l~O 0
+
++
0.00 0.04 0.08

axial stress [GPa]
Figure 1. Axial and hoop stresses of the pipe plies at the beginning of the exposure (t=O,
T=295K, w=0 - x), subjected to the temperarure only (T=393K, w=0 - 0) and the
saturated state (t=oo, T=393K and w=1.6% (w) - D)
o.~._--------------------_.
+
+
0
..
X
..
0
0
+
x
a. 0.15
o
o 0 0 +
r. )(
0
+
0 x +
0
o )( 0 +
+
~o/~o +
++
0.00 +------~--+--_I_"__----__+------_____l
0.00 0.04 0.08 0.12
axial stress [GPaj
Figure 2. Failure envelopes at the selected transient time in the plies 1 (+),2 (0), 3(x)
and 4 (D) after moisture ingress for 1.5 days; exposure conditions: T=393K, w( 00)= 1.6%
307
0.30 . . . - - - - - - - - - - - - - - - - - - - - - - - ,
• • .0.0
+1.5
0
0 • • A3.0
x6.0
x
012.0
.. 0.20 f-
.
024.0
0 • • 048.0
C- 0 ... 96.0
o
o .. X
• .192.0
.r. 0.10
•
•
•
•
0.00 +---. ' - - - - + - - - - - - _ _ < 1 1 - - - - - - - - - 1 ••
0.00 0.04 0.08 0.12
axial stress [GPa]
Figure 3. Variation of the failure envelo. of the weakest ply with time;
Exposure times [days] given by the legend, exposure conditions as in Fig. 2
308
4. Conclusions
- The existing approach for predicting composite failure has been

extended by accommodating hygrothennal stresses, quantifying the rate of mass
transport and accounting for the effect of moisture on the stiffness and strength
characteristics of the lamina.
- The developed model is based on simplifying assumptions and requires
a sizeable input for a computer programme that evaluates the hygrothennal
ageing process.
- Computational procedure has been illustrat~ by analysing
hygrothermal ageing of GRE pipes in terms of temperature/moisture effects,
behaviour of individual plies and the time dependence of the weakest ply.
References
1) Frost, S.R and Cervenka, A. (1994) Glass fibre-reinforced epoxy matrix

filament-wound pipes for use in the oil industry, Composite Manufacturing
5, 73-81.
2) Ellis, T.S. and Karasz F.E. (1984) Interaction of epoxy resins with water:
The depression of glass transition ter....,..:rature, Polymer 25, 664.
3) Bicerano, J (1993) Prediction of polymer properties, Marcel Dekker, Inc.,
Basel.
4) Micromechanics Model for Composite Materials, Centre for Composite
Materials, University of Delaware, Report CCM 87-29.
5) Cervenka, A.J., Bannister, D.l and Young, Rl Moisture absorption and
interfacial failure in aramid/epoxy composites, Composites, Part A -
accepted for publication.
6) Jones, RM. (1975) Mechanics of Composite Materials, Scripta Book, Co.,
London.
7) Raghawa, RS. (1988) Thermal expansion of organic and inorganic matrix
composites. A review of theoretical and experimental studies, Polymer
Composi tes 9, 1-11.
8) Adams, D.F. (1987) Environmental Effects on Composite Materials,
Technomic Publishing Co. (Handout for the Seminar held in Basel on
October 12-14th 1987.
9) Raghawa, RS. (1984) Polymer Composites 5,173 ..
10) Gadke, M (1993) Hygrothermomechanisches Verhalten Kohlen-
Stoffaserverstarkter Epoxidharze, Fortschritt-berichte VDI, Reihe 5: Grund-
und Werkstoffe, VDI Verlag, Dusseldorf.
11) Tsai, S.W. (1988) Composites Design, Think Composites, 4th Edition,
Dayton, Chapter 16.
12) Springer, G.S. (1981) Environmental effects in composites, Volume 1,
Technomic Publishing Co., Westport.
13) Springer, G.S. (1984) Environmental effects in composites, Volunle 2,
Technomic Publishing Co., Westport.
14) Crank, J (1956) The mathematics of diffusion, Oxford Univ. Press, London.
COMPOSITE PIPES BASED ON THERMOPLASTIC MATRICES
REfflFORCEDBYCONTINUOUSnBRES
A. CERVENKA
University ofManchester and UMISF, Grosvenor Street,
Manchester, MI 7HS, United Kingdom
Abstract
Experimental results have been obtained at the ambient temperature for

thermoplastic composite pipes fabricated by filament winding of thermoplastic
laminae on thermoplastic mandrels. A thermoplastic lamina (tape) was
manufactured by pultrusion involving 50010 (w) of a reinforcement. Two
composite types have been investigated: polyvinylidenef1uoride reinforced with
carbon fibres and polypropylene reinforced with E-giass fibres
A procedure for predicting the strain/stress response of thermoplastic
composites up to the ultimate failure has been developed. It is based on
recognised micro- and macromechanical modelling extended in such a way that
strong non-linear response due to matrix viscoelasticity is accounted for by
assuming strain softening of a thermoplastic matrix. The model has been found
useful to quantify the stiffness characteristics of a given pipe and to estimate its
ultimate failure stress.
Experimental data have been found to correlate well with the theory.
1. Introduction
Although composite pipes based on thermosets are widely used for transport of
pressurised fluids, composite pipes that would contain a thermoplastic matrix
might offer a number of advantages: no weepage due to matrix cracking prior
the ultimate failure, higher impact resistance, more rational jointing, elimination
of the reactive processing during manufacturing, longer reelable pipe segments
and a possibility to recycle. Regretfully, thermoplastic composite pipes are
commercially not available, their mechanical behaviour has not been evaluated
and thus theirPotential remains not fully exploited.
The presented contribution reports on a pioneering study that describes two
composite systems that were used for the pipe construction. It further deals with
evaluation of short-term properties of fabricated pipes and establishes a model
that allows experimental data to be predicted theoretically. The modelling
aspects concentration three distinct regimes of the pipe response, namely: i) the
initial slope of the stress/strain (cr/f:) of the pipe associated with low internal
309
310
pressures, ii) non-linearity in the cr/E dependence in the region of medium and
high pressures and iii) the ultimate failure crult of the pipe corresponding to the
burst pressure.
2. Experimental
The study concentrates on two matrix/fibre combinations: polvinylidenefluoride

reinforced with TJoo carbon fibre (PVDF/c) and polypropylene reinforced with
E-glass (pP/g). They can be considered to be material antipodes in the
costIperformance plane: The commodity PP/g system likely to satisfy
undemanding (low temperature) application and the PVDF/c speciality version
containing expensive, well performing (temperature, chemical resistance)
constituents.
All the materials, including PP and PVDF pipes used as mandrels were
provided by 'Laboratoire Central' of Solvay, Brussels. The Young's modulus and
the yield point characteristics of the matrices were 2.6 GPa, 55 MPa and 9% for
PVDF, 1.7 GPa, 36 MPa and 10% for PP.
The prepreg tapes were obtained by the pultrusion technique using the
technology proprietary to FLEX Line of lJ"Ordwijkerhout, The Netherlands. In
both cases, tapes containing 50% (w) of a reinforcement were 6 mm wide.
The pipes were manufactured by the University of Kaiserslautern, Germany
using their four axis winding machine [1] with a full computer control of
deposition of the carefully tensioned tape on the mandrel (internal diameter 110
mm, wall thickness 3.5 mm) at the ply angle a=57° and an excellent
consolidation (open flame) of the whole pipe. With respect to the testing
facilities available for the pipe testing (and bursting) the thicknesses of the
composite overwraps were 0.8 mm for PVDF/c and 2 mm for PP/g. Two pipes,
each 6 m long were fabricated for either composite system. After being cut into
I m long section, 12 specimens for each composite type were made available for
testing.
Testing of instrumented (strain gauges, LVDT) pipes was carried in the
Shell Laboratory Amsterdam. Pipes were pressurised at the ambient temperature
without axial restrain (thus the hoop stress exerted only) incrementing the
internal pressure in approximately 20 s steps. Some initial experimental
problems (retraction of cylindrical specimens from the closing plungers,
adhesion of iltrain gauges to thermoplastics) have been successfully overcome.
Experimentation with the PP/g pipes has been found to be more tnd>l~
than that with the PVDF/c composite. At least two sets of data relating the
internal pressure to the. hoop (Ey) and axial (Ex) strains have been generated. The
typical expeDmental4ata{discrete points) are shown in Fig. I for the PvDF/c
pipe.~J).d Fig. 2 for thel'P/g composite system.
311
3. Theoretical
3.1 CONSTITUENTS AND THEIR CHARACTERISTICS
A thennoplastic composite pipe produced by the filament winding process

consists of two building blocks: the mandrel (wall thickness 1m, internal radius ri)
and the load-bearing composite tape (thickness t.,) wound under a winding angle
a with respect to the axial direction of the pipe and restraining the mandreL The
extruded mandrel is based on one constituent only, i.e. a viscoelastic polymer of
a non-linear stress/strain response associated with parameters that are relatable
to the initial stiffness and the co-ordinates of the yield point. The restraining
tape is a two component material: In addition to the polymeric matrix identical
in its nature to that in the mandrel, the tape contains reinforcing fibres. Their
volume fraction is Vf. Alignment of the fibres is achieved by means of the
pultrusion process. Both constituents of the tape (lamina) are assumed to be
characterised by their own elastic constants and strength characteristics such as
the Young's modulus of a matrix Em, ultimate tensile strength of the fibre afu
and the yield stress of the matrix. amy.
Transversal isotropy of the lamina (pultruded tape) requires more stiffness
and strength characteristics to be use' These are: The longitudinal and
transverse moduli EI and E 2, the shear modulus G I2 and Poisson ratio Vl2 in the
lamina plane, the longitudinal and transverse strength in tension X and Y, and
the shear strength S. The stiffness characteristics can be either measured on
unidirectional laminates or calculated micromechanically using a computer code
[2]. The exact way of determining the strength characteristics is to utilise
specimens from sheets manufactured under the same conditions as used for the
pipe (filament winding of unidirectional race-track specimens). Alternatively,
their upper bound can be roughly assessed by means of empirical rules digested
from the behaviour [3,4] of polymeric composites. This pragmatic approach has
been used in the presented study and the highest possible values have been
assigned to both laminae: X=Vf*afu, Y=amy and S=O.03*G I2 .
3.2 ANALYSIS OF THE COMPOSITE PIPE
The pipe behaviour can be analysed taking into account the manner of its
construction. The construction is specified by the internal radius, thicknesses of
the mandrel and the composite re-wound and the winding angle. The analysis is
easily accomplished by using one of several computer codes [5] based on the
'Classical Laminate Theory'. Constructing a symmetrical laminate, the haJf-
stack is built of a single inner ply of the thickness 1m consisting of the isotropic
mandrel matrix and two outer plies, each tJ2 thick, arranged in the +aJ-a angle-
plied construction. Defining either a loading spectrum or constraints, the
principal stresses that develop in the pipe are also derived, namely the axial (ax),
hoop (ay) and the interlaminar shear (axy) stresses. A failure criterion (Tsai-Hill
312
in the case of [5]) is then used to ascertain whether the composite pipe remains
intact or has been breached. The stress level relevant to the type of restrainment
is then readily related to the ultimate pressure Pull sustainable by the cylindrical
structure:
p = CF c(t", + te> (1)

alt
r,. + t '" + t c
3.2.1 Stiffness characteristics ofthe pipe
The construction details in combination with the material properties yield
stiffness characteristics of the pipe: the longitudinal (EJ, transverse (Ey) and
shear (Gxy) moduli together with the Poisson ratiovxy.
3.2.2 Non-linear stress/strain behaviour

The non-linear stress/strain dependence can be quantified considering I) the
strain-softening of the matrix and 2) stress partitioning between the matrix and
the fibre. The former (under IJ IS associated with the viscoelastic nature of a
thermoplastic matrix and the fact that its modulus is strain dependent Em(e).
Postulating that stress Om in a matrix can ....; related to the matrix strain Em by a
three-parameter relationship of the type
As",
CF",= 2 W
l+Bs", +CS",
with the three parameters being simple functions of the matrix modulus Em and
the co-ordinates Emy and Omy of the yield point associated with Em(Emy)=O:
A=E",(O)
B=~-2JC (3)
CFmy
C=S-2
my
the dependence of the modulus of a matrix on the strain reads:
Concerning the latter (under 2), the presence of much stiffer fibres causes
that the level of stress exerted on a matrix as such, both in the liner and the
composite tape, will be lower than that associated with the internal pressure in a
composite pipe. Bearing in mind that the major stress component during pipe
pressurisation is the hoop stress 0y, we assume that the partition factor <l>
translating the overall composite stress o. (see eqn. I) to that of the matrix is the
ratio E/Em(e) at a given pressure. Thus the composite pipe during pressurisation
is assumed to find itself in many different states j= 1,2 ... n, each governed by its
313
own matrix stiffness EmU) , a specific set of the lamina characteristics E\U),
E2U), G l2 U), V12U) , pipe stiffness values E,U), EyU), GxyU), v,,,U) and
consequently, stress partition factors <1>U). Decrementing the matrix modulus, 3
repetitive algorithm generates the non-linear behaviour step by step for each
stage j.
3.2.3 The ultimate stress

As the internal pressure increases, a thermoplastic composite pipe passes
through different stages. However, the actual state at which the composite pipe
fails is not known. Thus when decrementing the matrix modulus, the pipe
failures are derived for all composite states keeping the strength characteristics
of the lamina and the mandrel constant. The repetitive algorithm yields however
a relatively narrow spectrum of failure stresses. Higher failure stresses are
predicted when j is small and the failure mechanism involves both building
blocks - restraining tape and the mandrel.
4. Results and discussion
Composite constituents used and their relative concentrations in fabricated

laminae can be micromechanically converted [2] to the lamina characteristics.
These, in terms of both the stiffness and strength, are given in Table I for the
two composites studied:
TABLE 1. Lamina characteristics of the two composite systems
EJ E2 G12 V12 X Y S
[GPaj [GPaj [GPa] [MPaj [MPa] [MPa]
PVDF/c 118.5 7.67 4.09 0.269 1685 55 120
PP/g 21.73 3.35 1.32 0.3 966 36 40
The higher stiffness and strength values associated with PVDF/c composite
is due to the presence of the stiffer fibre and its higher concentration.
Taking further into account the manner of constructing the pipes, the
stiffness characteristics of the pipe can be derived [4]. The calculated stiffness
characteristics are given in Table 2.
The PVDF/c composite pipe is seen to maintain its higher stiffness but the
angle-plied construction of the overwrap and its more significant thickness in the
case of the PPlg type cause the pipes not being so dramatically different in their
stiffnesses.
TABLE 2. Stiffness characteristics of the pipes
Ex [GPa] Ey [GPa] Gxy [GPa] V xy

PVDF/c 3.83 9.27 5.56 0.37
PP/g 2.17 4.34 2.26 0.37
314
Modelling of the stress/strain behaviour of composite pipes is accomplished

in Figs. 1 and 2. Fig. 1 summarises the situation for the PVDF/c pipe and Fig. 2
provides the same information for the PP/g system. Both deserve a further
discussion that concerns their initial slopes, aIr:. non-linearity and the ultimate
stress
Axial and hoop stiffness values of the composite pipes can be determined
from the initial slopes of the measured aIr:. dependencies as W !lax = -vxylE,.
and !lr:..J !lay = llEy• Relating the experimental values to the predictions (Table
2), the relative errors (predicted values used as the reference) are given in Table
3. Stressing that the Ex and Ey characteristics have been predicted from the '1st
principle', the agreement can be qualified as highly satisfactory.
Turning the attention to the 0/& non-linearity, strain induced softening of a
thermoplastic matrix causes changes in the Engineering constants of the
composite lamina which are further responsible for changes in the stiffness
values of a pipe. Fig. 3 illustrates this variation for the PVDF/c composite.
Whilst the fibre dominated properties such the Poisson ratio and the axial
stiffness of the pipe are seen to be practically invariant with E.,.(a), the matrix
dominated characteristics - the hoop stiffness and the shear modulus - fall
rapidly with decreasing stiffness of a matrix.
TABLE 3. Relative errors between experimental and predicted stiffness values

for the pipes
Ex Ey
PVDF/c -9.7 -5.3
PP/g -4.0 +16.4
With the lamina strength characteristics constants as given in Table 1, the

information contained in Fig. 3 is subsequently used as the input for predicting
the interval of failure stresses. This is, again for the PVDF/g system, illustrated
in Fig. 4 that allows the most probable failure stress to be estimated as 170 MPa
with an accuracy of approximately 7010. The predicted range agrees extremely
well with the observation. A similar pleasing relationship between the
experimental and theoretically derived data has been also obtained for the PP/g
composite pipe as demonstrated in Fig. 2.
5. Conclusions
- Short term properties of thermoplastic pipes have been determined

experimentally at the ambient temperature and modelled as the strain/stress
response of composite pipes from the initial linear elastic response to the
ultimate failure (burst pressure).
-Modelling has been based on the existing, however extended,
micromechanical and macromechanical algorithms. Two extensions have been
introduced, namely I) strain-sofiening of a thermoplastic matrix and ii) .partition
315
of the overall stress with respect to stiffnesses of the structural members - the
mandrel and the restraining tape - sharing the exerted load.
-Correlation of the experimental data with the theoretical analysis has
been successful; the agreement between predicted and determined values can be
qualified as highly satisfactory: The model yields a unique strain/stress traces,
the ultimate failure stresses are predicted as an acceptably (±'JOIo) narrow
interval.
- Two salient conclusions can be further drawn for thermoplastic
composite pipes: 1) They are structures of a high quality and 2) a pattern of their
generic behaviour is emerging.
316
3.0
2.5
2.0
1.5
#.
1.0
c 0.5
I'll 0.0
-D.5
tJj
-1.0
-1.5
-2.0
-2.5
-3.0
0 50 100 150 200
stress [MPa]
Figure 1. Correlation of experimental data (two different symbols used for independent
measurements) with theoretical predictions (lines) for the system PVDF/carbon fibre; the
arrow indicates ultimate ailure
4
o
3 o
o
2
#.
c
I'll 0
tJj -1
-2
-3
-4
0 20 40 60 80 100 120
stress [MPa]
Figure 2. Correlation of experimental data (three different symbols used for independent
measurements with theoretical predictions (lines) for the system PP/glass fibre; the arrow
indicates ultimate failure
317
0.4 0.5 0.6 0.7 0.8 0.9 1.0

normalised stiffness matrix
Figure 3. Changes in the stiffness characteristics of the PVDF/carbon fibre composite

pipe due to softening of the matrix during pressurisation
••..•••••••• E> ••••

180 o
o
o
o
175
o
o
o
o 170 MPa +1- 6.3 to
170
o
o
165 o
o
o
160 .... -0- ........... .
0.4 0.5 0.6 0.7 0.8 0.9 1.0

normalised stiffness matrix
Figure 4. Variation of the failure stresses [MPa] for PVDF/carbon fibre composite pipe
with the matrix stiffness
318
References
1. Haupert, F. and Friedrich, K. Processing related consolidation of high

speed filament wound continuous fibre/thermoplastic composite rings
(paper submitted to Composite Manufacturing).
2. Gillespie Jr., IW. and McCullough R.L. (1989) SMC: Micromechanical
model for composite materials, Report No. CCM-87-279, Center for
Composite Materials, University of Delaware, Newark.
3. Cogswell. F.N. (1992) Thermoplastic aromatic polymer composites.
Butterworth and Heinemann, London.
4. Tsai, S.W. (1986) Composite design. Think Composites, Dayton, Ohio.
5. Laminate Analysis Program (LAP). (1995) Imperial College of Science,
Technology and Medicine and Anaglyph Ltd, London.
IMPACT DAMAGE OF COMPOSITE STRUCTURES
M. DE FREITAS and L. REIS

Departamento de Engenharia Mecaruca, Instituto Superior Tecnico
Av. Rovisco Pais, 1096 Lisboa, Portugal
Abstract Damage due to out-of-plane loading in composite materials, by impact

ballistic or low velocity loading, is potentially dangerous since damage may be left
undetected and yields a considerable reduction in compressive residual strength. In
this paper, a review of the experimental methods used to characterize the sensitivity
of composite materials to low velocity impact loading is presented. A drop-weight
impact machine, an ultra-sonic C-scan device and a fully instrumented compression
after impact testing fixture were used. The effect of loading conditions during impact,
impact energy, impact angle, pre-loading and the effect of temperature in compression
after impact are studied.
1. Introduction
The utilization of generalized composite materials is increasing on a very wide range

of structures and mechanical components due to the advantages on thermal properties
and weight reduction. High performant composite materials have additional
advantages due to the possibility of an increased strength according to the stacking
sequences used on design, yielding a very high stiffness and strength in the loading
directions.
Nevertheless, laminated composite materials and sandwich structures, due to the
stacking sequences of lamina, have, nevertheless a very low stiffness and strength
through the thickness direction, when compared with the in-plane properties, since no
fibers may be present through the thickness. Attempts have been made to overcome
319
320
these difficulties with the so-called stitched composites but a great reduction of the in-
plane strength is obtained. Therefore, the sensitivity of laminated materials to out-of-
plane loading is of prime importance and is being studied with different theoretical
and experimental approaches [1, 4].
Damage due to out-of-plane loading, such as impact loading, may be caused either
by fiber failure or matrix cracking for very high velocity impact loading, such as
ballistic impact, but one of the most damaging situations is due to low velocity out-of-
plane impact loading which causes delamination on the opposite side of the impact
[1]. The strength reduction caused by the delamination is much higher in compression
than in tension loading and therefore the classic residual strength characterization is
the compression after impact test [2, 3]. These low velocity impact tests, impact and
compression after impact have been performed by aircraft industries and some
industrial standards are used such as Boeing and Airbus, using drop weight impact
tests on clamped specimens and damage is evaluated either by visibility of impact
damage or with delamination detection hy any NDT device, with residual strength
determined through Compression After Impact (CAl) tests. Nevertheless, the
conditions used during drop weight tests may affect the amount of damage, as well as
the environment conditions during compression after impact [5].
The objective of the present paper is to review the author's experience on
performing impact damage tests and about the influence of the conditions during
impact and compression after impact on damage growth.
2. Experimental Procedure
2.1. SPECIMEN AND MATERIAL
The specimen chosen in this study is the standardized one adopted by aircraft
industries (AlTM Airbus Industries Testing Method, 1.0010), a rectangular specimen
150x 100mm2, made of unidirectional prepregs of carbon fibers IM7 on an epoxy
resin 977-2 from Fiberite Co., with a nominal thickness of 0.135 mm for 60% volume
fiber (Vf) and a stacking sequence of32 plies [(45/0/-45/90)4ls-
The elastic properties of the laminae are listed in Table I, where directions 1 and 2
are respectively the longitudinal and transverse direction of each laminae.
321
TABLE 1 - Elastic properties of IM7/977-2

Material El E2 012 013 023 Vl2
(OPal (OPal (OPal (OPal (OPal
IM7/977-2 160.0 9.0 6.9 3.8 4.3 0.20
2.2. IMPACT TESTING
An instrumented falling weight impact machine was used, where the different norninal
impact energies were obtained through discrete masses of 2 to 10 Kg and a variable
height up to 2 m.
The instrumentation associated with the impact machine, measures the force and
time during impact. Force is measured discretely in time, through a force transducer
just above the impactor indentor. The system also provides for the measurement of
the velocity at the moment of impact, by measuring the time a flag of known width
takes to traverse a point. The velocit" ':luring impact is calculated by integrating
acceleration over time, where the acceleration is given by the force felt by the drop
mass, divided by its mass. The calculated velocity is then used to derive the
displacement of the impactor during impact (and consequently the deflection of the
specimen) using further integration. The energy is calculated by integrating force over
distance. With this measurement and calculation procedure, not only the actual
impact energy but also the absorbed energy by the specimen were calculated through
the energy/time curve.
The impactor used was a herni-spherical one made of steel with 16 mm diameter.
For the case of unloaded specimens, they were clamped in a standard impact testing
fixture with an open window of l25x75 mm2, according to Airbus Industries Test
Method (AlTM l.001O).
For the case of loaded specimens, they were placed in the Compression After
Impact testing fixture, replacing the previous impact testing fixture. The CAl testing
fixture was placed in a specially designed small testing frame equipped with a
hydraulic actuator, loading the CAl fixture described in the next section. The
instrumentation of the impact machine is still used during impact testing so plots of
Force, Velocity and Displacement are also available. Since the specimen is only pre-
loaded and guided (not clamped as in the previous testing fixture), the vibration of the
specimen during impact leads to a much more noisy acquired signal.
322
2.3. DAMAGE DETECTION
After testing, the specimen was placed in a plane surface and the indentation of depth
was measured through a dial gauge with 0.01 mm of accuracy and using the
procedure described in AlTM 1.0010. One of the objectives of this procedure is the
determination of the Barely Visible Impact Damage, i. e. the indentation depth
corresponding to 0.3 mm for composites made of unidirectional prepregs or 0.5 mm
for fabric.
For the evaluation of damage introduced by the Impact and Compression After
Impact tests, a non destructive testing method based on an ultrasonic device was
used, the so-called B and C-scan of the specimens. A Physical Acoustics equipment
was used, with a pulse/receiver transducer and an immersion scanning method, with
the following C-scan data:
- Scan frequency: 5 MHz
- Scan speed (X axis): 30 mm1sec
- Index axis : Yaxis
- Scan interval index: 0.05 mm
The software available, automatically stores images of B and C-scan either
measuring the Amplitude (AMP) of the defect signal (file. AMP) or measuring the
Time Of Flight (TOF) between the interface (entrance signal surface) and the defect
signal (file.TOF). The measurement of the delaminated area was carried out through a
home made image processing software, from the files of the C-scan in Time Of Flight.
2.4. COMPRESSION AFTER IMPACT TESTS
Compression after impact tests of the specimens were carried out on an universal
testing machine, in displacement control with a velocity of 1 mm1min, using the
testing fixture described in the Airbus testing procedure (AlTM 1.0010). In this
fixture and to prevent global buckling during compression, the specimens are clamped
at both ends and simply supported at the side edges. In order to establish the buckling
behavior of the delaminated area during testing up to final failure, two LVDT were
attached to the specimen: one at the center of the impact zone and the second one at
the opposite side. A third LVDT was used to monitor the vertical displacement of the
323
specimen during the test, in order to obtain, with the load cell data, the
load/displacement curves. Four strain gauges were bonded to both sides of the
specimens, at a distance far from the delaminated area and strains during testing were
recorded with a MUltipoint Measuring Unit. A schematic view of the instrumented
Compression After Impact (CAl) testing fixture with a specimen mounted and the
three LVDT in testing position, can be seen in references [6, 7]. In order to study the
effect of temperature on residual strength of impacted specimens, CAl tests were
carried out at room temperature (RT), -SO°C and 70°C. On low (-SO°C) and high
(70°C) temperature Compression After Impact tests, the testing fixture with the
specimens was placed inside a temperature chamber (temperature range from -ISO°C
to +3S0°C), coupled with the testing machine.
3. Results and Discussion
3.1. IMPACT TESTS
A complete series of tests for a wide range of impact energies around Barely Visible
Impact Damage energy up to Perforation of specimens energy levels were carried out.
The objective was to create a wide range of delaminated areas, from BVID to
perforation energy levels.
80
70
5: 60
~ 50
. ..
.."
~ 40
..
GO
...
i 30
~ 20
10
0
0 2 3 4 5
Indentation deph (mm)
Figure 1 - Preliminary tests for BVID and Perforation energy levels detennination
324
Figure I shows the preliminary results obtained for unloaded specimens where the
indentation depth is plotted against the absorbed energy, calculated through the
instrumentation of the impact machine.
The obtained BVID for this material and lay-up, is 30' J of impact energy (20.8 J of
absorbed energy) and Perforation is obtained at 78.7 J of impact energy. The
delaminated area has a circular shape since we have a quasi-isotropic lay-up, with the
delamination at the interfaces growing in area from the impact side to the opposite
side.
Additional tests were carried out in order to determine the influence of the impact
angle, i.e., the angle between the directions of impact loading and the normal to the
specimen plane. Three angles were chosen, respectively, 15°, 30° and 45° and two
energy levels, BVID and Perforation energies, determined on the unloaded specimens
for an angle of 0°, perpendicular impact tests. The results of the delaminated areas
obtained as a function of the impact angle, for both energies are shown in figure 2.
The delamination area is reduced considerably when the impact angle increases and
for an impact angle of 45° no perforation was obtained.
• BVID-15 • BVID-30 ~ BVID-45 [J PERFO-15 <> PERFO-30 A PERFO-4
4000 [J
[J
N 3000
iii
.!! <>
$"'"
FI 2000
•
.~ <>
.a
~., 1000 •
A ••
0
0 15 30 45
Angle CO)
Figure 2 - Delamination areas as a function of impact angle

325
As mentioned before, some low velocity impact tests were carried out with pre-
loaded specimens but with a different experimental procedure, concerning the
specimen clamping, and therefore the results must be analyzed carefully. The same
methodology was used, as for the unloaded specimens, i. e. a wide range of impact
energies were applied from BVID up to Perforation energies level. The impact tests
of pre-loaded specimens were carried out, according to the described procedure with
a compressive pre-strain of -0.21%, considered as a design data for this material.
Despite the limited results available, some results may be present. Figure 3 shows the
comparison between the delaminated areas obtained with impact tests of unloaded
and pre-loaded specimens. For this material and pre-load, no significant differences
were obtained, but notice that for higher pre-loads, an immediate compression failure
may be present, depending of the delaminated area, compressive pre-load and residual
strength of the material.
I.·
90
80 • unloadec
• • <>
70 <> loaded • <> •
5: 60
~ 50
.11..,
iF'4 40
.s 30
20 ••
10
0
0 1000 2000 3000 4000 5000
Delaminated area (mm2)
Figure 3 - Effect of a compressive pre-load on the delaminated area
3.2. COMPRESSION AFTER IMPACT TESTS
After impact testing and damage measurement, either by indentation depth or

delaminated area, the specimens were tested in compression in order to determine the
326
residual strength, as a function of the delaminated area and testing temperature. The
damage growth during compression, corresponds to a typical buckling failure where
the damage growth appears suddenly leading to complete failure of the specimen. The
CAl testing fixture, with the associated instrumentation allows the establishment of
the failure mechanisms of the specimen due to the buckling of the individual
delaminated laminae or global buckling of the specimen. These results were presented
elsewhere. In this paper we will present only the general results concerning the
residual strength.
• BVID-15 [J PERFO-15 • BVlD-30 o PERFO-30 • BVID-45 A PERFO-45
200
~
150 •
C
....
i
!; 100
• A·A
~
'" • 0
•0 [J [J
'"'"
j):l 50
0
0 500 1000 1500 2000 2500 3000 3500 4000
Figure 4 - Influence of impact angle on compressive residual strength
Figure 4 and 5 show the residual strength obtained for the unloaded impacted
specimens, respectively non-perpendicular impact angle tested at room temperature
(23 0 C, 50% HR) and perpendicular impact angle at three temperatures. A direct
correlation is observed between the residual strength and the delamination area. Note
that the delaminated area is associated with the absorbed energy but may be
dependent of the stacking sequence used in the design of the composite materials. The
specimens impacted at increased impact angles, have a reduced delaminated area, so it
327
is expected to obtain increased residual strength, despite that the delaminated area is
now dissymetrical for higher impact angles.
The effect of temperature on residual strength, may be observed in figure 5, where
the residual strength of several specimens are presented for CAl tests at the three
temperatures on consideration.
4000
•
<>
• • -50°C
3500 <> RT
;(j' <> • +70°C
.!!.
.~
~
• <>i ~
t: 3000
~ • <>
~ ~
~
J.&.
2500
~
2000
1500 2000 2500 3000 3500 4000 4500 5000
Figure 5 - Effect of temperature on residual strength of delaminated composites
For the same energy level, between BVID and Perforation, an increase of the residual
strength measured either in strain to failure or stress to failure, is obtained with a
lower temperature and a decrease of the residual strength for higher temperatures of
CAI testing. For example, for an impact energy level of BVID, the residual strength
decreases 50% for a testing temperature of -50°C, 55% for RT. and 60% for +70°C,
when compared with intact specimens.
No significant additional residual strength was obtained in CAI, when the
specimens were impacted with the compressive pre-load. This result was somewhere
expected, since for this material and pre-load no significative increase or decrease in
the delaminated area was observed, confirming that the residual strength is mainly a
function of the delaminated area.
328
4. Conclusions
Low velocity impact and compression after impact tests of a quasi-isotropic laminate
composite panels were carried out and allow the following conclusions:
- the delaminated area due to low velocity impact is a function of the impact
energy and of the angle of the impacting load.
- unstable damage growth, was obtained by compression after impact due to a
buckling mechanism of the laminae in the delaminated area.
- the residual strength is influenced by the delaminated area which is a function of
the impact energy, and depends of the testing temperature, increased temperatures
yielding lower residual strength.
5. Acknowledgments
The authors wish to thank to Min. Defesa Nacional and OGMA, SA which have
sponsored this research through the EUCLID Program, RTP 3.1, Aeronautical
Applications Technology, Impact and Damage Tolerance.
6. References
l. Madan, R.C. (1991) Influence oflow-velocity impact on composite structures,

Composites Materials: Fatigue and Fracture (Third Volume). ASTM STP
1110, T.K. O'Brien, Ed., 457-475.
2. Dost, E.F., llcewicz, L.B. and Coxon, B.R. (1991) Effects of stacking sequence
on impact damage resistance and residual strength for quasi-isotropic laminates
Composites Materials: Fatigue and Fracture (Third Volume). ASTM STP
1110, T.K. O'Brien, Ed., 476-500.
3. Delfone, D., Poursartip, A., Coxon, B.R. and Dost, E.F. (1995) Non-
penetrating impact behavior of CFRP at low and intermediate velocities,
Composites Materials: Fatigue and Fracture (Fifth Volume). ASTM STP
1230, R.H. Martin, Ed., 333-350.
329
4. Soutis C. and Curtis, PT (1995) Prediction of the post-impact compressive

strength of CFRP laminated composites, Composite Science and Technology,
Vol. 56,677-684.
5. Ishikawa, T., Sugimoto, S., Matsushima, M, Hayashi, Y. (1995) Some
experimental findings in compression-after-impact (CAI) tests of CFIPEEK
(APC-2) and conventional CFIEPOXY flat plates, Composites Science and
Technology, Vol. 55, 349-363.
6. Reis, L. and Freitas, M. (1997) Damage growth analysis of low velocity
impacted composite panels, Composite Structures, 38, 1-4,509-515.
7. Freitas, M. and Reis, L. (1998) Failure mechanisms on composite specimens
subjected to compression after impact, Composite Structures (in press).
REMARKS ON LINEAR AND NONLINEAR
VARIATIONAL DESIGN SENSITIVITY ANALYSIS
F.-J. BARTHOLD
Institute for Structural Mechanics and
Computational Mechanics, University of Hannover,
AppelstrajJe 9A, D-30167 Hannover, Germany
Abstract. Theoretical formulations and numerical methods for variational

design sensitivity analysis have been investigated since almost two decades
leading to two major formulations, i.e. the material derivative approach, see
e.g. [1], and the domain parametrization approach, see e.g. [2]. Recently, the
author described a third • iewpoint which takes elements from both fore-
mentioned approaches and yields an advanced presentation of variational
design sensitivity analysis within -- -,nlinear computational mechanics [5].
An integrated formulation of direct and design sensitivity analysis can
be obtained by tracing and separating geometrical and physical mappings.
This approach is based on a differential geometry setting of continuum me-
chanics using convected coordinates. The fundamental ideas are outlined.
Furthermore, a nonlinear elastic, isotropic damage law as well as the
subsequent sensitivity analysis are described in order to demonstrate the
efficiency of the proposed theoretical and computational formulations.
1. Continuum Mechanical Background

1.1. CONFIGURATIONS AND MAPPINGS
In a differential geometry setting of continuum mechanics both undeformed

and deformed bodies, i.e. no andnt, respectively, are considered to be
differentiable manifolds embedded into the 3-dimensional euclidean space
1E3 , see e.g. [6]. The different spaces are equipped with the cartesian base
vectors {EI' E2, E3} and {el' e2, e3} in case of no andnt, respectively, and
{Zl' Z2, Z3} in case of the parameter space Te, see figure 1.
331
332
\_~--L!L
e'1~/
81
Figure 1. Manifolds and mappings
For any point X E no there exists an open set Uo (X) c no and a

mappping a o such that Uo is mapped onto an open set le,o c IR3. Similarly,
for any point x E nt there exists an open set Ut(x) c nt and a mapping at
such that Ut is mapped onto an open set le,t C IR3. The Coo mappings a o
and at are invertible. Without loss of generality, we consider those mappings
with identical parameter sets, i.e. where Ie := le,o = le,t. Overall, a finite
number of sets and mappings, i.e. charts, is necessary to cover each body.
These charts are collected to be an atlas of the body.
Conversely, any atlas is seen to be a definition of a body, i.e. the bodies
;p
can be defined locally by the inverse mappings := a;;-1 and ip := a t 1 ,
respectively. Thus, we conclude that the undeformed body no is locally
defined by so-called (local) geometry mappings
(1)
and nt is locally defined by so-called (local) deformation mappings
(2)
For simplicity, we exclude indices to count the charts and furthermore con-
centrate on an atlas with a single chart.
The well known deformation mapping tp between undefor!lled and de-
formed body can be described by composing the mappings t/J and ip, i.e.
tp: no f-7 nt is defined by x = tp(X) := (ip 0 t/J )(9).
- -1
333
Furthermore, both mappings are parametrized either by time t in case
of the deformation mapping or by scalar valued design variables Si in case of
the geometry mapping. This similarity has been used for sensitivity analysis
within the material derivative approach.
This basic viewpoint yields a representation of continuum mechanics
which is fundamentally based on the convected coordinates 8 E Te and on
two independent m~pings with identical mathematical structure, i.e. the
geometry mapping tP and the deformation mapping ip, respectively.
Furthermore, a displacement mapping iI : Te M 1E3 can be defined by
u = iI(8) := ip(8) - tj,(8) = x-X. The structure of the embedding space
1E3 is necessary for this definition as well as for all subsequent variations.
1.2. GRADIENT OPERATORS AND STRAIN MEASUREMENTS
Three different gradient operators, i.e. grad x , Gradx and GRADe, w.r.t.
e
the independent variables x E nt, X E no and E Te can be defined and
applied to any function a = a(x) = a(X) = 0(9). Of special interest are
the local gradients of the geometI., and the deformation mapping, i.e.
Ke := GRADe tj, = Gi (8) Zi and Fe:= GRADe ip = gi (8) Zi, (3)
where Gi = atj,(8)/a8 i and gi = aip(8)/a8i denote the curvilinear base

vectors. Every tangent mapping can be decomposed into tangents w.r.t.
convected coordinates, e.g. the material deformation gradient is given by
Two length measurements,l i.e. metrices of the different tangent spaces, are
defined by the mappings tP and ip, respectively. The corresponding metric
tensors can be derived using Ke and Fe, i.e.
T
M,p := Ke Ke = Gij zt.(8) .
ZJ and
T
Mcp:= Fe Fe = gij zt.® .
ZJ (5)
for the undeformed and deformed configuration, respectively. The funda-

mental (local) strain tensor E is defined by the difference of the introduced
metric tensors, i.e.
Pull back and push forward transformations between current configuration

nt, reference configuration no and parameter space Te can be performed
using Ke,Fe and F to derive all other strain measurements, see [11].
334
2. Variations for Direct and Design Sensitivity Analysis

All variations within djrect and design sensitivity analysis are depending on
the variations oX = ot/J(9) of design and ox = ocp(9) of deformation. Using
the displacement mapping ii and its variation ou = oii(9), the variations
are related by ox = oX + ou. An enhancement of the theoretical foundation
to variations of the constitutive model is given in [9].
2.1. VARIATIONS OF GRADIENTS AND STRAINS
The variations of all geometrical quantities can be obtained efficiently by

a pull back ~ variation ~ push forward procedure based on the outlined
decomposition into geometry and deformation mappings defined on Te. For
example, the variation of the material displacement gradient H = Gradx u
yields oH = ouH + oxH, where Ou and Ox denote partial variations w.r.t.
displacement and design, Lf).
ouH = Gradx ou and oxH = - Grad x u Gradx oX. (7)

The variations of other geometrical quantities such as F, Grad x '1 and det F
can be calculated similarly, see [5] for details.
2.2. THE VARIATION OF THE WEAK FORM OF EQUILIBRIUM
The equilibrium conditions have to be fulfilled for every design, i.e. any
total variation of the weak form of equilibrium condition at any time tn
reads oRn = oR~nt + R;"xt = O. For notational simplicity we assume the
external loads to be constant for any design, i.e. oR;"xt = 0, and omit the
subscript n. The remaining total variation of the internal part
can be derived easily using the outlined technique for variations, i.e.
oRint r FT Gradx'1 : oS dV
Jn e
+
Jnr e
Gradx '1: {Grad x ou - Grad x u Gradx oX} S dV
r
Jne
FS: Gradx'1 Grad x oX dV
+ r FS: Gradx'1 Divx oX dV.

