(Barbero, Ever J.) Finite Element Analysis of Comp (B-Ok - Xyz)

Materials Science/Mechanical Engineering/Civil Engineering
FINITE ELEMENT ANALYSIS Barbero SECOND EDITION
FINITE ELEMENT ANALYSIS OF COMPOSITE MATERIALS USING ANSYS®

OF COMPOSITE MATERIALS
USING ANSYS®
SECOND EDITION
Ever J. Barbero
Designing structures using composite materials poses unique challenges,
especially due to the need for concurrent design of both material
and structure. Professors are faced with two options: textbooks that
FINITE ELEMENT ANALYSIS

teach the theory of advanced mechanics of composites, but lack
computational examples of advanced analysis, and books on finite
element analysis that may or may not demonstrate very limited
applications to composites. But there is a third option that makes the
other two obsolete: Ever J. Barbero’s Finite Element Analysis
of Composite Materials Using ANSYS®, Second Edition.
The Only Finite Element Analysis Book on the Market Using
USING ANSYS
ANSYS® to Analyze Composite Materials
By layering detailed theoretical and conceptual discussions with fully

®
developed examples, this text supplies the missing link between theory
and implementation. In-depth discussions cover all of the major aspects of
advanced analysis, including three-dimensional effects, viscoelasticity, edge
effects, elastic instability, damage, and delamination. This second edition of
the bestseller has been completely revised to incorporate advances in the
state of the art in such areas as modeling of damage in composites. In
addition, all 50+ worked examples have been updated to reflect the newest
version of ANSYS®. Including some use of MATLAB®, these examples
demonstrate how to use the concepts to formulate and execute finite element
analyses and how to interpret the results in engineering terms. In addition,
the source code for each example is available to students for download
online via a companion website featuring a special area reserved for
instructors. A solutions manual is available upon qualifying course adoption.
Ever J. Barbero
Cementing applied computational and analytical experience to a firm
foundation of basic concepts and theory, Finite Element Analysis of
Composite Materials Using ANSYS®, Second Edition offers a
modern, practical, and versatile classroom tool for today’s engineering
classroom.
SECOND
EDITION
K15077
ISBN: 978-1-4665-1689-2
90000
9 781466 516892
K15077_COVER_final.indd 1 11/8/13 9:13 AM

i i
“K15077” — 2013/11/3 — 21:45

i i
i i
i i
USING ANSYS ®
SECOND EDITION
K15077_FM.indd 1 11/8/13 1:10 PM

Composite Materials: Analysis and Design
Series Editor
Ever J. Barbero
PUBLISHED
Finite Element Analysis of Composite Materials Using ANSYS,® Second Edition,
Ever J. Barbero
Smart Composites: Mechanics and Design, Rani El-Hajjar, Valeria La Saponara,
and Anastasia Muliana
Finite Element Analysis of Composite Materials with Abaqus, Ever J. Barbero
FRP Deck and Steel Girder Bridge Systems: Analysis and Design,
Julio F. Davalos, An Chen, Bin Zou, and Pizhong Qiao
Introduction to Composite Materials Design, Second Edition, Ever J. Barbero
Finite Element Analysis of Composite Materials, Ever J. Barbero
K15077_FM.indd 2 11/8/13 1:10 PM

USING ANSYS ®
SECOND EDITION
Ever J. Barbero
K15077_FM.indd 3 11/8/13 1:10 PM

MATLAB® is a trademark of The MathWorks, Inc. and is used with permission. The MathWorks does not warrant the
accuracy of the text or exercises in this book. This book’s use or discussion of MATLAB® software or related products
does not constitute endorsement or sponsorship by The MathWorks of a particular pedagogical approach or particular
use of the MATLAB® software.
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
© 2014 by Taylor & Francis Group, LLC
CRC Press is an imprint of Taylor & Francis Group, an Informa business
No claim to original U.S. Government works

Version Date: 20131219
International Standard Book Number-13: 978-1-4665-1690-8 (eBook - PDF)
This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have been
made to publish reliable data and information, but the author and publisher cannot assume responsibility for the valid-
ity of all materials or the consequences of their use. The authors and publishers have attempted to trace the copyright
holders of all material reproduced in this publication and apologize to copyright holders if permission to publish in this
form has not been obtained. If any copyright material has not been acknowledged please write and let us know so we may
rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or uti-
lized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopy-
ing, microfilming, and recording, or in any information storage or retrieval system, without written permission from the
publishers.
For permission to photocopy or use material electronically from this work, please access www.copyright.com (http://
www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923,
978-750-8400. CCC is a not-for-profit organization that provides licenses and registration for a variety of users. For
organizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for
identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
i i
“K15077” — 2013/11/3 — 21:45