1ne
(8)
The variation of the weak form in case of linear elasticity is given in [9].
335
2.3. DESIGN SENSITIVITY AND HISTORY DEPENDENT BEHAVIOUR
The variation of the second Piola-Kirchhoff stress tensor S at time tn needs

extra investigations depending on the actual material behaviour. Whereas
only the strains at time tn are introduced in the geometrical part, the his-
tory of deformation is introduced by the total variation of the stress tensor.
Thus, all displacement fields Us for all times ts < tn may contribute to the
stresses as well as to their variations. The design sensitivity analysis has
to be consistent with the projection algorithm used for structural analysis.
Thus, discrete time steps [tn-I, t n] must be considered leading to a recursive
update strategy both for structural and sensitivity analysis.
The key points of our approach for the combined direct and sensitivity
analysis of history dependent material behaviour are as follows:
1. The variation of the geometry mapping oX = 01/1(9) is given.
v
2. The variations of all displacement mappings Us = s (9) for all times
ts < tn are determined by an incremental update procedure combined
with the structural analysis, i.e. both Us and oU s = As oX have been
calculated at time ts.
3. The equilibrium condition R.n(9,tj,(9),v s(9),v n(9)) = 0 at time tn
v
is solved to update the displacement mapping Un = n (9).
4. The variation of the equilibrium condition Rn = 0 is solved to relate
the known variation oX of design with the unknown variation oun.
The last step needs further discussions. The total variation of the weak form
of equilibrium condition can be split up into variations w.r.t. the actual
displacement field, w.r.t. all past displacement fields and w.r.t. design
(9)
The variation of Rn w.r.t. Un yields a bilinear form k(fJ,ou n ), which is
termed 'tangent stiffness' and defined by
(10)
Similarly, the partial variation w.r.t. design yields the so-called 'tangent
design sensitivity' bilinear form s(fJ, oX) defined by
(11)
Finally, the variations w.r.t. all past displacements Us are collected into the
so-called 'tangent history sensitivity' bilinear form
(12)
336
where the sensitivities of all past displacements t5u s can be expressed by

t5X, i.e. t5u s = A s t5X. This general description of h('1, X) must be specified
for the actual constitutive modeL For example, the internal variables within
the Prandtl-Reufl constitutive model for elasto-plasticity are important for
describing history dependencies and the corresponding design sensitivity,
see e.g. [8]. For a damage theory see the discussion below.
Overall, eq. (9) can be rewritten as
The bilinear form h vanishes in case of elasticity, see [9].
2.4. SENSITIVITY ANALYSIS FOR OBJECTIVES AND CONSTRAINTS
Any objective function or constraint has generally the same functional com-
plexity as the considered weak form, i.e. an = an (9, ¢(9), vs (9), vn (9)).
Three different linear forms must be supplied where a(t5un ) denotes the
partial variation w.r.t. the actual displacement, b(t5X) denotes the history
variation as outlined above and c\vX) denotes the partial variation w.r.t.
design. Overall, the total variation is given by
and can be calculated on the discrete level once t5X and t5u n are known.
3. Discretization Concepts and Computer Implementation

The derived continuous sensitivities must be discretized using e.g. an isopa-
rametric finite element modeL The corresponding discrete matrix formula-
tions for stiffness and design sensitivity depend on the nodal displacements
and nodal coordinates only, see [10, 9, 7] for further details. These infor-
mations are computed on the element level and are assembled into global
'tangent stiffness' and global 'tangent design sensitivity' matrices. So far,
no 'design velocity field' is considered, i.e. the loop over all finite elements
and the loop over design variables are separated.
Considering history dependent material behaviour this ideal decompo-
sition is not further applicable because sensitivity information w.r.t. each
chosen design variable must be stored at every integration point. An ade-
quate minimal interface for design variables entering the element loop for
computing the variation of the weak form must be developed, see e.g. [8].
The final sensitivity expression for any special design variable can be cal-
culated by multiplying the derived matrices with the corresponding 'design
velocity field'. The implementation of the derived expressions are briefly
described in [7].
337
4. Variational Design Sensitivity for Damage

In many practical problems structures under repeated loading exhibit a
softening material behaviour. From a phenomenological point of view this
loss of stiffness is often described by a damage mechanism. For these ma-
terials the sensitivity expressions depend on the response and the response
sensitivities of all previous locations and times.
4.1. CONSTITUTIVE MODEL
A finite strain constitutive model incorporating an irreversible damage

mechanism with a scalar valued internal variable is used [3]. This model is
characterized by an uncoupled volumetric and deviatoric response over any
range of deformation, achieved by a local multiplicative decomposition of
the deformation gradient into volume preserving and dilational parts. The
model was enhanced by empl~ying continuum damage mechanics to develop
a simple three-dimensional damage mechanism, see [4] for first concepts.
The loss of stiffness experimenta]1v observed in polymers when subjected
to levels of strain below the maximum strain previously attained by the
specimen can be described. The crucial assumption is that the maximum
strain attained by the specimen up to the present time completely controls
the damage process, see table 1 and [3] for further details. We adopt this
formulation here.
TABLE 1. Finite-Strain Elasticity with Isotropic Damage
1. kinematics: Et ~(Fr F t - 1)
2. equivalent strain: St J2w o (Et}
3. max. equiv. strain: ~s max J2WO(Es)
== sE(-oo,t)
4. free energy function: W g(Ss) wO(E t )
5. equivalent stresses: St g(Ss) ~t wO(Et}
6. damage criterion: q, St - S. :$ 0
7. evolution: ~s )../(g' (S) St)
8. loading and unloading: ).. > 0, )...q, == 0
This constitutive model was enhanced to incompressible, transversely

isotropic hyperelastic materials in order to model biological soft tissues [12].
This extended constitutive model will be discussed further.
338
4.2. DESIGN SENSITIVITY OF STRESSES
The variation of 8S n (line 1 in eq. (8)) has to be computed for the cho-
sen material law. A detailed analysis has been outlined in [13]. Using an
incompressible, transversely isotropic hyperelastic material with isotropic
damage as model problem the total variation is given by
aS n aS n ~ aS n ass
8S n = aE : 8En
n
+ a'=~s : 8:::::,s = <en : 8En + a'=~s : -a 8E s ·
Es
(15)
The consistent tangent operator <en already known from structural ana-
lysis is required to calculate the partial derivatives of the equivalent stresses
w.r.t. En, see eq. (15). The total variation of the Green-Lagrangean strain
tensor 8En at time tn can be computed from 8un and 8X.
The history dependencies are reduced in this case due to the hyperelastic
material behaviour, i.e. the actual stress at any integration point is only
influenced by that time step ts yielding the maximum damage variable Ss.
But this relevant time step may be different for different integration points.
Thus, additional informations for L~.lsitivity analysis must be stored at time
ts in order to compute the partial derivative of the stress at time tn w.r.t.
the maximal damage variable Ss.
The total variation of the damage variable 8S s at time ts finally re-
lies on the total variation of the strain tensor Es at that time. An update
formula for this quantity has to be provided to compute the sensitivities
correctly. Either full informations of the sensitivities of Ss or Es w.r.t. all
scalar valued design variables must be stored within the history fields of
the corresponding integration points or the sensitivities of the displacement
vector 8u s must be stored on the global level. In this case all sensitivity
informations for the past time ts can be recomputed due to the underlying
elastic behaviour. The computational efficiency of the two approaches de-
pends on the number of design variables and the amount of damage within
the structure.
The computed numerical values of the sensitivities both for loading and
unloading situations have been compared with those of a finite difference
approach proving the correct implementation of the variational approach.
5. Numerical Examples
Numerical examples considering the introduced isotropic damage law are
presented. Both isotropic hyperelastic material behaviour as well as in-
compressible, transversely isotropic hyperelasticity, see [12], are considered.
Stresses, strains and displacements as well as the shape of load-deflection
curves are controlled by changing the design of the structure.
339
6. Conclusions
The proposed representation of variational design sensitivity describes an
integrated treatment of all necessary linearizations in structural analysis
and sensitivity analysis of large strain elasticity with isotropic damage.
It is directed towards an easily applicable variational design sensitivity
analysis and efficient numerical algorithm. The proposed methodology and
the investigated problems are in line with our research on variational design
sensitivity analysis partly described in [5, 7, 8, 11, 13].
References
1. Arora, J.S. (1993) An exposition of the material derivative approach for structural
shape sensitivity analysis, Computer Methods in Applied Mechanics and Engineering
105,41-62.
2. Tortorelli, D.A. and Wang, Z. (1993) A systematic approach to shape sensitivity
analysis, International Jour'fln.l of Solids & Structures 30 (9), 1181-1212.
3. Simo, J.C. (1987) On a Fully Three Dimensional Finite-Strain Visco-Elastic Damage
Model: Formulation and Computational Aspects, Computer Methods in Applied
Mechanics and Engineering 60, 15" ~73.
4. Kachanov, L.M. (1958) Time of the Rupture Process under Creep Conditions, tech-
nical report, Izv. Akad. Nauks SSR Otd. Tech. Nauk.
5. Barthold, F.-J. and Stein, E. (1996) A continuum mechanical based formulation
of the variational sensitivity analysis in structural optimization. Part I: Analysis,
Structural Optimization 11, 29-42.
6. Marsden, J.E. and Hughes, T.J.R. (1983) Mathematical Foundations of Elasticity,
Prentice-Hall, Inc., Englewood Cliffs.
7. Barthold, F.-J. and Stein, E. (1997) Implementing variational design sensitivity
analysis, in W. Gutkowski and Z. Mroz (eds.), Proceedings WCSMO-2, Vol. 1, In-
stitute of Fundamental Technological Research, Warsaw, Poland, pp. 181-186.
8. Barthold, F.-J., Wiechmann, K. and Stein, E. (1997) Optimization of elasto-plastic
structures using the finite element method, in W. Gutkowski and Z. Mroz (eds.),
Proceedings WCSMO-2, Vol. 2. Institute of Fundamental Technological Research,
Warsaw, Poland, pp. 1013-1018.
9. Barthold, F.-J. and Mesecke, S. (1998) Remarks on robust design, in C.A. Mota
Soares (ed.), NATO ASI on Mechanics of Composite Materials and Structures.
10. Barthold, F.-J. and Mesecke, S. (1998) Remarks on computing the energy release
rate and its sensitivities, in C.A. Mota Soares (ed.), NATO ASI on Mechanics of
Composite Materials and Structures Kluwer Academic Publishers, Dordrecht.
11. Barthold, F.-J. (1998) Grundlagen und Algorithmen zur adaptiven Analyse und Op-
timierung von Ingenieurstrukturen, Habilitation, to be submitted.
12. Weiss, J.A., Maker, B.N. and Govindjee, S. (1996) Finite element implementation of
incompressible, transversly isotropic hyperelasticity, Computer Methods in Applied
Mechanics and Engineering 135, 107-128.
13. Firuziaan, M. (1998) Herleitung und Implementation der Sensitivitatsanalyse fur
Schadigungsmaterialien, Diplomarbeit, Institute of Structural Mechanics and Com-
putational Mechanics, University of Hannover.
REMARKS ON COMPUTING THE ENERGY RELEASE RATE
AND ITS SENSITIVITIES
F .-J. BARTHOLD AND S. MESECKE

Abstract. A numerical scheme for computing the maximal energy release

rate and its sensitivities is formulated. Herein, variations of design as well
as variations of the constitutive law are considered.
1. Introduction
We focus our attention on the static case of linear elastic material behaviour
with free energy function lJ!(E) = ~E : <C : E and constitutive law u = <C : E,
where E denotes the linear strain tensor, u the stress tensor and <C the tensor
of elastic moduli. The total potential energy II is defined by
(1)
with displacement vector u and the external load vector to'

II is minimized in case of equilibrium and the weak form of equilibrium
vanishes, i.e. R := 8u Il = 0, see e.g. [16].
The role of the energy release rate" in linear elastic fracture mechanics
is well understood, see e.g. [1, 3]. In case of 2-D test situations like e.g.
DENT (double edge notched tension, see e.g. [1, 12]) the energy release
rate is defined by " = -dIll da, where a denotes the length of the crack.
The crack will grow if the maximal energy release rate "max exceeds a
material dependent critical value "crit, see e.g. [8, 11] for the maximum
energy release rate criterion. "max has to be determined numerically in
general. The corresponding movement of the crack tip does not necessarily
coincide with the actual orientation of the crack.
341
342
2. Theoretical Investigations of the Energy Release Rate

The growth of cracks can be considered as special design variations of the
body leaving the overall mass unchanged but creating new free boundaries.
This observation in mind we use a virtual crack extension method, see e.g.
[2,4, 5, 8, 13], and reformulate the energy release rate as total variation of
the potential energy w.r.t. a virtual crack opening field 8X. This technique
is well known in linear elastic fracture mechanics, see e.g. [7, 6, 9, 14]. Using
the notation and the theoretical background of variational design sensitivity
analysis, see e.g. [15], we conclude
9 = -8ll = -8u ll - 8x ll = -8xll = _8xllint +8xllext = _8x ll int (2)
since R = 8u ll = 0 and 8x llext = 0, because 8to = 0 and 8dA = 0 on 80,0.

Finally, using the variations obtained in [15], the energy release rate reads
9= r
10. 0
(1: Gradx u Gradx 8X dV - ~
210.
r 0
(1: e Div X 8X dV (3)
where 8X denotes any given virtual crack opening field. The energy release
rate can be considered as linear form g(8X) if the displacement vector u
is fixed. On the other hand, if e.g. a large number of load cases shall be
investigated, a trilinear representation g( u, u, 8X) is possible.
3. Computing Energy Release Rates for Fixed Crack Directions

A cartesian coordinate system is considered and Einstein's summation con-
vention is applied on twice occuring indices.
3.1. DISCRETE FORMULATION OF THE ENERGY RELEASE RATE
In this paper and in [15, 16] an isoparametric bilinear element formulation

with nodal shape functions N[ is chosen where nel and ndi denote the
number of nodes and the spatial dimension, respectively. The domain of
each element and the displacements are described by
nel nel
X = XiEi = L N[ X}Ei and u = ui Ei = L N['u} Ei· (4)
[=1 [=1
The virtual crack opening field 8X 1 as well as any variation of design 8X 2

and the subsequent variation of the displacement field 8u2 are given by
nel nel nel
8X 1 = LN[8Xf1Ei' 8X 2 = LN[8X~1Ei' 8u2 = LN[8u~[Ei. (5)
[=1 [=1 1=1
343
The finite element discretization of the continuum mechanical setting

of the energy release rate, see eq. (3), in case of a trilinear form yields
(h =
N E nel nel nel ndi ndi ndi [
U L L L LLL
e=1 1=1 J=1 K=l r=l 5=1 t=l
1ne Tfjk dV
]
ft; 8ft; 8X[<, (6)
where the coefficients Tfjk of the nodal contributions are
A numerical scheme for implementing trilinear forms is discussed in [16]

in combination with second variations. Here, we concentrate on a reduced
representation for a fixed displacement vector u, i.e.
(h = UL L
NE nel ndi [
e=1 K=l t=l

1ne (O"~c~ Uv,t NK,w - O"vw Uv,w NK,t) dV
]
8X[<. (8)
The computation of this relationsluv is outlined below and can be compared

with the computation of the element residual. The element residual is the
derivative of the potential energy of the element w.r.t. the nodal displace-
ments while the 'energy release rate vector' fl, see below, is the derivative
w.r.t. the nodal coordinates. A vector ill with dimension ndi is computed
for every node of the element and assembled into a global vector iL, i.e.
NE nel
(h = U L iL'f 8X l = iLT 8X (9)
e=11=1
where the contribution of each node reads
(10)
The stress matrix Q: and the vector of the nodal shape functions L.l are
(11)
Finally, this global vector iL can be multiplied with any virtual crack
opening field 8X and the investigation of several cracks is reduced to simple
matrix multiplications gh = iLT 8X, i.e. gh is a linear form.
Thus, a natural decomposition of the energy release rate, i.e. into a
vector iL containing all information of the actual state of stress and into a
vector 8X describing the effect of a virtual crack opening field, is achieved.
344
3.2. COMPUTATIONAL EXPERIENCE AND NUMERICAL RESULTS
The described technique has been applied to standard plane stress test
examples like DENT (double edge notched tension) as well as on large
scale problems. Some remarks on computational details are given.
The numerically obtained value of the energy release rate 9h depends
on the quality of the chosen finite element discretization. Adaptive mesh
refinement with residual based error indicators is applied on a coarse initial
mesh. The zone around the crack tip is refined and a stopping criterion must
be applied due to the singularity of stresses. An asymptotic convergence of
the energy and of the energy release rate with increasing number of elements
can be observed. Comparing regular and adaptive mesh refinement methods
the latter shows significant better efficiency.
Furthermore, 9h is independent of the shape and of the size of 15X, see
section 4 for an example of 15X. This result corresponds with the relation-
ship between 9 and the path independent J R-integral, see e.g. [9, 12].
4. Computing the Maximal J;"Ilergy Release Rate

The energy release rate is considered to be maximal in the decisive direction
of the virtual crack extension field, i.e.
(12)
where the global vector 9 contains the system information and is therefore
constant for every direction of the crack extension.
4.1. OVERALL STRATEGY
The virtual crack extension field has to be orientated in the direction of the
decisive crack extension. A finite dimensional model space of all admissible
virtual crack extension fields has to be defined and the maximal energy
release rate has to be determined corresponding to that space.
An initial discretization is canonically given by the chosen isoparametric
model and the mesh. Thus, a finite dimensional space for 15X consisting of
all nodal coordinate variations is given which must be reduced further.
For simplicity, an angle 0' between the old and the new crack direction
and a mapping 15X = 'IjJ(0') is introduced. Thus, the admissible space is
reduced to one dimension.
The solution 15X max can be computed by solving the necessary condition
d
-d
0'
9h = ( -d
d A)T
g
0'-
A AT d
15X + -g -d
0'
A AT d
15X = -g -d
0'
A
15X = 0, (13)
for the angle O'max where 9.. is independent of the angle 0'.
345
4.2. ALGORITHMIC DETAILS
The mapping 8X = 'ljI(a) must be specified further, see figure 1. In this

model problem, four patches with corner nodes Yl' ••. , Y 10 are used to
describe 'ljI. The origin Y10 is placed on the crack tip. The initial direction of
the virtual crack advance ~ is in line with the crack orientation. The corres-
ponding normal direction can be calculated and the corner nodes Yi (a) of
each patch can be determined. For simplicity, only .t, is considered to be
dependent on the angle a and is therefore effected by any variation of the
crack, i.e. node L can be shifted on the straight line between Y 2 and L.
Figure 1. Model space of admissible virtual crack opening fields
During an iterative process the angle a is improved to yield the direction

of the maximal energy release rate. A transformation matrix T is used to
determine the actual vector of the crack extension Qr = T Qo and its normal.
With this in mind an update of the corner nodes Y i can be calculated.
The derivatives w.r.t. a are denoted with a prime. The derivative of the
actual crack extension vector w.r.t. a is Q~ = T'Q o and the derivatives of
Yi are calculated using a modified transformation matrix.
A loop over all finite element nodes X takes only those into consideration
that are within one of the four patches and therefore make a contribution
to the energy release rate. Thus, 8X can be described by the mapping 'ljI,
i.e. for all nodes within a patch we get
4 4
X = L Ni(8) Yi(a) and thus 8X = L Ni(8) 8Y i (a). (14)
i=1 i=1
The local coordinates 8 w.r.t. the intervall [0,1] can be computed using
a local Newton iteration. Several different starting points for 8 must be
chosen in order to guarantee convergence.
The global cartesian coordinates X of any finite element node are fixed,
but the local coordinates 8 change if the angle a is modified.
346
The total derivative de/do: can be determined by solving the equation
dX _ ax de + ax _ 0 I.e. de = _ (ax)-l ax (15)

do: - ae do: 00: -- do: ae 00: .
Furthermore, the virtual crack extension field oX is influenced by the
angle 0:, see eq. (13). The total derivative of oX w.r.t. 0: is given by
(16)
Finally, a fixed point method for eq. (13) is performed to evaluate O:max.
4.3. NUMERICAL EXPERIMENTS
Parameter studies are pell\Jrmed investigating the effect of changing fibre

angles, ply thicknesses as well as Young's moduli and Poisson's ratio on the
maximal energy release rate and v~le corresponding crack orientation. A
rigorous treatment needs an efficient sensitivity analysis as outlined below.
5. Computing Sensitivities of the Energy Release Rate

A sensitivity analysis of the energy release rate g can be performed w.r.t.
two different kinds of parameters. Variations w.r.t. design X can be devel-
oped using the technique described in [15], i.e.
(17)
The notations OU2 for the virtual displacement field and oX 2 for the virtual
geometry field are used. Furthermore, variations of the constitutive model
om, see [16] for details, are considered, i.e.
(18)
The variation oX l indicates the virtual crack extension field as introduced
above. For simplicity we assume that oX l is not explicitly dependent on
either oX 2 or om2, see [16] for a more general setting.
5.1. DISPLACEMENT VARIATIONS OF THE ENERGY RELEASE RATE
The partial variations w.r.t. U2 can be expressed by using the design sen-
sitivity bilinear form as quoted in [16, eq. (4)], i.e.
347
The discretization of eq. (19) yields ,su

(h = -,sxi ,sv
s...T 2 , where s...
denotes the global design sensitivity matrix, see [16, eq. (8)]. This matrix
is necessary for sensitivity analysis w.r.t. all types of variations. Further
results depend on the special design variable, i.e. either design or material.
5.2. DESIGN SENSITIVITIES

The variation of the energy release rate g w.r.t. X is
8x g = - r Gradx u Gradx 8X <C : Gradx u Gradx 8X

lno 1 : 2 dV
- lno
r Gradx u Gradx 8X Gradx 8X dV
iT: 2 1
- lno
r Gradx u Gradx 8X Gradx 8X dV
iT: 1 2
+r Gradx u Gradx 8X Div x 8X dV

iT: 1 2
lno
+ r Gradx u Gradx 8X Div x 8X dV
lno
iT: 2 1
+ r \lI Gradx 8X Gradx 8X 1 dV

1 2 :
lno
- lnor \lIDivx8XlDivx8X2dV. (20)
A discretization yields
with nodal contributions
EXIJ = - r HT BJCIfu H dV
lne
- lner LJ LJ ftT H dV - lner HT ftLI IJ dV
+ r
lne
HT ftLJLJ dV + r
lne
LIIJ g? H dV
+ r \lI LJ LJ dV - lner \lI LI L.) dV.

lne
(22)
The introduced stress matrix Q. and the vector of the shape functions L.I
are defined in eq. (11). The matrices of the material displacement gradient
H, the elastic moduli CI and the B-Matrix B are defined as usual.
348
The sensitivity of the displacement field 6V 2 w.r.t. design variations
s..
6X 2 is given by 6V 2 = _1(-1 6X 2 , see [16, eq. (12)), i.e. overall
On the element level local matrices are assembled and are inserted into the
global matrices, where 1( is stored in a banded structure and and Ex ins..
a compact form, see [16] for further remarks.
The solution procedure is split ted up into two parts. First, the quantities
and (24)
are computed. The necessary number of matrix-vector multiplications de-

pend linearly on the number nl of introduced virtual crack opening fields
6X 1 and on the number n2 of considered design variations 6X 2. Finally,
min(nl, n2)-times forwart: Jackward substitutions have to be performed in
order to compute the desired sensitivity of gh.
5.3. MATERIAL PARAMETER SENSITIVITIES

The variation w.r.t. the constitutive model 6m yields
6m g= Ino tJ m (1: (Gradx u Gradx tJX - ~E Divx 6X ) dV (25)
and the subsequent discretization can be realized as
NE net nmat "T "T

tJm gh = UL L 6X 11 EmIJ 6m2J = 6X 1 Em 6m2, (26)
e=1 [=1 J=l
with contributions
EmIJ = { HT Bf !is!.." dV - { Ia dlJ! T dV. (27)

In. In.
The matrices dlJ! T and du T denote the partial derivatives of the free energy
function lJ! and of the stress matrix Q. w.r.t. the material parameter mJ.
The sensitivity of the displacement field V2 w.r.t. the constitutive model
6m2 can be described by using 61'S = _1(-1 M 6m2, see [16, eq. (12)],
i.e. the sensitivity of gh w.r.t. scalar valued material parameters m2 reads
The matrix-vector multiplications are organized as above.

349
5.4. FURTHER REMARKS
The virtual crack opening vector is assumed to be fixed in the last sections
and the final sensitivities are 5 f:h = 5X~ 5g, see eq. (9) as well as eq. (23)
and eq. (28). Further investigations have to be made in order to describe
the sensitivity of the maximal energy release rate (h,max and the sensitivity
of the corresponding angle ct max .
For simplicity, we assume the virtual crack opening field 5X 1 to be
additionally dependent on ct, like explained in section 4, but to be explicitly
independent of any variation of design or of the constitutive model. See also
the discussion in [16].
Thus, the enhanced sensitivity of the maximal energy release rate f:h,max
is equal to the already obtained results, i.e.
5f:h,max = 5fl8Xmax + ~T 8(5Xmax) = 5~T 8X max , (29)

since 5(8X max ) = d(5Xmaxi/dct 5ct and the scalar product between this
vector and iL vanishes due to eq. (13).
Last but not least, the sensHlvity of the corresponding angle ct max
should be considered. The necessary condition for maximum energy re-
lease rate, i.e. eq. (13), has to remain valid for all variations of design and
material parameters. Thus, any variation of eq. (13) yields
(30)
and the variation of the angle ct max can easily be computed. Additionally,
second partial derivatives of 8_:\' W.r.t. ct must be derived for the introduced
one-dimensional model of admissible virtual crack opening fields.
5.5. NUMERICAL EXAMPLES
The derived variational approach has been implemented in our research

analysis tool INA-OPT (INelastic Analysis and OPTimization) and the
overall algorithm has been applied on standard test examples like DENT
using adaptive mesh refinement. The computed sensitivities were compared
with those obtained by a finite difference approach showing the correctness
of the implementation.
The additional storage requirements for sensitivity analysis of the energy
release rate are moderate because the matrices Ex, Em can be stored in
compact form.
Furthermore, the virtual crack opening field 8X Can be applied to any
given finite element mesh due to the separation of physics in the vector iL
and the change of geometry in the vector 8X.
350
6. Conclusion
An algorithm for computing the maximal energy release rate has been out-
lined. Parts modelling the actual state of stress and the virtual crack open-
ing field have been separated. Variations of the energy release rate w.r.t.
design and the constitutive model have been investigated. Discrete matrix
formulations for sensitivity analysis have been derived and computational
details have been discussed. The efficiency of the proposed computational
scheme has been proved by numerical experiments.
References
1. Anderson, T.L. (1991) Fracture Mechanics-Fundamentals and Applications, CRC
Press, Florida.
2. Barbero, E.J. and Reddy, J.N. (1990) The Jacobian Derivative Method for Three-
Dimensional Fracture Mechanics, Communications in Applied Numerical Methods
6, 507-518.
3. Broek, D. (1986) Elementary Engineering Fracture Mechanics, Martinus Nijhoff
Publishers, Dordrecht, 4. Edition.
4. deLorenzi, H.G. (1982) On the rel ..ase rate and J-integral for 3-D crack configura-
tions, International Journal of Fracwre 19, 183-193.
5. deLorenzi, H.G. (1985) Energy release rate calculations by the finite element
method, Engineering Fracture Mechanics 21, 129-143.
6. Fancello, E.A., Taroco, E.O. and Feijoo, R.A. (1993) Shape sensitivity analysis in
fracture mechanics. Structural Optimization 93, Proc. of the World Congress on
Optimal Design of Structural Systems, Vol.2, University of Rio de Janeiro, Rio de
Janeiro, pp. 239-248.
7. Haber, R.B. and Koh, H.M. (1985) Explicit expressions for energy release rates using
virtual crack extensions, International Journal of Numerical Methods in Engineering
21,301-315.
8. Hellen, T.K. (1975) On the method of virtual crack extension, International Journal
of Numerical Methods in Engineering 9, 187-207.
9. Lee, T.W. and Grosse, LR. (1993) Energy release rate by a shape design sensitivity
approach, Engineering Fracture Mechanics 44, 807-819.
10. Li, F.Z., Shih, C.F. and Needleman, A. (1985) A comparison of the methods for
calculating energy release rates, Engineering Fracture Mechanics 44, 405-421.
11. Lin, S.-C. and Abel, J.F. (1988) Variational approach for a new direct-integration
form of the virtual crack extension method, International Journal of Fracture 38,
217-235.
12. Mesecke, S. (1998) Anmerkungen zur Energiefreisetzungsrate - Berechnung und
Optimierung, Diplomarbeit, Institute of Structural Mechanics and Computational
Mechanics, University of Hannover.
13. Parks, D.M. (1974) A stiffness derivative finite element technique for determination
of crack tip stress intensity factors, International Journal of Fracture 10, 487-502.
14. Saurin, V.V. and Banichuk, N.V. (1997) Adaptive methods for structural analysis
and structural optimization of composites. Institute of Structural Mechanics and
Computational Mechanics, University of Hannover, internal report.
15. Barthold, F.-J. (1998) Remarks on linear and nonlinear variational design sensi-
tivity analysis, in C.A. Mota Soares (ed.), NATO ASI on Mechanics of Composite
Materials and Structures Kluwer Academic Publishers, Dordrecht.
16. Barthold, F.-J. and Mesecke, S. (1998) Remarks on robust design, in C.A. Mota
Soares (ed.), NATO ASI on Mechanics of Composite Materials and Structures.
Kluwer Academic Publishers, Dordrecht.
SOME REMARKS ON ROBUST DESIGN
F.-J. BARTHOLD AND S. MESECKE

Abstract. Designing structures with high insensitiveness w.r.t. parameter

uncertainties in e.g. material data and geometry is of high practical im-
portance in engineering science. These robust structures show a reduced
tendency to failure in case of uncertainties, i.e. they have a high reliability.
A special definition of robustness is given in terms of minimized first sen-
sitivities w.r.t. parameter variations. Thus, overall second variations have
to be supplied for a gradient based nonlinear programming approach and
an efficient first and second order sensitivity analysis is of high importance.
Details for computing the necessary variations for general nonlinear formu-
lations are elaborated in this paper. Full sensitivity information is given for
linear elastic material behaviour in order to reduce the complexity.
1. Introduction
Several definitions of robustness and reliability are given in literature.

An important treatment of reliability-based system design investigates
the stochastic behaviour of systems, see e.g. [10]. In this classical approach
a system is said to be reliable if the probability offailure is acceptably low.
The interaction with structural optimization and practical examples as well
as numerical schemes are discussed in e.g. [11, 4, 12].
Limitations of a probabilistic theory of reliability analysis are discussed
e.g. in [2, chapter 7]. He argues that 'probability is not the only viable option
for quantifying uncertainty in reliability analysis, nor is it even always the
most desirable mathematical model. In many complicated and important
technological situations, the inability to verify the details of a probabilistic
model may lead to significant inaccuracy in the reliability analysis'.
351
352
In an alternative approach to the quantification of reliability, Ben-Haim

[2] introduced the concept of robust reliability based on the evaluation of the
robustness of the system to uncertainties. The mathematical formulation of
robust reliability is derived from the theory of convex models, see e.g. [3]. In
this concept a system has high reliability when it is robust with respect to
uncertainties. It has low reliability when even small amounts of uncertainty
entail the possibility of failure.
The concept of robust reliability employs three components, i.e. (1) a
mechanical model, (2) a failure condition and (3) a model of the uncertainty.
Thus, the reliability is assessed as the greatest amount of uncertainty con-
sistent with successful operation of the structure.
Different convex models of uncertainty are discussed, e.g. energy-bound
models are used for initially straight beams and geometric imperfections in
case of stability failure. If the maximal energies of uncertain geometrical
shapes are below a critical value then no stability failure can occur and this
critical value is the quantitative measure for robust reliability, see [2].
Our research is directed towards convex models of uncertainty which are
directly based on first sensitivitie~ fthe failure condition w.r.t. important
parameter variations. This definition tackles directly the dependency of the
failure criterion from system parameters. Thus, the model is e.g. the convex
hull of the most significant variations. The most significant modes can be
investigated in an automated numerical procedure.
In order to optimize the structure, i.e. to enlarge the reliability, the sen-
sitivity of the reliability w.r.t. design parameters must be computed. Thus,
overall first and second variations of the failure condition must be derived,
implemented and computed. First and second order sensitivity analysis has
been considered in a variety of publications over the last decades, see e.g.
[5] and the references therein for further informations. First order sensi-
tivities in form of either overall numerical, semi-analytical or fully analyti-
cal/variational approaches are widely used. Deriving second order sensitiv-
ities is a much more challenging task and efficient numerical schemes for
the efficient computation are hardly available.
The actual contribution concentrates on deriving, implementing and
computing second order sensitivities w.r.t. variations of design and mate-
rial data. Our approach to variational sensitivity analysis is based on a
differential geometry setting of continuum mechanics, see [5] for further
informations.
The derived technique is applied to composites in order to optimize
their reliability w.r.t. uncertainties in geometry and material data, i.e. fibre
angle and ply thicknesses.
353
2. First Variation of Weak Form for Linear Elastic Materials

The investigations correspond with [5, 6] as far as design is considered.
Furthermore, variations of the constitutive model are taken into account.
2.1. CONTINUUM MECHANICAL FORMULATION
The weak form n( u, TJ) = 0 in case of linear elasticity reads
(1
where = <C : € denotes the stress tensor, € the linear strain tensor, <C the
tensor of elastic moduli, TJ the test function, Pobo the body forces and to
the tractions. We concentrate on n int and assume next to be deformation
independent and constant for any considered design, i.e. 8n ext = o.
The total variation of the weak form of the equilibrium can be separated
additively in partial variations w.r.t. the virtual displacement field 8u, w.r.t.
the virtual geometry field 8X and .:.r.t. the constitutive model 8m, i.e.
The 'tangent stiffness' bilinear form k(TJ, 8u) := 8un is defined by
k( TJ, 8u) = { Gradx TJ : <C : sym {Gradx 8u} dV. (3)

lno
Similarly, the 'tangent design sensitivity' bilinear form s(TJ,8X) := 8xn is
s(TJ,8X) = (Gradx TJGradx 8X - Gradx TJ Divx 8X) dV

lno
- { {1:
- (GradxTJ:<C:sym{GradxuGradx8X}dV. (4)
lno
Finally, variations 8m tT w.r.t. the constitutive model 8m are considered
yielding the 'tangent material sensitivity' bilinear form m(TJ, 8m) := 8m n
m(TJ,8m) ={
k
8m {1: Gradx TJdV ={
k
Gradx'1: :(1m :8m dV. (5)
Thus, eq. (2) can be rewritten using the introduced bilinear forms
8n('1, 8u, 8X, 8m) = k('1, 8u) + s(TJ, 8X) + m('1, 8m) = O. (6)
This equation implicitely defines 8u once 8X and 8m are known.

354
2.2. DISCRETE FORMULATION
An isoparametric finite element method is introduced, see [6] for the used
notation. The 'tangent stiffness' bilinear form k(f1, 6u) is approximated by
kh(f1h, 6uh) =
NE nel net
e~l r;.f;!Ii
[
ke BT CJJ. fu dV
]
610 = !IT K 6V, (7)
where K denotes the global stiffness matrix.

Similarly, the 'tangent design sensitivity' bilinear form Sh (f1h, 6Xh) reads
NE net nel
Sh (f1h , 6Xh) = UL: L:!Ii 8L = !IT S.6X,
§.iJ (8)
e=l 1=1 J=1
where S. denotes the global design sensitivity matrix with contributions
§.i J = [
JOe [- r,! (];. B J H - ~ 14 IJ + ~ L.I IS] dV. (9)
Finally, the 'tangent material ~aTJ.sitivity' bilinear form is
mh(f1h, 6m h)
NE nel nmat
= U L: L: fii
e=II=1 J=l
1 BT!lsLJ
[
Oe
]
dV 8mJ = fiT M 15m, (10)
where M denotes the global material sensitivity matrix. Only parameter

variations remain for a fixed constitutive model, i.e. the stress matrix Q: and
the matrix of elastic moduli (];. undergo certain variations d a J and dO:: J, re-
spectively. The constitutive model for a fibre-reinforced composite shell is
given e.g. in [8] and all partial derivatives w.r.t. the scalar valued macro-
scopic material parameters can be derived in a lengthy but straightforward
differentiation process, see e.g. [9].
The discrete form of eq. (2) is
(11)
Usually, 6V is computed either w.r.t. a given variation of design 6X or

w.r.t. a given variation of the constitutive model 15m, i.e. either
or (12)
3. Second Variation of Weak Form for Linear Elastic Materials

The usefulness of second variations w.r.t. robust design has been outlined
above. In this section a detailed analysis as well as hints to numerical com-
putation of second variations of the weak form of equilibrium condition in
case of linear elastic material behaviour are given.
355
3.1. CONTINUUM MECHANICAL FORMULATION
Any variation of eq. (6) must vanish due to the equilibrium condition, i.e.
02'R(fJ, 02u, OUI, OU2, oX 1, oX 2, omt, om2, 02X I ,2, 02 ml ,2) = O. (13)
This multilinear form depends on displacement, design and constitutive

model variations. The subscripts indicate the correlation between virtual
displacement vector and given variation of design and material. Further-
more, variations 02X I ,2 and 02m1 ,2 exist, if the given variations of design
oX I and oX 2 (or omi and om2, respectively) are not independent. This
might occur if e.g. oX I and oX 2 are derived from higher design models.
Thus, 02u is implicitely defined by 02'R = o. A rigorous approach yields
02'R ouu'R + Oxu'R + omu'R

+ oux'R + oxx'R + omx'R (14)
+ Oum::' + oXm'R + omm'R = 0
and all partial variations of the we::> 1<- form must be examined. This multilin-
ear form can be written using the subsequently introduced bilinear forms.
The second partial variation of'R w.r.t. displacements yields
since ouu'R = 0 due to the linearity of the considered problem.

All mixed partial variations w.r.t. displacements and w.r.t. design, i.e.
oux'R = a(fJ, ~UI, oX 2) and oXu'R = a(fJ, OU2, oX I ), are given by
a( fJ, OU, oX) := [ [ - Gradx fJ : <t : Gradx OU Gradx oX

Jn a
- Gradx fJ Gradx oX : <C : Grad x ou

+ Gradx fJ : <C : Gradx OU Div X oX] dV. (16)
All mixed partial variations w.r.t. displacements and w.r.t. material, i.e.
oum'R = b (fJ, OUI, om2) and omu'R = b (fJ, OU2, omI), are given by
b(fJ,ou,om):= [GradXfJ:om<C:sym{Gradxou}dV. (17)

Jna
The mixed partial variations w.r.t. design and w.r.t. the constitutive
model are 'Rxm = C (fJ, oXt, om2) and 'Rmx = c(fJ, oX 2, 8mt} with
C .- -
J[no Gradx fJ : om<t : sym{Gradx U Gradx oX} dV
- [ OmiT: [Gradx fJGradx oX - Gradx fJ Divx oX] dV. (18)
Jn o
356
The second partial variations w.r.t. design are given by
where the 'tangent design sensitivity' bilinear form s refers to eq. (4) and
the trilinear form d (1'/, 8X, 8X) is defined by
d .- r [- Gradx 1'/ : (]:; : Grad x u Gradx 8X

ino 1 Div X 8X 2
+ Grad x 1'/ Grad x 8X 2 : (]:; : Grad x u Gradx 8X 1
+ Gradx 1'/ : (]:; : Grad x u Gradx 8X 2 Grad x 8X 1
+0" : Gradx 1'/ Gradx 8X2 Gradx 8X 1
Gradx 1'/ Gradx 8X 1 Divx 8X 2
- 0" :
-1/20": Gradx 1'/(Gradx 8X 1 Gradx 8X 2 : 1)

+ 1/20" : Gradx 1'/ Divx 8X 1 Div x 8X 2] dV. (20)
Finally, the second variation v.... t. the constitutive model is given by
where the trilinear form e (1'/, 8m, 8m) reads
e(1'/, 8m, 8m):= r8

ino mm O": Gradx 1'/dV. (22)
3.2. DISCRETE FORMULATION
A discrete formulation based on an isoparametric finite element scheme of

all introduced trilinear forms a, b, c, d, e has been derived and implemented.
Although all trilinear forms as well as all functions and variations are finite
dimensional approximants of those introduced above, the same notation is
used for simplicity.
The general scheme is motivated in case ofthe trilinear form a(1'/, 8u, 8X)
which can be written as
a = e~l EE'fl?; ~ ~
N E nel nel nel ndi ndi ndi [
I AI1K dV
1
r,; 871:: 8X[<, (23)
where N E denotes the number of elements, nel the number of nodes on the
element level and ndi the dimension of the problem. Furthermore, r,!,8711
357
and 8xf are the nodal values of f'/, 8u and 8X, respectively. Einstein's sum-
mation convention is assumed for each contribution, i.e. repeated indices
sum up to the number of dimensions ndi. The significant nodal values are
Arst
IJK = NI,t NJ,v NK,w(J',svrw - NI,t NJ,w NK,v(J',rvsw (24)
+ NI,v NJ,w NK,t(J',rvsw. (25)
For each nodal combination (I J K) a field of dimension (ndi)3 is com-

puted on the element level. This field is inserted into an element field of
dimension (net x ndi)3 which is stored for latter usage. The same procedure
can be applied to all other trilinear forms b, c, d, e.
The dimension correlates with the approximation of geometry and dis-
placements introduced with the finite element model, i.e. the isoparametric
concept. The described trilinear forms are part of the variations w.r.t. the
full approximation space. P. more sophisticated approximation for design
using e.g. higher order models may reduce the overall storage requirements.
In this case it is more suitable to assemble all element contributions into a
global field. Thus, sensitivity analysis on the element level does not interact
with the actual design variables, e.g. 'design velocity fields'.
Each trilinear form a (f'/, 8U1, 8X 2) can be represented by a vector a1,2
once the variations for displacement and design (or material) have been cho-
sen, i.e. a = fiT ·a1,2. Thus, for any combination of variations (8U1' 8X 2) the
element fields of the trilinear form have to be multiplied with the (8V l' X2)
in order to derive ~h,2. Finally, ah = '!]T !!.1,2 and equivalently for all other
trilinear forms.
The second variation of the weak form w.r.t. design variations 8X 1 and
variations 8m2 of the constitutive model is given by
82 R 8uu R + 8um R + 8xu R+ 8xm R

k(f'/,8 2 u)
+ b(f'/, 8U1, 8m2) + a(fJ, 8U2, 8Xt} + c(fJ, 8X 1, 8m2) (26)
and the corresponding second variation of the discrete displacement vector

82V in case of design variations (8X l' 8m2) can be computed by using the
introduced matrices, i.e. with W := Q1,2 + !!.2,1 + £1,2
(27)
The same hold true for any other combination of design variables. Thus,
only matrix-vector multiplications outside the elements are necessary once
the element contributions have been computed and stored.
358
4. First and Second Variations of Objectives and Constraints

A similar technique as described above can be used to derive and implement
first and second order sensitivities of an objective function or constraint ¢
w.r.t. variations of design or the constitutive law, see eq. (2) and eq. (14).
Furthermore, linear and bilinear forms and the subsequent matrices are
used to describe these variations and the corresponding discrete formula-
tions, i.e. 8u ¢ =: f(8u) = fT 8Y and 8x ¢ =: g(8X) = gT 8X as well
as 8m ¢ =: h (8m) = !J? 8 min case of first sensitivities. Thus, the overall
sensitivity analysis in case of geometrical design variables using eq. (12)
reads
8¢ = LT 8Y + r/8X = - (LT [(-1 [i + 8X rl)(28)
and equivalently in case of material parameter design variables
The efficiency of the numerical computation depends on the number of

objective functions ¢ and design "o,liables 8X (or 8m, respectively), see the
discussion on the direct or the adjoint approach to sensitivity in e.g. [13].
Considering second sensitivities the following variations, bilinear forms
and matrices are obtained:
•T •
8uu ¢ A (8Ub8u2) =8V 1 A8V 2
A T A
8u x¢ B (8U1,8X2) = 8V 1 B 8X 2
AT
8um ¢ C (8U1, 8m2) = 8V 1 C 8 m 2
A T A
8xx¢ D(8X 1,8X 2)=8X 1 D8X 2

AT
8xm¢ E (8Xb 8m2) = 8 Xl E 8 m2
8mm ¢ F (8mt,8m2) = 8mi F 8m2'
All second order sensitivities can be computed using the introduced

matrices. For exam pIe, mixed sensitivities w.r. t. design 8X 1 and material
data 8m 2 , i.e. 82¢ = 8u¢ 82u + 8uu ¢ + 8um ¢ + 8xu¢ + 8Xm¢, are given as
T T
82¢ = LT 82V + 8 Vi A 8 V 2 + 8 Vi C c5 m 2
A A A •
•T A AT
+ 8 V 2 B 8 Xl + 8 Xl E 8 m2 (30)
and can be presented in various forms using e.g. the already computed first
variations 8Y 1 = _[(-1 8X 1 and 8Y 2 = _[(-18m2' see eq. (12), and the
second order variation 82y = _[(-1 w, see eq. (27).
359
5. Application of Sensitivity Analysis to Robust Design
The concept of robust reliability as introduced in [2] is essentially based

on a failure condition and a convex model of uncertainty. Our contribution
to robustness and reliability combines failure and uncertainty by means of
sensitivity analysis. The central ideas are as follows:
The variations of geometry and material data which contribute most
to the sensitivity of the failure condition are chosen to describe the
convex model of uncertainty.
The reliability of the structure is furthermore measured by some weigh-
ted norm of all computed first sensitivities of the failure condition.
Thus, the structure is reliable w.r.t. the chosen failure condition and
the chosen variations if this norm is suitably small.
A mathematical programming approach is used to minimize the norm
of the first sensitivities, i.e. reliability is improved.
This concept is highly dependent on a general theoretical foundation for
deriving first and second order ser.".tivities as well as on efficient numerical
algorithms for computing these sensitivities.
The representation of the sensitivities in form of linear, bilinear and tri-
linear forms as described above is seen to be a general approach. Further-
more, the variations are obtained prior to any discretization of displacement
or of geometry on a continuum mechanical level. Thus, adaptive techniques
for design can be chosen in order to suitably modify the optimization prob-
lem for improving the reliability.
The definition of the convex model of uncertainty based on sensitivity
information formulates the robustness in a direct manner. Using numerical
schemes this approach can be applied to a large range of engineering prob-
lems with high complexity because the convex model can be numerically
computed without additional theoretical investigations.
Numerical examples will be investigated in order to show the efficiency
of the overall approach. Hints on interesting test cases as well as important
technical applications can be found in [2]. Furthermore, this approach is
applied on robust design of composites.
6. Conclusion
The robust reliability concept for mechanical structures introduced by Ben-

Haim [2] was enlarged w.r.t. a sensitivity based formulation of convex mod-
els of uncertainty. The importance of a highly sophisticated first and second
order sensitivity analysis was outlined and a theoretical formulation as well
as a numerical implementation was described and discussed.
360
References
1. Barthold, F.-J. and Stein, E. (1996) A continuum mechanical based formulation
of the variational sensitivity analysis in structural optimization. Part I: Analysis,
Structural Optimization 11, 29-42.
2. Ben-Haim, Y. (1996) Robust reliability in the mechanical sciences. Springer, Berlin,
Heidelberg, New York.
3. Ben-Haim, Y. and Elishakoff, !. (1990) Convex models of uncertainty in applied
mechanics. Elsevier, Amsterdam.
4. Kleiber, M., AntUnez, H., Hien, T.D. and Kowalczyk, P. (1997) Parameter sensitivity
in nonlinear mechanics. John Wiley & Sons, Chichester.
5. Barthold, F.-J. (1998) Remarks on linear and nonlinear variational design sensitivity
analysis, in C.A. Mota Soares (ed.), NATO Advanced Study Institute on Mechanics
of Composite Materials and Structures Kluwer Academic Publishers, Dordrecht.
6. Barthold, F.-J., Mesecke, S. (1998) Remarks on computing the energy release rate
and its sensitivities, in C.A. Mota Soares (ed.), NATO ASI on Mechanics of Com-
posite Materials and Structures Kluwer Academic Publishers, Dordrecht.
7. Reddy, J.N. (1997) Mechanics of laminated composite plates, CRC Press.
8. Stein, E. and TeBmer, J. (1998) Theory and computation of multilayer composites,
in C.A. Mota Soares (ed.), NATO ASI on Mechanics of Composite Materials and
Structures. Kluwer Academic Publishers, Dordrecht.
9. Barthold, F.-J. (1998) Variations of multilayer composites due to parameter varia-
tions of the constitutive model, Insf" te of Structural Mechanics and Computational
Mechanics, University of Hannover.
10. Ditlevsen, O. and Madsen, H.D. (1996) Structural reliability methods, Wiley.
11. Khot, N.S. and Abhyankar, N.S. (1993) Integrated optimum structural and control
design, in M.P. Kamat (ed.) Structural Optimization: Status and Promise, Progress
in Astronautics and Aeronautics, Vol. 150, AIAA, pp. 743-767.
12. Schueller, G.!" Gasser, M., Hartl, J. and Lener, G. (1997) Reliability based op-
timization - a tool for practical design, in A.D. Belegundu and F. Mistree (eds.)
Optimization in Industry, ASME, pp. 169-176.
13. Dems, K. and Haftka, R.T. (1989) Two approaches to sensitivity analysis for shape
variation of structures, Mechanics for Structures £3 Machines 16 (4), 501-522.
OPTIMIZATION OF THIN-WALLED STRUCTURES
E. STEIN AND F .-J. BARTHOLD

Appelstrafle 9A, D-30167 Hannover, Germany
Abstract. In optimizing three-dimensional structures such as thin-walled

composite shells the design variables depending on the cross-section, the
material or the geometry are modified to obtain a minimum of the for-
mulated objective functi<.u, e.g. weight, eigenvalues, potential energy, limit
load etc. These different objective functions can be combined using different
weights to obtain a Pareto optimized structure. Designing in an economic
and safe way requires the consideration of geometrical and material nonlin-
earities in the structural analysis and in the optimization problem. Thus, we
reach a construction with several close or similar eigenvalues, i.e. the eigen-
values cluster. The postcritical behaviour becomes malicious, whereby it is
unavoidable to consider the influence of imperfections in the optimization
process.
1. Introduction
Some significant topics within structural optimization of thin-walled com-

posites are shortly described and stability phenomena and geometrical mod-
elling are more elaborated in the subsequent sections.
1.1. GENERAL REMARKS ON STRUCTURAL OPTIMIZATION
The structural optimization process can be subdivided into three parts, i.e.
(1) the optimization problem, (2) the mathematical programming algorithm
and (3) the structural analysis.
After choosing the design variables, the objective functions and the con-
straints a first solution of the starting design is required. This structural
analysis has to consider all nonlinear effects and furthermore the geomet-
361
362
rical imperfections. The sensitivity analysis determines the values of the
objective functions and the constraints delivers their gradients and, if nec-
essary, their Hessian matrices in an efficient and accurate way. With this
information new design variables are calculated by the optimization algo-
rithm. The new design variables are transformed to structural information
in the following interface.
1.2. OBJECTIVES AND DESIGN VARIABLES
In structural optimization of fibre reinforced thin-walled structures mainly

stability phenomena of the plates and shells as well as local failure such as
delamination are of practical importance. Controlling the eigenfrequencies
and eigenvectors as well as critical loads within the optimization problem is
a general area of research. Furthermore, the state of stress within compos-
ites must be controlled in order to avoid stress concentration which might
initialize micro failure. CJ '. ~ria like Hashin, Tsai- Wu are under considera-
tion.
The most important design vl:'~;l,bles within structural optimization of
composites are the design of the layered structure of the shells. Thus, fibre
angles as well as ply thicknesses are important design variables. Addition-
ally, the overall design of the structure will be investigated in industrial
applications with increasing interest.
1.3. MATHEMATICAL PROGRAMMING ALGORITHM
The following optimization problem including constraints is considered

min <p(x) (1)
xEIRn
subject to the constraints

xi ~ Xi ~ xi for i = 1, ... ,n ,
gj(x) ~ 0 for j = 1, ... ,m ,
hk(X) =0 for k = 1, ... ,I .
The objective function <P, the inequality constraints 9 and the equality con-
straints h, being explicit and implicit functions of the design variables x,
imply deflections, stresses, yield functions, eigenvalues, weight, limit load
etc. The necessary conditions for x* to be the solution of the above opti-
mization problem (1) are the Kuhn - Tucker relations.
A sequential quadratic programming algorithm [4] is used to obtain the
solution of the optimization problem. In each iteration step a subproblem
using a quadratic approximation of the objective function and a linear ap-
proximation of the constraints has to be solved. This BFGS-method requires
363
the values and the gradients of the objective functions and the constraints.
The Newton-Method using the exact second derivative of the Lagrangian
function in each iteration step requires a high theoretical and numerical
effort.
The solution of the quadratic subproblem is taken as a search direction
within the inaccurate line search algorithm of Armijo. Considering integer
(discrete) optimization problems, the change of the design variables is cal-
culated assuming a continuous distribution. Once a proposal for the change
of the design variables has been made by the line search algorithm, we take
the new profiles to be feasible and closest to the continuous solution, see
e.g. [5]. The modified search direction has to be a descent direction of an
appropriate merit function [4].
1.4. STRUCTURAL AND SENSITIVITY ANALYSIS
A suitable mechanical modt~ ~nd an efficient numerical algorithm for struc-

tural analysis of thin-walled structures is a central prerequisite for any
optimization. Furthermore, realis" • models for the occuring damage and
failure phenomena must be considered. Further hints on these modelling
and computational topics are given e.g. in [1, 2].
The optimization problem needs an extra mathematical and physical
modelling. First of all, the adequate choice of design variables e.g. in shape
optimization is of central importance, see section 2 for further remarks.
Additionally, the physical phenomena must be interpreted correctly in order
to define the adequate model for any optimization procedure.
Finally, an efficient sensitivity analysis must be supplied, see [3] for
further details. The investigation the sensitivities of stability phenomena
including the postcritical behaviour of structures is an area which needs
further research activities.
1.5. GEOMETRICAL MODELLING IN STRUCTURAL OPTIMIZATION
A natural approach for the treatment of shape optimization problems is

based on the definition of geometrical design variables. Whereas in former
times the coordinates of finite element nodes were chosen as design variables
nowadays CAD oriented formulations predominate in the field of shape
optimal design. A description of various possibilities for the definition of
design velocity fields based either on CAD or FEM models is given in [6].
With the development of CAD and CAGD (Computer Aided Geometric
Design) much effort has been focussed on the integration of CAD and FEM,
see e.g. [7, 8] and especially the implementation of adaptive FEM using
CAD methods is an actual research topic, see [9, 10, 11, 12] and [13] in the
field of topology optimization.
364
The basic idea of the CAD oriented approach is to introduce an adequate

parametrization of the given geometry of the considered structure. In this
case, geometrical quantities such as control parameters of Bezier curves
and surfaces, b-splines or NURBS (non uniform rational b-splines), see
[14], are chosen as design variables. It is well established in industry to use
CAD for the design of a structure and especially CAGD for the geometrical
modelling of structures.
Within the overall optimization algorithm, the finite element method
is used to compute the state variables and their sensitivities and a gradi-
ent method, e.g. Sequential Quadratic Programming (SQP), is utilized for
solving the optimization problem. The following problems arise:
1. The choice of design variables: In most cases the CAD oriented design
variables have only local influence on the shape. In order to achieve an
overall variation of the design, a large number of design variables must
be introduced.
2. The restriction of the number of design variables: The sensitivity anal-
ysis consumes the main part of cpu time. Therefore only few powerful
design variables should be deuued if possible.
3. The control of the FE mesh: Using the h-method of mesh refinement
the distortion of the finite elements must be controlled.
4. The integration of CAGD, FEM and the optimization algorithm: The
integration of the mentioned numerical tools is a necessary prerequisite
for an efficient overall optimization procedure.
2. Hierachical Design Modelling

Different approaches for reducing the complexity of the geometrical
model and the number of design variables were proposed in the literature.
A decomposition technique to minimize the numerical effort for the op-
timization of several sub domains and their subsequent assembly to a global
solution is discussed in [15]. Variable linking to decrease the number of de-
sign variables and to minimize the mesh distortion is described in [8]. In [16]
a hierarchical strategy for unconstrained problems is developed. A family of
embedded parametrizations for the smooth description of shapes are used
for optimization of flow problems. This leads to a multigrid procedure.
365
2.1. HIERARCHICAL DESIGN SPACE
A general methodology for the numerical treatment of shape optimal design

problems based on a given graduation of the design space and the succes-
sive activitation of design variables during the optimization procedure is
formulated. It is shown that the desired hierarchy of the design variables is
naturally given by CAD orientated modeling. The interest is furthermore
directed towards an efficient analytical sensitivity analysis incorporating es-
pecially the defined geometrical model and the finite element meshes. This
new methodology will be described in detail.
An hierarchical approach for shape optimization connected with an ef-
ficient sensitivity analysis is proposed. The main part of the computational
effort is consumed by the sensitivity analysis related to the number of de-
sign variables. Therefore one should limit the number of design variables
and define only few powerful design variables corresponding to the special
interests of the considered optimization problem. This procedure may lead
to a considerable limitation of the design space.
In order to achieve an efficien ~ensitivity analysis without loss of geo-
metrical flexibility we choose a hierarchical approach for shape optimiza-
tion. It is based on a brep model (boundary representation) and is character-
ized by the hierarchy: points - lines - patches - masterpatches. Master
patches are introduced for the description of main parts of the structure
combining several patches. This hierarchy leads to a graduation of design
variables. We distinguish global, regional and local design variables.
- Global design variables are coordinates of corner points of a master
patch. The variation of global design variables influences the geomet-
rical description of all patches within the master patch (global influ-
ence). Large global changes of the structure can efficiently be achieved
by using this subset of the design space.
- Regional design variables are the coordinates of corner points of patches.
The variation of regional design variables influences all geometrical ob-
jects of the considered patch (regional influence). This model is used
only for modifying smaller parts of the structure.
- Local design variables are the coordinates of control points of Bezier
curves or patches. Only one discrete point or control point is changed
thereby (local influence). Within this model local effects can be con-
sidered.
After defining the whole set of design variables and the described models
the optimization procedure is started with global design variables to get
an optimal design based on these few design variables. After a successive
activation of regional and local design variables an optimial design based
on the whole set of design variables is obtained.
366
2.2. EXAMPLES FOR GEOMETRICAL MODELLING
Examples of a shell structure (bridge construction) and of a stiffened can-

tilever demonstrate the advantages of the integrated model and the geo-
metrically nested optimization procedure, see also [17].
2.3. ADVANTAGES OF MULTILEVEL OPTIMIZATION PROCEDURES
The advantages of this multilevel optimization procedure are obvious.