i i
Dedicado a la memoria de mi padre, Ever Francisco Barbero

(1928–2007).
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Contents
Series Preface xiii
Preface xv
Acknowledgments xix
List of Symbols xxi
List of Examples xxix
Errata xxxi
1 Mechanics of Orthotropic Materials 1
1.1 Lamina Coordinate System . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Displacements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.4 Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 Contracted Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5.1 Alternate Contracted Notation . . . . . . . . . . . . . . . . . 5
1.6 Equilibrium and Virtual Work . . . . . . . . . . . . . . . . . . . . . 6
1.7 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.7.1 Traction Boundary Conditions . . . . . . . . . . . . . . . . . 8
1.7.2 Free Surface Boundary Conditions . . . . . . . . . . . . . . . 8
1.8 Continuity Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.8.1 Traction Continuity . . . . . . . . . . . . . . . . . . . . . . . 9
1.8.2 Displacement Continuity . . . . . . . . . . . . . . . . . . . . . 9
1.9 Compatibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.10 Coordinate Transformations . . . . . . . . . . . . . . . . . . . . . . . 10
1.10.1 Stress Transformation . . . . . . . . . . . . . . . . . . . . . . 13
1.10.2 Strain Transformation . . . . . . . . . . . . . . . . . . . . . . 15
1.11 Transformation of Constitutive Equations . . . . . . . . . . . . . . . 16
1.12 3D Constitutive Equations . . . . . . . . . . . . . . . . . . . . . . . . 17
1.12.1 Anisotropic Material . . . . . . . . . . . . . . . . . . . . . . . 18
1.12.2 Monoclinic Material . . . . . . . . . . . . . . . . . . . . . . . 19
vii
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
viii Finite Element Analysis of Composite Materials
1.12.3 Orthotropic Material . . . . . . . . . . . . . . . . . . . . . . . 20

1.12.4 Transversely Isotropic Material . . . . . . . . . . . . . . . . . 22
1.12.5 Isotropic Material . . . . . . . . . . . . . . . . . . . . . . . . 23
1.13 Engineering Constants . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.13.1 Restrictions on Engineering Constants . . . . . . . . . . . . . 28
1.14 From 3D to Plane Stress Equations . . . . . . . . . . . . . . . . . . . 29
1.15 Apparent Laminate Properties . . . . . . . . . . . . . . . . . . . . . 31
Suggested Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2 Introduction to Finite Element Analysis 37

2.1 Basic FEM Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.1.1 Discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.1.2 Element Equations . . . . . . . . . . . . . . . . . . . . . . . . 38
2.1.3 Approximation over an Element . . . . . . . . . . . . . . . . 39
2.1.4 Interpolation Functions . . . . . . . . . . . . . . . . . . . . . 40
2.1.5 Element Equations for a Specific Problem . . . . . . . . . . . 42
2.1.6 Assembly of Element Equations . . . . . . . . . . . . . . . . . 43
2.1.7 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . 44
2.1.8 Solution of the Equations . . . . . . . . . . . . . . . . . . . . 44
2.1.9 Solution Inside the Elements . . . . . . . . . . . . . . . . . . 44
2.1.10 Derived Results . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.2 General Finite Element Procedure . . . . . . . . . . . . . . . . . . . 45
2.3 Solid Modeling, Analysis, and Visualization . . . . . . . . . . . . . . 49
2.3.1 Model Geometry . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.3.2 Material and Section Properties . . . . . . . . . . . . . . . . . 52
2.3.3 Assembly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.3.4 Solution Steps . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.3.5 Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.3.6 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . 54
2.3.7 Meshing and Element Type . . . . . . . . . . . . . . . . . . . 56
2.3.8 Solution Phase . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.3.9 Post-Processing and Visualization . . . . . . . . . . . . . . . 56
3 Elasticity and Strength of Laminates 63

3.1 Kinematics of Shells . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.1.1 First-Order Shear Deformation Theory . . . . . . . . . . . . . 65
3.1.2 Kirchhoff Theory . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.2 Finite Element Analysis of Laminates . . . . . . . . . . . . . . . . . 71
3.2.1 Element Types . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.2.2 Sandwich Shells . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.2.3 Nodes and Curvature . . . . . . . . . . . . . . . . . . . . . . 74
3.2.4 Drilling Rotation . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.2.5 A-B-D-H Input Data for Laminate FEA . . . . . . . . . . . . 74
3.2.6 Equivalent Orthotropic Input for Laminate FEA . . . . . . . 77
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Table of Contents ix
3.2.7 LSS for Multidirectional Laminate FEA . . . . . . . . . . . . 82

3.2.8 FEA of Ply Drop-Off Laminates . . . . . . . . . . . . . . . . 83
3.2.9 FEA of Sandwich Shells . . . . . . . . . . . . . . . . . . . . . 87
3.2.10 Element Coordinate System . . . . . . . . . . . . . . . . . . . 88
3.2.11 Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.3 Failure Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.3.1 2D Failure Criteria . . . . . . . . . . . . . . . . . . . . . . . . 101
3.3.2 3D Failure Criteria . . . . . . . . . . . . . . . . . . . . . . . . 103
4 Buckling 113
4.1 Eigenvalue Buckling Analysis . . . . . . . . . . . . . . . . . . . . . . 113
4.1.1 Imperfection Sensitivity . . . . . . . . . . . . . . . . . . . . . 120
4.1.2 Asymmetric Bifurcation . . . . . . . . . . . . . . . . . . . . . 121
4.1.3 Post-Critical Path . . . . . . . . . . . . . . . . . . . . . . . . 121
4.2 Continuation Methods . . . . . . . . . . . . . . . . . . . . . . . . . . 126
5 Free Edge Stresses 133

5.1 Poisson’s Mismatch . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
5.1.1 Interlaminar Force . . . . . . . . . . . . . . . . . . . . . . . . 134
5.1.2 Interlaminar Moment . . . . . . . . . . . . . . . . . . . . . . 135
5.2 Coefficient of Mutual Influence . . . . . . . . . . . . . . . . . . . . . 141
5.2.1 Interlaminar Stress due to Mutual Influence . . . . . . . . . . 143
6 Computational Micromechanics 151

6.1 Analytical Homogenization . . . . . . . . . . . . . . . . . . . . . . . 152
6.1.1 Reuss Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
6.1.2 Voigt Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
6.1.3 Periodic Microstructure Model . . . . . . . . . . . . . . . . . 153
6.1.4 Transversely Isotropic Averaging . . . . . . . . . . . . . . . . 154
6.2 Numerical Homogenization . . . . . . . . . . . . . . . . . . . . . . . 157
6.3 Global-Local Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 172
6.4 Laminated RVE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
7 Viscoelasticity 179
7.1 Viscoelastic Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
7.1.1 Maxwell Model . . . . . . . . . . . . . . . . . . . . . . . . . . 181
7.1.2 Kelvin Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
7.1.3 Standard Linear Solid . . . . . . . . . . . . . . . . . . . . . . 183
7.1.4 Maxwell-Kelvin Model . . . . . . . . . . . . . . . . . . . . . . 183
7.1.5 Power Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
7.1.6 Prony Series . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
x Finite Element Analysis of Composite Materials
7.1.7 Standard Nonlinear Solid . . . . . . . . . . . . . . . . . . . . 185

7.1.8 Nonlinear Power Law . . . . . . . . . . . . . . . . . . . . . . 186
7.2 Boltzmann Superposition . . . . . . . . . . . . . . . . . . . . . . . . 187
7.2.1 Linear Viscoelastic Material . . . . . . . . . . . . . . . . . . . 187
7.2.2 Unaging Viscoelastic Material . . . . . . . . . . . . . . . . . . 189
7.3 Correspondence Principle . . . . . . . . . . . . . . . . . . . . . . . . 190
7.4 Frequency Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
7.5 Spectrum Representation . . . . . . . . . . . . . . . . . . . . . . . . 192
7.6 Micromechanics of Viscoelastic Composites . . . . . . . . . . . . . . 192
7.6.1 One-Dimensional Case . . . . . . . . . . . . . . . . . . . . . . 192
7.6.2 Three-Dimensional Case . . . . . . . . . . . . . . . . . . . . . 193
7.7 Macromechanics of Viscoelastic Composites . . . . . . . . . . . . . . 197
7.7.1 Balanced Symmetric Laminates . . . . . . . . . . . . . . . . . 197
7.7.2 General Laminates . . . . . . . . . . . . . . . . . . . . . . . . 197
7.8 FEA of Viscoelastic Composites . . . . . . . . . . . . . . . . . . . . . 198
8 Continuum Damage Mechanics 207

8.1 One-Dimensional Damage Mechanics . . . . . . . . . . . . . . . . . . 208
8.1.1 Damage Variable . . . . . . . . . . . . . . . . . . . . . . . . . 208
8.1.2 Damage Threshold and Activation Function . . . . . . . . . . 210
8.1.3 Kinetic Equation . . . . . . . . . . . . . . . . . . . . . . . . . 211
8.1.4 Statistical Interpretation of the Kinetic Equation . . . . . . . 212
8.1.5 One-Dimensional Random-Strength Model . . . . . . . . . . 213
8.1.6 Fiber Direction, Tension Damage . . . . . . . . . . . . . . . . 218
8.1.7 Fiber Direction, Compression Damage . . . . . . . . . . . . . 222
8.2 Multidimensional Damage and Effective Spaces . . . . . . . . . . . . 226
8.3 Thermodynamics Formulation . . . . . . . . . . . . . . . . . . . . . . 228
8.3.1 First Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
8.3.2 Second Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
8.4 Kinetic Law in Three-Dimensional Space . . . . . . . . . . . . . . . . 235
8.4.1 Return-Mapping Algorithm . . . . . . . . . . . . . . . . . . . 238
8.5 Damage and Plasticity . . . . . . . . . . . . . . . . . . . . . . . . . . 244
9 Discrete Damage Mechanics 249

9.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
9.2 Approximations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
9.3 Lamina Constitutive Equation . . . . . . . . . . . . . . . . . . . . . 256
9.4 Displacement Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
9.4.1 Boundary Conditions for ∆T = 0 . . . . . . . . . . . . . . . . 258
9.4.2 Boundary Conditions for ∆T 6= 0 . . . . . . . . . . . . . . . . 259
9.5 Degraded Laminate Stiffness and CTE . . . . . . . . . . . . . . . . . 260
9.6 Degraded Lamina Stiffness . . . . . . . . . . . . . . . . . . . . . . . . 261
9.7 Fracture Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Table of Contents xi
9.8 Solution Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263

9.8.1 Lamina Iterations . . . . . . . . . . . . . . . . . . . . . . . . 263
9.8.2 Laminate Iterations . . . . . . . . . . . . . . . . . . . . . . . 264
10 Delaminations 273
10.1 Cohesive Zone Method . . . . . . . . . . . . . . . . . . . . . . . . . . 276
10.1.1 Single Mode Cohesive Model . . . . . . . . . . . . . . . . . . 278
10.1.2 Mixed Mode Cohesive Model . . . . . . . . . . . . . . . . . . 281
10.2 Virtual Crack Closure Technique . . . . . . . . . . . . . . . . . . . . 290
A Tensor Algebra 297

A.1 Principal Directions of Stress and Strain . . . . . . . . . . . . . . . . 297
A.2 Tensor Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
A.3 Matrix Representation of a Tensor . . . . . . . . . . . . . . . . . . . 298
A.4 Double Contraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
A.5 Tensor Inversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
A.6 Tensor Differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . 300
A.6.1 Derivative of a Tensor . . . . . . . . . . . . . . . . . . . . . . 300
A.6.2 Derivative of the Inverse of a Tensor . . . . . . . . . . . . . . 301
B Second-Order Diagonal Damage Models 303