-+- Rough changes of the design based on only few global design variables
are achieved in only few iteration steps.
- The definition of bounds for global design variables is very simple
because all geometrical objects in the concerned master patches are
changed automatically.
- An overall analytical sensitivity analysis is performed based on the
variation of the geometrical model corresponding with the adapt ion of
the finite element mesh.
- The multilevel optimizati()ll procedure related to a successive expan-
sion of the design space leads to a considerable reduction of cpu-time
for sensitivity analysis.
3. Structural Optimization and Stability Phenomena

Stability phenomena in structural and sensitivity analysis are of high im-
portance both in research and in industrial applications. Maximizing the
critical load is a central engineering task but the reliability of the structure
is more often dominated by the post-critical behaviour.
3.1. QUALITATIVE REMARKS ON STABILITY AND OPTIMIZATION
Some qualitative remarks are made w.r.t. the pre- and post-critical be-
haviour of shell strcutures. These structures show a high sensitiveness w.r.t.
geometrical imperfections which often is enlarged within optimization. Three
different stability phenomena are discussed.
- Considering bifurcation, e.g. in case of straight beams and in plane
loaded plates, the critical load can be maximized using optimization
techniques. The load-deflection curve in case of bifurcation still has a
positive gradient and thus a stable equilibrium is obtainable. The post-
bifurcation path allows to enlarge the external forces without overall
collapse. The optimized structure still has this feature but with a re-
duced intensity. These structures are seen to be rather insensitive w.r.t.
geometrical imperfections and optimization does not increase the sen-
sitiveness drastically.
367
- A truss structure under compression is investigated to demonstrate

the effects occuring in case of the 'snap-through' phenomenon. The
system has a quite large hidden capacity to enlarge the external loads
in case of uniform ticknesses for each member. For an optimized struc-
ture this capacity is reduced, the overall stress distribution within the
framework changes and the critical load for 'snap-through' situations
can be increased significantly. But due to optimization the local buck-
ling of single truss members as well as the overall buckling of the whole
structure will occur simultaneously. Thus, in case of simultaneous fail-
ure of the structure as well as of several substructures an equilibrium
point can only be obtained if the external load is reduced.
- A third phenomenon can be observed in case of axially loaded cylindri-
cal shells. These structures have a large number of critical points with
corresponding post-bifurcation paths due to several different buckling
modes. The post-bifurcation path is instable by its nature and a severe
loss of load bearing capacity can be observed if pre- and post-critical
situations are compared. This effect is enlarged in case of optimiza-
tion, i.e. the critical load ca__ be maximized but simultaneously the
post-buckling minima are minimized. Thus, optimization will generate
highly sensitive structures with a low reliability w.r.t. imperfections.
Especially in case of 'snap-through' phenomena and vicious buckling
problems any optimization will enlarge the severeness of this problem, i.e.
the reliability of the structure w.r.t. imperfections is reduced. Thus, the
optimization model must include pre- and post-critical behaviour.
3.2. EXAMPLES FOR STABILITY AND OPTIMIZATION
Numerical examples will demonstrate the above mentionend phenomena

considering isotropic as well as fibre-reinforced shells. The influence of com-
posite material, i.e. fibre-angle and ply thicknesses on the overall stability
behaviour are explained.
4. Conclusion
Two important issues of structural optimization of thin-walled structures

have been discussed.
Firstly, an efficient theoretical approach to hierarchical design models
for an automated CAD orientated optimization approach has been formu-
lated. The implications on the realization and the interaction with the finite
element method have been outlined.
Secondly, important stability phenomena have been discussed in con-
junction with structural optimization regarding robust and reliable designs.
368
References
1. Stein, E. and TeBmer, J. (1998) Theory and computation of multilayer composites,
in C.A. Mota Soares (ed.), NATO ASI on Mechanics of Composite Materials and
Structures. KIuwer Academic Publishers, Dordrecht.
2. Stein, E. and TeBmer, J. (1998) Theory and computation of damage and failure
of composites, in C.A. Mota Soares (ed.), NATO ASI on Mechanics of Composite
Materials and Structures. KIuwer Academic Publishers, Dordrecht.
3. Barthold, F.-J. (1998) Remarks on linear and nonlinear variational design sensitivity
analysis, in C.A. Mota Soares (ed.), NATO Advanced Study Institute on Mechanics
of Composite Materials and Structures. KIuwer Academic Publishers, Dordrecht.
4. Schittkowski, K. (1981) The Nonlinear Programming Method of Wilson, Han and
Powell with an Augmented Lagrangian Type Line Search Function, Numerische
Mathematik 38 (I), 83-114.
5. Becker, A. (1992) Strukturoptimierung stabilitatsgefahrdeter Systeme mittels ana-
lytischer Gradientenermittlung, PhD-dissertation F 92/2, Institute for Structural
Mechanics and Computational Mechanics, University of Hannover.
6. Choi, K.K. and Chang, K.W. (1994) A study of design velocity field computation
for shape optimal design, Finite Elements in Analysis and Design 15, 317-341.
7. Braibant, V. and Fleury, C. (1986) Shape optimal design and CAD orientated for-
mulation, Engng. with Compo 1, 193-204.
8. Bletzinger, K.-U., Kimmich, A. and Ramm, E. (1991) Efficient modeling in shape
optimal design, Computing System - ~n Engineering 2, 483-496.
9. Hinton, E., Rao, N.V.R. and Ozakc;a, M. (1991) An integrated approach to struc-
tural shape optimization of linearly elastic structures. Part I: General methodology,
Computing Systems in Engineering 2 (I), 27-40.
10. Bugeda, G. and Oliver, J. (1993) A general methodology for structural shape opti-
mization problems using automatic adaptive remeshing, International Journal for
Numerical Methods in Engineering 36, 3161-3185.
11. Banichuk, N.V., Barthold, F.-J., Falk, A. and Stein, E. (1995) Mesh refinement for
shape optimization, Structural Optimization 9, 46-51.
12. Falk, A. (1994) Adaptive Verfahren fiir die Formoptimierung von Fliichentragwerken
unter Beriicksichtigung der CAD-FEM-Kopplung, Technical Report F 94/1,
Forschungs- und Seminarberichte aus dem Bereich der Mechanik der Universitiit
Hannover.
13. Maute, K. and Ramm, E. (1994) Adaptive techniques in topology optimization, 5
th AIAA/NASA/USAF/ISSMO Symposium - Multidisciplinary Analysis and Op-
timization, AIAA, pages 121-131.
14. Farin, G. (1993) Curves and Surfaces for Computer Aided Geometric Design, Aca-
demic Press, London, 3. edition.
15. Bremicker, M., Eschenauer, H. and Post, P. (1988) On a decomposition technique
within the scope of a design optimization process, in H. A. Eschenauer and G. Thier-
auf (eds.) Discretization Methods and Structural Optimization - Procedures and Ap-
plications, pages 87-94. Springer - Verlag, Berlin Heidelberg New York.
16. Beux, F. and Dervieux, A. (1994) A hierarchical approach for shape optimization,
Engineering Computations 11, 25-48.
17. Falk, A. and Barthold, F.-J. (1995) Ein hierarchisches Verfahren fiir die Formopti-
mierung, Zeitschrijt for angewandte Mathematik und Mechanik 75, S589-S590.
THEORY AND COMPUTATION OF MULTILAYER
COMPOSITES
E. STEIN AND J. TESSMER

Abstract. An hierarchical concept for the analysis of thin-walled com-

posite shells is presented within the finite element method for nonlinear
deformation of thin-wallen composite structures. For non-disturbed subdo-
mains of the structure an effective 4-node shell element with 5 or 6 d.oJ.
per node is used for the whole laminate. For the analysis of 3D-stress states
a multidirector shell element along the thickness with piecewise polynomi-
als is presented. n physical layers are approximated by N numerical layers,
discretized with hierarchical trial- a,Ild test functions. The coupling of both
elements is performed by special transition elements. An example shows
the technique and efficiency of coupling both types of elements.
1. Introd uction
Composites are typically used for light-weight structures, and in many
engineering fields there are attempts to replace components with clas-
sical materials (steel, concrete) by fiber reinforced materials. Therefore
we focus on the computation of thin-walled laminated composites. Due
to anisotropy and inhomogenity these structures show rather complicated
states of stresses and strains such that several adequate mechanical models
for thin composite structures have been developed in recent years, see [5,9].
Different FE-methods were applied with respect to special purposes
(global deformation analysis, stress analysis, first ply failure criteria, local
stress singularities, crack analysis), see e.g. [2, 8]. By admitting a priori
delaminations between plies and regarding their growth, load depending
disturbances of the perfect geometry of cross-sections are growing and di-
minish the critical load within stability analysis. We present an hierarchical
369
370
multi-layer shell model as a framework for the efficient nonlinear analysis

of delamination and failure process, focusing on basic topics of composite
theory and computation. In [10] the failure analysis for this model is shown.
For calculating a complete 3D stress state in disturbed subdomains a mul-
tidirector shell kinematic with deformation modes in thickness direction is
used and implemented in FEM. Within this kinematics, the interpolation
in thickness direction with hierarchical polynomials per layer is indepen-
dent from in plane interpolation. Since dominant parts of composite shells
usually reside in 2D stress states with little transvere shear deformation, a
conventional Reissner-Mindlin kinematic is used for regular parts. In these
areas normal stresses S33 are much smaller than normal in-plane stresses.
Therefore, S33 is neglected in the classical laminate theory. For coupling
both kinematic types within a finite element method a transition element
is applied.
2. Kinematics
The considered thin-walled composite structures consist of a layerwise build

up with n physical layers j of thickness hj and N numerical layers i which
can collect or subdivide the physical layers for numerical calculations, fig.
1 and 2
The position vector Xo of the reference surface So is parametrizised by
convected coordinates 8°. An orthonormal basis system tk (8°) is attached
to this surface where t3 is a normal vector and e 3 the coordinate in thick-
ness direction. The transformation between the different base systems is
given by tk(8°) = Ro(8°) ek where Ro is a proper orthogonal tensor.
Depending on the desired accuracy and numerical effort two different
types of kinematics are applied. One is the standard shell kinematic with
one director, yielding a 2D-FE-formulation with 5 or 6 d.oJ. per node,
and the other is a 3D-multi-director kinematic which still yields a 2D-like
finite element data structure by a priori CO-continuous hierarchical shape
functions in thickness direction, such we get the numerical complexity of a
3D-brick formulation.
2.1. INEXTENSIBLE ONE-DIRECTOR KINEMATICS
The position vectors of the reference and the current configuration of the
shell body are given by
X(8°,83) Xo(8°) + 8 3 t3(8°) , -hu S 8 3 S ho , (1)

x(8°,8 3) xo(8°) + 8 3 d(8°) ,
371
Figure 1. One-Director-Kinematic
Xo(8°) + uo(8°) ,
a3 = Rt3, IIdll = II t 311 = 1,
(2)
f (!:l{3) E 50 3 , !:l{3 = rotational vector,
v [uo; !:l{3jT .
Uo is the displacement vector of the reference surface, d the director vec-
tor, and R is an orthogonal tensor generated by application of the Euler-
Rodriguez-formula on the rotational vector !:l{3, see e.g. [1].
The Green-Lagrangian strain tensor E = EijG i ® Gj is formulated
with respect to the tangential base vectors Gk = X,k = aX/ a8 k for the
in extensible thinwalled structure.
Eij = ~(X'i X,j -X,i X,j) (3)
with the approximations
co{3 O.5(x,0 x,{3 -X,o X,{3) ,
/0 2e03 = (xo,o . d - Xo,o . t3) , (4)
K,0{3 O.5[(xo,0 . d,{3 +d,o ·xo,{3) - (Xo,o . t 3,{3 +t 3 ,0 ·X o,{3)] .
2.2. EXTENSIBLE MULTIDIRECTOR KINEMATICS
For the description of 3D-stress-states an extended multidirector kinematic

is used. The position vectors of the reference and the current configuration
of the shell body are given by
X(8°, ( 3) Xo(8° ) + 8 3 t3(8°)
x(8°,8 ) = X(8°,8 ) + u(80, ( 3) .
3 3
372
Figure 2. Multi-Director-Kinematic for N numerical layers
A multiplicative decomposition of u is introduced as

k
u(8 a ,83 ) = E iil(8 a ) wI(8 3 ) = (>(8 3 ) ii(8 a ) ,
1=1
ii(8°) = liib ii2,..... , Um) ... ,-Us+ll Us+:".'" Us+m)"., UN(m-1)+lY (6)
Ii 1 (layer 1) Ii i (layer i)
where k is the number of interpolation points over the thickness, WI are

shape function (piecewise linear for k = N + 1), s = (i - 1) (m - 1) is a
layer index, and m is the polynomial degree, (2 ::; m ::; 4). In thickness
direction hierarchical shape functions are defined for each numerical layer
i as (>i(8 3 ) •
ui (8°,8 3 ) = (>i(8 3 ) ui (8°) , iii(8°) = [iiL ii~, ,ii~Y,
,4>~1],
4>i ((i) = 0.5(1- (i) , 4>1((i) = 1- (i2, (i = 28 3 jh i

(7)
4>~((i) = 0.5(1 + (i) , 4>~ ((i) = (1 _ (i2)(i .
This leads to an a priori discretization in thickness direction within the
analytical shell theory, analogous to the analytical thickness integration of
stress resultants in classical shell models. These shape functions are inde-
pendent from the in-plane trial-functions, see sec. 5.2. For linear plates see
[8, 9]. A layerwise linear interpolation is realized by using 4>i and 4>~ only.
In contrast to eq.(4) the Green-Lagrangian strain tensor is computed
and applied in its complete form, eq.(3).
373
3. Material Equations on Macro Level

3D-material equations are formulated for each physical layer of the com-
posite laminate. These have to be transformed from the fiberoriented basis
to the local basis of the shell, see [5]. In case of standard shell kinematics
they are integrated a priori over the thickness of the overall thickness.
3.1. UNI-DIRECTIONAL TRANSVERSAL ISOTROPIC LAYERS
The constitutive equations for unidirectional (UD) layers oflaminated struc-

tures are derived under the assumption of small strains and the validity of
St. Venant Kirchhoff material equations with linearized strains.
Hence, the free energy W(E) = ~E . CE is formulated as a quadratic
function of E. The material tensor C is constant with respect to the in-
variants trE, tr(E2), ao' Eao, ao· E2ao. Using the elastic parameters
A, I-"T, I-"L, a, (3, the free energy W, the stresses S and the constant material
tensor C follow as
W(E) + +
~A(trE)2 I-"Ttr~~2) a(ao . Eao)trE
+
+2(I-"L - I-"T)(aO' E2ao) ~(3(ao' EaO)2, (8)
aW(E) a2W~E)
with S= aE = C . E, C= aE .
For UD-Iayers with the fiber direction ao = el = (1,0,0) and by compu-

tation of CC- 1 = 1 the components of the compliance matrix C- 1 result
with the common elasticity constants in
I/E1 -V12/E l -vld E I 0 0 0
-vldEI I/E2 -V23/ E2 0 0 0
C- 1 -vldEI -v23/E2 I/E2 0 0 0
0 0 0
(9)
I/G I2 0 0
0 0 0 0 I/G I2 0
0 0 0 0 0 I/G 23
3.2. HYPERELASTIC ISOTROPIC INTERMEDIATE LAYERS
For composite structures with macroscopicly seperated parts of anisotropic

and isotropic material, e.g. tires, also a hyperelastic isotropic material equa-
tion is formulated with respect to the invariants of the Cauchy-Green tensor
C = FTF = 2E + 1. The stored energy by 'Blatz & Ko', [7], yields
1
W = 21-"0[J(J1 - 3) + (1 - f)(h - 3)+
1 2 ~ (10)
1) + (1- f)(J;-2I1Q
-211Q
( - vO ){f(J;-2I1Q - - I))],
Vo
374
with J 1 = II, h = hi h, h = VI; = J ;

II = trC, h = O.5[(trC)2 - trC 2] , 13 = detC .
The 2. Piola-stress tensor S and the material tensor C follow as
8W 8
S 8E = 2 8C W(ft (C), h(C), h(C)) ,
(11)
82
4 8C2 W(ft (C), /2(C), h(C)).
===> S JLo[j + ~~ (1 - 1)]1 - ~~ (1 - J)C + JLofltC- 1 ,

8C-l
2~~ (1- 1)(1 ® 1 - I) + 2JL0,81 ac
2vo h (12)
2vo,82 + 2JLo J2 (1- I)]C- ® C-
1 1
+ [JLO 1 _
+ 2~~ (1 - I)(C- 1 ® C + C ® C- 1 )
- 2ft ~~ (1 - I)(C- 1 ®1 + 1 ® C- 1 ) ,
with ,81 -IJI-2f.'O

~
+ (1-I)JI-2f.'O
~
-J2 (1-J)
-2~ 2~
,82 IJI-2f.'O +(1_J)JI-2f.'O

8C- 1
-C- 1 11C- 1 .
8C
3.3. RESULTANT FORMULATION FOR ONE-DIRECTOR SHELL
Since the material equations for one-director shells are formulated in resul-
tants of stresses and strains, an adequate transformation and integration
of the material law, eq.(9), is conducted. With the assumption of 8 33 = 0
the matrix C 6X6 can be statically kondensed to C~X5 • Such, we get for the
fiberorientated basis
SR
F
tnR ER
= 'lUFF . h tnR
WIt 'IU F = [C~
0 C;0 1
_'
C~l1 = Et!(1- v2~), C~13 = 0, eftl = G 12 ,
C~12 = E 2vl(1- v2~), C~23 = 0, eft2 = 0, (13)
C~22 = E2/(1- v2~), C~33 = G 12 , C~2 = G 23
SR = [8 11 ; 8 22 ; 8 12 ; 8 13 ; 8 23], ER = [E l1 ; E 22 ; 2E12 ; 2E13 ; 2E23] .
For the integration over the thickness all matrices for each layer j must be
transformed to the local basis of the reference surface, see [4].
C iR
L -
- tiT CRti
F
-- [C~,
0 0.
CJ
s
1 (14)
375
with S [NIL, N22., N 12 ,. M11., M 22., M 12 ". Q1. Q2]T ,

E [E11; E22; 2E12; i'i:11; i'i:22; 2i'i:12; '11; '12f
~j
.
ctn
hj ,
h3
Dmb ~j ctn (zsh)j ,
+ 12-)1 ,
. . .
~j c~ (z;h Ds ~j C~hJ
4. Variational Equations
In this section the weak form of equilibrium (principle of virtual work) is

given. Linearization yields the consistent tangential operator.
4.1. ONE-DIRECTOR FORJvllJLATION
The principle of virtual work is ,,;"en in the material description for the
inextensible shell in thickness direction and external loads f at the reference
surface, using (15)
G(u,11) = f S· 8E dV - f f· 11 df (J = 0
(V) (ro-)
= f (N ex /3 8cex/3 + M ex /38i'i: ex /3 + Qex8'1ex)dfl - G ext (f, 11) (16)
(n)
f S· 8E dfl - G ext (f,11).
(n)
For constant material matrix t and conservative loads the linearization
(Gateaux-derivative) of (16) yields
DG(U,11) ·~u= f 8E·t·~Edfl+ f S~8Edfl. (17)
(n) (n)
4.2. MULTIDIRECTOR FORMULATION
The laminate structure is loaded at the bottom- and top-surface fl(J with
i> = fik ek . Hence, the virtual work reads in index notation
G(U,11) =f [f Skl8Eki Jd8 3 ] d8 2 d8 1 - f if 'f/k dfl(J = 0, (18)
(n) (8 3 ) (no-)
with J(8 i ) = (Xll XX,2 ),X,3. Note the special splitting of the volume inte-
gral in two parts. Through this split it is possible to apply the same interpo-
lation functions with respect to the reference surface for the multi-director
formulation and for the one-director kinematics. Linearization yields
DG(U, 11) . ~u = f [f (8Eklcklmn ~Emn + Skl~8Ekl) Jd8 3 ] dfl. (19)
(n) (8 3 )
376
5. Finite Element Discretization

The four node quadrilateral Q1-element with isoparametric bilinear shape
functions is used in the reference surface for all kinematic quantities. For
one-director kinematic shell elements with 5 or 6 d.o.f. per node and for
multi-director kinematics shell elements with more than 6 d.o.f. per node
are applied. For coupling both types of finite elements a special transition
element is used.
5.1. ONE-DIRECTOR ELEMENT
The approximation of geometry and displacements reads

4 4
X Oh = I: NK(~,'fJ)XOK, and vh = I: NK(~,'fJ)VK' (20)
K=l K=l
where VK = [uo; d,8]I is the nodal displacement vector; it consists of the
displacement components of the reference surface and the increment of the
rotational vector. Depending on the place of the node (within flat areas
or at intersections) the rotational . xtor is parametrisized with respect to
the local basis (5 d.o.f.) or the global basis (6 d.o.f.). Following eqs.(1)-(3),
all necessary kinematic values are computed. With the differential operator
matrix B, see [5], we get
_ T 4
[oe,OK-,o'Y] =I:BKOVK. (21)
K=l
Applying the virtual work principle (16) and linearization (17) yields
4 T TA A
G(v, OV) = I: OVK OK, with OK = J (BKS - NKt)dQ,
K=l 4 4 ne
DO(v, OV)dv = I: I: 8vkKKLt:.VL , (22)
K=lL=l
with KKL = J (BICBL + OKL)dQ.
ne
5.2. MULTIDIRECTOR ELEMENT
One major advantage of the multi director formulation in contrast to a

standard 3D-finite element is the 2D-data structure with bilinear shape
functions with respect to the reference surface, which allows for an easy
coupling with standard shell elements. The approximation of the geometry
and the displacement field of layer i, eqs. (7)-(5), follows from
(23)
377
The number of components for UK is 3(N + 1) + 3N(m - 2). Again, with

4
the differential operator B-matrix, see [4], aE = 2: BK aUK. Putting into
K=l
(18) and (19) yields
4
G(u, "1) = L JUk GJ{ ,
f d= J f
K=l
with GK = Bk SJ~ d(j] dQ e - NKPJ dQea,

(fle) ]=1 (8 3 ) (flea-)
nel nel (24)
DG(u, T])tw = L L JUk KKL ~UL,
K=1 L=1
with KKL = ft !
(fle) )=1 (83)
(Bk Cj BL + Gj{L) J~ d(j dQ e .
5.3. TRANSITION ELEMENT
For coupling standard one-director with multi-dirctor shells a transition

element is used. To prevent sensible disturbances of the 3D-stress state
each layer i has to be allowed for a constant thickness strains E~3 at the
multi-director side of the transition element and CO~continuity only in the
mid-surface at the other one-director side, see [6]. Such a thickness jump
of the deformed laminate is admitted there.
5.4. SHEAR-LOCKING
All presented elements use a special interpolation of transverse shear strains

to prevent "shear-locking", namely an assumed natural strain (ANS) method,
see [3],
l' = [ "f( ] = ~ [ (1 - 17h(B + (1 + 1]h(D ] (25)
"f7J 2 (1 - ~h7JA + (1 + ~h7]c .
In this formulation shear strains in the mid-side nodes (M = A, B, C, D)
are com pu ted by the standard bilinear shape functions N K (~, 1]) of the
element.
6. Example
Steel-Cord-Reinforced Rubber Beam under vertical load.

Cross-sec.: width/height = 100/20; Layersequenz [iso/20° /iso/- 20 0 /iso].
378
50 31M) 100
Element Equat. CPU
I I
1. One-Dir .-Shell 720 1.3
..
2. Multi-Dir.-Shell, N=15 6912 135.5
test'\pecimen [1]=1111111
3. Coupling 1. and 2. 2925 39.8

Figure 3. Composite-Beam in A and B
-20
j
-40
:;:
....
"
Q) -60
"
Q)
u
-80
~
0-
Ul
·rl
-100
"
-120
Figure 4. Displacement curves

References
1. Betsch, P., Menzel, A., and Stein, E. (1998) On the Parametrization of Finite Ro-
tations in Computational Mechanics. A Classification of Concepts with Application
to Smooth Shells, Compo Meth. Appl. Mech. Engrg., 155, 273-305.
2. Cochelin, B., Damil, N., and Potier-Ferry, M. (1994) Asymptotic-numerical methods
and pade approximants for non-linear elastic structures, Int. J. for Numerical Meth.
in Engrg., 37, 1137-1213.
3. Dvorkin, E. and Bathe, K-J. (1984) A Continuum Mechanics Based Four-Node
Shell Element for General Nonlinear Analysis, Eng. Comput., 1, 77-88.
4. Gruttmann, F. (1996), Theorie und Numerik diinnwandiger Faserverbundstruk-
turen, Habilitation, Bericht-Nr. F 96/1, Institut fiir Baumechanik und Numerische
Mechanik der Universitiit Hannover.
5. Gruttmann, F. and Wagner, W. (1994) On the numerical analysis of local effects in
composite structures, Composite Structures, 29, 1-12.
6. Gruttmann, F. and Wagner, W. (1996) Coupling of 2D- and 3D-composite shell
elements in linear and nonlinear applications, Comput. Methods Appl. Mech. Engrg.,
129, 271-278.
7. Levinson, M. and Burgess, I. (1971) A comparison of some simple constitutive rela-
tions for slightly compressible rubber-like materials, Int. J. Mech. Sci., 13, 563-572.
8. Reddy, J. (1987) Generalization of two-dimensional theories of laminated composite
plates, Communications in Applied Numerical Methods, 3, 173-180. ,
9. Robbins, D. and Reddy, J. (1993) Modeling of thick composites using ~ layerwise
laminate theory, Int. J. Numer. Methods Eng., 36, 655-677. '
10. Stein, E. and TeBmer, J. (1998) Theory and Computation of Damage and Failure
of Composites, in C. A. Mota Soares (ed.), NATO ASI on Mechanics of Composite
Materials and Structures, Kluwer, Dordrecht.
THEORY AND COMPUTATION OF DAMAGE AND
FAILURE OF COMPOSITES
E. STEIN AND J. TESSMER

Appelstrafte 9A, D-30167 Hannover, Germany
Abstract. Following the theoretical and numerical aspects given in [10]

the formulation for the multidirector shell element is extended for the de-
scription of failure, esppr;ally the onset and propagation of delaminations
in thin-walled composite structures. Therefore, stress based failure criteria
are presented as a necessary condition for initiation of delaminations and
energy based criteria as a sufficient condition. For the kinematic expansion
of the applied shell model for delaminated zones, additional degrees of free-
dom are introduced. They are Hut adjoined in advance to a distinct interface
of the delamination process. An algorithm for the onset and propagation
of delaminations is presented based on a virtual crack closure technique.
Examples show the correctness and effective usability of this algorithm.
1. Introduction
Consideration of damage and failure in composite structures is of greater

importance than for other engineering materials. The in general good-
minded properties of classical materials, like plastification of steel, are not
valid for these applications due to the brittle behavior of composites. There-
fore, structures have to be constructed regarding these failure mechanisms.
Special strategies for theory and computation of composites have to be
developed since incompatible material properties lead to 3D stress states
at free edges and cut outs with singularities, predicted by purely elastic
theories. Usually the global deformation and stress analysis is calculated
with different mechanical models from crack analysis in subdomains. Yet
for some types of cracks, especially delaminations, a strong coupling of lo-
cal failure and global load capacity is aparent. Therefore, we present failure
379
380
criteria within a global mechanical model of a structure and show the com-
putation of onset and growing of delaminations in thin-walled composite
structures. Other phenomena like fiber and matrix cracking are not con-
sidered here. The basic computational model for deformation and stress
analysis is described in [10]. Based on a multidirector shell element with
deformation modes in thickness direction an extended kinematics for delam-
inations is introduced through additional d.oJ. 'so Onset and propagation of
delaminations are described by necessary stress based and sufficient energy
based criteria. For applications within the finite element method a robust
algorithm for onset and propagation of delaminations is presented.
2. Physical and mathematical modeling of delaminations

For the physical and mathematical modelling of failure mechanics we sug-
gest a central criterion as
F = f((1, g) (1)
which means that every crack development is based on a local stress field
and the potential energy release Vu,,,les. All common concepts of failure cri-
teria fit into this central formulation. In special fields of interest only parts
of this criterion are considered, e.g. in [11]. For example, in the analysis of
complex laminate structures, only stress and strain based criteria are ap-
plied to predict first ply failure cases, and in fracture mechanical approaches
to local crack phenomena, only energy based criteria are applied. Yet, for
the application of computational algorithms for whole composite structures
with consideration of delaminations, stress and energy based criteria have
to be fullfilled. Therefore we describe a stress criterion as necessary con-
dition for initiation of failure and an energy based criterion as sufficient
condition for propagating failure.
In the following computations of delaminations we restrict to
- interpolation of the displacement field in thickness direction of the
laminate with piecewise linear shape functions, see [10]
- small strains
- delaminations arise in one layer boundary only.
2.1. STRESS BASED FAILURE CRITERIA
As a necessary stress criterion we use the formulation of HASH IN [3] for

F
interlaminar stresses u = [8 13 ,8 23 , 8 33
(8 13 )2 + (8 23 )2 (833 )2
F(u) = R2 + - p - 1, (2)
o 0
with Zo and Ro as normal strength and shear strength in thickness direction
of the laminate. To get the interlaminar stresses the Gauss point values
381
have to be interpolated to the layer boundaries. This is done under the

assumption that (T is CO-continuous along the thickness direction, see [2].
The following algorithm checks the stress criterion at the layer bound-
aries:
- Loop over all layer boundaries (i = 1, n)
• Interpolation of interlaminar stresses at nodes K
uk = II (u i )
• Selection of maxL.um value of all boundaries

FK(uk) = mil-x
l<,<n
{F(ukn
End of loop over all boundaries
- Computation of set K by checking the stress criterion for boundary 1
SO -+ KgK
FK(uk) { (3)
>0 -+ KEK
2.2. ENERGY BASED FAILURE CRITERIA
One major drawback of stress criteria within elastic computations results

from the singularities, which occur e.g. at free edges of composites, crack
tips, cut outs and holes. By extension of the physical modell through intro-
duction of inelastic deformations, e.g. with damage formulations, stresses
can be restricted effectively, see [5]. These approaches are efficient for pre-
dicting crack areas but make problems by describing great crack opening
displacements as they occur when sublaminates buckle in stability analysis.
Another possibility to overcome the problem of stress singularities is the
formulation of additional energy based criteria.
We choose the energy release rate as primary energy based sufficient
failure criterion for areas where the stress state exceeds the stress criterion.
Besides energy release rates also stress intensity factors or the J-integral are
frequently used to calculate fracture criteria. In contrast to the J-integral,
seperation of modes (I, II and III) is possible by energy release rates (err),
see [6, 7, 8]. In comparison to stress intensity factors (sif) energy release
rates or J-int. (as energy integrals) are approximated by FEM one order
better than stresses. Such, for the choice of method, namely err, J-int.
382
or sif, the computational tool (here the FEM) and the structure of the
problem (here shells) is most important, but within the theory of linear
elastic fracture mechanics all methods can be transformed into each other.
The energy release rate is defined as
g = - :: 7r = elastic potential of the structure
(4)
A = crack area.
The critical value go for propagating cracks has to be determined by
experiments as a material parameter. In the numerical framework of a vir-
tual crack closure method (VCCM) the energy release rate is computed on
a finite crack area ~A as
.6.7r 1
g=-- with ~7r=--pu·g. (5)
~A 2
For computation of ~7r the crack opening displacement vector g and
the nodal force vector P u are determined, see fig. 1. Puis calculated as
hi
Pu
epnK
=~
e==l (ne)
J [~ JiR
J==l (83)
BkSJ2d(i]dne, (6)
where epnK is the number of elements at node I<, and jR is the layer number
beneath the delamination. The area ~A can be denoted as the partial node
area ~AK which 'belong' to node I<. Defining nel as the number of nodes
per element yields
epnJ(
~AK = L .6. AKe , ~Ae = elem. area part.(7)

e==l
K K
Figure 1. Crack opening displacement vector and topology change at the onset
of crack
3. Extended Multidirector Kinematics

Once the fracture criterion is fullfilled the geometric boundary conditions
change. Thus the nodal displacement vector is extended by three additional
383
parameters which allow to describe crack opening displacements. The vec-

tor of crack opening displacement gK is introduced as the difference of
displacements of adjacent nodes k and k + 1 for an assumed crack at the
boundary between layers i and i + 1
g(X) := Uk+l - uk, (g(X) == 0 without crack) (8)
The extended nodal displacement vector for layer i of the multi director
shell element, see fig. 2, reads
4
ui = L N]{(~, 1])~k UK,
[1UK"'" UK"'" k+ 1]
K=l
.t h 'UK = k n
(9)
WI UK' UK '
an d k+1
UK = uK'
k'f
I gK =0 .
CD ®
Figure 2. Displacement vector in a cracked element by delamination
4. FE-Algorithm for Onset and Growth of Delamination

In the following section we propose a robust algorithm for the computation
of onset and propagation of delaminations within the standard non-linear
static analysis of composite shell structures. It starts out as a system with-
out cracks or prescribed 'old' cracks. During the incremental increasing of
load the stress criterion is checked at all layer boundaries. If this criterion
384
is fullfilled, another iteration for the calculation of energy release rates is

initiated. The algorithm is outlined in figure 3.
g K = 0 : initial system without crack
p =p + .1.P : next load increment

O(u, 8U) = 0 : nonlin. static analysis
no
Compute g in nodes K
81t K= 0.5 P "Kg K
...... 11.. . .
Figure 3. Flow chart of the crack propagation algorithm
After computing ~7r, three different cases have to be distinguished. For

all cases the sets JC and JC* are defined as
JC {All nodal points K within a trial crack area},

determined by eq. (3). (10)
JC* {K IKE JC 1\ YK = ~AI< ~7rK >_ YC } JC* c JC •
case 1: The energy criterion is not fullfilled for any node of the trial crack area.
The trial crack is closed and the structure remains unchanged for the
next load step
YK < gc V (K E /C) ==} /C* = 0. (11)
case 2: The energy criterion is fullfilled for every node of the trial crack area.
The boundary conditions are definitely changed and the stress criterion
has to be checked again with the new topology.
gk ~ gc V (K E /C) ==} /C* = /C. (12)
ri ~ rt , the crack produces a new surface.
case 3: The energy criterion is fullfilled for some nodes of the trial crack area.
{ :3 K E /C I gK ~ go} ==} /C* c /C. (13)
385
Global Energy Release condition (GER-condition):
If ""' Ll7r"K > fie set I\,

L...J LlA V"*
= I\,
V"
--+ proceed wit h Step 2,
K K
else close nodes K E {K \ K*} and compute the energy release rate for
the remaining trial crack area.
The GER-condition in eq.(13) is necessary to get mesh independency. It
extends the local criterion at node K by a global criterion with respect to
the whole crack surface.
5. Examples
5.1. TENSION STRIP
A tension strip under prescribed strains is shown in fig. 4, and 9, see also
[1]. The cross section has a width/hight-ratio of w/h = 20 and the fiber
orientation is [0 0 /90 0 /90 0 / u-]. Each physical layer is devided into 2 nu-
merical layers for computation. The stress curve for 8 33 shows the typical
singularity responsible for edge dt ....minations. It is also shown that for the
application of the stress criterion a proper coupling of different element
types is important.
Figure 4. Laminate strip under uniform strain with edge delamination
5.2. GLULAM BEAM
We consider a curved Glulam beam under prescribed deflection w of the

apex as shown in fig. 5. For the numerical analysis one half of the system
is discretized due to symmetry. We apply the deflection in 20 load steps.
Results are presented in fig. 6, 7 and 8. The load-deflection curve F{w),
see fig. 6, shows the influence of growing cracks which stops at point E. Six
386
meshes with 10 numerical layers are discretized (mesh 1 (10 elements) to

mesh 6 (320elements)). A comparison of fig. 7 and 8 shows the necessity of
the GER-contidion in eq. (13) to get mesh independency.
E.= llOOO MN I m**2

E t= 300 MN I m**2
G = 500 MN I m**2
V= 0,3
Zo= 0,2 MN I m**2
Ro= 0,9 MN I m**2
G c= 0,001 MN I m
. . cY I,Om
1I,1m
tO,Om
++
Figure 5, System of a curved Glulam Beam with material data, displacement con-
trolled loading at apex, The crack begins at point P and grows in both directions
until the points Q and R.
Glulam Beam
-0,3
Mesh 6~ without crack.-
Mesn 6, wi th crack -+-_.
Mesh 6, energy release rate ·B··
-0.25
-0,2
~
r.. -0.15
'0
111
0
..:I -0.1
-0.05
0
0 -0.05 -0.1 -0.15 -0.2 -0.25 -0,3
w(s=O) 1m]
Figure 6. Load-Deflection curve for GIlliam Beam without crack, with prescribed
crack and with calculated crack-propagation until the points Q and R in fig, 5
which corresponds to point E,
387
Glulam Beam
o
~ -0.15
~
§
'r!
t
Q)
-0.1
rl
~
Q)
't:l
't:l
1l
.r!
-0.05
1-1
U
III
Q)
1-1
0.
°0~----~1~----~2------~3------~4-------5~----~6
crack lenght 1 [rn)
Figure 7. Energy-Release-Rate-algorithm for Glulam Beam with GER-condition

in (13)
Glulam Beam
-0.2.------.-~------nr---,-.------~,--,--.------.
+
+ .;~ *~ ~
~
Mesh
Mesh
Mesh
1 --
2 .+---
3 'B"
+ ;.
I
~:Ie Mesh 4 -M-----
o
+ ...i. j< .,/
i
Mesh 5 ..... -
+ t
I .
~ -0.15 ~/ ¥~
~ t ~ * t
_________________.-."4.~,, --'-",:-r---
A- til ~_~
~
o ____________________ t-r +~,<~:~
'r! ._-_._-_.---------- ill
t
Q)
-0.1 : If.i
rl .., ..+ ......._..................................~Jb.-...... -K
~
~.::~;;;~:~:~:~:~::~ ______________ • _____ -El ...
.gj
't:l
1l
.r!
-0.05
1-1
U
III
Q)
1-1
0.
o~----~------~------~------~----~------~
012 3 4 5 6
crack lenght 1 [rn)
Figure 8. Energy-Release-Rate-algorithm for Glulam Beam without