B.1 Effective and Damaged Spaces . . . . . . . . . . . . . . . . . . . . . 303
B.2 Thermodynamic Force Y . . . . . . . . . . . . . . . . . . . . . . . . 304
B.3 Damage Surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
B.4 Unrecoverable-Strain Surface . . . . . . . . . . . . . . . . . . . . . . 307
C Software Used 309

C.1 ANSYS Mechanical APDL . . . . . . . . . . . . . . . . . . . . . . . . 309
C.1.1 ANSYS USERMAT, Compilation and Execution . . . . . . . 311
C.2 BMI3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
C.2.1 Stand-Alone BMI3 . . . . . . . . . . . . . . . . . . . . . . . . 314
C.2.2 BMI3 within ANSYS . . . . . . . . . . . . . . . . . . . . . . . 314
References 317
Index 329
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Series Preface
Half a century after their commercial introduction, composite materials are of

widespread use in many industries. Applications such as aerospace, windmill blades,
highway bridge retrofit, and many more require designs that assure safe and reli-
able operation for twenty years or more. Using composite materials, virtually any
property, such as stiffness, strength, thermal conductivity, and fire resistance, can
be tailored to the users needs by selecting the constituent material, their proportion
and geometrical arrangement, and so on. In other words, the engineer is able to
design the material concurrently with the structure. Also, modes of failure are much
more complex in composites than in classical materials. Such demands for perfor-
mance, safety, and reliability require that engineers consider a variety of phenomena
during the design. Therefore, the aim of the Composite Materials: Analysis and
Design book series is to bring to the design engineer a collection of works written
by experts on every aspect of composite materials that is relevant to their design.
Variety and sophistication of material systems and processing techniques has
grown exponentially in response to an ever-increasing number and type of applica-
tions. Given the variety of composite materials available as well as their continuous
change and improvement, understanding of composite materials is by no means
complete. Therefore, this book series serves not only the practicing engineer but
also the researcher and student who are looking to advance the state-of-the-art
in understanding material and structural response and developing new engineering
tools for modeling and predicting such responses.
Thus, the series is focused on bringing to the public existing and developing
knowledge about the material-property relationships, processing-property relation-
ships, and structural response of composite materials and structures. The series
scope includes analytical, experimental, and numerical methods that have a clear
impact on the design of composite structures.
Ever Barbero, book series editor

West Virginia University, Morgantown, WV
xiii
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Preface
Finite Element Analysis of Composite Materials Using ANSYS R

deals with the anal-
ysis of structures made of composite materials, also called composites. The analysis
of composites treated in this textbook includes the analysis of the material itself,
at the microlevel, and the analysis of structures made of composite materials. This
textbook evolved from the class notes of MAE 646 Advanced Mechanics of Com-
posite Materials that I teach as a graduate course at West Virginia University.
Although this is a textbook on advanced mechanics of composite materials, the use
of the finite element method is essential for the solution of the complex boundary
value problems encountered in the advanced analysis of composites, and thus the
title of the book.
There are a number of good textbooks on advanced mechanics of composite ma-
terials, but none carries the theory to a practical level by actually solving problems,
as it is done in this textbook. Some books devoted exclusively to finite element
analysis include some examples about modeling composites but fall quite short of
dealing with the actual analysis and design issues of composite materials and com-
posite structures. This textbook includes an explanation of the concepts involved in
the detailed analysis of composites, a sound explanation of the mechanics needed to
translate those concepts into a mathematical representation of the physical reality,
and a detailed explanation of the solution of the resulting boundary value problems
by using commercial Finite Element Analysis software such as ANSYS Mechanical
APDL. Furthermore, this textbook includes more than fifty fully developed exam-
ples interspersed with the theory, as well as more than seventy-five exercises at the
end of chapters, and more than fifty separate pieces of ANSYS APDL code used
to explain in detail the solution of example problems. The reader will be able to
reproduce the examples and complete the exercises. When a finite element analysis
is called for, the reader will be able to do it with commercially or otherwise available
software. A Web site is set up with links to download the necessary software unless
it is easily available from Finite Element Analysis software vendors. ANSYS and
MATLAB R
code is explained in the examples, and the code can be downloaded
from the Web site as well. Furthermore, the reader will be able to extend the ca-
pabilities of ANSYS by user material subroutines, as demonstrated in the examples
included in this textbook.
Chapters 1 through 7 can be covered in a one-semester graduate course. Chap-
ter 2 contains a brief introduction intended for those readers who have not had
a formal course or prior knowledge about the finite element method. Chapter 4
xv
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
xvi Finite Element Analysis of Composite Materials
(Buckling) is not referenced in the remainder of the textbook and thus it could be
omitted in favor of more exhaustive coverage of content in later chapters. Chapters
7 (Viscoelasticity) and 8 (Continuum Damage Mechanics) are placed consecutively
to emphasize hereditary phenomena. However, Chapter 7 can be skipped if more
emphasis on damage and/or delaminations is desired in a one-semester course. The
foundations for the analysis of damage is laid out in Chapter 8 followed by a chap-
ter on Discrete Damage Mechanics (Chapter 9). Either or both chapters could be
omitted for the sake of time, for example if the instructor desires to cover Chapter
10 (Delaminations) as part of a one-semester course.
The inductive method is applied as much as possible in this textbook. That is,
topics are introduced with examples of increasing complexity, until sufficient phys-
ical understanding is reached to introduce the general theory without difficulty.
This method will sometimes require that, at earlier stages of the presentation, cer-
tain facts, models, and relationships be accepted as fact, until they are completely
proven later on. For example, in Chapter 7, viscoelastic models are introduced early
to aid the reader in gaining an appreciation for the response of viscoelastic mate-
rials. This is done simultaneously with a cursory introduction to the superposition
principle and the Laplace transform, which are formally introduced only later in
the chapter. For those readers accustomed to the deductive method, this may seem
odd, but many years of teaching have convinced me that students acquire and retain
knowledge more efficiently in this way.
It is assumed that the reader is familiar with basic mechanics of composites as
covered in introductory level textbooks such as my previous textbook Introduction
to Composite Material Design–Second Edition. Furthermore, it is assumed that
the reader masters a body of knowledge that is commonly acquired as part of a
bachelor of science degree in any of the following disciplines: Aerospace, Mechanical,
Civil, or similar. References to books and to other sections in this textbook, as
well as footnotes, are used to assist the reader in refreshing those concepts and to
clarify the notation used. Prior knowledge of continuum mechanics, tensor analysis,
and the finite element method would enhance the learning experience but are not
necessary for studying with this textbook. The finite element method is used as
a tool to solve practical problems. For the most part, ANSYS is used throughout
the book. Computing programming using Fortran and MATLAB is limited to
programming material models and post-processing algorithms. Basic knowledge of
these programming languages is useful but not essential.
Only three software packages are used throughout the book. ANSYS is needed
for the finite element solution of numerous examples and suggested problems. MAT-
LAB is needed for both symbolic and numerical solution of examples and suggested
problems. Additionally, BMI3 c
, which is available free of charge on the book’s Web
site, is used in Chapter 4. Several other programs such as Abaqus TM , LS-DYNA R
,

R
MSC-MARC , and SolidWorks TM are cited, but not used in the examples. All the
APDL code for the examples in this textbook is available on the book’s Web site
http://barbero.cadec-online.com/feacm-ansys/.
Composite materials are now ubiquitous in the marketplace, including extensive
applications in aerospace, automotive, civil infrastructure, sporting goods, and so
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Preface xvii
on. Their design is especially challenging because, unlike conventional materials

such as metals, the composite material itself is designed concurrently with the com-
posite structure. Preliminary design of composites is based on the assumption of a
state of plane stress in the laminate. Furthermore, rough approximations are made
about the geometry of the part, as well as the loading and support conditions. In
this way, relatively simple analysis methods exist and computations can be carried
out simply using algebra. However, preliminary analysis methods have a number of
shortcomings that are remedied with advanced mechanics and finite element anal-
ysis, as explained in this textbook. Recent advances in commercial finite element
analysis packages, with user friendly pre- and post-processing, as well as power-
ful user-programmable features, have made detailed analysis of composites quite
accessible to the designer. This textbook bridges the gap between powerful finite
element tools and practical problems in structural analysis of composites. I expect
that many graduate students, practicing engineers, and instructors will find this to
be a useful and practical textbook on finite element analysis of composite materials
based on sound understanding of advanced mechanics of composite materials.
Ever J. Barbero, 2013
Abaqus TM and SolidWorksTM are registered trademarks of Dassault Systèmes. Abaqus is de-
veloped by SIMULIA, the Dassault Systèmes brand for Realistic Simulation www.simulia.com.
ANSYS R
is a registered trademark of ANSYS Inc. www.ansys.com
LS-DYNA R
is a registered trademark of Livermore Software Technology Corporation www.
lstc.com.
MATLAB R
is a registered trademark of The MathWorks, Inc. For product information, please
contact: The MathWorks, Inc. 3 Apple Hill Drive, Natick, MA 01760-2098 USA Tel: 508-647-7000
Fax: 508-647-7001. E-mail: [email protected] Web: www.mathworks.com
MSC-MARC R
is a registered trademark of MSC Software. www.mscsoftware.com
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Acknowledgments
I wish to thank Raimondo Luciano and Elio Sacco for their contributions to Chap-
ter 6 and to Tom Damiani, Joan Andreu Mayugo Majo, and Xavier Martinez, who
taught the course in 2004, 2006, and 2009, making many corrections and additions
to the course notes on which this textbook is based. Also, I wish to recognize Adi
Adumitroaie, who taught invited lectures on progressive damage analysis and co-
hesive zone models, thus providing the basis for Chapter 10. I am also grateful
to those who reviewed parts of the manuscript including Enrique Barbero, Grama
Bhashyam, Guillermo Creus, Fabrizio Greco, Luis Godoy, Paolo Lonetti, Severino
Marques, Pizhong Qiao, Timothy Norman, Sonia Sanchez, and Eduardo Sosa. Fur-
thermore, recognition is due to those who helped me compile the solutions manual,
including Hermann Alcazar, John Sandro Rivas, and Rajiv Dastane. Also, I wish
to thank Ruth Roman for helping me update the examples to Release 14 of ANSYS
and to my colleagues and students for their valuable suggestions and contributions
to this textbook. Finally, my gratitude goes to my wife, Ana Maria, and to my
children, Margaret and Daniel, who gave up many opportunities to bond with their
dad so that I might write this book.
xix
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
List of Symbols
Symbols Related to Mechanics of Orthotropic Materials