GER-condition in (13)
388
MPa 30.0
I
I I
I I
20.0 ------,------- ------,-------

I I
I I
rr----
I I
I I
I I
I I
I I
______ .J ______ _
10.0 I
I
I
I
I
I
-
I
0.0 ~---
I I
Without Coupling
Coupling with E33 = 0
Coupling with E33 = const
-10.0
0.0 5.0 10.0 15.0 20.0 mm
y
Figure 9. Laminate strip with edge singularity du to incompatibility of deformation
References
1. Gruttmann, F. and Wagner, W. (1996) Coupling of 2D- and 3D-composite shell
elements in linear and nonlinear applications, Comput. Methods Appl. Mech. Engrg.,
129, 271-278.
2. Gruttmann, F. and Wagner, W. (1996) Delamination Analysis of Thin Composite
Structures Using a Multi- Director Formulation, in B. H. V. Topping (ed.), Advances
in Analysis and Design of Composites, CIVIL-COMP Ltd, Edinburgh, pages 51-59.
3. Hashin, Z. (1980) Failure criteria for unidirectional fiber composites, J. Appl. Mech.,
47, 329-334.
4. Hellen, T. (1975) On the method of virtual crack extensions, Int. J. for Numerical
Methods in Engineering., 9, 187-207.
5. Kroplin, B. and Weihe, S. (1997) Constitutive and Geometrical Aspects of Fracture
Induced Anisotropy, in Owen et. al. (ed.), Computational Plasticity, Fundamentals
and Applications, Barcelona, pages 255-279.
6. Kruger, R. (1996), Delaminationswachstum in Faserverbundlaminaten, Bericht
Nr.13-96 d. Instituts fiir Statik und Dynamik der Luft- und Raumfahrtkonstruktio-
nen, Universitiit Stuttgart.
7. Rikards, R., Buchholz, F.-G., and Wang, H. (1995) Finite element analysis of delam-
ination cracks in bending of cross-ply laminates, Mechanic of composite materials
and structures, 2, 281-294.
8. Rybicki, E. F. and Kanninen, M. F. (1977) A Finite Element Calculation of Stress
Intensity Factors by a Modified Crack Closure Integral, Engineering Fracture Me-
chanics, 9, 931-938.
9. Stein, E. and TeBmer, J. (1997) Delamination, instability and failure of multilay-
ered composites, in M. Predeleanu and P. Gilormini (eds.), Advanced Methods in
Materials Processing Defects, Laboratoire de Mecanique et Technologie, Elsevier,
Amsterdam, pages 405-414.
10. Stein, E. and TeBmer, J. (1998) Theory and Computation of Multilayered Compos-
ites, in C. A. Mota Soares (ed.), NATO ASI on Mechanics of Composite Materials
and Structures, Kluwer, Dordrecht.
11. Tsai, S. (1988) Composites Design, Think Composites, Dayton, Ohio.
OPTIMAL DESIGN OF COMPOSITE STRUCTURES WITH INTEGRATED
PIEZOELECTRIC LAMINAE
VICTOR M. FRANCO CORREIA", CRISTOVAO M. MOTA SOARESb

and CARLOS A. MOTA SOARES b
a ENIDH - Escola Nciutica Infante D. Henrique
Av. Eng. Bonneville Franco, 2780 Pafo de Arcos, Portugal
b IDMEC _ Instituto de Engenharia Meclinica - Instituto Superior Tecnico
Abstract
A family of nine-node Lagrangian finite element models based on higher order

displacement fields is applied to the optimal design of laminated composite plates with
embedded and/or surface bonded piezoelectric layers and/or patches. The described
formulation introduces, for the electric behavior, one electric potential degree of
freedom for each piezoelectric layer in each element. In the optimal design, the design
variables are the ply orientation angles III the orthotropic layers, the ply thickness of
both the substrate and piezoelectric layers and the electrical potentials applied to the
piezoelectric actuators. The design objectives are the minimization of structural weight
and/or the minimization of specified displacements subjected to behavioral constraints.
1. Introduction
Developments in design and manufacturing technologies have increased in the recent

years the use of advanced reinforced composite materials with embedded and/or
surface bonded piezoelectric layers and/or patches. The piezoelectric materials have
the property of converting mechanical energy in electrical energy and vice-versa. An
electric charge can be generated under mechanical loads or imposed deformation.
Applying an electrical field to the material results in mechanical strains or stresses.
Hence these materials can be used as sensors and actuators on advanced structural
applications where the adaptive capabilities of the laminate are required.
Accurate and efficient structural analysis, design sensitivity analysis and optimization
procedures are very important for the design of this type of structures to obtain
important weight reductions or to efficiently control the response of the structure.
Most of the developed models for laminated composite plates are based in classical
laminate plate and first order shear deformation theories [1]. It has been demonstrated
that these theories can lead to substantial errors in the prediction of stresses of highly
389
390
anisotropic and/or moderately thick composite plates [2-4]. A similar behavior has
been observed for composite plates with embedded or surface bonded piezoelectric
layers [5,6].
Models based on higher order displacement fields involving higher order expansions of
the displacement field in powers of the thickness coordinate [7,8] are able to accurately
account for the effects of transverse shear deformation yielding quadratic variation of
out-of-plane strains, not requiring the use of artificial shear correction factors, and are
suitable for the analysis of highly anisotropic plates ranging from high to low length-
to-thickness ratios. Critical reviews of higher order models applied to the analysis of
laminated structures have been carried out by several researchers [3-4,9-10).
Crawley [11] presented an overview of the technology leading to the development of
intelligent structures incorporating actuators and sensors for the aerospace industry.
An extensive review of 128 papers is presented. Variational methods and finite element
models for piezoelectric solids have been reported by Allik and Hughes [12]. Tzou and
Tseng [13] presented a finite element formulation for plates and shells containing
integrated distributed piezoelectric sensors and actuators applied to measurement and
control of advanced structures. Ha et al [14] developed an eight-node three-
dimensional composite brick finite element for modeling the dynamic and static
response of laminated composites , .......taining distributed piezoelectric ceramics
subjected to mechanical and electrical loadings. The electrical potential is taken as a
nodal degree of freedom, leading to an element with four degrees of freedom per node.
Suleman and Venkayya [15] presented a simple finite element formulation for
laminated composite plates with piezoelectric layers. The finite element is based in a
four-node quadrangular Mindlin plate/shell element with one additional degree of
freedom per each piezoelectric layer. Chattopadhyay and Seeley [5] used a finite
element model based on a refined higher order theory to analyze piezoelectric materials
surface bonded or embedded in composite laminates. The displacement field accounts
for transverse shear stresses through the thickness and satisfies the boundary
conditions at the free surfaces. Through numerical examples they have shown that the
refined theory captures important higher order effects that are not modeled by the
classical laminate theory which give significant errors. Further works using higher-
order theories for piezoelectric laminates can be found in Mitchell and Reddy [18] and
Jonnalagadda et al [19], among others.
Xu et al [20] presented analytical three-dimensional solutions for the coupled
thermoelectroelastic response of multilayered hybrid rectangular composite plates
containing fiber-reinforced cross-ply and piezothermoelastic layers. The analytic
sensitivities of response quantities with respect to material, lamination and geometric
parameters of the plate are addressed. Donthireddy and Chandrashekhara [21] studied
the influence of boundary conditions, ply orientation and actuator voltage on the
change in shapes of laminated beams with piezoelectric actuators, by using a one-
dimensional finite element beam formulation. The effect of piezo-actuator thickness
on the vibration control of a cantilever beam was investigated by Kim and Jones [22],
using the Euler-Bernoulli beam tlteory. The optimal thickness that maximizes the
static beam deflection was analyzed. Meric and Saigal [23] derived shape sensitivity
391
expressions for linear piezoelectric structures with coupled mechanical and elastic
field. The adjoint variable method of optimization and the material derivative
formulation of shape variation were used in a systematic procedure to evaluate the total
variation of a general performance criterion with respect to shape variation.
Recently, Silva et al [24] proposed an optimal design method of piezocomposite
microstructures using topology optimization techniques and homogenization theory.
The objective is to find the distribution of material phase and void phase in a periodic
unit cell, that optimizes the performance characteristics, subject to constraints such as
property symmetry and stiffness. Batra and Liang [25] used a three-dimensional linear
theory of elasticity to find the optimal location of an actuator on a simply-supported
rectangular laminated plate with embedded PZT layers, that will result in the
maximum out-of-plane displacement for a given distribution of the applied voltage.
The optimal design is obtained by fixing the applied voltage and the size of the
actuator and moving it around in order to find the maximum out-of-plane
displacement. No gradient information is used for the optimization process. Liang et
al [26] proposed a model for the optimization of the induced-strain actuator location
and configuration for active vibration control based on the actuator power factor that
guarantees the highest energy efficiency for single frequency and broad-band
applications. Kang et al [27] iD,,,,.,tigated the optimization of the sensor/actuator
placement for structural vibration control of laminated composite plates. The damping
and stiffness of the adhesive layer and the piezoceramics are taken into account in the
finite element formulation and the effects of the outer-layer fiber orientation of the
laminated plate on the optimum sensor/actuator placement are investigated
numerically and experimentally. Leeks and Weisshaar [28] studied a self-strained
actuator mounted on one side of a panel to find the features of the actuator that
produces the largest deflection of simply-supported rectangular panels using a
Rayleigh-Ritz model associated to classical plate theory.
It has been shown [4,29-30] through numerical illustrative applications that higher
order models are able to accurately predict the behavior of highly anisotropic andlor
low length-to-thickness ratio laminated plates with reasonable advantage over first
order models. Additionally, classical laminate models (CPT) are unable to predict the
behavior of highly anisotropic andlor low length-to-thickness ratio laminated plates.
Also the use of first order theory models in optimization problems with first ply failure
criterion can lead to unsafe designs. In the following sections a family of CO nine-node
Lagrangian finite element models based in high-order and first-order displacement
fields are applied to the design sensitivity analysis and optimal design of laminated
composite plates with embedded andlor surface bonded piezoelectric layers andlor
patches. The assumed design variables are the fiber orientation angles in the
orthotropic layers, the ply thickness of both the substrate and piezoelectric layers and
the electrical potentials applied to the piezoelectric actuators. The design objectives are
the minimization or maximization of generalized displacement components andlor the
minimization of structural mass subjected to behavioral constraints. The sensitivities of
the response functions with respect to the design variables are evaluated analytically.
392
A few examples are presented in section 6, to illustrate the perfonnance of the

proposed models.
2. Displacement field and constitutive equations
In order to approximate the three-dimensional elasticity problem to a two-dimensional

laminate problem, the displacement components u, v and w at any point in the laminate
space in the x, y and z directions, respectively, are expanded in a Taylor's series
powers of the thickness coordinate z. Each component is a function of x, y, z and t,
where t is the time. The following higher order displacement field can be written
[3,4,7,8]
2 * 3 *
u=uo+z8 x +z uo+z q>x
V=Vo +z8 y +z 2 Vo* +z3 q>y

* (1)
w = wo + Z q> Z + Z 2 wo*
where uo,vo,wo are the displacements of a generic point on the reference surface,
8 x,8 y are the rotations of normal to the reference surface about the y and x axes
respectively and q>z,u~,v~,w~,q>:,q>~ are the higher order terms in the Taylor's series
expansions, defined at the reference surface.
The linear piezoelectric constitutive equations coupling the elastic field and the electric
field, can be written as [12]
o=Q&-eE (2)
D=eT &+p E (3)
where o={oxxoy.yozzoxyoyzo.xzV is the elastic stress vector and

& ={&xx& y'y&zz 'Y xy 'Y yz 'Y.xz}T is the elastic strain vector. Q is the elastic constitutive
matrix in the laminate (x,y,z) coordinate system, e is the piezoelectric stress
coefficients matrix in the same coordinate system, E the electric field vector, D the
electric displacement vector and p the dielectric matrix.
In the present work, within the context of the plate theory, it is assumed that the
electric field vector E is applied through the thickness direction and thus
(4)
where E Z is the electrical field in the thickness direction.

393
The stresses for an arbitrary ply k, written in the laminate (x,y,z) coordinate system are
evaluated by
(5)
where the non-zero element of E k vector for kth layer is given by
(6)
and ~k is the electric voltage applied across the kth layer and tk is the thickness of
the kth layer. The piezoelectric stress coefficient matrix in the (x,y,z) coordinate
system, for the kth layer, is
0 0 e31
0 0 e32
0 0 e33
ek = (7)
e14 e24 0
e15 e25 0
0 0 e36 k
The ej terms required for the present models are obtained from the et piezoelectric
moduli of the kth lamina in its material coordinate system by [1]
- 2+e32
e· sm 2e ,
- .2 2e
e31=e31cos e32 = e31 sm e + e32 cos
e36 = (e31 -e32)cosesine, e33 = e33 (8)
Equation (5) can also be written as
(9)
where dk is the piezoelectric strain coefficient matrix in the (x,y,z) coordinate system,
for the kth layer, related with ek by
(10)
The terms of constitutive matrix Qk of the kth layer referred to the laminate axes
(x,y,z) are also obtained from the Qk matrix referred to the material coordinate
system (1,2,3) with the usual coordinates transformation given explicitly in [1].
394
3. Higher order finite element models
The finite element model based on the displacement field represented by Eq. (1) is
referred to as Q9-HSDT11P. This model has 11 degrees of freedom at each node for
the elastic behavior and one additional electric potential degree of freedom for each
piezoelectric layer for the piezoelectric behavior. By deleting the higher order terms in
the w component of the displacement field we can obtain the Q9-HSDT9P higher
order model with 9 degrees of freedom per node [4] plus the electric potential degrees
of freedom. Also the first order model Q9-FDSDT5P with 5 degrees of freedom per
node, plus the electric potential degrees of freedom, can be obtained by deleting all
high order terms and introducing the adequate shear correction factors on the
transverse shear terms of constitutive matrix [1].
The elastic strain components of the finite element formulation for this displacement
field applied to nine-node Lagrangian quadrilateral elements is fully described in [4].
For the mechanical behavior the strain components are given by
(11)
where
* _ { Guo avo Guo + avo Guo* avo• Guo* + avo* 8e x -2.

8e 2w *
I
Ebm - ax' ay ,CPz' ay ay' ax ' ax ' ay ax' ax ' ay' 0,
T (12)
8e x + 8e y aq,: aq,; 8CP: + aq,;
ay ax'ax'ay'ay ax
and Zbm and Zs are matrices defined in accordance with displacement fields and
strain relations given by
Z
1 0 0 0 z2 0 0
zOO 0 z3 0
= [ 0 1 0 0 0 z2 0
0 zOO 0 z3
0
0
j Z = [1 0 Z 0 z2 0 ] (14)
bm 0 0 1 0 0 0 0 0 0 zOO 0 0 ' sOl 0 z 0 z2
o 0 0 1 0 0 z2 0 0 0 zOO z3
The generalized strains in Eqs. (12-13) can be represented as

395
(15)
where Bbm and Bs are the strain-displacements matrices, respectively, for bending
r,
and membrane and transverse shear, relating the degrees-of-freedom of the element
with the strain components and u e = {. .. u f ... i = 1, ... ,9 , is the vector of mechanical
degrees of freedom of the element and u ~ is the mechanical displacements vector for
node i.
Using CO Lagrangian shape functions [31] the displacements and generalized
displacements defined in the reference surface are obtained within each element,
respectively, by
(16)
where N i are the Lagrange shape functions of node i and the Zm matrix for the
displacement field represented by Eq. (1) is given by
1 0 0 zOO z2 0 0 z3 0
Zm = [ 0 1 0 0 zOO z2 0 0 z3
1 (17)
o 0 1 0 0 zOO z2 0 0
In these models the electrical potentials ~ are assumed constant for each piezoelectric
layer within each element. Hence there is one additional potential degree of freedom
for each piezoelectric layer to represent the piezoelectric behavior and the vector of
electrical degrees of freedom is
~
e = {"~j'"
e}T , }=l,
. ... ,NPL (18)
where NPL is the number of piezoelectric layers in a given element. The vector of
degrees of freedom for the element, q e is
e {e e
q = ul··· u 9
,j,e}T
'I' (19)
The electric field vector E in Eq. (4), can be represented for an arbitrary element, by
(20)
where B. is the electric-potential matrix relating the electric field vector with the
potentials, given by
(21)
and tl •.. tNPL are the thicknesses of the piezoelectric layers.

The principle of virtual work, applied to an arbitrary element of volume V.,
considering the mechanical and electrical energies, can be represented as [12]
ffi • {&lQe-8IleE-8E e e-8E PE+p8u ii}dV

T T T T e- 8Wu +8W. =0 (22)
where 8Wu and 8W. are, respectively, the virtual work done by the mechanical
external forces and the applied surface charge. Substituting Eqs. (4), (7) and (11) into
Eq. (22) for all values of the virtual displacements and assembling the contributions of
all the finite elements in the domain, one obtains for the discretized laminate
composite plate with embedded piezoelectric layers the system equilibrium equations
(23)
where Muu is the mass matrix Kuu the elastic stiffness matrix, K.. is the dielectric
'stiffness' matrix and are the coupling matrices between elastic
mechanical and electrical effects. {u, <jlY is the vector of the unknown
electromechanical response of the structure, that is, the system generalized
displacements and voltages at sensors. {Fu(t),F+(t)V is the force vector of the
structure including the mechanical loads and the applied electrical charges and t is the
time. For static type situations we have
(24)
In practice, voltage may also be specified as input to the actuators. In this case the
following equilibrium equation. for static type situations, can be written
397
(25)
where Fact = Ku~~ is an additional force vector due to the voltage ~ applied to the
actuators. The system matrices K uu , K u+, K++ and M++ are obtained by
assembling the corresponding element matrices K~u' K~+, K+. and M~u of the
finite elements derived from Eq. (22) and evaluated, respectively, by
(27)
where Qzm and QZ are, respectively, the membrane and bending and shear terms of
constitutive matrix in the (x,y,z) laminate axes for the kth ply, I is the identity matrix
of dimension equal to the number of piezoelectric layers in the eth element, N is the
matrix of the Lagrange shape functions and Pk the material density of kth layer. NL is
the total number of layers in the laminate, hk is the vector distance from the middle
surface of the laminate to the upper side of kth ply. Finally, det J is the determinant of
the Jacobian matrix of the transformation from (,; 11) natural coordinates to element
(x,y,z) coordinates.
4. Sensitivity analysis for static type situations
Due to space limitations only static cases are discussed. For static type situations
represented by the equilibrium equations (24) or (25) the sensitivities with respect to
changes in the design variables, b, of a generic function <P j = <P j (q, b) are given by
(28)
398
where 'V j is the adjoint displacements vector obtained from
ap.
K'V·J =-'
aq (29)
and matrix K is the stiffness matrix containing the mechanical, electrical and
coupling submatrices as in Eq. (26), F is the force vector including the mechanical
and electrical components as well as the force vector due to the applied voltages at the
actuators Fact and q is the displacement vector. The function cP j can represent
either the objective function of the optimization problem or a constraint equation. The
sensitivities in Eq. (28) are evaluated by calculating the derivatives at the element level
and adding the contributions of all the elements that have the design variable bi . In
the present work the design variables are the fiber angles in the orthotropic layers, the
ply thickness of both the substrate and piezoelectric layers and also the electric
potentials applied to the piezoelectric actuators. The sensitivities aKe/ abi at the
element level are evaluated analytically by
~[K~U
ab K
K:. 1=+1f +1f [B bm aD.0 jT
e Ke
j +u .. -1-1 0 --
ab j
~(~ f [Zbm O]T [~zm

abo
hk
e
e 1[Zbm 0] dz] [Bbm
0 I T pOI 0 a~+ 1 detJ dSifrl + (30)
I k=1 hk_1
ab j
The derivatives of the load vector in order to the present design variables are equal to
zero for the mechanical component, ie
(31)
and for the actuators, Eq. (25), they are evaluated analytically by
(32)
399
5. Optimal design
The structural optimization problem can be stated as
minimize n(b)
subject to: bi s bi s br i = 1, ... ,ndv
'¥j(q,b)sO j=l, ... ,m (33)
where nCb) is the objective function, '¥j(q,b) are the m inequality constraint
equations, bI and bi" are respectively, the lower and upper limits of the design
variables and ndv is the total number of design variables. The objective function can
be the weight/mass of the structure or the generalized displacement in a specified node.
The constraint equations can be a maximum or minimum displacement constraint
imposed in a given node of the discretized structure, a maximum stress constraint or
the Hoffman's stress failure criterion [4]. By setting the lower and upper limits of the
design variables, bI and br, we can limit, for example, the electric potentials applied
to the actuators according to the coercivt: field of the piezoelectric materials to be used
in a given application or limit the ply thickness variation to feasible values in practice.
The constrained optimization problems are solved by using the method of feasible
directions [32] and the unconstrained problems are solved by using the BFGS
(Broydon-Fletcher-Goldfarb-Shanno) or the DFP (Davidon-Fletcher-Powell) methods
[32].
6. Numerical Applications
6.l. CANTILEVER PIEZOELECTRIC ACTUATOR/SUBSTRATE PLATE
The first numerical application is based on the experiments conducted by Crawley and
Lazarus [33]. The experiment specimen consists of a cantilevered laminate composite
graphite/epoxy substrate plate with distributed G-1l95 piezoelectric actuators bonded
to the top and bottom surfaces as represented in Figure l. A constant voltage with an
opposite sign is applied to the piezoceramic actuators on each side of the plate.
The plate GraphitelEpoxy substrate has six layers of equal thickness with a lamination
sequence of [00/+45°/-45°]s and a total thickness of 0.83 mm. The following material
properties are assumed: E)= 150 GPa, E2=E3= 9 GPa, G12=G]3= 7.1 GPa, G23 = 2.5
GPa, V12 = V23 = V]3 = 0.3.
The G-1l95 piezoelectric actuators have 0.25 mm thickness each and the following
material and piezoelectric properties are used: E) = E2 = E3 = 63 GPa, G)2 = G13 = G23
= 24.2 GPa, V12 = V23 = V]3 = 0.3, d31 = d32 = -166xlO-12 m/V, P33 = 15xlO-6 F/m. The
400
d33 coefficient for the Q9-HSDTllP model is assumed to be d33 = 360xl0-12 mJV,
according to the Piezo Systems, Inc. product catalog of G-1195 piezoceramic material.
In the experiment the deflection of the plate was measured at the points A, B and C
(Figurel). The following non-dimensional displacements are defined
WC-WA
w2 = b '
where b is the width of the plate, WB is the tip deflection at the center of the plate, WA
and We are the tip deflections at the outer transverse edges of the plate. The non-
dimensional displacement Wi represents the longitudinal bending, w2 the twist
deformation and W3 the transverse bending.
I'" 51 ~I
hr~~~~~~~~~~=-~~-=~~~=A
b=152
L=292
+
(mm)
Figure 1. Cantilever piezoelectric actuator/substrate plate
TABLE 1. Comparison of cantilever plate non-dimensional results
Experiment Ritz solution Q9-FSDT5P Q9-HSDT9P Q9-HSDTlIP

[33J [33J
WI 0.034 0.028 0.03488 0.03474 0.03461
W2 0.0038 0.0036 0.004065 0.004040 0.004051
W3 0.0028 0.0032 0.003521 0.003494 0.003506
WI- longitudinal bending, W2 - twist deformation, W3 - transverse bending
The results obtained with the present models for an applied field of 315 Vlmm are
compared in Table 1 with Crawley and Lazarus [33J experimental results and also
with a Ritz solution [33] based in a Mindlin shear deformation theory. The agreement
with the experimental results is reasonably good for the longitudinal bending Wi and
401
also for the twist w2. The agreement with the experimental results is not so good for
the transversal bending w3. However, in this last case the agreement with the Ritz
solution is better. Both first order and higher order models gave very similar results,
which is an expected result because we have here a relatively high side-to-thickness
ratio.
6.2. STRESSES IN A SQUARE PLATE WITH PIEZOELECTRIC LAYERS
In order to compare the stress accuracy of the present models with respect to the side-
to-thickness aspect ratio, the problem of a clamped square plate is considered. The
plate has three substrate high-modulus graphite/epoxy with a lamination sequence
[90%°/90°] and one surface bonded piezoelectric PXE-52 layer on the top (Figure 2).
The side dimension of the plate is a= 1 m and two side-to-thickness aspect ratios are
considered: aIh = 100 corresponding to a thin plate situation where the thickness of the
graphite/epoxy layers is equal to 0.0032 m and the thickness of the piezoelectric layer
is 0.0004 m; aIh= 10 corresponding to a thick plate situation where the thickness of
the graphite/epoxy layers is equal to 0.032 m and the thickness of the piezoelectric
layers is 0.004 m. The material properties for the high-modulus graphite/epoxy are:
EI = 220 GPa, E2 = E3 = 6.9 GPa, GI2 = GI3 = G23 = 4.8 GPa, V12 = V13 = V23 = 0.25.
The material and electric properties for the piezoelectric PXE-52 are: EI = E2 = E3 =
62.5 GPa, GI2 = G13 = G23 = 24 GPa, V12 = V13 = V23 = 0.3, d31 = d32 = -280xl0-12 mIV,
d33 = 700xlO-12 mIV, P33 = 3.45xlO-8 F/m.
A constant voltage of 150 V is applied at the piezoelectric layer. A 8x8 finite element
mesh is used. The stresses are evaluated through the thickness at the point A
represented in Figure 2.
In Tables 2 and 3 are compared the stresses in the piezoelectric layer and in the
graphite/epoxy layers, respectively, obtained with the higher order models and first
order models. It can be seen that the agreement between higher order and first order
models is reasonably good for the thin plate situation (aIh= 100) but high errors are
obtained with the first order model for the thick plate situation (aIh= 10), both in the
piezoelectric and substrate layers.
Piezoelectric
~PXE-52
a
:t~
3 Ply [0°/90%°]
Graphite/Epoxy
a
A
Figure 2. Square plate with surface bonded piezoelectric layers
402
TABLE 2. Stresses in the piezoelectric layer (MPa)
a/h z (IDOl) ax ay aX)'

0.05 0.241 0.103 -0.410
Q9-HSDTllP 10 0.046 0.167 0.0717 -0.363
0.005 3.91 1.67 -1.74
100 0.0046 3.51 1.51 -1.64
0.05 0.200 0.0601 -0.479
Q9-HSDT9P 10 0.046 0.147 0.0440 -0.422
0.005 3.70 1.11 -1.94
100 0.0046 3.34 1.00 -1.82
0.05 0.0827 0.0248 -0.376
Q9-FSDT5P 10 0.046 0.0747 0.0224 -0.351
0.005 3.53 1.07 -1.84
100 0.0046 3.24 0.971 -1.76
TABLE 3. Stresses in the substrate layers (MPa)
a/h z (IDOl) ax ay aX)'

10 0.046 0.440 0.00460 -0.0726
Q9-HSDTlIP -0.05 -0.620 -0.00649 0.0237
100 0.0046 9.23 9.65 -0.329
-0.005 -11.4 -0.119 0.0367
10 0.046 0.471 0.00369 -0.0844
Q9-HSDT9P -0.05 -0.710 -0.00557 0.0292
100 0.0046 10.7 0.0841 -0.365
-0.005 -12.8 -0.100 0.0391
10 0.046 0.240 0.00188 -0.0703
Q9-FSDT5P -0.05 -0.370 -0.00290 0.0480
100 0.0046 10.4 0.0814 -0.351
-0.005 -12.3 -0.0965 0.0616
6.3. OPTIMAL DESIGN OF A PIEZOELECTRIC ACTUATOR BEAM
It is considered the illustrative problem of the optimal design for nummum

weight/mass of two types of piezoelectric bending actuators. The geometry of the
cantilever bending actuator is represented in Figure 3. First we have a 3-Ply actuator
beam consisting of an aluminum center layer and two piezoelectric PXE-5 actuator
layers (Figure 3.a). The second actuator type is a 4-Ply actuator beam consisting of a
2-Ply S-glasslepoxy composite substrate and two piezoelectric PXE-5 actuator layers
(Figure 3.b). A voltage of +180 V is applied to the bottom actuator layer and -180 V
403
to the top actuator layer. The two bending actuators are designed for minimum mass,
subject to the following constraints: Hoffman's stress failure criterion [4], minimum
ply thickness of 0.2 mm both on piezoelectric and center layers and required actuator
tip deflection 1.95mm :::; w :::; 2mm for the applied voltage. A finite element mesh with
20x I equal quadrangular elements is used.
Tip deflection w
a) 3-PLY Actuator beam b) 4-PLY Actuator beam
I~L
Piezoelectric
Actuator PXE-5 ::--------- ~
----~I Aluminum I ~
2 PLY [0°/90°]
S-Glass/Epoxy
Figure 3. Optimal design of a piezoelectric actuator beam
The following material and piezoelectric properties are assumed for the piezoceramic
PXE-5 layers: EI = E2 = E3 = 66.6 GPa, G I2 = GI3 = G23 = 25.6 GPa, VI2 = VI3 = V23 =
0.3, p = 7800 Kglm3, d31 = d32 = -215xlO- 12 e/N, P33 = 2100xpo, Po = 8.85xlO- 12 F/m.
The strength properties, are: XT=YT=ZT= 80 MPa, Xc=Yc=Zc= 600 MPa, R=S=T=46
MPa, where X T , Y T , ZT are the lamina normal strengths in tension along the 1,2,3
directions, respectively, Xc, Y c , Zc are the lamina normal strengths in compression
along the 1,2,3 directions, respectively and R, S, T are the shear strengths in the 23,
13, and 12 planes, respectively.
The assumed material and strength properties for S-glasslepoxy are: EI = 55 GPa, E2 =
E3 = 16 GPa, G12 = GI3 = G23 = 7.6 GPa, V12 = VI3 = V23 = 0.28, P = 2000 Kglm3,
X T=1620 MPa, YT=Zr= 40 MPa, Xc=690 MPa, Yc=Zc= 140 MPa, R=S=T=80 MPa.
Finally, the material and strength properties for the A1MgIMni aluminum alloy are:
E 1=E 2=E3= 69 GPa, G12=G 13 =G23= 25.94 GPa, VI2 = VI3 = V23 = 0.33, P = 2720 Kglm3,
XT=YT=ZT= 155 MPa, Xc=Yc=Zc= 155 MPa, R=S=T=89.5 MPa.
404
TABLE 4. Optimal design of a 3-Ply piezoelectric actuator beam
Initial Desi8!! Optimal Design

Ply Deflection Mass Ply Deflection Mass
thickness w (mm) (kg) thickness w (mm) (kg)
(mm) (mm)
t) 1.0 0.45869 0.0594 t) 0.2 1.98445 0.0224
Q9- t2 3.0 t2 2.140
HSDT9P t3 1.0 t3 9
0.2
t) 1.0 0.45857 0.0594 t) 0.2 1.98414 0.0224
Q9- t2 3.0 t2 2.140
FSDT5P t3 1.0 t3 7
0.2
TABLE 5. Optimal design of a 4-Ply piezoelectric actuator beam
Initial Design Optimal Desi8!!

Ply Deflection Mass Ply Deflection Mass
thickness w (mm) (kg) Thickness w (mm) (kg)
(mm) (mm)
t) 1.0 0.53799 0.0540 t) 0.2923 2.00451 0.0244
Q9- t2 1.5 t2 1.0671
HSDT9P t3 1.5 t3 1.5424
t4 1.0 t4 0.2923
t) 1.0 0.53830 0.0540 t) 0.2928 2.00434 0.0245
Q9- t2 1.5 t2 1.0729
FSDT5P t3 1.5 t3 1.5361
t4 1.0 t4 0.2928
The design variables are the thicknesses of each layer ti both piezoelectric and center
layers. The initial and optimal design data for the 3-ply bending actuator and for the 4-
ply bending actuator are presented, respectively, in Tables 4 and 5 for Q9-HSDT9P and
Q9-FSDT5P models. During the optimization process the active constraints were the
tip deflection constraint and the minimum ply thickness side constraints. Important
mass reductions were achieved: 62% in the 3-Ply actuator and 55% in the 3-Ply
actuator. A very good agreement between both finite element models is found.
405
6.4. MINIMUM DEFLECTION DESIGN OF A COMPOSITE PLATE WITH

SURFACE BONDED PIEZOELECTRIC ACTUATORS
The last illustrative example is the optimal design for minimum mid-point deflection
of a simply supported rectangular (axb) composite plate with two surface bonded
piezoelectric PXE-52 layers. The plate is subjected to a uniform mechanical load of 10
KN/m2 • The lamination sequence of the plate is [p/S/-S/Slp], where p represents the
PXE-52 piezoelectric layers with the properties given in example 6.2, and the three
substrate inside layers are made from S-glass/epoxy with the properties given in
example 6.3. The thickness of each piezoelectric layer is 0.0002 m and the thickness
of each S-glass/epoxy layer is 0.0004 m. The side dimension b is 0.1 m.
In a first stage, the design variable for the optimization process is the ply angle S in the
substrate layers where a design variable linking is introduced to enforce the sequence
[S/-S/S]. The optimal ply angles that lead to the minimum mid-point deflection, for a
few aspect ratios alb, are presented in Table 6. The results obtained with the present
finite element models are compared and a good agreement between all models is found.
TABLE 6. Optimal ply angles for minimum mid-point deflection
alb 1 1.2 1.5 2

Q9-HSDTllP Sopt 45.0 52.5 75.0/104.2 90.0
Woft 0.22091 0.30668 0.41274 0.51309
Q9-HSDT9P Sopt 45.0 52.4 75.3/104.7 90.0
Woft 0.21885 0.30376 0.40925 0.51007
Q9-FSDT5P Sopt 45.0 52.3 74.3/105.7 90.0
Woft 0.21865 0.30352 0.40913 0.50998
SOPI - optimal ply angles (O),WOpl - minimum mid-point deflection (mm)
In a second stage, the design variables for the optimization process are the electric
potentials applied in the surface bonded piezoelectric actuator layers. The objective is
to find the voltage that should be applied to the bottom and top piezoelectric actuators
in order to have a maximum specified mid-point deflection of 0.01 mm, thus reducing
the deflection produced by the mechanical load. The initial design for this second
stage optimization is the previously obtained design with the optimal ply angles for
minimum mid-point deflection (Table 6). According to the coercive field for this type
of piezoelectric material, two side constraints of -260 V and +260 V, are imposed in
the electric potentials during the optimization process. The optimal electric potentials
that should be applied to the top and bottom piezoelectric actuators, ~top and ~bott are
presented in Table 7. Significant reductions in the plate mid-point deflection are
achieved in the optimal design.
406
TABLE 7. Optimal electric potential (V) applied at the actuators for a specified mid-
point deflection of 0.01 mm.
alb 1 l.2 1.5 2

Q9-HSDTlIP ~bott 153.77 184.94 219.49 255.72
~tol! -153.77 -184.94 -219.49 -255.72
Q9-HSDT9P ~bott 15l.94 18l.84 216.05 253.08
~tOI! -15l.94 -18l.84 -216.05 -253.08
Q9-FSDT5P ~bott 15l.94 18l.84 216.03 253.05
~top -15l.94 -18l.84 -216.03 -253.05
7. Conclusions
A family of eO Lagrangian finite element models based on a refined shear deformation

theory is applied to the optimal design of composite laminated plates with embedded
and/or surface bonded piezoelectric layers/patches.
These models have been incorporated in an optimization package in order to obtain the
structural sensitivities of response with respect to changes in design variables and to
carry out the structural optimization of laminated plates with piezoelectric laminae,
considering static constraints and/or objective functions.
The finite element model based on a first order displacement field has shown poor
accuracy on the stress prediction for plates with low length-to-thickness ratios. In this
case the finite element models based on higher order displacement fields are more
adequate.
Two illustrative examples of optimization of composite plates with integrated
piezoelectric laminae have been presented. In both cases important mass savings
and/or improved structural performance have been achieved.
Acknowledgements
The authors wish to thank to the Funda~iio Gulbenkian, FIAD - Funda~iio Luso
Americana para 0 Desenvolvimento and Funda~iio para a Cieneia e Teenologia, for
their financial support.
References
[1] Reddy, 1. N., Mechanics of Laminated Composite Plates - Theory and

Applications, eRe Press, Boca Raton, USA, 1997.
[2] Pagano, N. 1. and Hatfield, S. 1., Elastic behavior of multilayered bidirectional
composites, AIAA Journal, 10 (1972) 931-933.
407
[3] Mallikatjuna and Kant, T., A critical review and some results of recently
developed refined theories of fiber-reinforced laminated composites and
sandwiches, Composite Structures, 23 (1993) 293-312.
[4] Mota Soares, C. M., Mota Soares, C. A. and Franco Correia, V. M., Optimization
of multilaminated structures using higher-order deformation models, Computer
Methods in Applied Mechanics and Engineering, 149 (1997) 133-152.
[5] Chattopadhyay, A. and Seeley, C. E., A higher order theory for modeling
composite laminates with induced strain actuators, Composites Part B, 28B
(1997) 243-252.
[6] Robbins, D. H. and Reddy, 1. N., Analysis of piezoelectrically actuated beams
using a layer-wise displacement theory, Computers & structures, 41 (2) (1991)
265-279.
[7] Lo, K. H., Christensen, R M. and Wu, E. M., A high-order theory of plate
deformation. Part 1: Homogeneous plates, Journal of Applied Mechanics (1977)
663-668.
[8] Lo, K. H., Christensen, R. M. and Wu, E. M., A high-order theory of plate
deformation. Part 2: Laminated plates, Journal of Applied Mechanics (1977) 669-
676.
[9] Noor, A. K. and Burton, W. S., Assessment of computational models for
multilayered anisotropic plates, Composite Structures, 14 (1990) 233 - 265.
[10] Reddy, 1. N. and Robbins Jr., D. H., Theories and computational models for
composite laminates, Applied Mechanics Revue, 47 (6) (1994) 147-169.
[11] Crawley, E. F., Intelligent structures for aerospace: A technology overview and
assessment, AIAA Journal, 32 (8) (1994) 1689-1699.
[12] Allik, H. and Hughes, T. 1. R, Finite element method for piezoelectric vibration,
Int. 1. for Numerical Methods in Engineering, 2 (1970) 151-157.
[13] Tzou, H. S. and Tseng, C. I., Distributed piezoelectric sensor/actuator design for
dynamic measurement/control of distributed parameter systems: A piezoelectric
finite element approach, Journal of Sound and Vibration, 138 (1) (1990) 17-34.
[14] Ha, S. K., Keilers, C. and Chang, F., Finite element analysis of composite
structures containing distributed piezoceramic sensors and actuators, AIAA
Journal, 30 (3) (1992) 772-780.
[15] Chandrashekhara, K. and Agarwal, A. N., Active control of laminated composite
plates using piezoelectric devices: A finite element approach, J. of Intelligent
Material Systems and Structures, 4 (1993) 496-508.
[16] Suleman, A. and Veokayya, V. B., A simple finite element formulation for a
laminated composite plate with piezoelectric layers, 1. of Intelligent Material
Systems and Structures, 6 (1995) 776-782.
[17] Tzou, H. S. and Ye, R, Analysis of piezoelastic structures with laminated
piezoelectric triangle shell elements, AIAA Journal, 34 (1) (1996) 110-115.
[18] Mitchell, J. A. and Reddy, 1. N., A refined hybrid plate theory for composite
laminates with piezoelectric laminae, Int. 1. Solids Structures, 32 (16) (1995)
2345-2367.
408
[19] Jonnalagadda, K. D., Tauchert, T. R. and Blandford, G. E., Higher-order

displacement formulation for a piezothermoelastic laminate. Mechanics of
Electromagnetic Materials and Structures, ASME, AMD-1611MD-42 (1993) 145-
156.
[20] Xu, K., Noor, A. K. and Tang, Y., Three-dimensional solutions for coupled
thermoelectroelastic response of multilayered plates, Computer Methods in
Applied Mechanics and Engineering, 126 (1995) 355-37l.
[21] Donthireddy, P. and Chandrashekhara, K., Modeling and shape control of
composite beams with embedded piezoelectric actuators, Composite Structures,
35 (1996) 237-244.
[22] Kim, S. 1. and Jones, 1. D., Influence of piezo-actuator thickness on the active
vibration control of a cantilever beam, 1. of Intelligent Material Systems and
Structures, 6 (1995) 610-623.
[23] Meric, R. A. and Saigal, S., Shape sensitivity analysis of piezoelectric structures
by the adjoint variable method, AIAA Journal, 29 (8) (1991) 1313-1318.
[24] Silva, E. C. N., Fonseca, 1. S. O. and Kikuchi, N., Optimal design of
piezoelectric microstructues, Computational Mechanics, 19 (1997) 397-410.
[25] Batra, R. C. and Liang, X. Q., The vibration of a rectangular laminated elastic
plate with embedded piezoelectric sensors and actuators, Computers &
Structures, 63 (2) 203-216, 1997.
[26] Liang, C., Sun, F. P. and Rogers, C. A., Determination of design of optimal
actuator location and configuration based on actuator power factor, 1. of
Intelligent Material Systems and Structures, 6 (1995) 456-464.
[27] Kang, Y.K., Park, H.C. and Agrawal, B., Optimization of piezoceramic
sensor/actuator placement for vibration control of laminated plates, AlAA
Journal, 36 (9) (1998) 1763-1765.
[28] Leeks, T. 1. and Weisshaar, T. A., Optimizing induced strain actuators for
maximum panel deflection, Proceedings of AlANASME Adaptive Structures
Forum, AlAA-94-177 4-CP (1994) 378-387.
[29] Franco Correia, V. M., Mota Soares, C. M. and Mota Soares, C. A., Design
sensitivity analysis and optimal design of composite structures using higher order
discrete models, Engineering Optimization, 29 (1997) 85-11l.
[30] Franco Correia, V. M., Mota Soares, C. M. and Mota Soares, C. A., Higher order
models on the eigeofrequency analysis and optimal design of laminated
composite structures, Composite Structures, 39 (3-4) (1997) 237-253.
[31] Zienkiewicz, O. C., The Finite Element Method in Engineering Science, 3rd
edition, McGraw-Hill, London, 1987.
[32] Vanderplaats, G. N., Numerical optimization techniques for engineering design:
with applications, McGraw-Hill Inc., New York, 1994.
[33] Crawley, E. F. and Lazarus, K. B., Induced strain actuation of isotropic and
anisotropic plates, AlAA Journal 29 (6) (1991) 944-95l.
OPTIMIZATION OF COMPOSITE STRUCTURES USING RESPONSE
SURFACE APPROXIMATIONS
R.T. HAFTKA (haftka@ufl.edu), R. Vitali and B.V. Sankar

Department of Aerospace Engineering,
Mechanics&Engineering Science
University of Florida
Gainesville, Florida, USA 32611-6250
1. Introduction
Response surface techniques have been originally developed for improving de-
signs based on testing in the laboratory or the field. These techniques fit simple
functions. typically low order polynomials, to the experimental data, using least
square fit, in a process known as regression. The least square fit filters out some
of the variablility and experimental errors, especially if the number of experi-
ments is substantially larger than the number of unknown coefficients in the
fitted function. The design optimization process then can employ the simple
fitted function, and requires minimal effort.
This approach is based on evaluating a large number of designs as a batch
for constructing the response surface, while methods for analytical design opti-
mization typically evaluate one design at a time. Because experiments can be
conducted more easily and inexpensively in batches rather than singly, the re-
sponse surface approach has been a mainstay of experimental design optimiza-
tion. A related approach, Taguchi methods, does not explicitly fits functions to
experiments, but does it indirectly to find the relative influence of design vari-
ables on the design objective function. Information on response surface tech-
niques is available in many textbooks, such as Box and Draper [I], and Myers
and Montgomery [2].
In the past few years, there has been growing interest in the use of response
surface techniques for analytical design optimization, motivated in part by
growing similarity between experimental evaluation of designs and evaluation
via numerical simulations. First, large discretized numerical models, run with
complex software on modern computers, often exhibits numerical noise due to
the discretization process, incomplete convergence of iterative processes, or
small human errors. Second, the effort associated with creating complex nu-
merical models, and the availability of parallel computers, create an incentive
for performing complex analyses in batches rather than singly. In addition, the
use of response surface techniques also obviates the need for close integration of
409
410
analysis software with optimization software. Once a response surface is created

it is easily connected to optimization software. Finally, response surfaces can be
prepared by specialists familiar with complex simulation codes, and given to
designers in a manner similar to design curves in the Be (before computers) era.
Designers can then use response surfaces to get a complete view of the design
space instead of just the points visited by an optimization procedure. More in-
fonnation about the relationship between experimental and analytical optimiza-
tion may be found in [3].
One major limitation of response surface techniques is a manifestation of the
curse of dimensionality. The cost of response surface techniques increases rap-
idly with the dimensionality of the problem, and they are therefore limited to a
small number of design variables, typically less then 10 or 20. For experimental
optimization, this limit is rarely an issue. However, in analytical optimization,
and especially the optimization of structures, this limit is indeed severe. For this
reason, applications of response surface techniques in engineering optimization
have been limited to problems such as vehicle configuration design (e.g., [4],
[5]), or component design (e.g., [6]).
For many composite stiffened panel design problems, the number of design
variables allows the use of response surface techniques. However, some of the
response characteristics for such problems, such as damage propagation, require
large computational resources for accurate modeling, so that the cost of per-
forming optimization with or without response surfaces can be prohibitive. For
such cases, response surface techniques allow the user to take advantage of sim-
pler, less expensive and less accurate models, which are often available for the
same response quantities. This is done by using response surfaces to correct the
results obtained from the simple models instead of approximating directly the
results obtained from the more complex and expensive ones. This approach,
often called multifidelity or correction response surfaces, has been applied suc-
cessfully to a variety of problems and is discussed in detail in this paper.
2. Response Surface Approximations
Response surface approximations usually employ simple functions. such as

polynomials to fit data so as to minimize the root-mean-square (RMSE) error.
All the applications described in the present work utilize quadratic polynomials,
that is, a response quantity y is approximated as a function of a vector x, with
components Xi as
(1)
with the polynomial coefficients to be estimated by the least squares procedure.