Strain tensor
εij Strain components in tensor notation
α Strain components in contracted notation
eα Elastic strain
pα Plastic strain
λ Lame constant
ν Poisson’s ratio
ν12 In-plane Poisson’s ratio
ν23 , ν13 Interlaminar Poisson’s ratios
νxy Apparent laminate Poisson’s ratio x-y
σ Stress tensor
σij Stress components in tensor notation
σα Stress components in contracted notation
[a] Transformation matrix for vectors
ei Unit vector components in global coordinates
e0i Unit vector components in materials coordinates
fi , fij Tsai-Wu coefficients
l, m, n Direction cosines
u
e(εij ) Strain energy per unit volume
ui Displacement vector components
xi Global directions or axes
x0i Materials directions or axes
C Stiffness tensor
Cijkl Stiffness in index notation
Cα,β Stiffness in contracted notation
E Young’s modulus
E1 Longitudinal modulus
E2 Transverse modulus
E2 Transverse-thickness modulus
Ex Apparent laminate modulus in the global x-direction
G=µ Shear modulus
G12 In-plane shear modulus
G23 , G13 Interlaminar shear moduli
Gxy Apparent laminate shear modulus x-y
xxi
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
xxii Finite Element Analysis of Composite Materials
Iij Second-order identity tensor

Iijkl Fourth-order identity tensor
K Bulk modulus
Q0ij Lamina stiffness components in lamina coordinates
[R] Reuter matrix
S Compliance tensor
Sijkl Compliance in index notation
Sα,β Compliance in contracted notation
[T ] Coordinate transformation matrix for stress
[T ] Coordinate transformation matrix for strain
Symbols Related to Finite Element Analysis
∂ Strain-displacement equations in matrix form

Six-element array of strain components
θx , θy , θz Rotation angles following the right-hand rule (Figure 2.8)
σ Six-element array of stress components
φx , φy Rotation angles used in plate and shell theory
a Nodal displacement array
uej Unknown parameters in the discretization
B Strain-displacement matrix
C Stiffness matrix
K Assembled global stiffness matrix
Ke Element stiffness matrix
N Interpolation function array
Nje Interpolation functions in the discretization
Pe Element force array
P Assembled global force array
Symbols Related to Elasticity and Strength of Laminates

0
γxy In-plane shear strain
γ4u Ultimate interlaminar shear strain in the 2-3 plane
γ5u Ultimate interlaminar shear strain in the 1-3 plane
γ6u Ultimate in-plane shear strain
0x , 0y In-plane strains
1t Ultimate longitudinal tensile strain
2t Ultimate transverse tensile strain
3t Ultimate transverse-thickness tensile strain
1c Ultimate longitudinal compressive strain
2c Ultimate transverse compressive strain
3c Ultimate transverse-thickness compressive strain
κx , κy Bending curvatures
κxy Twisting curvature
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
List of Symbols xxiii
φx , φy Rotations of the middle surface of the shell (Figure 2.8)

c4 , c5 , c6 Tsai-Wu coupling coefficients
tk Lamina thickness
u0 , v0 , w0 Displacements of the middle surface of the shell
z Distance from the middle surface of the shell
Aij Components of the extensional stiffness matrix [A]
Bij Components of the bending-extension coupling matrix [B]
Dij Components of the bending stiffness matrix [D]
[E0 ] Extensional stiffness matrix [A], in ANSYS notation
[E1 ] Bending-extension matrix [B], in ANSYS notation
[E2 ] Bending stiffness matrix [D], in ANSYS notation
F1t Longitudinal tensile strength
F2t Transverse tensile strength
F3t Transverse-thickness tensile strength
F1c Longitudinal compressive strength
F2c Transverse compressive strength
F3c Transverse-thickness compressive strength
F4 Interlaminar shear strength in the 2-3 plane
F5 Interlaminar shear strength in the 1-3 plane
F6 In-plane shear strength
Hij Components of the interlaminar shear matrix [H]
IF Failure index
Mx , My , Mx y Moments per unit length (Figure 3.3)
M
cn Applied bending moment per unit length
Nx , Ny , Nx y In-plane forces per unit length (Figure 3.3)
Nbn Applied in-plane force per unit length, normal to the edge
Nbns
Applied in-plane shear force per unit length, tangential
Qij k Lamina stiffness components in laminate coordinates, layer k
Vx , Vy Shear forces per unit length (Figure 3.3)
Symbols Related to Buckling
λ, λi Eigenvalues
s Perturbation parameter
Λ Load multiplier
Λ(cr) Bifurcation multiplier or critical load multiplier
Λ(1) Slope of the post-critical path
Λ(2) Curvature of the post-critical path
v Eigenvectors (buckling modes)
[K] Stiffness matrix
[Ks ] Stress stiffness matrix
PCR Critical load
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
xxiv Finite Element Analysis of Composite Materials
Symbols Related to Free Edge Stresses
ηxy,x , ηxy,y Coefficients of mutual influence

ηx,xy , ηy,xy Alternate coefficients of mutual influence
Fyz Interlaminar shear force y-z
Fxz Interlaminar shear force x-z
Mz Interlaminar moment
Symbols Related to Micromechanics
α Average engineering strain components

εij Average tensor strain components
0α , ε0ij Far-field applied strain components
σα Average stress components
Ai Strain concentration tensor, i-th phase, contracted notation
2a1 , 2a2 , 2a3 Dimensions of the representative volume element (RVE)
Aijkl Components of the strain concentration tensor
Bi Stress concentration tensor, i-th phase, contracted notation
Bijkl Components of the stress concentration tensor
I 6 × 6 identity matrix
Pijkl Eshelby tensor
Vf Fiber volume fraction
Vm Matrix volume fraction
Symbols Related to Viscoelasticity
ε̇ Stress rate
η Viscosity
θ Age or aging time
σ̇ Stress rate
τ Time constant of the material or system
Γ Gamma function
s Laplace variable
t Time
Cα,β (t) Stiffness tensor in the time domain
Cα,β (s) Stiffness tensor in the Laplace domain
Cbα,β (s) Stiffness tensor in the Carson domain
D(t) Compliance
D0 , (Di )0 Initial compliance values
Dc (t) Creep component of the total compliance D(t)
D0 , D00 Storage and loss compliances
E0 , (Ei )0 Initial moduli
E∞ Equilibrium modulus
E, E0 , E1 , E2 Parameters in the viscoelastic models (Figure 7.1)
E(t) Relaxation
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
List of Symbols xxv
E 0 , E 00 Storage and loss moduli