In this case y is a linear function of the coefficients, and the least square fit re-
411
quires the solution of a linear system of equations, with the process called linear
regression.
When the response is not a linear function of the coefficients to be fit, the
minimization of rms requires the solution of a nonlinear set of equations, and
the procedure is called nonlinear regression. Neural networks are possibly the
most popular form of nonlinear regression, and the process of estimating the
unknown coefficients for such networks is called training. While nonlinear re-
gression offers more flexibility in the choice of the functional form of the ap-
proximations, linear regression has advantages in the assessment of the predic-
tive power of the approximation, and in the selection of which coefficients to
retain and which to drop. Furthermore, linear regression also allows easy opti-
mization of the location in design space of the experiments in order to minimize
the error with a given number of experiments. Facilities for the selection of co-
efficients and design of experiments are currently available in many software
packages. The results reported in the present paper were obtained with the JMP
package [7].
Section 2.1 reviews error estimation and choice of coefficients to be re-
tained, and Section 2.2, reviews the choice of the location in design space of the
experiments, a process known as design of experiments. Section 2.3 introduces
the concept of multifidelity response surfaces for combating extrapolation errors
associated with response surfaces in high dimensions.
2.1 FITTING A MODEL TO GIVEN DATA
When the number of data points is substantially larger than the number of coef-
ficients to be fit, we not only filter out experimental or numerical noise, but also
obtain reliable measures of the predictive error in the approximations and the
errors in the coefficients. Most procedures for fitting data provide an estimate of
the rms error at points not included in the data, called estimated variance. An-
other useful measure of the goodness of the fit is denoted as R.', and is a meas-
ure of the percentage of the variance captured by the approximation. Fitting
procedures also provide the standard error or the t-statistic of each coefficient.
The former is an estimate of the standard deviation of the value of the coeffi-
cient, and the latter is the ratio of the coefficient and the standard error.
For example, assume that we fit the five (x.y) data pairs (-2,-1.5), (-1,-1.5),
(0,0), (1,1.25), and (2,1.75) as a quadratic polynomial y=ao+a~+a,.i. A linear
regression procedure will estimate the coefficients as ao=-O.1071. a l =O.925. and
a,=O.0536,
• with an estimated variance of 0.264 and Ra'=0.884. The corre-
sponding t-statistics for the coefficients are -0.30, 5.7, and 0.39. The low values
of the first and third t-statistics indicate that the error in the coefficients may be
substantially larger than their values, indicating that they may not be useful. The
value of the estimated variance is also high relative to the typical values of y.
In this example. if we use a linear model only, eliminating a" the fit im-
proves. The coefficients are now estimated as an=O, al =O.925, with an estimated
412
variance of 0.190 and Ra2=O.91O. The t-statistic of the first coefficient is not
defined, as its value is zero. However, the standard errors in the first two coeffi-
cients drop from 0.36 and 0.16 to 0.19 and 0.14, respectively. Thus, eliminating
a poorly defined coefficient often improves both the predictive capabilities of
the model, as well as the confidence in the values of the other coefficients.
Unfortunately, when we have several coefficients with low t-statistics, there
is no simple way to choose which to drop and which to keep. Rather, many
software packages provide procedures, called stepwise regression, which search
for an optimum set of coefficients to keep in order to maximize some measure
of predictive capability.
2.2 DESIGN OF EXPERIMENTS
Design of experiments is the name used for a set of procedures concerned with
choosing the location in design space of experiments in order to minimize the
predictive error of the response surface fitted to the experimental data. Most of
the work in design of experiments has been focused on filtering out random
noise, or reducing the so-called variance error, and for this purpose the best
strategy is to place the experiments as far as possible from one another. When
the experiments are numerical, modelling errors are often as important as vari-
ance errors. Modeling or bias errors arise due to the fact that the functional form
selected for the approximation cannot fit well the true function. There is some
limited amount of work on design of experiments for reducing modeling errors
(e.g. [8]), but it is of limited applicability.
When the region of interest in design space is box-like there are standard
optimal experimental designs available for linear and quadratic polynomials of
almost any dimensions. When the number of design variables n is small, these
designs typically include the vertices of the design domain. For example, for
fitting quadratic polynomials, one of the popular experimental designs is the
central composite design, which includes all the 2" vertices, the center of the
design domain, and 2n points obtained by perturbing only one design variable
(both in the positive and negative directions) from the central point. The total
number of design points thus obtained amounts to 2" + 2n +1, a graphical repre-
sentation in a three dimensional space is given in Fig 1.
413
Figure 1. Central composite design in three dimensions
However, even for 10 variables, the number of vertices in a boxlike design

domain is 1024, which is more than most applications can afford. For n=20, the
number of vertices is more than a million. Furthermore, the number of coeffi-
cients of a quadratic polynomials grows more slowly, as (n+1)(n+2)12, so that
there is no need for so many experiments. For example, for n=lO, a quadratic
polynomial has only 66 coefficients. Consequently, there are many experimen-
tal designs which do not include all the vertices, with orthogonal arrays being
the most popular for many applications.
For irregularly shaped domains, or when the number of experiments to be
performed does not tit any of the standard designs, it is common to numerically
search for the best set of points. Typically, a candidate set of points is generated
(e.g., on a grid), and then an optimal subset with a given number of points is
searched. The most popular optimality criterion for this search is D-optimality,
which maximizes the determinant of the matrix used for solving the least
squares problem. This criterion minimizes the volume of uncertainty around the
vector of estimated coefficients. Like other variance-based criteria, it tends to
place most of the selected points on the boundary of the region.
2.3 EXTRAPOLATION ERRORS AND MULTFIDELITY RESPONSE SUR-

FACES
While including all the vertices in the design domain is not needed for fitting a
quadratic polynomial, it is needed for avoiding extrapolation in a box-like do-
main. Since we usually cannot afford to evaluate the function at all the vertices,
414
we have to resign ourselves to using the response surface in an extrapolation

rather than interpolation mode. Unfortunately, extrapolation may lead to very
large errors.
To combat the large errors, which may be associated with extrapolation
from response surfaces, simple inexpensive models of the approximated re-
sponse quantity become extremely useful. Such models may have unacceptably
large errors, but they rarely have the huge errors associated with extrapolating
from. say quadratic polynomials.
Simple or lower fidelity models are first useful in limiting the size of the re-
gion where response surfaces need to be constructed. Even if their estimate of
the response is not accurate, they can still identify points where the value of the
response is so large or so small as to render the design totally unacceptable.
This use of simple fidelity models has been dubbed as the reasonable design
space approach [9].
A second usage for lower fidelity models is for the purpose of identifying
good functional relationships between the response and the design parameters.
This concept has been introduced by Schmit and coworkers (e.g., [to]) and
utilized extensively in local approximations for structural response, and is
known as the use of intervening functions or intervening variables. For example,
the buckling load of a composite laminate is roughly proportional to the cube of
the total thickness. Consequently, if the design variables are the number of plies
of different orientations, the buckling load normalized by the cube of the total
thickness will be more easily approximated by a quadratic response surface than
the buckling load itself. In addition, the lower fidelity model may help reduce
the number of terms included in the higher fidelity approximation, by identify-
ing unimportant terms (e.g., [5]).
Finally, the response surface may be fitted to a correction from the lower fi-
delity model to the higher fidelity one. This correction will typically vary over a
smaller range than the response itself. In addition, the correction may be limited
by knowledge of the magnitude of the error in the lower fidelity model. For
example, the value obtained from the response surface may be limited to be no
higher than twice the lower fidelity model and no lower than half of its value.
Such caps can greatly reduce the danger of runaway extrapolation errors.
3. Blended Wing Body Panel Design Example
3.1 DESIGN PROBLEM
A blended wing body (BWB) airplane capable of carrying about 800 is a wide
double deck center body which is blended in the wing. Due to the shape of the
airplane the pressured center-body region, which includes the passengers area
and the cargo area is non-circular. This type of configuration for the center-
body region is challenging from the standpoint of structural design. since the
415
upper and lower cabin surfaces carry both internal pressure loading and running
loads due to wing bending. To keep structural weight as low as possible, the
structure is to be built of composite materials.
Reference [11] describes work which is part of a BWB design study, led by
the McDonnell Douglas Corporation in collaboration with NASA and four uni-
versities (University of Florida, Stanford University, University of Southern
California, and Clark Atlanta University). The study involves the optimization
of one of the structural concepts considered for the design, a stiffened shell con-
cept for the upper cover panel of a typical passenger bay. The passenger area is
divided into several bays, separated by longitudinal ribs, to reduce instability
problems in the upper wing skin, which carries compressive wing bending
loads. Specifically, results are presented for a composite hat-stiffened skin con-
figuration.
The structural optimization problem is formulated using weight as the ob-
jective function, with constraints on buckling of the cover panel and maximum
stress in the cover panel. The structural analysis is performed by an analysis
code which does not have optimization capabilities and is difficult to connect to
an optimizer. For such situations, response surface techniques, which are very
useful.
The structural configuration considered in the optimization is a hat-stiffened
skin, upper cover panel of a typical passenger bay, as shown in the Fig. 1. The
panel is assumed to be 150 in. long in the spanwise, or x, direction and 900 in.
in the chordwise, or y, direction. The structure was first designed subject to a
uniform pressure load of 14.84 psi applied to the bottom skin surface, and an
end load, N, equal to 4319 lb./in. applied to the right end of the panel in the
spanwise direction. The left end of the panel is restrained from movement in the
x-direction. The right and left ends are clamped and the unloaded sides are sim-
ply supported.
Y (Chord·wIse.
Figure 2. Hat-stiffened skin upper cover

416
The skin and the hat stiffener are constructed from graphite-epoxy preforms.
Each preform is a stack, with a cured thickness of 0.055 in., equivalent to seven
layers of unidirectional prepeg with 44.9%, 49.2% and 12.9% of 0,45 and 90
degree fibers, respectively. The nominal stacking sequence of the preforms used
in the skin and in all of the components of the hat stiffener is [45/-45/0/90/-
45145]. Nominal material properties for a cured stack and the stress allowables
that are se in the designs are provided in Table 1.
TABLE I. Material properties for graphite-epoxy
Design Value
Specifications
EII(msi) 9.25
E,,(msi) 4.67
G,,(msi) 2.27
v" 0.397
&oJll 5.40 10·l
a.,(ksi) 50
't,,"(ksi) 18
p (Ib1inl) 0.0057
3.2 INITIAL STRUCTURAL DESIGN
The first step in the structural optimization was to use PANDA2 [12] to obtain
an initial design. PANDA2 is a program specifically designed to find minimum
weight design of laminated composite stiffened panels using simple analysis
models.
A panel module and the design variables used in the PANDA2 optimization
are shown in Fig 2. A module includes the cross section of a stiffener plus the
panel skin. with width equal to the spacing between stiffeners (b in Fig. 2). A
typical stiffened panel consists of several panel modules placed adjacent to each
other.
417
,f-l---wc::----f/
1\
tf _J--
/ /
/ /
/ /
b
Fi~ure 3. Design variables for the PANDA2 optimization
Aside from geometrical constraints, stress and buckling constraints were also
imposed by PANDA2. Initially, a factor of safety of 1.3 was applied to the
stresses, and a factor of 1.15 was applied to the buckling load factor. The opti-
mum design obtained by PANDA2 is provided Table 2. The active constraints
at the mid-length of the panel were fiber compression in the crown, web buck-
ling, and axial strain. At the ends of the panel, the active constraints were local
buckling of the panel skin between stringers, local buckling of the panel skin
under the hats, and fiber compression in the skin.
TABLE 2. PANDA2 Optimum design
Variable Original Rounded

Weight (lb/ft2) 4.65 4.86
b (in.) 13.83 13.83

b, (in.) 8.4 8.4
h (in.) 6.5 6.5
w(in.) 4.3 4.3
w,(in.) 5.8 5.8
t. (in.) 0.22 0.22
t,(in.) 0.34 0.33
t.(in.) 0.20 0.22
{(in.) 0.33 0.33
418
3.3. REFINED DESIGNS USING RESPONSE SURFACES
A more efficient design for the panel was obtained by relaxing the constraint
that the cross-section of the panel is constant in the span wise direction. The
panel was divided into three sections: two identical sections at the ends of the
panel and a section in the middle of the panel as shown in Fig. 3. Because the
skin buckled near the ends of the panel in the PANDA2 design. a layer of mate-
rial equal to the thickness of the flange was added to the panel skin between the
stiffeners and to the panel skin under the hat stiffeners. To prevent buckling of
the web crown at the mid-length point of the panel. the thickness of the crown
of the hat was increased.
An optimum design for the more complicated panel configuration shown in
Fig. 3 was obtained using response surface techniques. The structural analysis
was performed using STAGS (Structural Analysis of General Shells. [13]).
STAGS is a finite element code for non-linear analysis of stiffened shell struc-
tures of arbitrary shape and complexity. The STAGS analyses. an additional
factor of safety equal to 1.25 was applied to the end load. The end load applied
in the STAGS analyses was equal to 5398 Ib./in.
x skin
Figure 4. Division of the panel in three sections

419
Quadratic polynomial response surfaces were created and then used as con-
straint function in an optimization to minimize the panel weight. The optimiza-
tion was performed using a spread sheet program in Microsoft EXCEL[14],
which allows discrete design variables. the design variables used in the optimi-
zation are provided in Table 3.
TABLE. 3. Design variables used in STAGS Analyses
Design variable meaning

Variable
toO Skin Thickness near panel ends
1,. Skin Thickness in the middle of the panel
I", Crown thickness the near panel ends
t,... Crown thickness near the middle of the panel
D distance from panel ends to thickness discontinuity
The D-Optimal criterion as implemented in the JMP program [7] was em-
ployed for selection of design points. The implementation in lMP tinds a D-
Optimal set of points from a given set of candidate design points in the design
domain. The candidate design points used for constructing the response sur-
faces are provided in Table 4.
TABLE 4. Candidate points for Response Surface Construction
Design minimum step maximum

Variable (in.) (in.) (in.)
d 12 12 48
0.11 0.055 0.44
0.11 0.055 0.44
0.11 0.055 0.44
0.11 0.055 0.44
The size of the design domain was reduced to 740 feasible design points by
introducing the following considerations: (i) The lower and upper bounds to the
optimal weight were estimated to be 3.0 lb.lfe and 4.3 lb.lfe respectively. (ii)
The skin near the ends of the panel was expected to be thicker than the skin in
the middle of the panel (t...<t.). (iii) The thickness of the crown in the middle of
the panel was expected to be greater than the thickness of the crown near the
ends of the panel (teb <ton)'
Initially a linear response surface was tried based on 23 analyses, but its ac-
curacy was not satisfactory. With a quadratic polynomial in 5 variables having
21 coefficients, we added 10 analyses for a total of 33 points. We discarded
terms in the polynomial with low t-statistics as long as RaJ kept increasing. For
the final response surface, the variance estimate was 0.19 and R/=O.90 and the
lowest t-statistic was 3.15, indicating reasonable confidence in the coefficients.
The accuracy of the response surface was also checked by constructing response
420
surfaces with 32 points and then comparing the response surface predictions at
the point left out with the STAGS analysis predictions at that point (a procedure
known as PRESS, [2]). Based upon these results the quadratic response surface
predictions for the buckling load factor are expected to have less than 25% er-
ror.
A response surface was also constructed for the compressive axial stresses at
the change in cross-section. Here the accuracy of the fit was better with
Ra1=O.96 and rms=1924 psi. The t-statistics for all the coefficients except for
tsb were satisfactory. We also checked accuracy by the PRESS procedure which
indicated that a maximum error of 20% is expected. The response surface for
the buckling load factor A. in the box domain had errors of 25%. Therefore the
allowable buckling load factor was increased by 25%, from 1.15 to 1.4375.
Similarly for the compressive stresses s, we increased the safety factor by 20%
from 1.3 to 1.56.
TABLE 5. Optimization Problem Using Response Surfaces
Objective Function: min I weight Ib/ft') }

Design Variables: d I.... t tallt teb
t l .,
Constraints: A.~1.4375
0'.$ 32051
t,.-t.. ~O
t""-t,.. ~O
t• •t., .t.. t.. > 0.11
The optimization problem in the rectangular domain was formulated as

shown in Table 5. The optimization was performed using a generalized reduced
gradient optimizer available in Microsoft EXCEL. The optimum design ob-
tained, subject to the constraints specified in Table 5, is provided in Table 6.
Analysis of the design in Table 6 with STAGS gave a buckling load factor of
A.=1.379 and a maximum stress at the discontinuity 0=22,450 psi.
As shown by comparing the results in Table 6 with the results in Table 2,
using STAGS and response surface approach to perform the optimization, the
weight of the upper cover panel was reduced considerably.
TABLE 6. Optimum Design Obtained

using response surfaces
Variable Value
d (in.) 30
I .. (in.) 0.165
t,. (in.) 0.110
t"M (in.) 0.165
I,.. (in.) 0.110
Weighl(lblrf) 3.533
A. response surface 1.532
0' (psi) response surface 16796
421
3.4 OPTIMIZATION WITH NEW DESIGN LOADS
After the previous optimization was performed the design loads and the re-
quired safety factors were updated. based on changes to the overall design of the
airplane, as well as to refinement in the selection of load cases. Additionally,
the safety factors were reduced, mostly reflecting greater confidence in the de-
sign, and allowing local buckling between the limit and ultimate load factors
A new load case of pressure (p = 18.56 psi) and with no in-plane loads was
=
added, and the combined load case was changed to p 15.59 psi combined with
in-load of 2879 lb./in. These new load cases had a safety factor equal to 1.0.
To update the design for the new load case the starting point was the opti-
mum obtained in the previous design cycle. The new pressure-only load case
generates high stresses in the crown of the hat because the neutral axis of the
section near the ends is very close to the skin and far from the crown of the hat.
From the preceding optimization it was clear that the thickness of the skin in
the middle section. t,m' was likely to remain at its lower limit therefore it was
removed from design variable list. On the other hand, the webs accounted for
about the 40% of the total weight and so the thickness of these structural ele-
ments was introduced as a new design variable.
The thicknesses of the skin at the boundary and of the crown in the middle
of the panel were allowed to change by ±0.055 in. from the previous optimum.
The thickness of the crown at the boundary was allowed to be twice as thick as
it was for the initial load case and therefore it could take one of the three values
of 0.11,0.165 and 0.22 inches. The thickness of the web was allowed to take
only the values of 0.165 in. or 0.22 in., since a lower value was sure to violate
the buckling constraint, and a larger value would lead to very heavy designs.
Similarly the thickness of the crown near the ends was also limited to only two
values of 0.11 in. or 0.165 in .. The distance from the panel ends to the thickness
discontinuity, that is d, was permitted to change by ± 6 inches form the value of
30.0 inches obtained in the previous optimization.
Given these restrictions a total of 162 new design points around the previous
optimum design point were created. The JMP program was employed to select a
D-Optimal set of 25 new design points from the 162 candidate design points.
STAGS was used to conduct linear analyses of each of these 25 designs for the
internal pressure only load case and for the combined load case.
The stress calculations performed with STAGS indicated that the stress al-
lowable was exceeded only in the crown of the hat-stiffener, for the internal
pressure only load case. Furthermore, the STAGS analysis showed that for the
two load cases considered in the present analysis, a simple beam analysis pro-
vides a good approximation of the stress distribution along the panel length.
Since the stress calculations performed using the beam approximation were in-
expensive, the stresses in the crown of the hat-stiffener were calculated for all of
422
the 162 structures. The stress results from the approximate beam analysis were
compared with the STAGS stress results for 13 structures and the difference
ranged from 0.1 % to 6.5% justifying the use of the beam approximation.
The buckling calculations performed with STAGS for the 25 structures
showed that for buckling load factors lower than one or close to one. buckling
was mostly local and confined to a single element of the hat-stiffened panel.
Interactive buckling modes were also encountered, but the buckling load factor
for these modes was substantially greater than one. Based upon the observations
that the individual components of the stiffened panel mostly buckled independ-
ently for buckling load factors close to one, an approximate buckling analysis
was developed by representing a subregion of each component of the stiffened
panel as a simply supported plate. The dimensions of these subregions were
found using the width of each element of the stiffener (for example the width
w = 4.3 in. was used for the crown of the hat) and the length of the lowest
buckling half wave of a plate that was simply supported on all edges. The com-
ponent with the lowest buckling load determined the buckling load for the stiff-
ened panel.
The spanwise stresses in the panel induced by the internal pressure load
vary quadratically along the length of the panel. However, the stress is ap-
proximately constant over a single buckling half wave, since the length of a
single buckling half wave is much smaller than the length of the panel. There-
fore, the buckling stress for a single component of the stiffened panel was ap-
proximated from the buckling solution for a simply supported plate under con-
stant load, and was given by:
~
rr ab 2 ab 4
(jbuck=
tab
2 DII +2(D I2 +2D66 ) ( - ) +D22 ( - )
bb bb 1 (2)
where bb and t were the width and the thickness of the component of the hat-
stiffened panel under consideration, and a b was the length of a buckling half
wave, determined by minimizing the buckling stress. Minimizing this equation
for the length a b gave:
(3)
Substituting this last one and the material properties in Table I into the
former expression, we could find the approximate buckling stress for an indi-
[:.J
vidual component of the hat-stiffened panel:
"",. =2.25 10' (4)

423
An approximation for the buckling load factor was obtained by dividing the
approximate buckling stress by the approximate applied stress, determined from
the simple beam analysis discussed above. The approximate buckling load
factor is defined as
A. = (]buck
(5)
(]applied
A value of bh equal to 8.0 inches was used in the calculations of (]buclc for
the skin near the ends of the panel and for the skin at the thickness discontinuity
(b - w, in Fig. 2). To evaluate the buckling stress in the crown of the hat-
stiffener bb was set equal to 4.3 inches (w in Fig. 2).
Comparisons of the buckling load factor obtained using the approximate
analysis with the buckling load factor predicted by STAGS for the 25 structures
indicated that the approximate buckling load factor underestimated the resis-
tance to buckling. This result was to be expected since the neighboring plate
elements provided more boundary restraint than simple support. Also the
stresses were not constant and the buckling load factor was calculated on the
basis of the maximum stress over the length abo The buckling load factors given
by the approximation were up to 75% smaller than the buckling load factors
predicted by STAGS. To improve the buckling load factor predictions obtained
using the approximate beam-plate analysis the STAGS results were used to fit a
scale factor (as a single term response surface) to the buckling load factor in
each critical region of the panel. The computed scale factors are provided in
Table 7.
TABLE 7. Scale Factors for A,
Buckling Location Combined load factor
Cap in the interior section 1.804

Skin in the interior section 3.943
Skin at the end section 1.313
The combined load condition was critical for buckling for all the designs
considered, so the response surface was generated for this load case only. The
buckling load factors predicted by the approximate analysis combined with the
scale factors provided in Table 7, had a maximum error of 10% for structures
whose buckling modes were mainly localized in one element. The approximate
analysis generally overestimated the buckling load factor for structures that
show and interactive buckling mode. Fortunately these structures had buckling
load factors that were well above 1.0.
424
Using the previously illustrated simple methods we could afford to analyze

all of the 162 designs using beam theory to determine the stresses, and the
scaled plate-beam approximation to determine an estimate of the buckling load
factor, and then analyze the most promising design with STAGS. The optimum
candidate design found by inspection using the approximate analysis approach
implemented into Microsoft EXCEL is given in Table 8.
TABLE 8. Minimum Weight Structure
Design Variable Value
d (in.) 24.0
t,j) (in.) 0.165
!em (in.) 0.165
t..'b (in.) 0.220
tw (in.) 0.165
Weight 2.7551b/ft2
The approximate maximum spanwise stress was 44,101 psi, tension, in the
crown of the hat-stiffener, for the pressure-only load condition. The minimum
approximate buckling load factor was A. = 0.97 for the combined load condi-
tion. The design did not exactly satisfy the constraint but it was close enough to
test with STAGS. This buckling load factor corresponds to skin buckling in the
interior region of the panel. The STAGS linear bifurcation analysis gave a
buckling load factor A.=0.93 with the same type of buckling mode predicted by
the approximate analysis that was predominantly buckling of the skin at axial
location where the thickness changes. The maximum tensile stress in the crown
of the hat near the ends was 45,097 psi.
The design did not satisfy the buckling constraint however it was so prom-
ising that we also performed a non-linear analysis. The STAGS program com-
putes a non-linear static solution and then conducts a linear bifurcation analysis
with the eigenvalue applied to both loads, the in-plane load and the pressure
load. The buckling load factor found following this procedure was A.=1.l14.
The buckling load factor corresponding to the first buckling mode obtained
from the non-linear analysis was was interactive where the crown and the web
of the stiffener were buckling together and it was different from the buckling
mode obtained using linear analysis. The value of the estimated buckling load
factor is 14% higher in the non-linear case. The interactive buckling mode
found by the non-linear analysis corresponds to the ninth buckling mode found
by the linear analysis and its value was A. = 1.21.
425
4. Approximation of Stress Intensity Factor by Correction Response Sur-

faces
4.1 PROBLEM DEFINITION
The next generation supersonic transport planes are expected to fly at very high
altitudes and are very vulnerable to pressure loss due to large cracks, such as
due to turbine blade penetrations. Consequently accurate prediction of crack
propagation is a major issue in the structural design of such aircraft. Exact cal-
culation of crack propagation requires detailed structural models for calculating
stress intensity factors and is computationally too expensive to include in struc-
tural design optimization. This section demonstrates the use of correction re-
sponse surfaces for alleviating the computational burden based on [15].
A composite blade stiffened panel with a through-the-thickness crack sub-
jected to tensile loads (Figure 4) is investigated. Various stiffened panels were
created by changing three design variables, the distance between the blades,
2s, the length of the plate, 21, and the thickness, t, of the blades.
z H
t
20
Figure 5. Stiffened panel with a crack (dimensions are in inches)
A 4.0 inch long through-the-thickness crack running in the x-direction is lo-

cated in the middle of the plate. The panel is made of AS/3501-6 laminate ar-
=
ranged as [+45/90/0], and the total thickness is tp 0.04 in. The stiffeners are
also made of the same material and same lay-up, but their total thickness is
varied. The panel is simply supported along the sides parallel to the .v-direction,
and it is free along the sides parallel to the x-direction. The panel is loaded by
uniform load of 1.0 lblin in the x-direction, with the blades restrained to un-
dergo the same y displacement as the plate.
426
4.2 HIGH AND LOW FIDELITY MODELS
Vaidya and Sun [16] proposed that it is possible to establish a general fracture
toughness parameter, independent of laminate configuration, that controls the
propagation of a trough-the-thickness crack in a composite laminate. The ex-
perimental results presented in Ref. 16 indicate that while matrix cracks provide
stress relaxation, final fracture of the laminate is controlled by fiber breakage in
the zero-deg plies. Therefore, in a composite structure, cracks propagate when
the Stress Intensity Factor (SIF) in the zero-deg plies reaches the fracture tough-
ness of the material or
(6)
A method for calculating the stress intensity factor in the zero-deg plies. KO
from the equivalent orthotropic plate value K is described in [IS].
Two methods for evaluating the SIF in composite structures are considered:
a high-fidelity method and a low fidelity method. The higher-fidelity method
employs a finite element model of the structure under consideration including a
highly refined mesh near the crack (Figure 5). This higher-fidelity method esti-
mates the value of the stress intensity K from a least square fit of the stresses at
nodes close to the crack tip.
Figure 6. Detailed finite element model near the crack
The value of the SIF can also be estimated using a simpler procedure, which
we use as our low-fidelity-method. We start with a coarse finite element model
of the structure, which does not include the crack, and calculate the average
axial stress in the direction orthogonal to the crack axis at the location where the
crack is placed. The value of the average stress calculated is then applied to an
427
infinite plate with the same material, and the SIF is calculated for the infinite
plate.
Response surface techniques have been used to predict the value of the SIF,
K, as a function of the design variables using two types of response surfaces.
The first approach utilizes a response surface that directly fits values of K ob-
tained from the high-fidelity method, while the second approach employs a
correction response surface to the ratio of K obtained from the high-fidelity
method and the low-fidelity method.
4.3 RESULTS WITH CORRECTION RESPONSE SURFACE
The three design variables were discretized in a region of design space as

shown in Table 9, and 27 designs were selected using a slightly asymmetric face
centered central composite design. Two response surfaces were created based
on the 27 selected points.
TABLE 9. Design variable
Variable Max Min Step

s (inches) 8.0 2.5 0.5
I (inches) 10.0 4.0 2.0
t (inches) 0.056 0.020 0.004
First a quadratic polynomial was fit to the high-fidelity stress intensity factor
to obtain
Kf(S = 40.40 + 4.04 1-443.73 t -130 s-O.22 /' - 4.!XJ t 1+ 2005.52 t' +0.04 s / + 4.10 s t -0.16 s'
The goodness of the fit is measured by Ra = 0.998 and a percentage estimated

root mean square error of PRMSE = 0.52%, where PRMSE is defined as the
ratio of the RMSE to the average value of the response surface. The t-statistics
of the quadratic response surface coefficients ranged from 2.93 for the sf term
and 44.01 for the constant term ..
Next a correction quadratic response surface was fitted to the values of the
ratio, rp between the SIF found from the High-fidelity model and the
Low-fidelity model
K High-{uk/i'"
r =-.;;....;...-
P K /uw-fitk/i",
The polynomial obtained from the fit at the 27 points is given as
,
r =0.59+7.8010-' 1-1.30t +2.5510' 5-4.4810-' 1'+8.3810-· 51+1.2310-/5 t-2.73 10-' 5'
428
with an R. = 0.993 and a percentage root mean square error PRMSE = 0.54%.
The t-statistics ranged from 4.18 for the st term and 36.41 for the constant term.
With the correction response surface, the stress intensity factor of the panel
may be estimated as
Response surface values at points used in the construction of the response sur-
face tend to be more accurate then values at other points. For this reason, the
accuracy of the various approximations was also checked at several other points
not used in the construction of the response surfaces. The results for these 4
representative extra designs are presented along with the errors in Table 10.
From the table it is clear that the conventional response surface based on the
high-fidelity analyses is less accurate in predicting SIP for design points not
used to construct the response surface. The error of the high-fidelity response
surface is particularly high for points outside the region used to construct the
approximation, that is points 3 and 4.
For example, design point 4 has a blade thickness that is ten times the
maximum thickness in the design domain. The correction response surface per-
forms much better than the conventional response surface because it only needs
to correct the low fidelity method predicted value of the SIP. That is it has to
=
reduce the predicted value of the SIP by about 30% from Ku. 12.05 to Km. =
8.60.
rABLE 10. Response surfaces at test points
Point Lin. Tin. Sin. Kuo lew ~ Kas Error % Error %

~ Kas
8.0 0.056 2.5 44.13 40.75 41.61 41.42 2.11 1.67

2 5.0 0.048 4.0 46.08 42.62 41.91 41.99 1.65 1.45
3 4.0 0.112 2.5 34.06 27.40 30.27 27.15 10.48 0.89
4 4.0 0.56 2.5 12.05 8.60 431.10 8.60 4915.8 0.Ql
s. References
1. Box, E.P., and Draper, N.R. (1987) Empirical Model-building and

Response Surfaces, John Wiley.
2. Myers, R.H., and Montgomery, D.C. (1995) Response Surface Meth-
odoloy, John Wiley.
429
3. Haftka. RT .• Scott. E.P .• and Cruz. lR (1998) Optimization and Ex-

periments - a Survey, Applied Mechanics Reviews. 51(7) pp. 435-448.
4. DeLaurentis, D.A., Cesnik, C.E., Lee, l-M., Mavris, D.N., and
Schrage. D.P. (1996) A new Approach to Integrated Wing Design in
Conceptual Synthesis and Optimization, Proceedings of the 6th
AlAA.NASAIISSMO Symposium on Multidisciplinary Analysis and
Optimization. AlAA Paper 96-4000, pp. 1835-1843, Bellevue, WA.
5. Knill, D.L. , Giunta, A.,A.. Baker, C.A., Grossman, B., Mason, W.H.,
Haftka, RT., and Watson, L.T. (1998) HSCT Configuration Design
Using Response Surface Approximations of Supersonic Euler Aero-
dynamics. AIAA Paper 98-0905, 36th Aerospace Sciences Meeting
Reno, NY, January 12-15.
6. Venter, G., Haftka, RT., and Chirehdast, M. (1997) Response Sur-
face Approximations for Fatigue Life Predictions, AIAA Paper 97-
1331, Proceedings of the 38th AlAAIASMElASCElAHSIASC Struc-
tures, Structural Dynamics and Materials Conference, Kissimmee,
Apr 7-10, FL, Part 2, pp. 1383-1396.
7. SAS Institute, Inc. (1995) 1MP Users Guide, Version 3.1, Cary NC.
8. Venter, G., and Haftka, RT. (1997) Minimum-Bias based Experi-
mental Design for Constructing Response Surfaces in Structural Op-
timization, AIAA Paper 97-1053, Proceedings of the 38th
AlAAIASMEIASCElAHSlASC Structures, Structural Dynamics and
Materials Conference, Kissimmee, FL, April 7-10, Part 2, pp. 1225-
1238.
9. Balabanov. V., Kaufman, M., Giunta, A.A., Haftka, RT., Grossman,
B., Mason. W.H., and Watson, L.T. (1996) Developing Customized
Wing Weight Function by Structural Optimization on Parallel Com-
puters," AIAA Paper 96-1336, Proceedings of the 37th
AlAAIASMEIASCEIAHS Structures, Structural Dynamics and Mate-
rials Conference, Salt Lake City, UT, April 15-17, Part I, pp. 113-
125.
10. Lust, RV., and Schmit, L.A., Jr. (1986) Alternative Approximation
Method for Structural Synthesis with Reference to Space Structures,
AlAA 1. 24(10), pp. 1676-1684.
11. Vitali, R, Park, 0., Haftka, RT., and Sankar, B.V. (1997) Structural
Optimization of a Hat Stiffened Panel by Response Surface Tech-
niques," AIAA Paper 97-1151, Proceedings of the 38th
AlAAIASMElASCElAHSlASC Structures, Structural Dynamics and
Materials Conference, Kissimmee, FL, April 7-10, Part 4, pp. 2983-
2993.
430
12. Bushnell D. (1983) PANDA-Interactive Program for Minimum

Weight Design of Stiffened Cylindrical Panels and Shells" Comput-
ers and Structures, 16, pp. 167-185.
13. Brogen, A., Rankin, C.C., and Cabiness, H.D. (1994) Structural
Analysis of General Shell version 2.0, Lockheed Research Labora-
tory, Palo Alto CA.
14. Microsoft Corporation, EXCEL 5.0
15. Vitali, R, Haftka, R.T., and Sankar, B. V., (1998) Response Surface
Approximations for Crack Propagation Critical Loads," AlAA Paper
98-2047, Proceedings, 39th AIANASMEIASCEI-AHSIASC Struc-
tures, Structural Dynamics, and Materials Conference, Long Beach,
California, April 20 - 23, Part 4., pp. 2917-2922
16. Vaidya R. S. and Sun C. T. (1996) Fracture Criterion for Notched
thin Composite Laminates, AIAA Paper 96-1354, Proceedings
AIANASMEI- ASCEIAHSIASC Structures, Structural Dynamics and
Materials Conference, Salt Lake City, Utah, Part 1, pp.33 1-338.
GENETIC ALGORITHMS FOR OPTIMIZATION OF COMPOSITE LAMI-
NATES
R.T. HAFrKA (haftka@ufl.edu)

Department ofAerospace Engineering, Mechanics&Engineering Science
University of Florida
Gainesville, Florida, USA 32611-6250
Abstract
The design of composite laminates is a combinatorial problem well suited for solution
via genetic algorithms. However, the number of analyses required for genetic optimiza-
tion of even single laminates may be many thousands. Consequently, it is important to
tailor standard genetic algorithms to the special problems associated with composite
laminates. Two examples are used to demonstrate the importance of specialized codings
and operators for the efficiency of genetic optimization of composite laminates. An un-
stiffened panel example demonstrates the usefulness of repair for handling constraints.
A stiffened panel example shows the usefulness of specialized mutation operators.
1. Introduction
Ply orientations for composite laminates are often limited to a small set of directions,
such as 0°, ±45°, and 90°. Consequently, the design of that stacking sequence is a com-
binatorial problem, not amenable to stardard optimization algorithms. In addition, for
many design problems, most notably for ones where stiffness requires dominate, there
are multiple designs .with similar performance. These designs may have very different
stacking sequences but very similar A or D matrices, so that their in plane or bending
properties are nearly identical. In such cases, it is important for the optimization proce-
dure to produce all or most of these design alternatives.
Genetic algorithms are well suited for solving stacking sequence optimization prob-
lems. They deal easily with combinatorial problems, and they easily produce a multi-
tude of alternative solutions. Their main disadvantage is that they usually require thou-
sands or even millions of analyses. However, for many laminate design problems, the
computational costs of a single analysis are so low, that the large number of required
analyses is not a problem.
Genetic algorithms are search procedures, which imitate biological evolution and
work with a population of designs. Their development is usually credited to Holland
[I], and early applications to structural optimization are due to Goldberg and Santini [2],
and Hajela [3]. Genetic algorithms have been applied to composite laminate and stiff-
ened panel design by several researchers (e.g., [4]-[12]). In the process, these algo-
rithms have been tailored to work better for the design of composite panels.
431
432
The objective of the present paper is to describe some of the improvements intro-
duced into genetic algorithms for the design of composite panels, and some of the ap-
plications where these algorithms have been employed.
2. Genetic Algorithms
Genetic algorithms imitate biological evolution, working on populations of designs and

evolving them towards ever better properties. The attributes that define a particular
genetic algorithm include coding, selection. crossover, and mutation, each having
many forms of implementation, resulting in a great variety of possible genetic algo-
rithms. The following descriptions reflect choices made by the author and his cowork-
ers. The reader is advised to consult textbook on genetic algorithms, such as Goldberg'S
[13], and Michalwicz [14] for alternatives.
2.1 CODING
The primary design variables for the design of composite laminates are the number of
plies and their orientations. For stiffened panels. geometrical dimensions may also be
part of the design. The description here is mostly focused on coding the stacking se-
quence of a laminate, with a brief discussion of coding dimensional varibales.
For historical reasons, most genetic algorithm applications employ binary numbers
to code design variables. However, for coding composite laminates it is more conven-
ient and more efficient to use a more natural alphabet. With n different available ply
orientations, one may code them by the first n integers. For example, if the possible ply
orientations are 0°, ±45°, and 90°, we may assign these angles the codes 0" - 1, 45° - 2, -
45° - 3. and 90° - 4. Then the laminate (451-45190/0/90/-45/45) is coded as
(213/41111/41312). Furthermore, in applications which require symmetric laminates, only
half of the plies need to be coded, so that the previous laminate, (451-45190/0)s' is coded
as (2/3/411).
In biology, the amino acids that are the basic coding units in DNA strings are called
codons. a sequence of codons which codes for a single protein is called a gene, and a
collection of genes occupying a distinct spatial location is called a chromosome. In de-
sign applications, the smallest unit is often called a gene, and the entire string is called a
chromosome.
The coding becomes more complicated when the number of plies in the laminates is
not fixed, but instead it is part of the design. Most procedures for mating designs (a
process called crossover) assume that all designs are coded with strings of the same
length. For laminates of variable thickness this can be accomodated by coding a chro-
mosome with the highest possible size, and denoting empty plies with an additional
symbol, say O. For example, if the laminate (451-45190/0)s' is part of a population of
designs, where we allow a maximum of 12 plies, it could be coded as (0/012/3/411).
In many applications it is further required that the laminate is balanced, so that for
the examples above, there is a _450 ply for each 45° ply. Enforcing this balance condi-
tion via a constraint tends to slow down the progress of the algorithm. An alternative is
433
to have two-ply stacks of ±Bas the basic gene. For example, the 16-ply laminate
(0!90/±4S)" may be coded as (1/3/312)
This two-ply coding excludes some laminates from consideration, and may not be
advisable for thin laminates. However, for thick laminates, the effect on performance
will be minimal, and the algorithm will work much faster because the design space is
greatly reduced in size. For example, with a 48-ply laminate made of On, ±4Sn, and 900
plies, there are 4 48=7.9x 10'8 possible designs. A requirement of symmetry reduces the
design space to 424=2.8xI014 designs, and the use of two ply-stacks reduces it further
to 312=S3144I possible designs.
When the design variables include continuous dimensions, they are often discretized
for the purpose of coding. We can use the value of the variable directly as the gene it-
self, but that would require different crossover and mutation operators for the part of
the chromosome that codes for the dimensions. It may be more attractive to use the
same alphabet for the ply orientations and the dimensions. With an alphabet of size n
(that is each gene having n alleles) using m genes for each dimension, we can divide the
range of variation of a dimension to n"-l inervals. For example, with 3 possoble ply
stacks, we may want to use trinary numbers to model dimensions. Then if we have a
variable x varying in the range [1,10], if we use 4 digits to code this variables, we di-
vide the range into 80 intervals, with 0000 (trinary 0) coding for 1.0, 2222 (trinary 80)
coding for 10, and, for example, 0120 (trinary IS) coding for 1.687S.
In some applications, the number of plies of each orientation is determined by over-
all stiffness requirements for the laminate, and the only freedom is to change the stack-
ing sequence. For this case, normal coding may not be efficient, and special permuta-
tion coding have been devised (e.g. [1S]).
Finally, in some cases it is worthwhile to modify the translation of the chromosome
into a laminate for the purpose of satisfying some constraints on the laminate. Such con-
straints can be accomodated through the use of a penalty function, as discussed in Sec-
tion 2.2. However, this often leads to slower convergence. Satisfying constraints by
changing the decoding from a chromosome to a laminate is called repair, and one com-
mon application is to satisfy limits on the number of contiguous plies of the same ori-
entation. (e.g. [16])
2.2 FITNESS AND SELECTION
The process of selecting design for reproduction begins with evaluating their per-
formance through analysis. Then probabilities of selection as parents are determined
by processes biased towards superior designs. When there no constraints, ranking of
designs is based on the objective function. When constraints are present, the evaluation
of a design should take into account constraint violations and constraint margins. Most
applications perform this process through some form of a penalty function. Penalty
function are also often used for other optimization algorithms, with often strict require-
ment of continuity and differentiability of the penalty function. For genetic algorithms,
the best penalty function is often discontinuous.
In the following we assume that the objective function! is minimized (the negative
of ! may be used if it is maximized), and that the penalty for constraint violation is
determined only on the basis of the most critical constraint. We denote the violation in
434
the most critical constraint by v, or the margin in the same constraint by m, so that
alway mv=O. Both III and v are positive, and should be normalized so that a value
of 0.1 correspojnds to 10% violation or \0% margin. An augmented objective function
f , used to rank the designs is defined as
f*=f + pv-bm+sign(v)Ll .
where p is a penalty parameter for constraint violations, b is a bonus parameter for

constraint margins, and Ll is a small fixed penalty for infeasibility. The penalty pa-
rameter p should be chosen so that the penalty pv is comparable to the increase in the
objective fneeded to rectify the violation. Choosing very large values of p is undesir-
able, because it does not allow the search to take shortcuts through infeasible regions of
the design space. Often the feasible design domain is convoluted, and infeasible de-
signs can provide a good stepping stone to improvement in performance. High values
of p can reduce the fitness of such infeasible designs to the point where they have al-
most no chance of being selected as parents of future designs.
The small infeasibility penalty.1 helps us avoid large values of p. Without 4 we
may be forced to select a high value of p to avoid convergence to an infeasible design.
For example, consider a case where f is half of the number of plies in a symmetric
laminate. We may have, say, three designs with (j=12, v=0.01), (j=J3, m=0.05), and
(j=14, m=O.25). We would like the middle design to have the lowest augmented objec-
tive, and this limits the value of the bonus parameter b to be less than 5. Then without
using 4 the penalty parameter p would have to be at least 75. Such a high value of p
means that any design with 5% constraint violation will have a penalty of about 4 plies,
lowering greatly its chances to be selected. If we use .1=0.5, we can reduce the value
of p below 30.
Once an augmented objective has been determined, most genetic algorithms trans-
late it to an intermediary called fitness. The fitness of a design can be the value of the
objective function normalized over the range of variation of that objective. A popular
alternative is to use ranking of designs to determine fitness. For example, with n, de-
signs, the fitness may be defined to be n, -1 minus the rank. Thus with four designs, the
best will always have a fitness value of 4 the poorest will always have a fitness of one.
Once fitness has been assigned to all designs, the probability of selection as a parent
of future design is often set equal to the value of the fitness over the sum of all fitnesses.
For example, with 4 designs, and rank based fitness, the probabilites of selection, of the
four designs are (0.4,0.3,0.2,0.1). One may liken the process of selection to spinning a
roulette wheel with the wheel divided into pie slices proportional to the fitness, so this
process is often referred to as a roulette wheel selection.
Most genetic algorithms repeat the selection process to obtain a pair of parent de-
signs, which are then mated in a process of crossover described in the next section, and
mutated to procduce one or several child designs. Some genetic algorithms replace par-
ent designs piecemeal with child designs. However, most create as many child designs
as there are parent designs, by repeatedly selecting pairs of parents and mating them.
Then they replace the entire parent generation with the child generation. For these algo-
rithms, the number of analyses required is the number of generations times the popula-
tion size. Finally, many genetic algorithms clone the best design of the parent generation
435
into the child population, in a process called elitist selectioll. This has not been shown to
improve performance, but it guarantees monotonic convergence.
2.3 CROSSOVER
Crossover is the process of combining genes from two parents to obtain a child design.
The simplest crossover is a single-point crossover, where a the chromosomes of the two
parents are cut at some random point, and then spliced together. For example, given two
parent designs [0/±45/902], and [±45/021., coded as [1111212/3] and [212121211], respec-
tively, a cut after the fourth gene would produce
Parent 1 [1I112121JJ.3]
Parent 2 [21212121111]
Child [ 111121213]
It is also possible to create another child from the same process, with the front segment
coming from the second parent and the rear segment from the first. Besides this one-
point crossover, two-point crossovers are also popular. In these chromosomes are cut at
two random points, and a child can be created by taking the first and third segment from
the first parent, and the second segment from the second parent. Finally, there is also a
uniform crossover, where each gene is selected randomly from one parent or the other.
Single point crossover is optimal for preserving sequences of plies that work well
together (so caIl schemata), and will not be likely to be dismembered by the process.
Uniform crossover, will dirupt the formation of such sequences, but it protects against a
problem called hitchhiking, where a poor gene persists for many generations because it
is located next to a good gene. For most composite design applications, the formation of
building block sequences is important, and so one-point or two-point crossovers are
commonly used.
When the thickness of the laminate is not fixed, so that some of the genes describe
empty stacks, more care needs to be taken in the process of crossover, as discussed in
[9]. Also, following crossover, empty plies may be moved between full plies, and the
chromosome needs to be compacted to eject the empty plies to the outside. When per-
mutation coding is used, normal crossovers do not work, and specialized crossovers
have been developed for that case [15]. Finally, crossovers are often applied with a
given probability (usually about 80%). When crossover is not applied to the pair of par-
ents, one or both are simply cloned into the child population. For composite laminate,
most experience indicates that crossover should be applied with 100% probability.
2.4. MUTATION
Crossover is often termed a process of exploitation in that it exploits design features

already present in the population. However, the optimal design features may not be pre-
sent in the initial population, or some may be lost if designs incorporating these features
are not selected as parents. Mutation introduced local random changes into the design
and allow us to create new features, in a process of exploration of the design space.
The most common form of mutation is to change a single gene randomly.
436
A common rule of thumb is that the probability of mutation occurring somewhere i

each child design should be over 50%. In some implementations one gene is selected
randomly from the chromosome with the target probability and only that gene is mu-
tated. In other implementations each gene is considered separately, and mutated with
very small probability. For example, if the desired overall proability of mutation is 80%,
and we have 10 genes, then the probability of mutating an individual gene should be
about 15%. When the chromosome includes empty plies, it is advisable to mutate only
the full plies, and have separate mutation operators that add and delete plies [9].
Another form of mutation, rarely used in other genetic algorithm applications, but is
common in composite laminates is the gene-swap. This consists of interchanging the
values of two genes. This mutation does not change the in-plane properties of the lami-
nate but only its bending properties, and has shown to be very beneficial for problems
involving bending. The optimal probability of this gene-swap is often stated as higher
than 100%, which means that it is applied more than once to each child design. As an
example of both types of mutation, consider a chromosome [111/2/2/3] which undergoes
a mutation of the first gene and a swap of the fourth and fifth genes to become
[2/1/2/3/2]. The corresponding stacking sequences change from [O/±45/9021. to
[±45/0/±45/90/±45],.
3. Convergence, performance and tuning
Genetic algorithms are random search procedures, and as such their performance is also
random. This creates difficulties in comparing performance of different alternatives and
in the selection of parameters such as population size and probabilities of mutation.
Furthermore, it is difficult to have a reasonable convergence criterion.
Some implementations use a convergence criterion based on the number of genera-
tions without improvement in the best design. However, it is quite common for genetic
algorithms to proceed for many generations without improvement, and then to spurt
ahead once a critical mutation has been introduced. For this reasons many implementa-
tions simply run the algorithm for a given number of generations. This approach has the
added advantage that it makes comparison of performance easier because the computa-
tional cost of the algorithm is fixed.
With fixed computational cost, the performance is measured by the probability of
reaching the global optimum. This probability, called the reliability of the search, is
calculated by repeating the genetic search many times. This is possible only for inex-
pensive problems, and therefore simple versions of the problem that need to be solved
are used to compare performance of different alternatives. For example, if we consider
the buckling of stiffened plates, we may test the algorithm for a simply supported un-
stiffened plate, where an analytical solution is readily available.
The apparent reliability of a genetic algorithm is the ratio of the number of runs that
found the optimum design to the total number of runs. If the true reliability (that is the
same ratio for an infinite number of runs) of the algorithm is r, it is easy to show that the
t(l-r)
standard deviation cr, of the apparent reliability over II runs is
CY, =
11
437
so that for 100 runs and r=0.63, we obtain a standard deviation of about 0.04S. When we
tunc parameters of a genetic algorithm, such as population size, by comparing the reli-
ability of the algorithm, it is essential to make enough runs so that differences in appar-
ent reliability are substantially larger than the standard deviation of the apparent reli-
ability.
When the reliability of the algorithm is too low, it is possible to improve it by run-
ning the optimization for more generations. Often, though, a better strategy is to plan to
run the genetic algorithm several times for each design study. If the reliability of a sin-
gle run is r, the reliability of m runs (that it the probability that one of these runs will
find the optimum) is 1-( 1-rt. So, for example, if the reliability of the algorithm for a
given number of analyses is 50%, the reliability in 5 runs is about 97%.
Finally, genetic algorithms are notoriously inefficient in completing convergence
from the neighbourhood of the optimum to the exact location of that optimum. For this
reason, it may be more efficient to complete the optimization with local searches started
from the best designs identified by the genetic algorithm. Also, it may be reasonable to
estimate reliability in terms of getting close to the optimum design rather than reaching
it exactly. This kind of reliability estimate is also called practical reliability.
4. Examples
4.1 UNSTIFFENED PLATE WITH GIVEN NUMBER OF PLIES
As an example of the importance of handling constraints efficiently we consider the

design of a simply supported unstiffened panel with a given number of 0°, ±45°, and 90°
plies from [17]. A symmetric and balanced stacking sequence is designed to maximize
the buckling load of the panel under normal in-plane loads N, and Ny' and a shear load
per unit length N",. Because the panel is simply supported an analytical equation is
available for the buckling load under bi-axial loading, an infinite long plate approxima-
tion is used for buckling under the shear loading, and an interaction formula is used for
the combined buckling load. These simplifications allow us to calculate the buckling
load very inexpensively. This allows [17] to perform hundreds of genetic optimization
for comparing different algorithms and for assessing the reliability of each algorithm.
The constraints applied to the panel are that the number of plies of each orientation
is given, and the number of contiguous plies of the same orientation should not exceed
four. The constraints that the laminate be symmetric and balanced were automatically
satisfied through the coding. Only half of the laminate was modeled, and the basic units
were 0" 90" and ±45 stacks.
The constraint on the number of plies was handled in three ways. First, a standard
genetic algorithm (SGA) used penalty function for this violation. Second, an SGA plus a
repair system for the constraint was used. Finally, two permutation genetic algorithms
were used, a partially mapped crossover (PMX) algorithm due to Goldberg and Lingle
[18], and a gene-rank algorithm (GR) developed in [17]. Similarly, the constraint on the
number of contiguous plies was dealt with by a penalty function and by repair.
Results were obtained for a 24-inch square graphite-epoxy plate with following
propcrties E1=18.5xI06 psi (127.59 GPa), E,=1.89xl0· psi (13.03 GPa), G1,=0.93xI0·
psi (6.4GPa), '1,,=0.005 in (0.0127 cm), and v 1,=0.3.
438
Table 1. Three Test Laminates
Case Nx (Ib) Ny (Ib) Nxv (Ib) Ilo n45 n90 ntota)

I 0 -16000 8000 8 16 8 32
2 15980 -14764 10160 9 8 13 30
3 -16657 1963 828 13 7 15 35
Table 1 shows the definition of the loads and number of plies for three laminate de-
sign problems. Table 2 shows the performance of the three algorithms without repair,
and it is evident that the elimination of the constraint on the number of plies through the
permutation coding greatly enhances performance. It is also clear that the contiguity
constraint exacts very high cost even from the permutation genetic algorithms.
Table 2. Comparison of Computational Efficiency of Three GAs for Three Thick

Laminates
Case Number of Analyses Required For 80% Reliability
SGA GR PMX
With Conti- Without With Conti- Without With Conti- Without

guity Con- Contiguity guity Con- Contiguity guity Con- Contiguity
straint Constraint straint Constraint straint Constraint
1 7984 5112 1328 480 1480 848
2 23544 2176 11840 360 5784 336
3 26230 5024 2216 296 2504 840
Table 3 shows the effect of repair on the three algorithms. The effect on the standard
GA is most dramatic, because we repair for both contiguity and number of ply con-
straints. However, even the permutation GAs show dramatic improvements.
Table 3: Number of analyses required for 80% reliability for three algorithms with
repair.
case SGA GR PMX

I 400 416 552
2 3512 288 336
3 48 352 696
439
4.2. BLADE STIFFENED PLATE
As an example of the use of a genetic algorithm for a computationally expensive prob-

lem, we take a stiffened panel design study from [12] . Figure I shows a four blade stiff-
ened panel subjected to combined axial and shear loading. In addition the panel is as-
sumed to have a longitudinal bow of 3% of the panel length.
(3.0736)
Figure 1: Blade stiffened panel with four equally spaced stiffener under compres-
sion and shear (dimensions are in inches).
440
The panel was designed subject to buckling and maximum strain constraints using
the PASCO program [19] for analysis. The stiffener blade and flange were constrained
to have identical laminates. The laminates were constrained to 0°, ±4So, and 90° plies,
and to reduce the likelihood of matrix cracks, the number of contiguous plies of the
same orientation was restricted to four or less. To improve damage tolerance the skin
laminate did not include 0° plies, and both the skin and blade laminate were required to
have ±4So plies on their surfaces. The optimization variables included the stacking se-
quences of the skin and the blade as well as the blade height.
The laminates were coded in terms of two-ply stacks of of O2, 902 or ±4S, coded as
1,2, and 3, with 0 designating empty stacks. The height of the blade was coded as a 9-
digit trinary number in the range (1.S,3.3), allowing a resolution of about 0.001 in.
A single analysis required about 30s of computer time, and so the number of analy-
ses that could be run by the genetic algorithm was limited to about 3000, representing a
24-hour run. The panel was originally optimized by using continuous optimization and
rounding in [20] resulting in a design with a mass of 26.08 lb. The panel was then op-
timized by a basic GA in [21], leading to 10 designs ranging in weight between 2S.19
lb., and 25.S8 lb. In [12] the genetic operators were improved by introducing special
mutations for deleting and adding stacks, as well as swapping stacks inside and between
laminates. In addition, the parameters of the genetic algorithm, such as probabilities of
mutation were tuned on a simpler and less expensive version of the same problem.
The pay-off from the improvements in the genetic algorithms was a significant im-
provement in the weight. Forty genetic optimizations were performed, and the 10 best
designs ranged from 24.14 lb. to 24.62 lb. Great variability in the designs was observed.
The skin laminate varied in thickness from 32 to 40 plies, the blade laminate from 68
plies to 80 plies, and the blade height varied from 2.92 in. to 3.17 in. This demonstrates
the potential of genetic algorithms to find a variety of similar performance designs that
a designer may choose from.
5. References
1. Holland, J.H. (197S) Adaptation in Natural and Artifcial Systems, University

of Michigan Press, Ann Arbor, Michigan.
2. Goldberg, D.E., and Santini, T.A. (1987) Engineering Optimization by a Ge-
netic Algorithm, Proc. 9th con! Electronic Computation, ASCE, pp. 471-
482.
3. Hajela, P., (1990) Genetic Search - an Approach to the Nonconvex Optimi-
zation Problem, AlAA JOllrnal26 (7), pp. 120S-121O.
4. Callahan, KJ., and Weeks, G.E. (1992) Optimum Design of Composite
Laminates using Genetic Algorithms, Composites Engi1leeri1lg, 2 (3), pp.
149-160.
441
5. Ball, N.R., Sargent P.M., and Ige, D.O. (1993) Genetic Algorithm Repre-
sentations for Laminate Layups, Artificial intelligence in Engineering, 8 (2),
pp.99-108.
6. Le Riche, R. and Haftka, R.T. (1993) Optimization of Laminate Stacking
Sequence for Buckling Load Maximization by Genetic Algorithm, AIM
Journal, 31 (5), pp. 951--956.
7. Kogiso, N., Watson, L.T., GUrdal, Z., and Haftka, R.T. (1994) Genetic Algo-
rithms with Local Improvement for Composite Laminate Design, Structural
Optimization, 7, (4), pp. 207--218.
8. Marcelin J.L., and Trompette, P. (1995) Optimal Structural Damping of Skis
using a Genetic Algorithm, Structural Optimization, 10, pp., 67-70,
9. Le Riche, R. and Haftka, R.T. (1995) Improved Genetic Algorithm for
Minimum Thickness Composite Laminate Design, Composite Engineering,
5(2) pp. 143-161.
10. Todoroki, A., Watanabe, K., and Kobayashi, H (1995) Application of Ge-
netic Algorithms to Stiffness Optimizaion of Laminated Composite Plantes
with Stress Concentrated Open Holes, JSME International Journal, Series A,
38 (4), pp. 458-464.
11. Yamazaki, K. (1996) Two-Level Optimization Technique of Composite
Laminate Panels by Genetic Algorithms, AIAA Paper 96-1539, Proceedings,
37th AlAAIASMEIASCEIAHS Structures, Structural Dynamics and Material
Conference, Salt Lake City, UT, April 15-17, 1996, Part 3, pp. 1882-1887.
12. Nagendra, S., Jestin, D., GUrdal, Z., Haftka, R.T., and Watson, L.T. (1996)
Improved genetic Algorithms for the Design of Stiffened Composite Panels,
Computers & Structures, 58 (3), pp. 543-555.
13. Goldberg, D.E. (1989) Genetic Algorithms in Search, Optimization and Ma-
chine Learning, Addison-Wesley.
14. Michalewicz, Z. (1992) Genetic Algorithms + Data Structure = Evolution
Programs, Springer-Verlag,.
15. Liu, B., Haftka, R.T., and Akgun, M.A. (1998) Permutation Genetic Algo-
rithm For Stacking Sequence Optimization, AIAA Paper 98-1830, Proc.
39th AlAAIASMEIASCEIAHSIASC, Structures, Structural Dynamics, and
Materials Conference, Long Beach, CA, Apr. 20-23, 2, pp.1141-1152.
16. Todoroki, A., and Haftka, R.T. (1998) Optimization of Composite Laminate
by using Genetic Algorithm with Recessive Gene like Repair Strategy, in
press, Composites B.
17. Liu, B., Haftka, R.T., and Akgun, M.A. (1998) Permutation Genetic Algo-
rithm For Stacking Sequence Optimization, AIAA Paper 98-1830, Proc.
39th AlAAIASMEIASCEIAHSIASC, Structures, Structural Dynamics, alld
Materials Conference, Long Beach, CA, Apr. 20-23, Part 2, pp. 1141-1152.
442
18. Goldberg, D.E. and Lingle, R. (1985) Alleles, loci, and Travelling salesman
Problem. Grefenstette, 1.1. (Editor), Proceedings of the First International
Conference on Genetic Algorithms, Lawrence Erlbaum Associates, Hillsdale,
NJ, pp 154-159
19. Stroud, W.J., and Anderson, M.S. (1981) PASCO: Structural Panalel Analy-
sis and Sizing Code: Capability and Analytical Foundations, NASA Techni-
cal Memorandum 80801.
20. Nagendra, S., Haftka, R. T., Gtirdal, Z., and Starnes, 1. H. (1991) Design of
Stiffened Composite Plate with a Hole, Composite Structures, 18, pp. 195-
219.
21. Nagendra, S., Haftka, R.T., and Gtirdal, Z. (1993) Genetic Algorithm Based
Design Procedure for Stiffened Composite Panels under Stability and Strain
Constraints," 10th DODINASAIFAA Conference on Fibrous Composites in
Structures, Hilton Head, Sc.
IDENTIFICATION TECHNIQUES
IN COMPOSITE LAMINATES
PAULI PEDERSEN
Department of Solid Mechanics
Technical University of Denmark
Building 404, DK-2BOO Lyngby, Denmark
Abstract
Combined experimental-numerical methods are presented with the goal of obtaining mate-
rial stiffness for composite materials. The identification is based on eigenfrequencies for
a free rectangular plate, because excellent agreement between measured and calculated ei-
genfrequencies can be obtained. The numerical identification problem is formulated as an
optimization problem and in one experiment we obtain all the involved moduli.
The total approach consists of experiment, numerical eigenvalue analysis, eigenvalue
sensitivity analysis, optimization, and error estimation. Recent results from these different
parts of the approach are commented and the general conclusion is that the method has sev-
eral significant advantages compared to traditional determination of material moduli. An
overview of the state of the art is intended.
1. Introduction
Traditional laboratory tests are designed to give the desired quantiti~s in a direct manner.
As an example, determination of material parameters of constitutive models leads to special
design of test samples and choices of applied load in an attempt to obtain homogeneous
stress-strain fields. These idealized tests are not easy to perform, and especially with com-
posite systems problems often arise. Furthermore, the tests have a local nature, e.g. the point
of the strain gauge. So for more general material information a series of tests are required.
The identification approach is an experimental strategy from an opposite point of view.
The experiment is made as simple as possible to give reliable results. On the other hand,
this often leads to complex inhomogeneous stress-strain fields where the desired quantities
are not among the directly measured quantities. This causes a complicated interpretation,
but nowadays this is comfortably taken care of by computer calculations. Although our aim
is to find static material data, we decide to measure eigenfrequencies of a free rectangular
plate because excellent agreement between measured and calculated eigenfrequencies can
443
444
be obtained. A structural eigenfrequency is an integrated quantity and we thus obtain mate-
rial quantities that are valid in the mean for the entire laminate.
In a pilot project by Markworth and Petersen (1987) this technique was found most
promising (some of the results can be found in Pedersen and Frederiksen (1992)). The same
conclusions was reported in the doctoral thesis by Sol (1986). The research groups in Brus-
sels, Belgium and in Eindhoven, The Netherlands have continuously refined the method
and obtained impressive results. The proceedings from an Euromech colloquium, edited by
Sol and Oomens (1997), give a picture of the state of the art. A recent doctoral thesis by
Kuttenkeuler (1998) shows the importance of these techniques for composite design.
In the authors home department the extended work of Per S. Frederiksen (1992, 1995,
1997, 1998) reports to a great depth about the method. Especially the work related to thick
plate formulations, Frederiksen (1995, 1997b), should be noted and also the recent work
related to parameter uncertainty, Frederiksen (1998), deserves special attention.
The research group in Portugal has concentrated on the combination with the finite ele-
ment method and has also applied thick plate formulation. Reference is given to the paper
by Mota Soares et at. (1993).
In these lecture notes we first formulate the identification problem from an optimization
point of view. Then the experimental set-up is described, followed by numerical eigenvalue
analysis and sensitivity analysis. The optimization algorithms will be shortly commented,
and then results from the literature are commented. Among these are the study of material
moduli in different environments.
2. Identification Formulation
A structure can be described in many ways, directly or indirectly. The directly approach can
be by parameters, like length, width, height, weight, topology and material parameters. Al-
ternatively, we can describe it by its response like static displacements or dynamic eigenfre-
quencies. Combinational approaches are also possible.
A laminated plate is a structure where the ply stacking is an important aspect. The thick-
ness of the plies and their positions in addition to ply orientations are the parameters that
describe this stacking. Also boundary conditions are of vital importance for the response
of the plate. The laminate membrane stiffnesses [A] as well as the laminate bending stiff-
nesses [D] are linearly depending on the material stiffness [C] for each ply, i.e. upon the
constitutive matrix.
We can describe the identification problem as a problem where we do not have informa-
tion enough about [C], but we have information about the structural response, say in terms
of measured eigenfrequencies Wi' Let us be specific and formulate the identification prob-
lem as an optimization problem.
A number of parameters Pk are unknown for the determination of the material stiff-
nesses
(2.1)
445
and therefore a calculated eigenfrequency response will also depend on these parameters
(2.2)
The functional relations (2.1) and (2.2) can be found in textbooks on laminates, say in Jones
(1975). As an alternative to the relation (2.1) we could formulate the identification problem
on the level of bending stiffnesses
(2.3)
which then eliminate the need for the stacking information, but on the other hand do not give
detailed information.
The optimization problem is to find best possible agreement between measured eigen-
frequencies Wi and calculated eigenfrequencies Wi. An objective could be to
= I
I
(Wi - wi
Minimize <P
i=l
w; (2.4)
i.e., a weighted least-squares estimator, but naturally other choices are possible. The solu-
tion must be obtained in an iterative process, where the parameters are simultaneously im-
proved
(2.5)
and constraints can be put on the absolute values of the parameters (side constraints) as well
as on the change of the parameters (move limits)
With these kinds of constraints we do not a priori need to distinguish very clearly between
known and unknown parameters.
In section five we shall comment on the optimization procedure. Here the choice of pa-
rameters will be discussed. For orthotropic two dimensional models, two parameters are
especially important. The level of anisotropy described by the parameter a 2
(2.7)
where E T , EL are the engineering moduli in transverse and fiber direction, respectively.
The relative shear stiffness is described by the parameter a 3
(2.8)
where vLT is the major Poisson's ratio and G LT is the shear modulus. The non-important
parameter a o is given by a o : = I - v~TET/EL , and a further not so important parameter
is a 4
(2.9)
446
i.e. only new information from the Poisson's ratio. Solving the problem (2.4) in the formula-
tion
(2.1O)
we can then invert the relations (2.7}-(2.9) and identify E L , ET , G LT , V LT . In terms of the
non-dimensional quantities the condition for [C] to be positive definite will be
(2.11)
3. The Experimental Set-Up
A detailed description of the experimental procedure and the critical aspects of this can be
found in Frederiksen (1 997a}. We shall here describe only the main aspects. As shown in
fig. 3.1 the set-up consists of six main components:
I} the test specimen, i.e. the laminated plate
2} the excitation source, i.e. an impact hammer or a loudspeaker
3} the response detector, i.e. an accelerometer, a microphone or a laser vibrometer
4} a power amplifier
5} a transient recorder and frequency analyzer
6) a personal computer, which includes the identification program(s}
D
B~
/ ~
5 6
Fig. 3.1: Scheme of the experimental set-up.

447
The test specimens in the works of Frederiksen were chosen as rectangular, orthotropic and
symmetric laminates. In the work by Kuttenkeuler (1998) also non-rectangular test speci-
mens were used, but then the numerical analysis needs to be based on finite element model-
ling. In the section on uncertainties we shall discuss the other physical aspects for the test
specimens.
As excitation source we mostly use a small impact hammer, which imparts a force with
a broad frequency range to the plate. This will simultaneously excite all the modes of vibra-
tion. The pulse frequency range can be altered by adjusting the tip-hardness and the ham-
mer weight. In the identification of complex moduli by Vissher et al. (1997) the excitation
by a loudspeaker or a disconnected shaker is also described. The joke is:"ask the plate for
its moduli", but do it loudly (in any language).
In the early research a light weight accelerometer with a mass of 2.2 gram was used
as response detector, and the concentrated mass is taken into account in the eigenfrequency
calculations. In more recent research, especially for the study of small and light plates, a
microphone is used as response detector. In Vissher et at. (1997) a laser vibrometer is shown
as response detector. In general we can state that good experimental repeatability is ob-
tained and reported by the different authors.
The power amplifier, the transient recorder and the PC equipment are standard ex-
perimental tools. The software separation between the transient recorder and the PC is not
very specific. Thus, the fast Fourier transformation could be contained in either of them.
4. Plate Eigenfrequency Analysis and Sensitivity Analysis
Plate eigenfrequencies can be calculated with different methods. For the numerical model-
ling as well as for the numerical procedure we have many possibilities and in this section
we shall comment on the ones used for the actual identification problem. For details see Fre-
deriksen (1995, 1997b).
The numerical modelling is performed either by the Rayleigh-Ritz method with global
expansion functions or by the finite element method with local expansion functions. Both
methods are applicable to rectangular plates, but for more complicated boundary shapes
only the finite element method will be practical. The main issue of the numerical modelling
is to choose the relevant plate theory, and it is generally agreed that higher order plate theo-
ries are often required. In the studies of Frederiksen (1997b) an estimation of the errors that
result from the use of the classical plate theory is included. The higher-order shear deforma-
tion theory by Reddy (1984), Levinson (1981) is a good compromise between accuracy and
complexity.
When the numerical modelling is chosen we end up with a generalized, linear eigenva-
lue problem
(4.1)
where [S] is a "stiffness" matrix, [M] a "mass" matrix, and wf, {d a squared natural
frequency and its corresponding eigenvector. In a Rayleigh-Ritz model the order of the ma-
448
trices are say 500, while in a finite element model the order may be say 20.000. (Compari-
sons of results from the two different methods can be found in Frederiksen (1997b)). The
assumption of linearity is justified by very small strains during the vibrations
(1-10 ~-strains) .
An effective numerical procedure to solve the matrix eigenvalue problem (4.1) is the
subspace iteration method, see Bathe (1982). By iteration this method determines the eigen-
vectors and then the eigenvalues in a selected domain of the spectrum, say the lowest eigen-
frequencies. The method is capable of dealing with multiple eigenfrequencies, and in gener-
al computer time is no problem.
For the identification we also need the partial derivatives of the eigenfrequencies with
respect to the material moduli, here symbolized by a parameter Pk' Results for this sensi-
tivity analysis can be found in the literature on optimal design. The general result for non-
multiple eigenfrequencies are
Td[S] Td[M]
{<pd dih {<pd - wf{ <pd dp;;" {<pd
(4.2)
{<pJ T [M]{ <pd
and normalizing with {<Pi} T[M] {<Pi} I we get for the actual problems
d(wf) = {rh}Td[S] {A} (4.3)

dpk 't'l dpk 't'l
because the material moduli have no influence on the mass matrix, i.e. d[M]/dpk = [0] .
For multiple eigenfrequencies a special calculation is necessary to obtain the eigenvec-
tors that decouple the sensitivities. With {<pd, {<pj) being two eigenvectors corresponding
to the same eigenfrequency Wi' we determine these vectors uniquely by the conditions:
(4.4)
{rh.}Td[S] {rh.) = <')d(Wf)

't'l dpk 't'J IJ dpk
where <')ij is the Kronecker delta.
5. Optimization Procedures and Comments on Results
In the different studies of identification techniques, different objectives and different opti-
mization algorithm has been applied. Different algorithms for the different steps has been
necessary to remove the necessity of good initial guesses. As local minima are an actual
problem Frederiksen (1992a) suggests initial steps based on rather rough estimation to get
good initial guesses before the more advanced optimization algorithms are applied. The ref-
erences given in this thesis should be consulted for more detailed advice on the optimization
algorithms.
449
Let us comment on results for weakly, moderately and strongly anisotropic materials.
AroUed aluminium plate is identified in the early work by Markworth and Petersen (1987)
which based on ten frequencies found EL = 68.9 GPa, ET = 66.0 GPa ,
G LT = 24.8 GPa, v LT = 0.343 and fairly good agreement with traditional uniaxial tests.
Frederiksen (1 997a) for another aluminium plate identified EL = 71.1 GPa,
ET = 70.7 GPa, G LT = 25.9 GPa, V LT = 0.336, G LZ = 27.9 GPa and
GlZ = 28.0 GPa, thus showing the possibility for identifying the out-of-plane shear
moduli G LZ and G lZ .
For glass-epoxy also the early studies report good agreement with static strain gauge
test results in the identification of the in-plane moduli. Based on 14 frequencies, Frederik-
sen (1997a) identified the following parameters EL = 42.4 GPa, ET = 11.6 GPa,
G LT = 4.68 GPa, v LT = 0.305, G LZ = 4.55 GPa, GTZ = 4.07 GPa , with details on
60 22
.....0 '0
58
e ea.
Q.
20
u:1 56 W
iii iii 18
:J :J
"5 "5
"0 504 "0
0 0
E E
01 01
-01 -01 16
C 52 c
:J :J
~ ~
50 14
0 20 40 60 80 100 120 140 0 20 40 60 SO 100 120 140
Temperature (Oe) Temperature (Oe)
10.0 0.40
1. 9.0 0.35
......
t!)
~
~
Js.o
iii
:J
e
c5
0.30
~ 7.0 -<II
C
0 0
E 01
... <II
·00.25
~ 6.0 a.
.t:
III
5.0 0.20
0 20 40 60 80 100 120 140 0 20 40 60 SO 100 120 140
Temperature COe) Temperature ("e)
Fig. 5.1: Temperature dependence for glass/epoxy ply obtained by identification.

450
iteration history and the extremely small residuals from experimental and calculated fre-
quencies.
Carbon-epoxy identification is properly of most interest and extended results are
available, in early as well as recent studies. Let us here again comment the result of Frede-
riksen (1997a), who reports EL = 113.0 GPa, ET = 8.50 GPa, G LT = 4.45 GPa,
VLT = 0.323, G LZ = 4.43 GPa , G TZ = 2.97 GPa . We find almost transversely iso-
tropic stiffness for this unidirectional laminate and the ratio GTZ/G LZ = 0.666 is in other
studies reported between 0.548 and 0.624. Note that the determination of out-of-plane
shear moduli is very difficult to perform by other experimental techniques.
Finally, we show in fig. 5.1 results from identification of the temperature dependence
for glass--epoxy moduli; mainly to illustrate the possibilities when knowledge on environ-
mental influence is important.
6. Uncertainties and Optimal Experiments
In the early thesis of Sol (1986) and also in the recent thesis of Kuttenkeuler (1998) assess-
ment of uncertainties was given priority. In the extended study of Frederiksen (1998) these
questions are analysed based on statistics, and we shall start this section by citing the ab-
stract of Frederiksen (1998):
"Abstract - This paper investigates an inverse technique for the identification of

orthotropic elastic constants from measured plate natural frequencies. In general,
the accuracy of the identified parameters depends on the method of estimation, mod-
elling errors and measurement errors. The paper addresses the parameter uncer-
tainty due to errors in the measurements. Based on assumptions of the measurement
errors, second-order statistics of parameters are approximated by linearization
schemes. The main focus is on the possibility of designing the experiment to mini-
mize the uncertainty of the estimated parameters. The uncertainty of each estimate
as function of the experimental design variables is investigated. Also the overall op-
timality of the experimental design defined as the hypervolume of the confidence re-
gion is considered. The results show that not all parameters are estimated with a suf-
ficient precision in the general case, but by carefully designing the experiment, the
parameter uncertainties can be greatly reduced. Both thin and thick plates are con-
sidered with focus on single-layer plates, but the results for laminates plates are also
discussed. "
The individual errors can be grouped into
a) Experimental errors
b) Physical modelling errors
c) Numerical modelling errors
d) Plate defects.
451
The experimental errors are reported as limitations of the instruments, and are general-
ly low. Frederiksen (1998) states the standard deviations of the frequency values to be less
than 0.2% . Measuring the thickness of a laminated plate may be non-unique and a method
based on volume measure in water is therefore suggested.
The physical modelling errors includes the assumptions related to linearity, damping,
surrounding air, gravity, suspension and frequency dependence. Most critical of these
sources might be the suspension, but with a free plate only hanged in rubberbands the
boundary conditions are well established. Other boundary conditions will normally cause
severe errors. Lowest and highest frequency in the identification often deviate by one order,
and if the material moduli are frequency dependent we should conclude that averaged prop-
erties in some sense are found.
The numerical modelling errors has been investigated in detail in Frederiksen (1995).
For the first 15 frequencies, the maximum error was found to 0.4% for a moderately thick
orthotropic plate. For most of the frequencies the error is considerably smaller.
Through many experiments the most essential problems were related to plate defects,
such as non-rectangular shape, non-uniform thickness, non-zero plate curvature, and most
important a non-homogeneity on the macro level. This last aspect has different influence
on the individual eigenmodes, eigenvalues and therefore often makes identification impos-
sible, which in a way is better than an artificial adjustment of the moduli.
7. Conclusion
The identification technique is a fascinating approach involving very close cooperation be-
tween experimental and numerical work. The studies have shown that reliable values of the
six most important moduli of orthotropic laminates can be determined from one experi-
ment, measuring plate frequencies.
Because of the simplicity of the free isolated specimens, environmental influence on
material moduli can easily be determined. Results from temperature dependence are shown
to illustrate these potentials. Naturally also other materials than composite laminates can
be studied with the method. With thick plate modelling these potentials are greatly enlarged.
As example we mention bone material (see early study by Thomsen (1990)), ceramics, and
sandwich combinations.
The design of optimal experiments for identification of specific parameters is an impor-
tant new aspect that follows from the statistical analysis for determination of parameter un-
certainties. Bounds related to the identified parameters on the laminate level as well as on
the ply (lamina) level are important additional information to obtain.
452
References
1. Bathe, K.J. (1982) Finite Element Procedures in Engineering Analysis, Pren-

tice-Hall.
2. Frederiksen, P.S. (1992a) Identification of material parameters in anisotropic plates
- a combined numericaVexperimental method, Ph.D. Thesis, Dept. of Solid Me-
chanics, Technical University of Denmark, Denmark.
3. Frederiksen, P.S. (1992b) Identification of temperature dependence for orthotropic
material moduli, Mechanics of Materials 13, 79-90.
4. Frederiksen, P.S. (1995) Single-layer plate theories applied to the flexural vibration
of completely free thick laminates, J. of Sound and Vibration 186, 743-759.
5. Frederiksen, P.S. (1997a) Experimental procedure and results for the identification
of elastic constants of thick orthotropic plates, J. Composite Materials 31, 360- 382.
6. Frederiksen, P.S. (1997b) Numerical studies for the identification of orthotropic
elastic constants of thick plates, Eur. J. Mech., A/Solids 16, 117-140.
7. Frederiksen, P.S. (1998) Parameter uncertainty and design of optimal experiments
for the estimation of elastic constants, Int. J. Solids Structures 35, 1241-1260.
8. Jones, R.M. (1975) Mechanics of Composite Materials, McGraw-Hill.
9. Kuttenkeuler, J. (1998) Aircraft composites and aeroelastic Tailoring, Thesis, Royal
Institute of Technology, Sweden.
10. Levinson, M. (1981) An Accurate, Simple Theory of the Statics and Dynamics of
Elastic Plates, Mech. Res. Commun. 7,81-87.
11. Markworth, N.J. and Petersen, C. (1987) Identifikation afmaterialeparametre for
fiberbaserede laminatplader (in Danish), McS Thesis, Solid Mechanics, DTU.
12. Pedersen, P. and Frederiksen, P.S. (1992) Identification of orthotropic material
moduli by a combined experimental numerical approach, Measurement 10, 113-
118.
13. Reddy, 1.N. (1984) A Simple Higher Order Theory for Laminated Composite Plates,
J. Appl. Mech. 51, 745-752.
14. Soares, C.M.M., de Freitas, M.M., Aranjo, A.L. and Pedersen, P. (1993) Identifica-
tion of material properties of composite plate specimens, Composite Structures
25(1-4),277-285.
15. Sol, H. (1986) Identification of anisotropic plate rigidities using free vibration data,
Thesis, Free University of Brussels, Belgium.
16. Sol, H. and Oomens, C.W.J. (Eds.) (1997) Material Identicication using mixed nu-
merical experimental methods, Proc. Euromech 357, Kerkrade, The Netherlands,
KluwerAcademic Publishers, Dordrecht, The Netherlands
17. Thomsen, J.J. (1990) Modelling human tibia structural vibrations, J. Biomechanics
23(3),215-228.
18. De Vissher, J., Sol, H., De Wiede, W.P. and Vantomme, 1. (1997) Identicication of
the complex moduli of thin fibre reinforced polymer plates using measured modal
parameters, Proc. Euromech 357 (Sol and Oomens, Eds.), Kerkrade, The Nether-
lands, Kluwer Academic Publishers, Dordrecht, The Netherlands 1-9.
SENSITIVITY ANALYSIS AND INVERSE PROBLEMS
FOR LAMINATES AND MATERIALS
PAULI PEDERSEN
Department of Solid Mechanics
Technical University of Denmark
Building 404, DK-2800 Lyngby, Denmark
Abstract
New materials such as fibre-reinforced laminates and other composites make it necessary
to look at a material point as a microstructure that we can tailor to specific needs by design.
From this follows that the application of tools from optimal designs has increased and is
important also for solving the inverse problems, i.e. finding the microstructure that offers
specified properties. Many different aspects are involved: design for elastic stiffness,
design for thermal expansion, design for strength, design for piezoelastic behaviour, design
for damage evolution, etc.
1. Introduction
Using the finite element method for analysis we can extend from homogeneous and/or iso-
tropic materials to heterogeneous and/or anisotropic description without serious problems.
However, we cannot in the same calculation operate at both the microstructural (material)
and the macrostructural (structure) level. The solution to this problem is to separate these
levels but, in reality, to apply identical techniques at both levels for analysis as well as for
design.
In this paper we concentrate on the microstructural level and in a broad sense describe
some recent results from analysis (homogenization) and from synthesis (inverse homogeni-
zation). The aspects of synthesis can have a number of different objectives: design for elas-
tic stiffness (constitutive parameters), design for thermal expansion, design for strength,
design for piezoelastic behaviour, design for damage evolution, etc. In general, the opti-
mization formulations and techniques applied are very similar, and we shall try to underline
this point of view. Design parameters are classified as size, orientation, shape and topology
parameters. Especially topology optimization has proven important, and many of the recent
developments in structural optimization are related to this. Tools and formulations (from
453
454
topology to size parameters) are presented in the monograph by Bends!ile (1995), which also
includes an extended list of references.
The introduction to the relations between energy densities and constitutive parameters
are based on the results in Pedersen (1998) and Pedersen and Tortorelli (1998). These
results allow for a general and uncomplicated verification of the broad class of results that
we shall deal with.
To be more specific, we then, in section three formulate the problem of designing a lami-
nate with a specified bending stiffness matrix. It is hopefully clear that the extension to more
general laminate stiffness problems is straightforward.
More general material design is then treated. Homogenization and the inverse homoge-
nization problem are presented, and a large number of interesting solutions can be found
in the papers by Sigmund (specific references to follow).Thermoelastic and piezoelectric
designs are natural extensions of this class of problems. For the thermoelastic problem, see
also the paper by Autio (1998).
In the thesis by Hammer (1997) and also in the paper by Hammer et al. (1997), stiffness
optimization is carried out with the use of lamination parameters. Especially the fact that
the strain energy is linear in these parameters gives rise to important analytical results. The
feasible design space is convex and is completely described for the pure membrane case as
well as for the pure bending case. However, the general coupled case is not yet completely
clarified.
2. Energy Densities and Constitutive Parameters
Anisotropic, non-linear elastic composites can be analysed in an iterative secant formula-

tion. In the modulus description with effective strain Ee we have
(2.1)
where non-linearity is given by the power p (p = I for linearity), [aJ is a non-dimen-

sional, symmetric and positive definite constitutive matrix with Frobenius norm equal to
one and E has the modulus dimension. In the compliance description with an effective
stress a e , this is given by
(2.2)
with n = lip, [~J = [ar' and a e = EE~ .

From the detinitions of energy density differentials, du : = {a} {dE} for strain energy
T
and du c : = {E} T {da} for stress energy. we get
u = _E_E P+ 1 (2.3)
P+ I e
which for these power-law materials, gives the relation
(2.4)
455
Based on this we can derive the relations between strain energy V = udV , external f
f
potential V ext = - (f TividA + PiVidV) and total potential rr
= V + V ext . In these
definitions, V is the domain volume, A is the domain external surface, Ti is surface trac-
tions, Pi volume forces and Vi the corresponding displacements. The rather simple rela-
tions are
V=lvc=_lrr=
p P - 1 +1 P V ext (2.5)
The principle of stationary total potential (virtual work principle for potential systems)
states orr = 0 for kinematically admissible displacement fields. From this it follows (see
Pedersen (1998) for details) that
dVc
~~ = - b(~~) fixed strain tield dh (a~C) fixed stress field (2.6)
for any parameter h that does not affect the external load. The general relation
(~~) fixed strain field = - (a~C) fixed stress tield (2.7)
agrees with the relation V C = pV that follows from the density relation (2.4).
In optimal design for global stiffness the simple results (2.6) represent a localised sensi-
tivity analysis for a global function. As such, analytical results for the optimal design can
often be obtained. However, it should be noted that, in optimal design with local constraints,
the sensitivity analysis is more complicated.
For material homogenization and design we need the relations between constitutive
parameters and the energy densities. We shall here restrict ourselves to linear elasticity, i.e.
in the above formulas, p = n = I . When the contracted stresses and strains are defined
by
(2.8)
they transform from one Cartesian coordinate system to a rotated one by the orthonormal
transformation matrix [T] ([T] -I = [T] T) , see Pedersen (1995). This implies that the
constitutive matrix [L] correspondingly transforms with an orthonormal transformation
matrix. In terms of the tensor components, the complete constitutive matrix from
{a} = [L]{E} is
456
[L] =
l"" (2.9)
symmetric
The invariant squared Frobenius norm F2 is
Let us write out the elastic strain-energy density u for linear elasticity
(2.11)
Using the above we can express the constitutive parameters in terms of strain-energy densi-
ties by defining only six elementary strain states (+ linear combinations of these) and corre-
sponding energies
U 100000 := U({E}T = (I 0 0 0 0 01) UIKKlHXl:= rr({E}T = (0 0 0 I 0 OJ)
UUHKKKl : = rr( {E} T = (0 1 0 0 0 OJ) UIKKXlHl:= rr({E}T = (0 0 0 0 1 OJ) (2.12)
UIX)HKXl: = rr( {E}T = (0 0 1 0 0 OJ) U1KKKKll:= rr({E}T = (0 0 000 IJ)
In this way, the constitutive parameters are given by

457
L 2233 = UOIHKIO - UOllXIOO - U!XIUKKI
U!XIOIIO - UO!XIlliO - UOOOO lil , 2L 1223 = U(XXllOl - U(X)OJ(X) - U(XXXXII
2L 1323 = U(XX)OlI - UOlXXIlO - U()(XXXll
(2.13)
Ii LII12 = UUXltml - UlO(X)(XI - U()()OlOO , Ii LII13 = UlOlXIIO - UUXX)(X) - U(XXXIlO
Ii L2212 = UOlOHX) - UO[(XIOO - U()(XIIOO , Ii L 2213 = UOHKIlO - UOUX)(X) - U(XXXIlO
Ii L 3323 = U(KllIXII - UOOlOOO - U()(XXXII
Note that the definitions in (2.12) for the shear strains include the factor Ii, say
Ii € 12 = I . Also note that the factors on L jjkl in (2.13) are chosen to agree with those in
the matrix of (2.10).
In the theory of homogenization the notion of mutual energy density is often introduced,
and the off-diagonal constitutive parameters can then be expressed more elegantly. Here,
we shall keep the physical energies and thus the differences of energy densities shown in
(2.13). To make the alternative approach clear we can show that L 1213 can be obtained with
{E}T = 10 0 0 1 00) to the left of [L] in (2.12) and {E} = (0 0 0 0 I 0) asthefac-

tor to the right of [L].
3. Solution of Inverse Problems in Laminate Design
It is a natural wish to strive towards optimal designs that are formulated in a general descrip-
tion by an objective with constraints gj, say by
Minimize 
(3.1)
subjected to gj ~ 0 for j = 1,2, ••• J
The objective can be to extremise a physical quantity like elastic energy, eigenfrequency
or buckling load, but a load factor, a specific displacement or the expected lifetime of a com-
458
posite, may also be the objective. Constraints can be bounds on other such quantities in
addition to size restrictions on design parameters Pi
plower :5 Pi :5 p~pper for 1 = 1, 2, ... I (3.2)
We may often interchange the objective with one of the constraints and perhaps in this way
make the problem easier to solve.
Experience from optimal design is used for the less ambitions - but perhaps more practi-
cal important - case of finding the solution to an inverse problem. By inverse problem we
mean: find a design which will give certain results for analysis, as an alternative to the tradi-
tional engineering problem of analysing a specified design. In this section we will discuss
such inverse problems, related to laminate stiffness or strength. An important aspect of the
inverse problems is that the existence of a solution is not normally guaranteed, and that
when there is a solution, there are normally several solutions. Engineering justification is
thus an integral part of the approach to solution of inverse problems.
Laminate stiffness is described by the membrane stiffness matrix [A], the coupling
stiffness matrix [B] and the bending stiffness matrix [D], all calculated from the constitu-
tive matrices [C k] for the K plies (k = 1,2, ... K) according to the well-known formulas
(see Jones (1975»
=I
K
[A] [Ck](Zk - Zk-l)

k=l
Ik=l [Ck]~(Z~ -
K
[B] = z~-ll (3.3)
Ik=l [Ck]~(Z~ -
K
[D] = ZLll
where Zk-l , Zk gives the position of ply k (thickness equal to Zk - Zk-l) and where the
constitutive matrix [Ck ] is highly dependent on the orientation 8 k of that ply. In the follow-
ing discussion the design parameters Pi could be the set Zk' 8 k for k = 1,2, ... K .
A quantity of importance to the laminate strength is, in the strain space described by the
matrix [G k] , also a symmetric matrix of order three and positive definite. We shall deal
with all these matrices in the same way and, just for illustration, give the description for the
bending stiffness matrix
[D) >- 0 (positive definite) (3.4)
{D}T = {Dllll D2222 2D12l2 ,fiD ll22 2Dll12 2D 2212 }

(contracted vector notation)
459
The factors fi and 2 are related to a moment vector definition {M} =

{Mil M22 fi M 12} and a curvature vector definition {'K} = f'Kll 'K 22 fi'K 12} thatimply
orthogonal rotation transformations and then, as an examp\e, invariant trace of [D]
(D IlII + D2222 + 2D12d· See Pedersen (1995) for detailed arguments.
The matrix [D] = [D(Pi)] depends on the design parameters Pi. Our goal is to find
parameters that give [D] = [D] where [Dj is a given matrix to be obtained - or at least
obtained as close as possible. This closeness is measured by the residual matrix [D] - [Dj ,
and to get an objective that is invariant we define as the squared Frobenius norm of this
matrix
+ 2(D1I22 i
- D Il2 + 4(Dll12 - Dllll + 4(D2212 - D22ul 2
It follows that ~ 0, and if = 0, we have located an exact solution [D] = [D]. A
solution approach is then to iteratively minimise <1>. In such an approach we need the gradi-
ents
(3.6)
where formulas for d{D}/dpi can be found, say in Pedersen (1987).