F [] Fourier transform
(Gij )0 Initial shear moduli
H(t − t0 ) Heaviside step function
H(θ) Relaxation spectrum
L[] Laplace transform
L[]−1 Inverse Laplace transform
Symbols Related to Damage

α Laminate coefficient of thermal expansion (CTE)
α(k) CTE of lamina k
αcr Critical misalignment angle at longitudinal compression failure
ασ Standard deviation of fiber misalignment
γ(δ) Damage hardening function
γ0 Damage threshold
δij Kronecker delta
δ Damage hardening variable
ε Effective strain
ε Undamaged strain
εp Plastic strain
γ̇ Heat dissipation rate per unit volume
γ̇s Internal entropy production rate
λ Crack density
λlim Saturation crack density
λ̇, λ̇d Damage multiplier
λ̇p Yield multiplier
ρ Density
σ Effective stress
σ Undamaged stress
τ13 , τ23 Intralaminar shear stress components
ϕ, ϕ∗ Strain energy density, and complementary SED
χ Gibbs energy density
ψ Helmholtz free energy density
∆T Change in temperature
Ω = Ωij Integrity tensor
2a0 Representative crack size
di Eigenvalues of the damage tensor
fd Damage flow surface
fp Yield flow surface
f (x), F (x) Probability density, and its cumulative probability
g Damage activation function
gd Damage surface
gp Yield surface
h Laminate thickness
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
xxvi Finite Element Analysis of Composite Materials
hk Thickness of lamina k
m Weibull modulus
p Yield hardening variable
pb Thickness average of quantity p
pe Virgin value of quantity p
p Volume average of quantity p
q Hear flow vector per unit area
r Radiation heat per unit mass
s Specific entropy
u(εij ) Internal energy density
A Crack area
[A] Laminate in-plane stiffness matrix
Aijkl Tension-compression damage constitutive tensor
Bijkl Shear damage constitutive tensor
Ba Dimensionless number (8.57)
C α,β Stiffness matrix in the undamaged configuration
Ced Tangent stiffness tensor
Dij Damage tensor
D1tcr Critical damage at longitudinal tensile failure
D1ccr Critical damage at longitudinal compression failure
D2tcr Critical damage at transverse tensile failure
D2 , D6 Damage variables
E(D) Effective modulus
E Undamaged (virgin) modulus
Gc = 2γc Surface energy
GIc , GIIc Critical energy release rate in modes I and II
Jijkl Normal damage constitutive tensor
Mijkl Damage effect tensor
N Number of laminas in the laminate
{N } Membrane stress resultant array
Q Degraded 3x3 stiffness matrix of the laminate
R(p) Yield hardening function
R0 Yield threshold
S Entropy or Laminate complinace matrix, depending on context
T Temperature
U Strain energy
V Volume of the RVE
Yij Thermodynamic force tensor
Symbols Related to Delaminations

α Mixed mode crack propagation exponent
βδ , βG Mixed mode ratios
δ Separation of the interface in the cohesive zone model (CZM)
δm Mixed mode separation
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
List of Symbols xxvii
δm0 Mixed mode separation at damage onset

δm0 Mixed mode separation at fracture
σ0 CZM critical separation at damage onset
` Delamination length for 2D delaminations
σ0 CZM strength of the interface
ψxi , ψyi Rotation of normals to the middle surface of the plate
Ω Volume of the body
ΩD Delaminated region
Πe Potential energy, elastic
Πr Potential energy, total
Γ̇ Dissipation rate
Λ Interface strain energy density per unit area
∂Ω Boundary of the body
d One-dimensional damage state variable
kxy , kz Displacement continuity parameters
[Ai ], [Bi ], [Di ] Laminate stiffness submatrices
DI , DII , DIII Damage variables for modes I, II, and III of CZM
G(`) Energy release rate (ERR), total, in 2D
G Energy release rate (ERR), total, in 3D
GI , GII , GIII Energy release rate (ERR) of modes I, II, and III
Gc Critical energy release rate (ERR), total, in 3D
GcI Critical energy release rate mode I
[Hi ] Laminate interlaminar shear stiffness matrix
K Penalty stiffness
K̃ Virgin penalty stiffness
KI , KII , KIII Stress intensity factors (SIF) of modes I, II, and III
Ni , Mi , Ti Stress resultants
U Internal energy
W Work done by the body on its surroundings
Wclosure Crack closure work
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
List of Examples
Example 1.1, 7 Example 3.7, 87 Example 6.3, 170

xxix
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Errata
For the most current errata, go to

http://forum.cadec-online.com/viewforum.php?f=6
xxxi
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Chapter 1
Mechanics of Orthotropic
Materials
This chapter provides the foundation for the rest of the book. Basic concepts of
mechanics, tailored for composite materials, are presented, including coordinate
transformations, constitutive equations, and so on. Continuum mechanics is used
to describe deformation and stress in an orthotropic material. The basic equations
are reviewed in Sections 1.2 to 1.9. Tensor operations are reviewed in Section
1.10 because they are used in the rest of the chapter. Coordinate transformations
are required to express quantities such as stress, strain, and stiffness in lamina
coordinates, in laminate coordinates, and so on. They are reviewed in Sections 1.10
to 1.11. This chapter is heavily referenced in the rest of the book, and thus readers
who are already versed in continuum mechanics may choose to come back to review
this material as needed.
1.1 Lamina Coordinate System