A redesign described by
(3.7)
will then give
(3.8)
and, since d<l>/dpi is only correct for small ~Pi' a possible strategy would be to
minimise ( )next

(3.9)
subjected to g = I (~pi - R =0
where R is a redesign limit. Problem (3.9) can be solved analytically from the Lagrangian
L = (<I»next + A.g and weget
(3.10)
where A. is a constant (the same for all i). From (3.8) and (3.6) we have
460
(3.11 )
and from (3.9)
(3.12)
Then (3.10) gives the solution (a = - 1If... a new constant)
Llpi = ad<P = a2({D}T _ (TIlT) d{D} (3.13)

d~ dpi
To determine the constant a, we have, with (3.13) inserted in (3.9),
(3.14)
or, with (3.13) inserted in (3.8), from
<P next =0 (<P next ;::: 0)

(3.15)
thus showing a < 0 and possibly restricting the numerical value of a from (3.14).
4. Material Design and Inverse Homogenization
We now turn our attention to design of microstructures, using optimization techniques. For
calculating effective material properties we use homogenization and assume the existence
of a base cell that, in a periodic distribution, represents the material. Subjecting this base
cell to any test strain field that satisfies the periodicity, we demand that the strain energy
of the homogenized base field to equal the strain energy of the detailed base cell (the micro-
structure).
For a homogeneous microstructure no calculations are needed and we shall simply set
up the relations between the strain energy U and the homogenized constitutive parameters
C~kl . In (2.13) we have listed the relation between energy densities and 3D--constitutive
parameters. From this and with V as the volume of the base cell, we have the homogenized
constitutive parameters for the 2D-case in terms of cell strain energies
C~222 = 2U!lIo/V (4.1)
for the diagonal terms (U(XII results from {Ell E22 nEI2} = (0 0 1)). The off-
diagonal terms are
461
(4.2)
For the inhomogeneous microstructure we have to perform a detailed analysis (often a

finite element analysis) to obtain the strain energy. Three different boundary conditions are
prescribed corresponding to the corner condition for the homogeneous microstructure
when evaluating U UK), UOlO and Uml , as illustrated in fig. 4.1. Note the assumed periodici-
ty (shape unknown) in the boundary displacements, which, for the homogeneous micro-
structure, are straight lines.
X2 X2
t---
or
XI
Fig. 4.1: Illustration of cases for homogenization analysis.
Except for the forced periodicity of the boundary conditions, the analysis of the inhomoge-
neous microstructure is exactly the same as for any macrostructure with load only from
forced displacements. The reason for not putting a superscript H on the strain energies in
(4.1) and (4.2) is that these energies are obtained from the micromechanical structure. From
a finite element analysis with the microstructural stiffness matrix [S], we get the displace-
ment {D](K)}' which gives the strain energy
(4.3)
We have now established the direct relation between the strain energy as obtained in a tradi-
tional finite element analysis (4.3) of the inhomogeneous microstructure and the homoge-
nized constitutive parameters (4.1) and (4.2).
The problem of micromechanical design is here understood as the problem of finding
a distribution of a given material (materials) in a base cell of homogenization in order to
obtain prescribed material parameters. For obvious reasons, the problem is also termed the
inverse homogenization problem or identification of the micromechanical structure. As
in the case of most identification problems, we cannot expect a unique solution, but, in fact,
this is an advantage and we can introduce penalties and/or priorities as exemplified by a
minimum mass micromechanic structure. Even so there may be several possible solutions.
Like most optimal design problems, the inverse homogenization problem can, be solved
by a number of different methods. Most of these will be based on an iterative scheme in
which each iteration includes analysis of a micromechanic structure, sensitivity analysis
462
and determination of an improved design. For the sensitivity analysis we need, for each
design variable Pe, (related to element e of the micromechanical structure), the change
in the homogenized constitutive parameter, i.e. dC~kJdPe. Using the results of section two
we can determine this in a localised calculation.
Take as an example C~Ill; then, from (4.1) and (2.6), follows for linear elasticity
(p = 1)
2(aU IIXI )
V aPe fixed strains
(4.4)
2(aU]{KI,e)
V aPe fixed strains
- ~ ( a( (IT~~~KIV e) ) fixed strains
where De is the mean strain energy density in element e. Behind the result (2.6) is the
assumption of design-independent load. It thus takes a more extended analysis to show that
the simple sensitivity analysis (4.4) also holds for the actual case.
In the thesis by Sigmund (l994a) and in the papers by Sigmund (1994 b, 1995), we find
2D as well as 3D solutions to inverse homogenization problems, often with minimum
microstructural volume as a solution priority. The ultimate optimal materials are also
obtained, as well as materials with negative Poisson's ratio. In recent papers by Sigmund
and Torquato (1996, 1997) the inverse homogenization problem is formulated and solved
for the thermoelastic problem. Properties of thermal expansion cannot be designed inde-
pendently of the constitutive elastic behaviour. Very interesting results for the bounds that
limit our possibilities are given in these papers. Many examples of interesting base cells are
shown in Sigmund and Torquato (1997), where one of the more interesting situations is a
homogenized material with negative thermal expansion, obtained by mixing two phases
with positive thermal expansion coefficients and void.
5. Conclusion
A rather broad view of the influence of optimization formulations in microstructurally

based material systems is presented. Details and examples can be found in the given refer-
ences.
Energy formulations show the simplicity of analysis as well as of sensitivity analysis.
From this point of view, anisotropy and non-linear elasticity (power law) only slightly com-
plicate the necessary procedures. It is also important to realise the close relations between
energy densities and constitutive parameters.
In order not to copy the vast number of papers on laminate optimal design, we have in
the present paper formulated the inverse problem of finding the laminate with a given bend-
ing stiffness matrix and a possible solution procedure is described.
463
Material design is then reported more generally and is treated as a problem of inverse
homogenization, where the relations between energy and constitutive parameters can be di-
rectly be applied.
Most of the results have been obtained within the last five years. Similar formulations
to those examined here can be applied to a number of other important problems, such as
bone-mechanics, compliant mechanisms, actuator and sensor design (see Sigmund et al.
(1998», etc.
It is therefore expected that the general subject of this lecture will be a very active
research area in the next ten years as well.
References
1. Autio, M. (1998) Coupled thermal-structural problems in the optimization of lami-

nated plates, Structural Optimization 15, 49-56.
2. Bendspe, M.P. (1995) Optimization of structural topology, shape and material,
Springer.
3. Hammer, VB. (1997) Design of composite laminates with optimized stiffness,
strength and damage properties, DCAMM S 72, Ph.D. thesis, DTU.
4. Hammer, VB., Bendspe, M.P., Lipton, R. and Pedersen, P. (1997) Parametrization
in laminate design for optimal compliance, Int. 1. Solids Structures 34(4), 415-434.
5. Jones, R.M. (1975) Mechanics of composite materials, McGraw-Hill.
6. Pedersen, P. (1987) On sensitivity analysis and optimal design of specially ortho-
tropic laminates, Eng. Opt. 11,305-316.
7. Pedersen, P. (1995) Simple transformations by prober contracted forms: can we
change the usual practice, Comm. in Num. Meth. in Eng. 11,821-829.
8. Pedersen, P. (1998) Some general optimal design results using anisotropic, non-lin-
ear materials, Structural Optimization 15, 73-80.
9. Pedersen, P. and Tortorelli, D.A. (1998) Constitutive parameters and their evolution
(submitted).
10. Sigmund, o. (l994a) Design of Material Structures using Topology Optimization,
DCAMM S 69, Ph.D. thesis, DTU.
11. Sigmund, o. (l994b) Materials with prescribed constitutive parameters: an inverse
homogenization problem, Int. 1. Solids Struct. 31(17), 23l3-2329.
12. Sigmund, o. (1995) Tailoring materials with prescribed elastic properties, Mech.
Materials 20(4),351-368.
13. Sigmund, o. and Torquato, S. (1996) Composites with extremal thermal expansion
coefficients, Applied Physics Letters 69(21), 3203-3205.
14. .Sigmund, O. and Torquato, S. (1997) Design of materials with extreme thermal
expansion using a three-phase topology optimization method, 1. of the Mechanics
and Physics of Solids 45(6), 1037-1067.
15. Sigmund, 0., Torquato, S. and Aksay, I.A. (1998) On the design of 1-3 piezo-com-
posites using topology optimization, 1. of Materials Research 13(4), 1038-1048.
ASYMPTOTIC APPROACH TO OPTIMAL STRUCTURAL
DESIGN WITH BRITTLE-FRACTURE CONSTRAINTS
Part 1: Prototype Problem
N.V. BANICHUK
The Institute for Problems in Mechanics Russian Academy of
Sciences, Moscow, Russia
Abstract. Questions raised in this paper are related to an important class

of problems concerning shape optimization of brittle or quasi-brittle struc-
tures with cracks. Considered optimization problems consist in finding the
boundary of the body or some geometrical parameters such that the cost
functional (the weight or material volume) attains a minimum, while sat-
isfying prescribed bounds on stress intensity factors.
1. Introduction
A substantial literature has been developed in the field of optimal structural

design problems with constraints on strength, stiffness and buckling (see, for
example, [1-3]). Majority of investigations was characterized by application
of deterministic approaches for formulation of shape optimization prob-
lems and determination of optimal solutions. Nondeterministic approaches
(game approach, probabilistic approach, mixed approaches) has been for-
mulated and applied to some optimal design problems when the internal
conditions (external forces) were taken as a random variables ([2,4]). Some
parametric probabilistic problems of optimal structural design with limit-
ing state and buckling constraints have been formulated and investigated
in [5,6]. A few studies have been devoted to important class of problems of
brittle and quasi-brittle elastic bodies optimization on the basis of modern
fracture mechanics criterions ([7-11]). In accordance with fracture mechan-
ics representations it is necessary to consider all possibilities of the crack
appearance. But in the most cases the number of cracks, their positions, ori-
entations, sizes and modes (opening cracks, shear cracks, ... ) are unknown
beforehand and so the optimal design problems for brittle and quasi-brittle
465
466
materials intrinsically contain uncertainty or randomness. Note that the
problems in the design of optimal structural systems with incomplete in-
formation are essentially different, both in the formulation and in research
techniques. It is important to consider various concepts, criteria and ap-
proaches to optimization problems with incomplete information.
2. Optimization - based design for deformable bodies and struc-

tures made of brittle or quasi-brittle materials with allowance for
the origination of cracks
Let us consider the loading of a two dimensional elastic body that occupies
the domain n with boundary r . Portion r v of the boundary is what is being
sought and is considered to be a design variable. The remaining portion
r - r v is assumed to be fixed. The surface forces T act on the part r 0" of
the boundary r , and displacements are specified along the part r u of the
boundary (r = r u + r 0" ). We assume that there is a crack in the body; this
crack is simulated by a rectilinear cut, whose sides r f are not subjected to
internal forces. The stressed-strained state of body incorporating a crack is
defined by the following equations constituting a boundary-value problem
1
v .u = 0, u = C . '10, 10 = 2(Vu + (Vu)T) (1)
(u)ru = 0, (u· n}r.,. = T, (0"' n)r, = 0 (2)

The stress u, and strain 10, tensors and also the displacement vector u,
are determined from a solution to this problem. In (1) and (2) , C stands
for the tensor of the elastic moduli; n is a unit vector normal to the body
boundary; and the midlevel-line dot and two dots signify, respectively, scalar
and double scalar products. The following quantities can be determined on
the basis of the known solution for a cracked body: the stress intensity
coefficients Kl, K2 appearing in the asymptotic representations for the
stresses at the crack side
Kl K2
an = tte= + 0(1), O'nt = tte= + 0(1)
y21rr y21rr
and the energy release rate

g=--
arr
8l
during crack extension; this rate is related to the stress intensity coefficients
by the expression
467
where E is Young's modulus. The parameter fJ = 1 in a plane-stressed

state and fJ = 1 - v 2 for a plane strain, where v is Poisson's ratio. The t
and n -axes of a local system of coordinates are directed along the crack
and along the normal to the t-axis, respectively, furthermore the n-axis lies
in the xy plane that contains the domain n. The origin of this system of
coordinates coincides with the midpoint of the cut opening. In asymptotic
representations for stresses , the quantity r specifies the distance between
the crack tip and a given point on the t-axis.
It is convenient to assess the strained state and determine the stress
intensity coefficients if (following Bueckner) the representation for displace-
ments, strains and stresses is written as
(3)
Here the quantities uO, eO, and 0'0 define the stressed-stained state of a
cracked free body and are determined from the solution of the boundary-
value problem defined by (1), and (2) with the third condition in (2) ig-
nored. The quantities u f , e f , and O'f are the characteristics of the stressed-
strained state of a cracked body that is not subjected to external loads with
forces applied to the crack sides that are in opposite direction to forces aris-
ing in the corresponding areas of the crack-free body. The quantities u f ,
e f , O'f satisfy the boundary-value problem given by
Thus , decomposition (3) separates the initial fields u, e, 0' into the
smooth component uO, eO and 0'0 and the singular components u f , ef and
O'f which define the stress intensity coefficients and the energy release rate
during crack propagation. Let the analysis problem given by (1) and (2)
(with the third boundary condition in (2) discarded) and by (4) and (5)
be solved sequentially and let the expression for the stress-intensity factors
be derived. In this case, the condition that the structure is not fractured is
written as
(6)
for normal-tear crack (i=l) and shear -induced cracks (i=2) or is given by
(7)
468
when the strained state at the crack tips involves both tearing and shearing
stresses, The given brittle-strength constants are designated as KiC and Gc
Let us assume that the length of rectilinear crack I is small compared to

the characteristic size L of the body (I « L) and that the crack is located
away from the body boundaries. In this case the problem given by (4) and
(5) admits of an exact solution [12]. The stress intensity factors are given
by
a 1
K1 = J an'
f(t 0)(2'«a+»)2dt
I a-t
-a
(8)
a 1
K2 = J ant'
f (t 0) (2'«a+p)2dt
I a-t
-a
where a = 1/2. If the stress fields in the problem for a crack-free body vary
insignificantly along the crack length and can be regarded as approximately
uniform
a!(t,O) = a~ = const, a!t(t, O) = a~t = const

equations (8) yield
(9)
In the case of a surface crack directed perpendicularly to the unloaded

boundary of the body, the assumption that the crack length is much smaller
that the curvature radius R of the boundary and the characteristic size L
of the body (I « R and I « L) make it possible to reduce the analysis of
the stressed-strained state of an elastic medium in the vicinity of the crack
to the problem of the theory of elasticity of a half-plane with a cut at the
surface. The boundary conditions at infinitely distant points correspond to
the half-plane being exposed to tension in direction parallel to the half-
plane boundary. The coefficient Kl is defined ([13]) as
Kl = ,..a~y';j, ,.. = 1.12. (10)

Expression (10) is generalized to the case of a surface crack in a strip
of the width h , which is subjected to tensile stresses a~ applied at infinity,
where the crack length may be comparable to the strip width h. In this
case the quantity,.. appearing in (10) is defined ([14]) as
,.. = <p('\), where ,\ = l/h and
<p('\) == 1.12 - 0.23'\ + 10.6,\2 - 21.7,\3 + 30.4,\4, (11)

469
In contrast to this , if the strip is subjected to stresses acting at

infinity and corresponding to the problem of pure-bending strain as =
a~(/(h/2), -h/2 ~ ( ~ h/2 where ( is the coordinate varying along the
direction perpendicular to the strip boundaries,the relevant expression for
K1 is given by (10) with ([13,14]) "( = 'l/J(>"), where
(12)
The above expressions (8)-(12) for the stress-intensity factors in cases of

surface and interior cracks can be used to advantage in an optimization-
based design.
The problem of the optimal design consists of finding the boundary
r v, such that the cost functional J (volume or weight of the body) attains
minimum
(13)
while satisfying strength constraint
(14)
where the energy release rate g is determined for all admissible variants
of crack location. Analysis of this variants and finding the variant with
maximum energy release rate is very complicated problem.
Some additional assumptions concerning admissible variants of crack
location and additional data concerning the most dangerous parts (subdo-
mains) of the structure (in the sense of cracks appearance) will be discussed
in Part 2 and give the possibility to find optimal design. In this context it is
important to note that we can effectively use the presented above formulas
for the stress intensity factors as the asymptotic expressions if the crack
is internal one and located far from the boundary or if the crack located
nearly the body surface.
3. Optimal Design of Bimaterial Layered Beams
We suppose that the beam lies along the x-axis and that it has the length
L a rectarlgular cross-section with constant (along the x-axis) height hand
width b. The beam is supposed consists of two layers, with heights H 1, H2
(h = H1 + H2), cross-sectional areas 8 1 ,82 (8 = 8 1 + 82), Young's moduli
E 1 , E2 and specific weights "(1, "(2 (per unit volume). The neutral line of the
beam concides with the x-axis. The beam is subjected to pure bending in
the plane xy. The cross-section, composed of two rectangular areas 8 1 ,82 ,
470
is rotated with respect to the z-axis perpendicular to the plane of bending

and passing through the neutral axis of the beam. The distances between
the neutral line (y = 0) and the outer surfaces (y=const in plane of bending
xy) of the beam are denoted by hI and h2 (h = hI + h2) . The values hI
and h2 are determined with the help of the condition of zero resultant force
applied to the beam cross-section
0= / (1xd8 = R- l (EI / yd8 + E2 / yd8) (15)

8 81 82
where (1x, R are respectively the stress component and the curvature radius.
Using this relation and performing necessary transformations we will have
hI = E I H r + E2 H 2(H2 + 2Hd , h2 - E2H~ + EIHI (HI + 2H2) (16)

2(EIHI + E2H2) - 2(EIHI + E 2H2) ,
The equilibrium condition, written for the bending moment, gives us
the following relation
M = / (1xydS = R- l (E l it + E212) (17)

8
Here II and 12 are the moments of inertia of the cross-sectional areas

8 1 , 82 with respect to the z-axis. Hence the stress distribution is described
by the formulas ([15])
(1 -
x -
1L§ -
R -
yM E 1
E1I!+E212' -
h 1 -< Y -< H 1 - h 1
(18)
(1 -
x -
JJ.§ -
R -
yM E2
ElI! +E212'
H1 -
h1 <
-
Y<
-
h2
The maximum of the absolute value of the considered stresses is attained

either at
y = -h l ((11 =1 (1x I) or else at y = h2((12 =1 (1x I). We have the following
expressions to compute the values (11, (12:
it = t[h~ + (HI - hd 3], 12 = t[h~ + (H2 - h2)3]

(19)
Elit + E212 = t(EIH r+ E2H~) + bEIHl hl (h l - Ht}+
We make allowances for the possibility of normal-tear cracks origination

at the surfaces y = -hI and y = h2 of the beam. It is assume also that
471
the lengths I of rectilinear cracks under consideration do not exceed a given

limiting value 1m , i.e.
(20)
Furthermore, it is presumed that the crack direction is perpendicular to

the boundary. In this case the expression for evaluation of the maximum
value of the stress intensity factors can be written in explicit form as
Kl = IWiV7rlm ('" = 1.12; i = 1,2) (21)
The problem of optimal beam design consists in finding an admissible

heights HI, H2 that provides the minimum of the beam-weight function
(22)
and satisfies both the strength constrains
(23)
and the geometrical conditions
(24)
where HP > 0,Hg > 0 are given parameters and the given brittle-strength
constants are designated as (K1C h, (K2C b
For the sake of convenience, we shall introduce the following dimension-
less variables and notations
J-
-
J
bLHP'Yl
(25)
In these dimensionless variables (with tilde omitted) the expression for

the cost function of the formulated optimization problems (22)-(24) and
considered constraints can be written as
J = HI (1 + rt) (26)
HI ~ 1, HI ~ k'IjJ(t) (27)
HI ~ (31'IjJ1(O:, t), HI ~ (32'IjJ2(O:, t) (28)

472
o 2 3 4
t
Figure 1.
Here
1/J(t) = t, 1/Jl(a, t) = CPl(a, t)cp(a, t), 1/J2 = CP2(a, t)cp(a, t)
CPl(a, t) = (6(1 + 2at + at 2))1/2, CP2(a, t) = (6a(1 + 2at + at 2))1/2,
(29)
Using the introduced function the optimization problem can be formu-
lated in more compact form
J = H1 (1 + rt) --+ min (30)

Hl,t
The operation max in (31) results in the selection of the larger of the
four quantities inside the braces. If the maximum in (31) is realized for the
473
o 2 3 4
t
Figure 2.
first or the second term we can conclude that the corresponding geometrical
constraint from (24) is active for optimal design. If the maximum in (31)
is realized for the third or the fourth term we obtain that the fracture
mechanics constraint (23) is active for y = -hi or for y = h 2 . Functions
Hi = 1, Hi = k'IjJ(t), Hi = (31'IjJ1(t), Hi = (32'IjJ2(t) and dependences Hi =
J /(1 + rt) with J = 2,3,4,5 are shown in Figure 1 by the lines 1,2 ... 8 for
the case r = 1, a = 0.5, k = 1, (31 = 1, (32 = 2 and in Figure 2 for
the case r = 0.5 (other parameters are the same as for previous case).
If we exclude the geometrical constraints from consideration we reduce
the previous optimization problem (30), (31) to the new one with
(32)
474
Figure 3.
Optimal solution
t* = 5.22087, J* = 2.109988
of the problem (30), (32) was obtained for the case r = 0.5, a =
0.1, /31 = 1, /32 = 2. Active strength constraint and the values of cost
function J at the point of the constraint curve are shown in Figure 3 by
the curves 1,2 for 0 :::; t :::; 8.
The research reported in this paper was supported by Volkswagen-
Stiftung under grant N 1/71113 and by Russian Federal Programme "Inte-
gration" under grants N 368 and 480.
References
1. Haug, E.J., Arora, J.S. {1979} Applied Optimal Design, Wiley, New-York.
2. Banichuk, N.V. {1983} Problems and Methods of Optimal Structural Design, Plenum
Press, N.-Y.
3. Haftka, R.T., Gurdal, Z. (1993) Element of Structural Optimization, Kluwer Academic
Publishers, Dordrecht,The Netherlands
475
4. Banichuk, N .V. (1990) Introduction to Optimization of Structures, Springer- Verlag

New-York.
5. Augusti, G., Baratta, A., Casciati, F. (1984) Probabilistic Methods in Structural En-
gineering, Chapman and Hall.
6. Augusti, G., Casciati, F. (1979) Further Studies on Structural Design for Maximum
Expected Utility, Proc. 3rd Int. Conf. On Applications of Statistics and Probability
in Soil and Strural Engineering (ICASP3), Sydney,Vol. 2, pp. 676-700.
7. Wang, J., Karihaloo B.L. (1995) Optimum in the Situ Strength Design of Composite
Laminates, Proc. of the First World Congress of Structural and Multidisciplinary
Optimization, Pergamon, N- Y., pp. 259-266.
8. Yu, X., Choi, KK, Chang, K-H. (1997) A Mixed Design Approach for Probabilistic
Structural Durability, Structural Optimization, Vol. 14, pp. 81-90.
9. Banichuk, N.V. (1998) Problems of Optimum Design Based on the Mechanics of Quasi-
brittle Fracture, Physics-Doklady, Vol. 43 No.1 Translated from Doklady Akademii
Nauk a Journal of the Russian Academy of Sciences, Vol. 358, No 1 (1998).
10. Banichuck, N.V. (1996) Free Boundaries Optimization Under Fracture Mechanics
Constraints, Analele Stiintifice, Universitatii, "OVIDIUS", Constanta, Vol. 5, N 1,
pp. 13-20.
11. Banichuk, N.V., Ragneda, F.K, Serra, M. (accepted for publication) Probabilis-
tic Approaches for Optimal Beam Design Based on Fracture Mechanics, Meccani-
ca,International Journal of the Italian Association of Theoretical and Applied Me-
chanics.
12. Banichuck, N.V. (1998) Optimal Design of Quasi-Brittle Elastic Bodies with Cracks,
Int. J. Mechanics of Structures and Machines, Vol. 26, N 4, pp. 365-376.
13. Sih, G.S., Liebowitz H., Fracture, Liebowitz, H., Ed. (1971) Mathematical Funda-
mentals, New York, Academic, Vol. 2
14. Hutchinson, J.W. (1979) A Course on Nonlinear Fracture Mechanics. Technical Uni-
versity of Denmark, Lyngby.
15. Hellan, K (1984) Introduction to Fracture Mechanics, New York, Mc Graw-Hill.
16. Timoshenko, S.P., Gere, J.M. (1972) Mechanics, Van Nostrand Reinhold Company,
New-York.
ASYMPTOTIC APPROACH TO OPTIMAL STRUCTURAL
DESIGN WITH BRITTLE-FRACTURE CONSTRAINTS
Part 2: Deterministic and Stochastic Problems
N.V. BANICHUK
The Institute for Problems in Mechanics Russian Academy of
Sciences, Moscow, Russia
Abstract. Shape optimization problems for brittle and quasi-brittle struc-

tures with cracks are characterized by incomplete information concerning
crack sizes, location and orientations. In this context the paper presents
some new basic formulations of optimal structural design problems based
on guarantee minimax and probabilistic theories. Using these theories op-
timal designs of elastic bodies with surface and internal cracks has been
found.
1. Model Assumptions and Minimax Optimization Problems
Consider a deformed quasi-brittle elastic body occupying a two dimensional

domain n with boundary r v , that is taken as a design variable while anoth-
er part r - r v is supposed to be fixed. The body contains a crack modelled
by a rectilinear notch r f with a traction free boundaries. It is important for
the subsequent considerations that the possible location of the crack, aris-
ing from manufacture or exploitation, are unknown beforehand. In context
of structural design this leads to the essential complications caused by the
necessity to analyze the variety of crack locations and orientations and to
solve corresponding boundary value problems of the theory of elasticity for
cracked body. Let us characterize the crack by the vector w = {xc, Yc, l, a},
where (xc, Yc) - coordinates of the crack midpoint, l and a are the length
of the crack and the angle setting the crack orientation with respect to the
global coordinate system x, y as shown in the Figure 1.
Consider the following model assumptions:
1. The perfect brittle or quasi-brittle body contains a crack modelled
by a rectilinear notch r f;
477
478
/"
----r----
(/ /~ "\
\ ./ C(x ~y) J
'" /" c c
/'-- 0 .,/
a. -- -1!. _ --- r
o
Figure 1. Quasi-brittle body with cracks
2. The notch is traction free;

3. The length of considered cracks does not exceed given limit value lm;
4. The body has only one crack, but this crack can be characterized by
any vector w from given set A.
Note that it is possible to admit the assumption 4 in general case if
the distances between the originating cracks are large enough to exclude
the mutual influence of cracks. In what follows some additional asymptotic
assumptions concerning admissible variants of crack location are taken into
account. In the case of the internal cracks it is supposed that the limit
length lm of considered cracks does not exceed given parameter Lo
l ~ lm «Lo = min{ (x - xc)2 + (y _ y~P/2 (1)
(x, y) E r, (xc, Yc) End
This condition means that the cracks are located far from the boundaries
of the body. Subdomain nd C n is considered as given.
In the case of surface crack it is supposed that the crack is orthogonal
to the boundary and the crack length is much smaller than the radius 01
curvature R of the boundary and characteristic size L of the body
l :::; lm «min(R,L), L = minLBB'
(2)
(Xb, Yb) E r
479
Here B' (Xb, Yb) - the point of intersection of the crack line and the boundary
r (Figure 1).
Accepted assumptions and available additional data concerning the
most dangerous parts (subdomains) of the structure (in the sense of cracks
appearance) give us the possibility to consider the set A of admissible cracks
as given
wEA (3)
The problem of optimal structural design consists of finding the part
of the boundary r v, such that the cost functional J (volume of the body)
attains minimum
J = mesO -t min (4)
rv
while satisfying strength constraint
max 9 = max{ E(3 (K{

w w
+ Ki)} ~ Gc (5)
where the energy release rate 9 is regarded as being applied and therefore
a functional of geometry and load.
On the other hand, G c is regarded as a material parameter. Parameter
(3 = 1 for plane stress state and (3 = 1 - 1./2 for the case plain strain.
Another optimization problem consists in finding admissible shape of
contour r v such that the maximum of the energy release rate 9 is minimized
J* = minmax{E(3 (K{ + Ki)} (6)

rv W
2. Solution of the Shape Optimization Problems

Consider the problem of optimal design of a beam taking into account the
possibility of crack appearance at the beam surfaces. We assume that the
beam of length L lies along the x-axes (0 :::; x :::; L) and that it has a
rectangular cross-section with height h = h(x) and constant width b.
Here we consider the statically determinate loading case (simply sup-
ported beams or cantilever beams subjected to the actions of transverse
loads), with bending moment function M(x) and shear load function Q(x)
acting on the cross-section of the beams. We regard these functions as
known for 0 :::; x :::; L. The maximum of stretching stresses is attained on
the elongated beam surface for any cross-section x = ~, i.e. 1( 1= h(O/2,
where the coordinate ( measures the distance from the center of the cross-
section and varies in the interval -h/2 :::; ( :::; h/2. Therefore, we have
(7)
480
Suppose that
1m «hmin ~ maxh(x) « L (8)
where h min - given positive constant. Use the expressions K1 = K,0'~..,fiil,
K, = 1.12 ([1,2]) to evaluate the stress intensity factor K1. The function
h = h(x) determining the shape of the beam is the unknown quantity.
The problem of optimization consists in finding an admissible thickness
distribution such that the volume of the beam is minimized, while satisfying
strength and geometrical constraints
J
L
J= bh(x)dx ~ minh (9)

o
6 1 M(~) 1 J7f1(~) )
(K 1 )max = mrcmrx ( bh2(~) K, ~ KlC (10)
h(~) ~ hmin, 1(~) ~ 1m (11)

The solution of the weight optimization problem for the case of opening
surface cracks is written in the explicit form
1/2
h( x ) -- max {h. ( 61 M(x) 1~ ) } (12)
mm, bK 1C K,
The operation max in (12) results in the selection of the larger of the
two quantities inside the braces. If the maximum in (12) is realized for some
subdomains by the second term inside the braces, we can derive that the
maximum of O'x takes the constant value KlC/K,J7f1 m for these subdomains.
On the remaining segments of the beam's length, where the height of the
beam is constant: h(x) = hmin , the max O'x is smaller than KlC/K,J7fl m .
Consequently the applied minimax or "guaranteed" approach gives us the
beam design with partially uniform strength if the admissible surface cracks
length is small enough.
Consider the optimal design of brittle and quasi-brittle beams with low
limiting value of shear intensity factor K2C . Shear stresses are maximal at
the neutral axis of the beam. We use the expression max(O'xy) = 3 1 Q(~) 1
/2bh(~) to obtain a direct functional dependence of maximal shear stresses
on the variables b, h and Q(~). Suppose that the internal shear crack is
located at the axis x and it's length is much smaller than the minimum
thickness of the beam and the characteristic size of the domain of shear
stress variation for the uncracked beam. In accordance with the formula
([3]) for K2 we shall have
K2 = 3 1 Q(~) 1 J7fI(~)/2V2bh(~)
481
The problem of finding an optimal design consists of finding a thickness

distribution from the condition of minimizing the beam volume (9) while
satisfying geometric conditions (11) and the fracture mechanics constraint
(
K
2)max -
- max max
~ I
(3 1Q(O 1J1J([5) <
2V2bh(~) -
K
2C (13)
where ~ is the coordinate of the crack midpoint. The solution of this problem
can be written in the form
-
h( x ) -max {h. 3 1 Q(x) 1 ~} (14)
mill, M
2y 2bK2C
Consider now the optimal design of a beam taking into account the pos-
sibility of crack appearance at the beams surfaces (mode I-opening cracks)
and at the neutral line (mode II - shear cracks). Taking into account that
the maximum of stretching stresses is attained on the elongated beam sur-
face and the shear stresses are maximal at the neutral axis of the beam,
we suppose that the internal shear cracks are located at the axes x and
the edge opening cracks are arisen at the beam surfaces. The problem of
finding an optimal thickness distribution consists in volume minimization
and has the following solution
( )_ {6 1 M(x) 1 ~ 31 Q(x) 1 ~ (15)

h x - max K, bK1C ' hmin}
2V2K2C
Let us consider the problem aroused when we wish to determine opti-
mum shapes of the holes in infinite plates with surface cracks having the
smallest stress intensity factor. Assume that the admissible length of the
cracks satisfies the constraint l ~ lm, the boundary of the hole r v has no ex-
ternalloads applied to it and the plate is subjected to tensile forces applied
at infinity, i.e.
(O"n)rv = (O"ns)rv = 0 (16)
(O"x)= = 0"1\ (O"y)= = 0"2
where 0"1 and 0"2 are given positive constants and nand s denote, respec-
tively, the directions normal and tangent to the curve r v. We suppose that
the hole contour is smooth, cracks are orthogonal to the contour and their
length is much smaller the hole curvature radius, i.e. l S lm < < min R( P).
Here the minimum is taken with respect to PEr v. The optimization prob-
lem consists in finding of admissible shape of the contour r v such that the
maximum of the stress intensity factor Kl is minimized
(17)
482
Using conventional decomposition the analysis of a stressed and de-

formed states is reduced to the solution of the following two boundary
value problems. First we solve a boundary value problem of the plane theo-
ry of elasticity for uncracked plate with hole and loading conditions at the
infinity. The solution of this problem is approximated by the homogeneous
stresses fields in the vicinity of the boundary point P (tension stresses in
tangent to the curve r v direction). Then we solve the second problem for
the plate with hole and crack, which boundaries are loaded by tractions
aroused in the first problem (Js = -(J~, and obtain the expression for stress
intensity factor Kl = /'i,(J~.;;z (/'i, = 1.12). The problem (17) takes the
form
Thus the optimization problem for plate with hole and surface cracks,
aroused at it's boundary, is reduced to the problem of tension stresses
minimization (or stress intensity minimization) which solution is presented
in [4J.
3. Probabilistic Problem of Optimal Structural Design Based on

Fracture Mechanics
This presentation follows research results of Ref. 5. We will consider one

crack parameter as a random variable ~ with known density function f(~),
while the others crack parameters will be considered as given or will be
found with the help of guaranteed approach of the game theory (design for
the worst case). Let us suppose that the shape of the body is described by a
function h of space coordinate that is taken as unknown design variable and
that we consider the crack mode i = I, I I. If we introduce small positive
parameter E and reduced meaning of limiting value of stress intensity factor
Kif = KiC - E, E ~ 0 then we can write the fracture mechanics constraint
in the following form
(18)
In general case the stress intensity factor Ki(~' h) depends implicitly
(through the dependence on crack parameters and solution of equation
of the theory of elasticity) on random variable ~, but for particular cases
it is possible to simplify the dependence of Ki on ~ and to express the
formula for Ki in some explicity functional form. The probability density
distribution for variable ~ permits us, at least in principle, to determine the
moments of the random quantity Ki and in particular, their mathematical
expectation and dispersion. This permits us, in turn, to control the proba-
483
bility of violating the assigned constant (18) on the strength and to study
the following problem of optimal design, consisting in finding a function h
that satisfies the system of inequalities
J Ki(~' h)f(Od~ ~ Kif

00
Ki = E(Ki) = (19)
-00
J(K(~,
00
D(Ki) = E((Ki - Ki)2) = h) - K)2 f(~)d~ ~0 (20)

-00
and minimized the functional (volume or weight of the structural element)
v= V(h) -+ minh (21)
while taking care of the equations of the theory of elasticity. Here E and D
denote the mathematic expectation and dispersion, respectively, for a ran-
dom variable, and 0 denotes a sUUldently small number, that is supposed
to be given. Inequalities (19), (20) imply that some conditions must ac-
company a choice of the design variable h . The brittle fracture mechanics
constraint given in (18) must be satisfied in "the average" sense, and the
deviation of the quantity Ki - Ki must not exceed some chosen sufficiently
small value.
Consider now another possible probabilistic approach for formulation
of structural optimization problem using the theory of probability. Let us
suppose additionally that Ki(~' h) is a monotonic function of the ~ and the
equation
(22)
has the unique solution ~o = ~o(h, Ko) for fixed hand Ko. Then the prob-
ability P of satisfying of the inequality Ki ~ Ko can be estimated by the
following relation
Jf(~)d~
~o
P{Ki ~ Ko} = P{~ ~ ~o} = (23)

o
Probabilistic expression for fracture mechanics condition for crack not go
ahead can be written in the form
Jo f(Od~ ~
~.
P{Ki ~ Kif} = 1- II (24)

484
where Kif = Kic - E, 1/ -small enough given positive value (1/ > 0) and ~f
is the solution of the equation
(25)
An optimization problem based on fracture mechanics and probabilis-

tic approach can be formulated as a problem of minimization of the cost
functional (21) under constraint (24), taking into account the behavior
equations of the theory of elasticity for designed structural elements.
Consider the problem of optimal design of a beam subjected to the
action of transverse loads, taking into account the possibility of crack ap-
pearance at the beam surfaces. We assume that the beam of length L lies
along the x - axis (0 ~ x ~ L) and that it has a rectangular cross-section
with height h = h(x) and constant width b. Bending moment function M(x)
and shear load function Q(x) acting on the cross-section of the beam are
independent of the elastL .,:>roperties of the shape of the cross-section and
considered as known for 0 ~ x ~ L. The maximum of stretching stress
is attained on the elongated bear lurface for any cross-section and varies
in the internal -h(x)/2 ~ ( ~ h(x)/2. Therefore, we have for maximum
stresses ax = 6IM(x)l/bh 2(x). Stress intensity factor KI for surface crack
at any cross-section can be evaluated using the formula
KI = 1.12ax~<, = tp(x)~, () _ ki I M(x)

tp x - bh 2 (x)
I (26)
where ki = 6.72fo and the length of the surface crack ~ is a random

variable characterized by the known density faction f(~). For mean value
of KI (mathematical expectation E(Kd ) and the variance D(Kd we will
have
-
E(Kd == - Jf.)
J Ktf(~)d~ = tp(x)E(
o
(27)
D(KI )
=(KI
=J -
- E(Kd)2 f(~)d~ = tp2(x)D( Jf.)
o
Using the expression (26), (27) we can present the fracture mechanics
constraints (19), (20) in the form
h(x)
(28)
h(x)
where KIf = KlC - E, E - small parameter.

485
The optimization problem consists in finding an admissible thickness

distribution such that the volume of the beam is minimized, while satisfying
strength constraints (28). The solution of this problem is written in explicit
form
tcA) > (DCf)) 1/2
{ A,(x) if Kl,
h(x) = (29)
AD(X) if KletcA) < (DCf)) 1/2
or, in more compact form
h(x) = max{At(x), AD(X)}
Take into account the geometrical constraint on the thickness distribu-

tion h > hmin . Then the optimal solution can be written as
h(x) = max{h min , At(x), AD(X)} (30)
Consider now the problem thc~ is characterized by the monotonic de-

pendence of stress intensity factor on random variable. Let the beam of
length L lies along the x-axis (0 S x S L) and has a rectagular cross-
section with height h = h(x) (design variable) and given constant width b.
Suppose that the beam is loaded by transverse loads and that the crack of
random length ( can be appeared at the beam surfaces. Density function
for ( is supposed to be known and we will take the expression (26) for K I .
Taking into account the monotonicity of the KI with respect to random
variable ( we obtain the unique solution ( of the equation KI = KlC for
every cross-section of the beam
b2 Kr"h4(x)
~,,= krM2(x)
For the considered case the probabilistic inequality corresponding to brittle
fracture mechanics criteria can be written in the form
J
~,
P{KI SKI,,} = P{~ s ~E} = f(~)d~ == F(~,,) ~ 1 - v (31)

o
where v > 0 - given small positive value, F - distribution function deter-
mined with the help of analytical expression or tabulized data for many
known density function f.
Considered optimization problem consists in finding thickness distri-
bution such that the volume of the beam is minimized, while satisfying
strength constraint (31). Optimal solution of the problem is corresponded
486
to the case of rigorous equality in (31) and can be written in the following
manner
h(x) = (kl 1bKl€
M(x) I) 1/2 S1/4(1 _ v) (32)
where S(l-v) -inverse function with respect to function F, i.e. S(F(~)) = ~

for any~.
Consider now the general case of nonmonotonic dependence of stress in-
tensity factor on random variable. We investigate shape optimization prob-
lem for beam with low K 2c material constant and suppose that the center
coordinate ~ of shear crack is a random variable with given density function
f(~). In this case we have the expression
for stress intensity factor K2 and can introduce the value
(33)
The length of the crack l is suppose to be given. The inequality l::.K2 > 0,
that guarantees for crack not go ahead,is written in the form
h ~ k2 1 Q(O 1 v'l == X(x)

bK2€
Corresponding probability can be estimated with the help of integration
of density function f(~) over the intervals where l::.K2 ~ 0 , written in the
following manner
00
P{K2 ::; K 2€} = J 9(~)f(Od~

o (34)
9 =1 if h ~ X, 9 =0 if h <X
Optimization problem consists in determination of the thickness distri-
bution function under the constraint
J9(~)f(~)d~ <
00
1- v (35)
o
such that h ~ hmin (hmin - given constant) and the optimized cost function-
al (21) (volume) is minimum. As it can be proved for considered problem
the diminishing of material takes place for the regions where the density
function is relatively small.
48
References
1. Kanninen, M.F. and Popelar,C.H. (1985) Advanced Fracture Mechanics, Oxford Uni
versity Press, N.-Y.
2. Hutchinson, J.W. (1979) A Course on Nonlinear Fracture Mechanics. Technical Uni
versity of Denmark, Lyngby.
3. Rice, J.R. (1968) Mathematical Methods in Fracture Mechanics, in Fracture (An Ad
vanced Treatise), Vol. II, Mathematical Fundamentals, Academic Press, N.-Y.
4. Banichuk, N.V. (1983) Problems and Methods of Optimal Structural Design, Plenur
Press, N.-Y.
5. Banichuk, N.V.,Ragneda, F.K., Serra, M. (accepted for publication) Probabilistic AJ:
proaches for Optimal Beam Design Based on Fracture Mechanics, International JOU7
nal of the Italian Association of Theoretical and Applied Mechanics, Meccanic.
SOME COMPOSITE STRUCTURES EXPERIENCES AT NASA LANGLEY RESEARCH
CENTER
James H. Starnes, Jr. and Damodar R. Ambur

Structural Mechanics Branch
NASA Langley Research Center
Hampton, Virginia 23681-2019, USA
Abstract
The results of an analytical and experimental study of the effect of impactor mass and plate radius of
curvature on the low-speed impact response of laminated composite plates is presented. Dropped-weight
and airgun impact responses, damage characteristics, and residual strengths of quasi-isotropic and
orthotropic plates are presented for a range of impact energy levels determined by systematically varying
the mass of the impactor. The flat plate responses are compared with the responses of curved plates with
three different radii of curvature. Measured contact forces and damage areas are also presented. The
results indicate that contact force and damage area are nonlinear functions of the impactor mass and vary
considerably over the entire range of impact energy levels considered. The different damage levels caused
by impactors with a given impact-energy level and with different masses significantly influence the
compressive residual strength of impact damaged plates. The trends in the results for contact force, damage
area, and compression-after-impact strength are clear and consistent when the data are expressed as a
function of the impactor momentum, The contact >v.ce, damage initiation, and residual strength results for
curved plates suggest that the low-speed impact response of plates is also a function of the radius of
curvature.
1. Introduction
The effect of low-speed impact damage on the compression strength of laminated composite structures has
been studied extensively by many researchers over the past several years (e.g., Refs. 1-8). A
comprehensive review of such work has been presented by Abrate in Reference 9. Test data show that the
compression strength of composite structures can be reduced significantly by low-speed impact damage,
even if the damage is not detectable by visual inspection. Current damage-tolerance design criteria for
compression-loaded composite airframe structures are related to the impact energy or to the indentation
depth in the specimen caused by the impact event. Many of the researchers who have investigated the
transient response of laminated composite plates used either an airgun-propelled projectile (e.g., Refs. 10
and 11) or a dropped weight to impact the specimens (e.g., Refs. 12 and 13). The specimen support
conditions were also different for many of the investigations. For an airgun impact test, a projectile of a
given diameter and material is propelled at a given speed with a compressed-air apparatus to generate an
impact condition that simulates the impact of an aircraft structure by runway debris or hail stones. The
dropped-weight impact test uses a weight of a given mass and material that is dropped from a pre-selected
height to generate an impact condition that simulates the impact of the structure by a dropped tool. While
the results of these studies all indicate that low-speed impact damage can degrade the compression strength
of composite structures, there are enough differences in the results to indicate that a consistent analytical
and experimental approach to understanding the plate impact response mechanisms is still needed. This
approach should account for different structural, impactor and laminate parameters. These parameters
include impactor mass, size and speed; specimen geometry including curvature, concept, materials,
laminate stacking sequence, and boundary or support conditions; and the relative magnitudes of the
impactor and structural masses.
489
490
Analytical results and experimental data are presented in the present paper which are used to help
explain and understand the low-speed impact response of flat and cylindrically-curved composite plates. In
general, thin flat plates with clamped boundary conditions exhibit a stiffening response when impacted, and
curved plates exhibit a softening response. The nonlinear analysis methods used to generate the analytical
results in the present paper account for such nonlinear effects, and are general enough to include all of the
significant structural and impactor parameters. Generalized beam functions are used to represent the
displacement of the plate such that the response of the plate can be obtained for combinations of simply-
supported and clamped boundary conditions. The plate equations of motion are presented in Ref. 14, and
include the stiffness of an elastic foundation. These equations can be used to determine the response of a
flat sandwich plate. A first-order shear-deformation theory is used in the analyses to represent properly any
local short-wave-length bending transient response phenomena that may occur. The contact force, out-of-
plane deflections, and in-plane stresses and strains at any plate location due to an impact force at any plate
location can be determined using this analysis method for both monolithic and sandwich plates. The effects
of large deformations for the flat and cylindrically curved plates are described using the experimental
results and the results from the nonlinear analysis. Impact responses of 8-, 16-, and 24-ply-thick quasi-
isotropic and orthotropic monolithic curved and flat plates, and 32- and 48-ply-thick quasi-isotropic plates
are discussed. A linear analysis method has been used to predict the impact response of flat sandwich
plates with a Nomex honeycomb core and with different face sheet thicknesses, core thicknesses and core
densities.
2. Analysis Methods
Two analysis methods were used to develop the analytical results presented in the present paper.
One method is a nonlinear analysis method for monolithic cylindrically curved composite plates (Ref. 15).
The other method is a linear analysis method f01 "omposite sandwich flat plates (Ref. 14). First-order
shear-deformation effects are included in both analysis methods.
3. Experiments and Methods
3.1. Test specimens
The specimens tested in this investigation were fabricated from commercially available
unidirectional graphite fiber tape material preimpregnated with an epoxy resin. The graphite-epoxy
material system used in this study consisted of Hercules, Inc. AS4 graphite fibers and Hercules, Inc. 3502
epoxy resin. The identification of commercial products in the present paper is intended to describe
adequately the test specimens, and does not constitute endorsement, expressed or implied, by the National
Aeronautics and Space Administration. The mechanical properties for this material system are presented in
Table l. Unidirectional tapes were laid up to form 8-, 16-, 24-, 32- and 48-ply (45/0/-45190)ns quasi-
isotropic laminated plates that were cured in an autoclave using the resin manufacturer's recommended
procedures. The resulting plates were ultrasonically inspected to establish specimen quality and then
machined into 5-in-wide by lO-in-long rectangular specimens.
3.2 Apparatus and Tests
Knife-edge supports were used to provide simply supported boundary conditions for the test
specimens. The knife-edge supports were attached to the edges of each specimen at a location that is 0.64
centimeters from each edge. Each specimen was impacted on one surface at the specimen center by either
an airgun-propelled or a dropped-weight impactor. The airgun used in the present study is based on the
airgun described in Reference 16. The airgun-propelled impactors were O.5-in.-diameter aluminum or steel
balls that were propelled at the specimens with a given speed. The dropped-weight impactor consisted of a
2.5-lb dropped-weight assembly with an instrumented tup and a O.5-in.-diameter hemispherical steel
impactor tip. For the dropped-weight impact tests, the dropped-weight assembly was raised to a prescribed
height and then released to impact the specimen. Specimens were instrumented with electrical resistance
strain gages. Force and strain gage data were recorded during the impact event using a digital high-speed
data acquisition system. Impactor weights and impact-energy levels for the tests are given in Table 2.
491
4. Results and Discussion
The analytical and experimental results of the present study are presented and compared in this
section. Unless otherwise noted, all strain results described herein are in the longitudinal direction (y-
direction) of the plate. Also, unless otherwise identified, the longitudinal strain data for the experiments are
from a strain gage rosette placed at the center of the plate.
4.1. INFLUENCE OF IMPACTOR MASS
The effects of impactor mass on impact contact force and damage area associated with low-speed impact
are reported in Refs. 17-22. Spherical steel projectiles that weigh 0.0022 Ib and 0.00132 lb and nylon
spheres that weigh 0.000374 lb were used in Ref. 17 to study the effects of impactor mass and impactor
material on impact damage initiation threshold. In this study (Ref. 17), it was shown that the impact energy
needed to initiate damage increases as the impactor mass increases. The effects of low-speed and high-
speed impacts with impactors of different masses were investigated in Refs. 17-21. Penetration threshold
impact energy levels were studied in Refs. 17 and 18 using 1.5-lb impactor weights for low-speed impacts
and 0.OO22-lb impactor weights for high-speed impacts. The studies in Refs. 20-21 identified the dynamic
effects associated with using different impactor weights. It was concluded in Ref. 21 that the damage area
decreases and the residual strength of the plate increases when the impactor velocity is increased for a
given impact energy. A similar study described in Ref. 22 suggested that the dynamics of the impact event
is a function of the masses of the plate and the impactor. It was concluded in Ref. 22 that an impact with a
small impactor mass at high speed results in greater forces on the plate than an impact with a large mass at
low speed. Most of these studies address impacts with a small projectile mass propelled at a high speed,
rather than impacts with a dropped-weight impactC'o
Dropped-weight impact tests are usually performed with different weights; these weights range
typically from 2.5 Ibs to 25 Ibs. The general practice for obtaining the compression-after-impact (CAl)
strength test data for a composite material or laminate is to perform a compression test on a flat plate which
has been impact damaged at its center by a dropped-weight impactor of a given mass. The information
generated using this method is then used to design wing and fuselage structures for impact damage
tolerance. The maximum contact force results for simply supported (45/0/-45190)2£ quasi-isotropic plates of
different radii subjected to a central impact are presented in Fig. I. The contact force results in Fig. I are
generated using the analysis method presented in Ref. 15 for an impact-energy level of 7 ft-Ibs with
different combinations of impactor mass and impactor speed. Since the impact-energy level is constant for
the results shown in this figure, an increase in impactor speed suggests a decrease in impactor mass for a
given impact-energy level. These results indicate that an impactor weight of less than 0.25 lb (which
corresponds to impactor speeds of less than 578 in.lsec) influences the magnitude of the maximum contact
force for plates which have large curvature. Similar results are presented in Fig. 2 for flat sandwich plates
with (45101-45190)2£ quasi-isotropic AS4/3502 graphite-epoxy facesheets and with different core stiffnesses.
The analysis approach presented in Ref. 14 has been used to generate these results. For the range of core
densities used in this limited study, it appears that impactor mass values of less than 0.25 lb have an
insignificant effect on contact force results. This insensitivity to impactor mass is due to the very local
nature of the impact response of facesheets that are supported over their entire surface area by the sandwich
core. These results suggest that the influence of impactor mass on impact response, damage characteristics
and compression-after-impact strength of composite flat plates is still not well understood. There is very
limited information in the literature that relates the residual strength of composite flat plates to low-speed
impact damage caused by different masses over a range of impact-energy levels.
An analytical effort to identify the influence of impactor mass on the contact force and surface
strains of (45/0/-45190)6.1 quasi-isotropic flat plates was reported in Ref. 16. The results presented in this
reference indicate that increasing the impactor mass results in a marginal increase in the maximum contact
force and surface straiil values for a given impact-energy level. This observation is consistent with the
maximum analytical contact force results presented in Fig. 3 for a 48-ply-thick flat plate. Since the contact
force associated with an impact event is a major contributor to the resulting damage and to the residual
strength of the plate, it is important to understand the influence of the impactor mass on the plate response.
A normalized impactor momentum approach for relating the analytical plate contact force and strain results
to impacts with different impactor weights was suggested in Ref. 16.
492
The effects of increasing impactor mass on contact force results are presented in Fig. 4. The maximum
contact force results corresponding to impacts with increasing energy levels are plotted in this figure for
different impactor weights. All of the data in the figure exhibit a nonlinear trend between the impact-
energy levels and the contact force. For impactor weights of 2.5 lbs and 5.0 lbs, the maximum contact
force magnitude increases as the impactor weight increases. This response is due to damage initiation in
the plate as the impact-energy level is increased. The initiation of damage results in a decrease in the value
for the maximum contact force that is developed due to impact. It has been shown in Reference 23 that a
5.0-lb impactor weight causes damage to initiate at approximately 12 ft-lbs for the plate specimen studied.
The contact force magnitudes for impacts with 10.0- and 20.0-lb weights indicate a trend which suggests
that the contact force magnitudes for impacts using these weights are smaller than those for the 2.5 and 5.0
lb impactor weights. The contact force results for both the 10.0- and 20.0-lb impactors also exhibit an
increasing and then a decreasing trend with the curves shifted along the abscissa as the impactor weight is
increased. If a decrease in the positive slope of the contact force result curves is an indication of damage
initiation in a plate, these results suggest that the impact-damage-initiation threshold-energy levels decrease
and then increase as the impactor mass is increased. These results have no consistently general trend and it
appears that the plate response for impacts with 10.0- and 20.0-1b weights is affected significantly by the
dynamic interaction of the plate and the impactor mass which results in a plate global bending response.
The damage area results determined from ultrasonic C-scan images are presented in Fig. 5 for
increasing values of impact-energy levels and impactor masses. The damage areas calculated from C-scan
images are qualitative indications of damage and cannot be used to interpret directly the compressive
residual strength of the plate. The resulting damage areas for the 2.5- and 5.0-lb impactor weights increase
with decreasing weight with the larger impactor mass causing more damage. This trend is similar to the
trend observed for the contact force results. When impacted with 10.0- and 20.0-lb impactor weights, a
larger damage area seems to develop as the impact-energy levels are increased. For impacts from 10.0- and
20.0-1b weights, the damage areas for impact-e..",gy levels less than 15 ft-Ibs are comparable to the
damage areas for impacts from 2.5- and 5.O-lb impactor weights for corresponding impact-energy levels.
For values of impact-energy levels greater than 15 ft-Ibs, the resulting damage area gradually increases
with increasing impact energy. A change in the dynamic interaction of the test specimen with the impactor
results in a change in the test specimen impact damage mechanisms as the impactor mass is increased.
Compression-after-impact (CAl) strength test results for the test specimens are shown in Fig. 6 as a
function of the impact-energy levels generated using different impactor weights. These results indicate that
the specimen failure loads generally decrease as the impactor energy is increased. There is an insignificant
difference in compression failure loads for specimens impacted at an impact-energy level of approximately
15 ft-Ib or less. For impact energy levels greater than .15 ft-lbs, this variation increases and reaches a
maximum value of approximately 20 percent for a 25 ft-lb impact energy. Plate specimens impacted at a
given impact-energy level with heavier impactor weights appear to have a higher CAl strength. Although
this result is consistent with the contact force and damage area results presented previously, these data do
not have a consistent trend over the entire range of impact-energy levels considered in this study.
These impact test data are presented in Figs. 7-9 in terms of impactor momentum as a parameter to
help understand the trends of the results. The contact force, impact damage area and the CAl strength
results are shown to be nonlinear functions of the impactor momentum. The nonlinear trend for the
maximum contact force results is not due to large plate deformations during the impact event (see Ref. 24),
but is a consequence of the changes in stiffness of the plate at the impact location. The trends of these
results can be used to identify the effect of impactor momentum on the plate impact response. These
results indicate that the heavier impactor results in a smaller impact damage area for the two weights
impacting the plate with a specified impact momentum. This observation is consistent with the findings in
Ref. 16 that indicate that increasing the projectile mass corresponds to a decrease in the impact damage
area. Similarly, the CAl strength data suggest that, for a given impactor momentum, the lighter impactor
results in a lower compression failure load. Scaling contact force and damage area results based on ratios
of impactor momentum is not applicable to results for impact energies which cause damage to the plates.
Contact force and damage area results are nonlinear functions of impact energy for these impact-energy
levels.
4.2. EFFECTS OF PLATE CURVATURE
Contact force test results are presented in this section for flat and cylindrically curved composite plates
that were impacted by dropped weights with impact-energy levels that do not initiate damage in the plates.
493
These results are compared with analytical results for curved plates impacted with dropped weights and
airgun-propelled impactors with the same impact-energy levels to identify the differences in the response
characteristics of plates impacted by airgun-propelled and dropped-weight impactors. Impact-damage
initiation and barely visible impact-damage threshold-energy results are also presented, and the effects of
varying plate radius and thickness on these results is discussed. Compression-after-impact residual-
strength results are presented for plates with barely visible impact damage, and a criterion based on these
results is also presented. The effects of varying plate radius and thickness on these results are discussed.
4.2.1. Contact Forces
The maximum contact force values measured by an instrumented dropped-weight test apparatus are
shown in Fig. 10 for 16-ply-thick plates with different radii and impacted with 1.5 ft-Ib of impact energy.
The experimental results are shown by the open circles in the figure and the corresponding analysis results
from the analysis of Ref. 23 are shown by the open squares. It is assumed that flat plate results can be
represented in this figure by the results for a plate with a 200-inch radius. These results indicate that the
value of the contact force decreases significantly for radii between 15 and 60 inches, and then increases to a
value that closely corresponds to the flat-plate contact-force result for radii greater than 60 inches. These
results suggest that curved plates have a nonlinear response for radii less than 60 inches. These
experimental and analytical contact-force results for dropped weights compare well. The analytical results
for the airgun-propeJled impactors with a 1.5 ft-Ib impact-energy level are shown in the figure by the open
triangles. The magnitudes of these contact forces are significantly greater than the contact forces for the
dropped-weight impact events, and the radius of the plate has an insignificant influence on the value of the
contact forces. Apparently, the high-speed transient nature of the airgun impact event results in a very local
response of the plate which does not affect the value of the contact forces for the plates considered in this
study.
4.2.2. Impact-Damage-Initiation and Barely Visible Impact-Damage Energies
The differences in the contact-force magnitudes described in the previous section influence the damage
characteristics of the plates. The damage-initiation impact-energy levels for curved plates were determined
by impacting a given specimen with impact-energy levels that were increased in increments of 0.25 ft-Ib
until damage initiation occurred. Damage initiation was determined nondestructively using the volumetric
scanning approach outlined in Ref. 16. The damage-initiation results for the 16-ply-thick (45/0/-45/9O)2s
quasi-isotropic plates impacted with dropped weights and airgun-propelled impactors are indicated in Fig.
II by the open circles and open squares, respectively, for different plate radii. The impact energy required
to initiate damage is significantly greater for the dropped weights than that of the airgun-propelled
impactors. As the plate radius increases to a value of approximately 30 inches, the impact energy required
to initiate damage increases, and then decreases for larger radii for both the airgun-propelled and dropped-
weight impactors. The dropped-weight impact-energy results are consistent with the trend of the maximum
contact-force results presented in Fig. 10. This observation suggests that a greater impact-energy level is
needed to generate the impact force required to initiate damage for the 30-in.-radius plates than for the
other plates considered in this study. The trend for airgun contact-force results presented in Fig. 10 is not
consistent with the nonlinear damage-initiation impact-energy results trend as the plate radius increases.
The damage-initiation results presented in Fig. II suggest that the damage-initiation mode associated with
the much larger contact forces for the airgun-propelled impactors could be different from those for the
dropped weights for plates with different radii, and might be responsible for the nonlinear trend observed in
the damage-initiation results.
The impact-energy levels required to cause barely visible impact damage on the impacted surface and
opposite or back surface of 16-ply-thick plates were determined by increasing the airgun and dropped-
weight impact-energy levels until damage was visible on one or both plate surfaces. Typical damage
modes for the impacted and back surfaces of 16-ply-thick flat and 15-in.-radius curved plates impacted
with dropped weights are shown in Figs. 12 and 13, respectively. The impact event causes a dent to form
on the impacted surface of the plate at the impact site, and causes the surface ply on the back surface of the
494
plate opposite to the impact location to crack and separate from the plate as shown in the figures. These
observations indicate that barely visible impact damage can be detected when the measured dent depth on
the impacted surface is approximately 0.05 in. for the l6-ply-thick specimens for both the airgun-propelled
and dropped-weight impactors. The dropped-weight impact-energy results presented in Fig. 14 suggest that
greater impact-energy levels are needed to develop visible damage in both the 8- and 16-ply-thick curved
plates with radii between 15 and 60 in. The airgun impact-energy level needed to develop barely visible
impact damage is a constant of approximately 9.0 ft-Ibs over the range of plate radii considered in this
study as indicated in Fig. 14. The 0.05-in. dent depth for the 8-ply-thick plate (which is 0.044-in. thick)
does not result in a through penetration of the plate. The damage modes for the 16-ply-thick. 15-in.-radius
curved plates that were impacted with airgun-propelled and dropped-weight impactors with 8.5 ft-Ibs of
impact energy are shown in Fig. IS. The damage due to the airgun-propelled impactor is more severe at the
impact site than that of the dropped-weight impactor. The airgun-propelled impactor causes all of the plies
to fail through the plate thickness. but the dropped-weight impactor only causes extensive delaminations in
the plate. The results presented in Fig. 14 suggest that radius is an important structural parameter that must
be considered when assessing the damage tolerance of thin, cylindrically curved plates.
The l6-ply-thick flat and curved plates were analyzed with an applied quasi-static load of 1,000 lbs
and the out-of-plane displacement results are presented in Figure 16. The nonlinear analysis was conducted
using the STAGS Structural Analysis of General Shells finite element code (Ref. 25). These results
suggest that the curved plate with 30-in. radius exhibits a pronounced nonlinear softening response
compared to the other plates in the load range (-750 lbs) which causes barely visible impact damage. This
nonlinear softening mechanism may nave contributed to the improved damage tolerance results observed
for the 30-in.-radius plate (Ref. 26). The nature and extent of the nonlinear softening response is a function
of the plate laminate stacking sequence, plate thickness, radius and side lengths. The influence of plate side
lengths on the nonlinear softening behavior is illu,ua(ed in Figure 17 by the nonlinear analysis results for a
12-in.-square plate. When the side lengths of the 16-ply-thick plate is increased to 12 inches, the plate with
60-in. radius appears to exhibit a pronounced nonlinear softening response and, hence, could exhibit
improved damage tolerance characteristics.
4.2.3. Compression-AJter-Impact Strength
Compression-after-impact residual strength results for damaged and undamaged 8- and 16-ply-thick
cylindrically curved plates are summarized in Fig. 18. These results indicate that the residual strength of
these plates is a function of the plate radius. The 15-in.-radius plates have the most severe reduction in
residual strength relative to the flat plates. The results for the 16-ply-thick curved plates indicate that the
plates with radii between 30.0 and 60.0 inches have higher failure loads than the flat plates, and the plates
with radii approaching 15.0 inches have lower failure loads that the flat plates.
Results from an analytical and experimental study of the effects of dropped-weight and airgun-
propelled impactor masses on the impact response and compression-after-impact residual strength of flat
and curved laminated composite plates have been presented. Contact force results obtained from an
instrumented dropped weight and the corresponding damage area obtained form ultrasonic C-scan images
are used to relate impactor parameters and response and damage characteristics. The experimental results
suggest that as impact energy increases, contact force values decrease for all impactor mass values
considered in this study. The decrease in contact force as impact energy increases are is to the initiation of
damage, and a change in damage modes in the specimens. As the impactor mass increases, the contact
force results increase with increasing impact-energy levels. This trend suggests that the damage area
decreases as the mass of the impactor increases, and the damage area results follow a trend that is
consistent with the contact force results. The compression-arter-impact test results indicate that the
compression strength of a plate varies when impactors of different weights arc used to impact the plate at a
given impact-energy level. This variation in results is determined to be as much as 20 percent over the
impact-energy range investigated in this study.
Scaling the contact force, damage area, and compression failure strength results using the impactor
momentum as a parameter is not appropriate since these results are nonlinear functions of the impactor
495
momentum. The nonlinearities occur when the impact energies are sufficient to cause damage initiation to
occur in a plate. Presenting the test results as a function of the impactor momentum, however, appears to
provide clear trends for plate response. The results presented included the results of a study of the effects
of low-speed impact damage on the response of thin, cylindrically curved, graphite-epoxy plates. These
results include quasi-isotropic plates with different radii of curvature and different thicknesses, and the
results are compared with results for the corresponding flat plates. The results of the study indicate that the
contact forces for cylindrically curved plates impacted with a given impact-energy level are a function of
the plate radius of curvature and thickness. Test results for a 1.5 ft-Ib impact-energy level indicate that the
magnitude of the maximum contact force developed for the curved plates decreases and then increases as
the radius of the plate increases. This contact-force trend is also predicted analytically. This decreasing
and then increasing trend for the contact forces associated with dropped-weight impactors is reflected in the
impact-energy results for both damage initiation and for the creation of barely visible impact damage. The
impact-energy levels for damage initiation in 16-ply-thick plates are different for airgun-propelled and
dropped-weight impactors, but the impact-energy levels required to produce barely visible impact damage
are not different. The contact forces predicted by analysis for an airgun-propelled impactor with 1.5 ft-Ibs
of impact energy are much greater than the corresponding forces predicted for a corresponding dropped-
weight impactor, and are not influenced by the plate radius. Apparently, the response of a plate impacted
by an airgun-propelled impactor is localized enough that curvature effects do not affect the plate response.
The compression-after-impact residual-strength results for flat and curved 8- and 16-ply-thick plates
suggest that these results are also a function of the plate radius. The residual strength of flat and 15-in.-
radius curved plates with barely visible impact damage is approximately 3 and 15 percent less that the
failure load for the corresponding uhu"maged plates, respectively. These results suggest that radius of
curvature is an important structural parameter that must be considered when determining the damage-
tolerance characteristics of thin, cylindrically curved. composite plates.
496
Table 1. Mechanical properties of AS4- 3502 graphite-epoxy unidirectional tape material.
Longitudinal modulus, EI (Msi) 20.00

Transverse modulus, E2 (Msi) 1.30
In-plane shear modulus, G 12 (Msi) 0.87
Transverse shear modulus, G23 (Msi) 0.51
Transverse shear modulus, GI3 (Msi) 0.87
Major Poisson's ratio, V 12 0.30
Table 2. Impactor weight and impact-energy level combinations for impact tests.
Impactor weight, Ib Impact energy, ft-Ib
2.5 7.5 10.0 12.5 15.0 20.0

5.0 10.0 12.5 15.0 20.0 25.0
10.0 10.0 12.5 15.0 20.0 25.0
20.0 10.0 12.5 15.0 20.0 25.0
497
~ Flat plale
--B - Curved plate: 30-in. radius
-~- Curved plate: IS-in. radius
1400
,A
i'6. 4
1300
~
1200 \
\
1100
Maximum
contact 1000
force,lb
900
800
2000 2500
Impact speed, in/sec
Figure 1. Influence of impactor weight on the cont- -< force results for (45/0/-45190)2s plates impacted with an
impact-energy level of 7 ft-Ibs.
~ Sandwich core stiffuess = 25875 Ib/in.

--a - Sandwich core stiffuess = 17250 Ib/in.
- &- - Sandwich core stiffness = 34500 lb/in.
2400 tk4--tJ. ~
"'tJ.lr ~ b
2200 LBt:I'II_o.....::L ....
2000
Maximum
contact
1800
1600
\
force, lb
1400
\
1200 \
1000
0
Impact speed, inlsec
Figure 2. Influence of impactor weight on the contact force results for sandwich plates made of (45/0/-
45190)2s facesheets with core of different stiffnesses and impacted with a 7 ft-lb impact-energy
level.
498
3800
3600
3400
3200
Maximum 3000
contact
force.lb 2800
2600
2400
2200
0 500 1000 1500 2000 2500
Impact velocity. in/sec
Figure 3. Influence of impactor weight on the contact force results for (45/0/-45190)68 flat plate impacted with
an impact-energy level of 9 ft-Ibs.
-e-- 2.51b
-B -
-A--
5.01b
1O.01b
4500 --,..-- 2O.01b
0
4000
lP
3500 .,.... 0/
I
, I
Maximum
3000
~ I
contact
force, Ib
2500 / \ ,,
J ,,
I
2000 \
,,\g- -A
1500 -<> -~-'
1000
5
Impact energy, ft-Ib
Figure 4. Experimental contact force results for impacts with increasing impact-energy levels and with
different impactor weights.
499
-e- 2.Slb
--B - 5.01b
-tr- 10.0 In
5 - -~-- 20.01b
3
Damage
. 2
area, In. 2
Figure 5. Experimental damage area results for imy __ .:s with increasing impact-energy levels and with
different impactor weights.
--e- 2.51b
---8 - 5.01b
-A-- 10.01b
45 - -G-- 20.01b
40
Compression 35
failure
load, kip
30
25~----~----~----~-----L----~
5
Figure 6. Compression-after-impact strength results for impacts with increasing impact-energy levels
and with different impactor weights.
500
-B-- 2.51b
-e - 5.01b
-A-- 10.01b
4500 - -~-- 20.01b
9
4000
0
3500 J
Maximum
contact
3000
A
IA\
2500
force,lb
\
2000 \
~ ..!v'
.J$
1500
1000
0
Impact momentum, lb-sec

Figure 7. Experimental contact force results for impacts with increasing values of impactor momentum and
with different impactor weights.
--e- 2.51b
--E3 - 5.01b
-A-- 1O.01b
5 --~-- 20.01b
~
I I
4
, I I
Damage
3 1\ "'6 I ,,,
/
,<>
I
I
I
~
2
area, in. 2 J
~-<>
,
,,
0
0
Impact momentum, lb-sec

Figure 8. Experimental damage area results for impacts with increasing values of impactor momentum and
with different impactor weights.
SOl
--e- 2.51b
-i3 - 5.01b
-A-- 10.0 III
45 - -0-- 20.01b
Compression 35
failure
load, Ib
40
\ .\ tfl"'t1
~~
O~
I
,
0._ -0..
.
~
~ \
30 I!I A- ~
I
\
IJ..
25
0
Impact momentum, Ib-sec
Figure 9. Compression-after-impact strength results for impacts with increasing values of impactor
momentum and with different impactor weights.
---G-- Experiment - Dropped weight

--B - Analysis - Dropped weight
1200 - A- • Analysis - Airgwi
1000 &--/1- -A - - - - - - -/1
800
Maximum
contact
force. Ib 600
.---.::: - - - [J
400
50 100
Plate radius, in.
Figure 10. Contact force results for l6-ply-thick, quasi-isotropic plates impacted by dropped weight sand
airgun propelled impactors with 1.5 ft-lbs of impact energy.
502
0 DroppcJ weight
3
0 Airgun
2.5 0
2 0
Impact
energy,
ft-Ib \.5 0 0
0
0 0
0
0.5
0
Plate radius, in.
Figure 11 . Damage-initiation impact energy levels for 16-ply-thick, quasi-isotropic plates impacted by
dropped-weight and airgun impactors.
____ Impact
location
Back surface
damage
a. Impacted surface b. Back surface
Figure 12. Photographs of a 16-ply-thick quasi-isotropic flat plates with barely visible impact damage.
503
Impact location
a. Impacted surface b. Back surface
Figure 13. Photographs of a 16-ply-thick quasi-isotropic curved plate with IS-in. radius with barely visible
impact damage.
o 8-ply - cropped weight

o 16-ply - cropped weight
II 6. 16-ply - airgun
o
10
o Airgun impact ~6
9
o
8
Impact 7
energy, Dropped-weight
ft-Ib impact
6
4
o o
30
Plate radius, in .
Figure 14. Dropped-weight and airgun impact-energy levels for barely visible impact damage in 8- and 16-
ply-thick quasi-isotropic plates with different radii.
504
~ Shear failure
.
a. Airgun impact damage b. Dropped-weigh! impact damage
Figure 15. Photomicrographs of 16-ply-thick quasi-isotropic 15-in.-radius curved plates subjected to

dropped-weight and airgun impacts with 8.5 ft-Ibs of impact energy.
--e-- Flat plate

--B - curved plate, 60-in. radius
- A- - Curved plate, 30-in. radius
1200 - -0 - - Curved plate, IS-n. radius
1000
800
Load,lb 600
400
200
0
0 0.5 1.5 2 2.5 3 3.5
DisplacementIPlate thickness
Figure 16. Analytical load versus displacement results for 16-ply-thick quasi-isotropic plates subjected to a
static load.
505
--e- Flat plate

-i3 - Curved plate, 15-in. radius
- tr- - Curved plate, 6O-in. radius
- -$ - - Curved plate, 30-in. radius
1000
)lJ ,'I
800 , 't!
,(1 I
" I
600
Load,lb
400
200
0
0 2 3 4 5
DisplacementIPlate thickness
Figure 17. Analytical load versus displacement results for 16-ply-thick, 12-in. square quasi-isotropic plates
subjected to a static load.
506
0 No damage
14
• Impact damage
13 • 0
12
•
•
0
Compression II
failure
load, 0
kip 10
9
•
8
0
Plate radius, in.
a. 16-ply-thick curved plates
0 No damage
• Impact damage
3.5
0 • 0
3
•
Compression
failure
load. kip 2.5
•
2~--~----~----~----~----~
o
Plate radius, in.
b. 8-ply-thick curved plates
Figure 18. Summary of residual strength results for curved plates with different radii.
507
6. References
I. Sun, C. T., and Chattopadhay, S., (1975) Dynamic Response of Anisotropic Laminated Plates Under
Initial Stress to Impact of a Mass, Journal of Applied Mechanics. Vol. 42, pp. 693-698.
2. Dobyns, A. L., (1981) Analysis of Simply-Supported Orthotropic Plates Subject to Static and Dynamic
Loads, AIM Journal, Vol. 19, No.5, pp. 642-650.
3. Ochoa, C. M., (1987) "Nondimensional Models for Low Velocity Impact of Laminated Composite
Panels," Proceedings of the AlANASMEJASCEJAHS 28th Structures, Structural Dynamics, and
Materials Conference, Monterey, CA, April 6-8, pp. 443-447. AIAA Paper No. 87-0802.
4. Shivakumar, K. N., Elber, W., and Illg, W., (1985) Prediction of Low-Velocity Impact Damage in Thin
Circular Laminates, A1AA Journal, Vol. 23, No.3, pp. 442-449.
5. Christoforou, A. P., and Swanson, S. R., (1991) Analysis ofImpact Response in Composite Plates,
International Journal of Solids and Structures, Vol. 27, No.2, pp. 161-170.
6. Olsson, R., (1992) Impact Response of Orthotropic Composite Plates Predicted from a One-Parameter
Differential Equation, AIM Journal, Vol. 10, No.6, pp. 1587-15%.
7. Chou, P. C., and Mortimer, R. W., (1976) Impact Behavior of Polymeric Matrix Composite Materials,
Report No. AFML-TR-76-242, Air Force Materials Laboratory, Wright-Patterson AFB, Ohio.
8. Qian, Y., and Swanson, S. R., (1990) A Comparison of Solution Techniques for Impact Response of
Composite Plates, Composite Structures, Vol. 14, No.3, pp. 177-192.
9. Abrate, S., (1973) Impact on Laminated Composite Materials, Applied Mechanics Reviews, Vol. 44, No.
4, pp. 155-190.
10. Ross, C. A., and Sirakowski, R. L., (1973) Studies on the Impact Resistance of Composite Plates,
Composites, Vol. 4, pp. 157-161.
11. Joshi, S. P., and Sun, C. T., 1985) Impact Induced Fracture in a Quasi-isotropic Laminate, Journal of
Composite Materials, Vol. 19, pp. 51-66.
12. Cantwell, W. 1., Curtis, P., and Morton, 1., (1986) An Assessment of the Impact Performance ofCFRP
Reinforced with high-strain Carbon Fibers, Composite Science and Technalogy, Vol. 25, No.2, pp.
133-148.
13. Greszczuk, L. B., (1982) Damage in Composite Materials due to Low Velocity Impact, Impact
Dynamics, edited by Z. A. Zukas, et aI., Wiley, New York, pp. 55-94.
14. Ambur, D. R., and Cruz, J. R., (1995) Low-Speed Impact Response Characteristics of Composite
Sandwich Panels, AlAA 95-J460-CP.
15. Ambur, D. R., and Starnes, J. H., (1997) Nonlinear Response and Damage Initiation Characteristics of
Composite Curved Plates Subjected to Low-Speed Impact, AIM 97-J343-CP.
16. Prasad, C. B., Ambur, D. R., and Starnes, J. H., (1993) Response of Laminated Composite Plates to
Low-speed Impact by Airgun-propelled and Dropped-weight Impactors, Proceedings of the
AIAAlASMEJASCEJAHS/ASC 34th Structures, Structural Dynamics, and Materials Conference, La
Jolla, CA, AIAA Paper No. 93-1402.
17. Cantwell, W. J., and Morton, J., (1989) The Influence of Varying Projectile Mass on the Impact
Response ofCFRP, Composite Structures, Vol. 13, pp. 101-114.
18. Cantwell, W. J., and Morton, J., (1989) Comparison of the Low and High Velocity Impact
Response ofCFRP, Composites. Vol. 20, No.6, 1989, pp. 545.
508
19. Delfosse, D., Poursartip, A., Coxon, B. R., and Dost, E. F., (1995) Non-~enetratin~ Impact
Behavior of CFRP at Low and Intermediate Velocities, Composite Mate~lals: Fatlg~e and
Fracture - Fifth Volume, ASTM STP 1230, R. H. Martin, Ed., American Society for Testmg and
Materials, Philadelphia, pp. 333-350.
20. Christoforou, A. P., and Swanson, S. R., (1990) Analysis of Simply-Supported Orthotropic Cylindrical
Shells Subject to Lateral Impact Loads, Journal of Applied Mechanics. Vol. 57. pp. 376.
21. Ghaseminejhad, M. N., and Parvizi-Majidi. A.. (1990) Impact Behaviour and Damage Tolerance of
Woven Carbon Fibre-Reinforced Thermoplastic Composites. Composites. Vol. 21, No.2, pp. ISS.
22. Cairns. D. S .• and Lagace, P. A., (1989) Transient Response of GraphiteJEpoxy and KevlarlEpoxy
Laminates Subjected to Impact, A1AA Journal, Vol. 27, No. II, pp. 1590. .
23. Ambur, D. R., Starnes, I. H., and Prasad, C. B., (1995) Low-Speed Impact Damage-Initiation
Characteristics of Selected Laminated Composite Plates, AIAA Journal, Vol. 33, No. 10, pp. 1919-
1925.
24. Ambur, D. R., Starnes, 1. H., and Prasad, C. B., (1993) Influence of Transverse Shear and Large
Deformation Effects on the Low-Speed Impact Response of Laminated Composite Plates, NASA TM-
107753.
25. Brogan, F. A., Rankin, C. C., and Cabiness, H. D., (I994) STAGS User Manual, LMSC P032594.
26. Ambur, D. R., and Starnes, J. H., (1995) Influence of Large-Deflection Effects on the Impact Response
of Flat and Curved Composite Plates, AIAA-95-1205-CP.
PROSPECTS OF SMART STRUCTURES FOR FUTURE AIRCRAFT
K. W. Dittrich
Daimler-Benz Aerospace AG, Military Aircraft
0-81663 Muenchen
1. Introduction
Smart structures, in which additional functions (e.g. vibration control, health monitoring,
sensing, shape change) beside the function of loadcarrying are integrated into the
airframe structure, are considered to be an enabling technology for future high per-
formance aircraft. They offer to provide additional degrees of freedom for structure and
equipment design, pushing overall system performance.
At present, there are numerous technology and even demonstration programs related to
this technology going on worldwide. It can be noticed that these programs are often
purely technology driven, starting from the question of technical feasibility, not of
economic viability. This is understandable from an scientific point of view, but may
hinder industrial acceptance.
In this lecture, a top down approach to the world of smart structures will be presented.
The lecture concentrates mainly on military aircraft, but some remarks on other appli-
cations are given. Starting from basic considerations with regard to requirements and
time frames of present and future military aircraft developments, a critical review on
different concepts for smart structures will be presented. Topics to be discussed are:
• Introduction to smart structures
• Description and critical review of important smart structure concepts under investi-
gation presently
• Basic considerations on applicability and realization time frame of smart structures
• Needs and requirements of the aircraft industry
• Overview over ongoing technology programs
509
510
2. Introduction to smart structures
According to a definition now widely accepted, smart structures can be defined as sys-
tems which combine the functions of sensing, control and actuation in a single structural
unit. Thereby, smart structures can sense external or internal conditions and act in
response to respective changes in the environment. In a more practical way, smart
structures can be understood as a combination of different functional elements in a
single structural unit: sensors, control units, actuators and the loadcarrying structure
itself.
The development of smart structures began in the late 1980s. [t was made possible by
the evolution of two different technologies to a reasonable technological maturity:
• So called "smart materials" like piezoelectrics or shape memory alloys (SMA) show
sensorial or aktuatoric functions, or even an combination of both. Thereby, these
materials allow to realize different functions in a pure bulk material, without the need
for complex mechanics.
• Fiber reinforced plastic composites (FRP) as an engineering material have two
advantages over previously used more or less homogeneous materials like metals:
First, the unique manufacturing process of FRPs, during which the final structural
material is formed in the manufacturing process itself, allows for the integration of
other components into the structure. Second, the mechanical and electromagnetic
properties of FRPs can be tailored in a way hardly achievable with bulk materials by
selecting appropriate fiber types and lay-up sequences.
The idea to combine both technologies leads to the concept of multifunctional struc-
tures, combining structural materials with sensorial and actuatoric functions. Examples
of multifunctional structures are quite old in the Aerospace world. Radomes for example
combine electromagnetic and structural functions in a single structure. Interesting
functions achievable with smart structures, in view of possible Aerospace applications,
are expected to be:
• Change of the structural properties
o Shape Change (e.g. adaption of the airfoil section, fine-tuning of antenna reflec-
tors for space applications)
o Vibration Suppression (e.g. buffet suppression)
o Vibration isolation
• Sensing
o Determination of the structural conditions (e.g. structural integrity)
o Determination of the external environment (e.g. flow conditions)
• Control of the environment
o Flow control
o Thermal control
• Electromagnetic interaction with the environment
o Integrated antennas (e.g. communication antennas, radar systems)
• Signature control (which will not be discussed further in this article)
511
The design challenge of smart structures is that a multidisciplinary approach is neces-

sary, as previously de-coupled technical disciplines, like structural mechanics and
electronics, have to design the same structure. This is a major challenge, but not totally
new, as the already mentioned example of the radome technology shows.
Smart structures are designed as a combination of structural and functional elements
within the structure. Typical functional elements under investigation are:
Elements Realization
Actuators Piezoelectric materials (PZT), Shape Memory
Alloys (SMA), Electrorheological fluids (ER),
pneumatics
Sensors Piezoelectric materials (PZT, PVDF), Fiberoptics
(strain sensors, temperature sensors)
Electronics Processors, transmit-/receive modules, amplifi-
ers, multiplexers, data links
Micro electro- Microsensors, Microactuators
mechanical de-
vices (MEMs)
Most work an smart structures concentrates presently on embedding actuators and

sensors into the structure. There are different smart materials for actuation and sensing
applications under investigation now.
Piezoelectrics like PZT (lead zirconate titanate) and PVDF (polyvinyldene flouride) can
be used either as actuator or as sensor. Piezoelectrics show a shape change when an
electrical field is applied (either contraction or expansion, depending on the field
polarity). This effect can be used to exert mechanical forces within a smart structure.
Unfortunately, the maximum strain levels are in the region of 200 ..300 microstrain only
for a PZT. On the other hand, when a shape change is imposed on the piezoelectric
material, a voltage can be measured, making the piezoelectrics good strain sensors,
especially for dynamic applications. PZT can be integrated quite nicely into an FRP
structure, as long as the strain compatibility with the structure is kept in mind (pure PZT
is limited to about I %0 strain, which can be pushed up to 3%0 within a real structure by a
careful design of the interface).
Magnetostrictors like Terfenol-D show a shape change when a magnetic field is applied.
The achievable strain levels of above 1.400 microstrain are much higher compared to
piezoelectrics, but the necessity to generate a magnetic field makes magnetostrictors
better suited for stand-alone actuators. As an example, magnetostrictors are very
interesting for vibration suppression systems.
512
Shape memory alloys (SMAs) are showing a solid to solid martensitic phase deforma-
tion, which allows for large, recoverable strains. Strains of up to 8% (or even 15% for
some copper-based SMAs) can be reversed by heating up a SMA to a temperature above
the phase transformation temperature. SMAs can be based on various metallic alloys,
like Nickel/Titanium (e.g. Nitinol, as for Nickel Titanium Naval Ordonance Labora-
tory), copper/aluminum/nickel, copper/zinc/aluminum or even some steel alloys. The
phase transition temperature is dependent on the alloy composition, typical upper values
of commercial Nitinol variants are about 90°C.
E1ectrorheological (ER) and magnetorheological (MR) fluids show a variation of the
rheologic properties when subjected to electric or magnetic fields. These fluids contain
small particles which form large chains when the exerting field is applied, resulting in an
increase in the apparent viscosity. Being very interesting for simple yet efficient valves
or clutches, an application of these materials for smart Aerospace structures has not yet
materialized.
The following table I lists some typical parameters of smart actuator materials:
TABLE I. Typical parameters of smart actuator materials

Smart material PZT PVDF Terfenol-D NiTiNol
Actuation principle Piezo- Piezo- Magneto- Shape
ceramic polymer strictor Memory
Alloy
Amax 190* 10- 12 23* 10- 12 >1,4%0 8%
mlV mN
Young's modulous [OPal 63 2 35 30
Martensit
90
Austen it
Tmax [0C] 360 80 - 120 380 90°C
Linearity + + 0
Hysteresis 10% > 10% 2% 5%
Temperature dependence %/0C] 0.05 0.8
Bandwidth High High Medium Low
3. Description of important smart structure concepts
The possible applications of smart structures in the Aerospace world (and beyond) have
still to be explored fully. On the other hand, a concentration process has happened in the
Aerospace industry in the past few years. As the development cycles are slowing down
and cost is becoming an increasingly critical issue, the main research is focused on a
513
limited number of applications which promise to show near- to mid-term benefit. These
important applications, as far as they are known publicly, are described now.
3.1 COMPOSITES HEALTH MONITORING
Composites structures are susceptible to a number of failure modes different from those
known from metallic structures. Most prominent problem of FRP structures are impact
damages, which may lead to barely visible impact damages. Therefore, one of the first
proposals for smart structures was to use integrated sensors to detect damages within the
structure itself. There are a number of different approaches under investigation now:
• Fiber optic sensors can detect stress concentrations or residual strains resulting from
an impact damage in the vicinity of a delamination. A close net of fiber optics may
monitor at least critical parts of the structure.
• The acoustic signature of impacts can be detected by integrated PZT or fiber optic
sensors. The characteristic acoustic signature of an impact can be filtered out of the
background noise by appropriate signal processing techniques.
• Lamb wave actuators on PZT basis can exitate lamb waves in the structure. The
propagation of these lamb waves is modified by delaminations in the structure. These
changes can be detected by lamb wave sensors.
• The dynamic behavior of the structure is changed by delaminations. Whereas the
classic modal analysis approaches turned out to be a dead end development in the
past, structural dynamic analysis techniques using a network of multiple strain sen-
sors are showing promising results in the identification of delaminations.
The expected benefits of the integration of an Composites Health Monitoring (CHM)
system into an FRP structure are:
• Reduction of inspection costs, as no scheduled inspection of the structure with costly
and time consuming non-destructive testing methods is necessary.
• The design allowables (e.g. the allowable strain levels within the structure) may be
increased, thus allowing for the design of lighter structures.
Unfortunately, the problems associated with CHM are manifold. First, for present FRP
structures which are in service or nearing service introduction (e.g. on Eurofighter
2000), there are no foreseeable requirements for scheduled inspections which could be
reduced by introduction of CHM. Second, an increase of the design allowables is
possible only for new designs. Third, the design allowables are not only limited by the
damage tolerance requirements. Thus, the potential weight and cost savings of a CHM
system are at least questionable. At the moment, the question is still open, whether or
not a CHM system is worth the effort
514
3.2 VIBRATION SUPPRESSION
Especially on military aircraft, high vibration levels may be encountered on the struc-
ture. These vibrations may be caused either by external forces (e.g. buffet, a forced
vibration of the structure, especially the fin, caused by strong turbulent flows exitated by
the aircraft forebody at high angles of attack) or by a coupling between aerodynamic
forces and structural elasticity (flutter).
By using actuators within the structure or at the interface between a structural compo-
nent (e.g. a fin or a flap) and the main structure of the aircraft, vibrations can be actively
dampened to a considerable degree. Practically, most work concentrates on buffet
suppression. The reason is that buffet mainly represents a problem of fatigue and ride
comfort, thereby a buffet suppression system must not necessarily be flight critical (but
it can be, if it can excite the structure itselfl). Flutter, on the other hand, is a problem of
instability which, if it occurs, can lead to a very fast catastrophic failure of the structure.
Therefore, the smart skins community is very reluctant at the moment to touch an active
flutter suppression system
As buffet occurs at frequencies of some 10Hz, only fast reacting actuators are feasible.
This effectively rules out SMAs for active systems. Most approaches known for buffet
suppression are using PZTs as an actor. Two ways of application seem feasible: struc-
tural integrated PZTs, acting to induce contra acting strain in the skin layers, or active
interfaces with integrated massive bulk PZT actuators. Technically, both ways seem to
be feasible.
Active vibration suppression systems have already been tested in wind tunnel models
and on larger demonstrator structures (e.g. at Dasa military aircraft). First flight tests of
a system are expected soon in the USA.
3.3 SHAPE CONTROL
Shape control of structures has different possible applications, ranging from quite large
deformations of airfoils to small adjustments of precision structures (e.g. space based
antenna reflectors). For military aircraft, the main interest at the moment is on the
variation of the airfoil section.
The shape of an airfoil section is so important for the performance of the aircraft, that
changing the airfoil of the structure during flight using integrated actuators was one of
the first ideas at the beginning of smart structures technology. There are different rea-
sons for changing the airfoil profile:
• Cruise camber control. The profile of an airfoil is a compromise between different
flight attitudes. A fine tuning of the profile can enhance the aerodynamic perform-
ance for a variety of flight attitudes.
• Maneuver load control. During high-g maneuvering, the wing root bending moment
increases rapidly. The structure must be designed to sustain the maximum bending
moment at the maximum g level. If the load distribution over the wing span during
515
high-g maneuvers is shifted inwards, the wing root bending moment can be reduced
on cost of some aerodynamic efficiency. A lower maximum wing root bending
moment results in a lighter structure. The shift of the load distribution is performed
by a spanwise variation of the profile.
• Maneuver enhancement. Instead of using a stiff wing with rudders for steering, the
structure may be flexible and "flown" to an appropriate angle of attack, thus acting
as a large control surface and improving maneuverability while maintaining a low
structural weight. Even small changes of the airfoil geometry can be used to induce
this controlled deformation of the wing.
There are a number of possible ways to achieve shape control on a structure. One basic
feature of shape controlled structures is that the structure should be flexible in the
direction of the planned movement, with the actuator stabilizing the structure in that
direction. In this way the force required for the actuator is minimized. The selection of
the actuator is determined by the required agility of the system: SMAs have low
bandwidth and are suited mainly for slow actuation requirements. If a faster response is
needed, PZTs, Magnetostrictors or pneumatics may be used.
Various investigations on shape control of airfoil sections are going on worldwide, one
of the front-runners presumably being the civil "Adaptive Wing" program of Dasa
Airbus, DLR and Daimler-Benz Research and Technology.
4. Basic considerations on applicability of smart structures
One of the basic lessons learned from the smart structures research during the past years
is that in most cases structures have to be designed for the "smart" functions from the
beginning on to fully exploit this capability. Therefore, it is less promising in general to
adapt a smart structures concept to existing structures. This leads to the question, under
which conditions we can realize smart structures on actual aircraft in the future. In this
context it is important to differentiate between new aircraft designs and retrofits of
existing aircraft.
The design steps of modem aircraft, both for new design and retrofits, can be described
as a stepwise process:
I. Concept Phase
2. Definition Phase
3. Development Phase
4. Production Investment Phase
One of the basic facts of aircraft development is that program managements are often
very conservative concerning new and untried technologies, which is very understand-
able. Therefore, concepts of smart structures have to be demonstrated before the concept
516
phase is started. In this context the time frame of foreseeable future developments is
worth looking on.
On the military side, a new combat aircraft (however it may look like), cannot be
expected to be launched until at least 20 lOin Europe. The time frame for the develop-
ment of smart structures is set by this assumption: until 20 I0, only retrofits are to be
expected which open up the possibility to introduce smaller applications of smart struc-
tures. CHM may be such an example (if it is needed). Larger applications, which require
a more massive ingress into the structure system (e.g. vibration suppression), must be
demonstrated until latest 20 10.
As brand-new aircraft are not on the agenda in the near future, the most practical ap-
plication of smart structures technology in near- to midterm is for improvements or
retrofitting of aircraft in production or already existing. The consequence is, that in this
case only a "local" approach is possible.
5. Needs and requirements of the aircraft industry
Considerable work is done on the exploitation of smart structures technology on the

scientific level. Still, there is a lack of basic understanding of some fundamentals of
smart structures. Material laws for the description of smart materials are still available
only in very preliminary depth. Another deficiency is that only limited FE-modules are
available for the design. Another problem is the question of aircraft compatibility, which
in the past has been poorly addresses by researchers.
To go ahead, more answers are necessary on questions like:
• Can the smart system be flight qualified?
• Has it been demonstrated yet (risk assessment)?
• Does it payoff?
The way to answer these questions is visible now. There is a new group of investigations
beside the pure research work forming during the past few years at the industry. Work
done asked in this context is:
• Establishing of clear specifications
• Qualification work
• Demonstrators
• Costlbenefit analysis
517
6. Conclusion
Smart structures can be considered to be an enabling technology for future high per-
formance aircraft. They provide additional degrees of freedom for structure and
equipment design, pushing overall system performance.
At present, there are numerous technology and even demonstration programs related to
this technology going on worldwide. It can be noticed that these programs are often
purely technology driven, starting from the question of technical feasibility, not of
economic viability. This is understandable from an scientific point of view, but to
achieve industrial acceptance more work is needed. on issues like cost/benefit analysis
and practicability
At the moment, there are signs of a more system oriented approach on this technology
visible, especially in industry. Especially demonstrator programs are needed in the near
future to transfer smart structures from academic research into real applications.

(NATO Science Series 361) J. N. Reddy (auth.), Carlos A. Mota Soares, Cristóvão M. Mota Soares, Manuel J. M. Freitas (eds.)-Mechanics of Composite Materials and Structures-Springer Netherlands (1999).pdf

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

(NATO Science Series 361) J. N. Reddy (auth.), Carlos A. Mota Soares, Cristóvão M. Mota Soares, Manuel J. M. Freitas (eds.)-Mechanics of Composite Materials and Structures-Springer Netherlands (1999).pdf

Hochgeladen von

Copyright:

Verfügbare Formate

Mechanics of Composite Materials and Structures

NATO Science Series

1. Disarmament Technologies Kluwer Academic Publishers

Series E: Mathematical and Physical Sciences - Vol. 361

Carlos A. Mota Soares

Manuel J.M. Freitas

ISBN 978-0-7923-5871-8 ISBN 978-94-011-4489-6 (eBook)

Printed on acid-free paper

Ali Rights Reserved

Mechanics of Composite Materials and Structures

Local Organizing Committee

List of Contributors xix

Theory and Analysis of Laminated Composite Plates

Micromechanics Analysis of Composite Materials

Mechanics of Tbree-Dimmsiooal Textile Structural Composites: Processing

Mechanics ofTbree-Dimmsiooal Textile Structw"al Composites: Analysis

Mechanics ofThree-Dimmsional Textile Structural Composites: Performance Modeling

Testing and Modeling of IneIastic Behavior of Fibrous Composites

Characterizatim of Strain Rate-Depmdmt Behavior in Polymeric Composites

A Treatmmt of Interfacial Cracks in the Presmce of Frictim

Impact and High Rate Testing of Composites: An Overview

Composite for Marine Applications

Composite for Marine Applications

Examples Co Non-Destructive Testing of Glass- and Carbon-Fiber-Reinforced

Towards Standardised InterIaminar Fracture Tests for Unidirectional Fiber-Reinforced

Advantages and Disadvantages of Thermoset and Thermoplastic Matrices for Continuous

Hygrothennal Ageing of Glass Reinforced Composite Pipes

Composite Pipes Based on Thermoplastic Matrices Reinforced by Continuous Fibres

Impact Damage of Composite Structures

Remarks 00 Linear and Nonlinear Variational Design Sensitivity Analysis

Remarks 00 Computing the Energy Release Rate and its Sensitivities

Some Remarks 00 Robust Design

Optimization of Thin-WaIled Structures

Theory and Computation of Multilayer Composites

Theory and Computation of Damage and Failure of Composites

Optimal Design of Composite Structures with Integrated Piezoelectric Laminae

Optimization of Composite Structures Using Response Surface Approximations

Genetic Algorithms for Optimization of Composite Laminates

Identification Techniques in Composite Laminates

Sensitivity Analysis and Inverse Problems for Laminates and Materials

Asymptotic Approach to Optimal Structural Design with Brittle-Fracture ConsIraints

Asymptotic Approach to Optimal Structural Design with Brittle-Fracture Constraints

Some Composite Structures Experiences at NASA Langley Research Center

Prospects os Smart Structures for Future Airaaft

Lisbon, April 1999

Carlos A. Mota Soares

Bamber R.K- BIaekmaa

Carlos A. Mots Soares

Joaquim Infante Barbosa

Maria Helena Barros

Wdoro Pinto Correia

Victor Reis Franco

Jose Miranda Guedes

Cbi Tsieb Liu

Antonio Torres Marques

Maria Augusta Neto

Miguel Matos Neves

Eduardo Ribeiro Pereira

Maria Amelia Ramos Loja

Mariam SAnchez de Ia Sierra

Manuel Gon~alves da Silva

Crist6viio M. Mota Soares

CIuistiaD Van VoaoIe

Bamber R.K. BIadmwI