A single lamina of fiber reinforced composite behaves as an orthotropic material.
That is, the material has three mutually perpendicular planes of symmetry. The
intersection of these three planes defines three axes that coincide with the fiber
direction (x01 ), the thickness coordinate (x03 ), and a third direction x02 = x03 × x01
perpendicular to the other two [1].1
1.2 Displacements
Under the action of forces, every point in a body may translate and rotate as a
rigid body as well as deform to occupy a new region. The displacements ui of any
point P in the body (Figure 1.1) are defined in terms of the three components of
the vector ui (in a rectangular Cartesian coordinate system) as ui = (u1, u2, u3 ). An
1
× denotes vector cross product.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
2 Finite Element Analysis of Composite Materials
Figure 1.1: Notation for displacement components.
alternate notation for displacements is ui = (u, v, w). Displacement is a vector or

first-order tensor quantity
u = ui = (u1, u2, u3 ) ; i = 1...3 (1.1)

where boldface (e.g., u) indicates a tensor written in tensor notation, in this case
a vector (or first-order tensor). In this book, all tensors are boldfaced (e.g., σ),
but their components are not (e.g., σij ). The order of the tensor (i.e., first, second,
fourth, etc.) must be inferred from context, or as in (1.1), by looking at the number
of subscripts of the same entity written in index notation (e.g., ui ).
1.3 Strain
For geometric nonlinear analysis, the components of the Lagrangian strain tensor
are [2]
1
Lij = (ui,j + uj,i + ur,i ur,j ) (1.2)
2
where
∂ui
ui,j = (1.3)
∂xj
If the gradients of the displacements are so small that products of partial deriva-
tives of ui are negligible compared with linear (first-order) derivative terms, then
the (infinitesimal) strain tensor εij is given by [2]
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Mechanics of Orthotropic Materials 3
Figure 1.2: Normal strain.
1
ε = εij = (ui,j + uj,i ) (1.4)
2
Again, boldface indicates a tensor, the order of which is implied from the context.
For example ε is a one-dimensional strain and ε is the second-order tensor of strain.
Index notation (e.g., = εij ) is used most of the time and the tensor character of
variables (scalar, vector, second order, and so on) is easily understood from context.
From the definition (1.4), strain is a second-order, symmetric tensor (i.e., εij =
εji ). In expanded form the strains are defined by

∂u1 ∂u1 ∂u2
ε11 = = 1 ; 2ε12 = 2ε21 = + = γ6 = 6
∂x1 ∂x2 ∂x1

∂u2 ∂u1 ∂u3
ε22 = = 2 ; 2ε13 = 2ε31 = + = γ5 = 5
∂x2 ∂x3 ∂x1

∂u3 ∂u2 ∂u3
ε33 = = 3 ; 2ε23 = 2ε32 = + = γ4 = 4 (1.5)
∂x3 ∂x3 ∂x2
where α with α = 1..6 are defined in Section 1.5. The normal components of strain
( i = j) represent the change in length per unit length (Figure 1.2). The shear
components of strain (i 6= j) represent one-half the change in an original right angle
(Figure 1.3). The engineering shear strain γα = 2εij , for i 6= j is often used instead
of the tensor shear strain because the shear modulus G is defined by τ = Gγ in
mechanics of materials [3]. The strain tensor, being a second order tensor, can be
displayed as a matrix
   
ε11 ε12 ε13 1 6 /2 5 /2
[ε] =  ε12 ε22 ε23  =  6 /2 2 4 /2  (1.6)
ε13 ε23 ε33 5 /2 4 /2 3
where [ ] is used to denote matrices.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 1.3: Engineering shear strain.
1.4 Stress
The stress vector associated to a plane passing through a point is the force per
unit area acting on the plane passing through the point. A second-order tensor,
called the stress tensor, completely describes the state of stress at a point. The
stress tensor can be expressed in terms of the components acting on three mutually
perpendicular planes aligned with the orthogonal coordinate directions as indicated
in Figure 1.4. The tensor notation for stress is σij with (i, j = 1, 2, 3), where the first
subscript corresponds to the direction of the normal to the plane of interest and the
second subscript corresponds to the direction of the stress. Tensile normal stresses
(i = j) are defined to be positive when the normal to the plane and the stress
component directions are either both positive or both negative. All components of
stress depicted in Figure 1.4 have a positive sense. Force and moment equilibrium
of the element in Figure 1.4 requires that the stress tensor be symmetric (i.e.,
σij = σji ) [3]. The stress tensor, being a second order tensor, can be displayed as a
matrix
   
σ11 σ12 σ13 σ1 σ6 σ5
[σ] =  σ12 σ22 σ23  =  σ6 σ2 σ4  (1.7)
σ13 σ23 σ33 σ5 σ4 σ3
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 1.4: Stress components.
1.5 Contracted Notation

Since the stress is symmetric, it can be written in Voigt contracted notation as
σα = σij = σji (1.8)
with the contraction rule defined as follows
α=i if i=j
α=9−i−j if i 6= j (1.9)
resulting in the contracted version of stress components shown in (1.7). The same
applies to the strain tensor, resulting in the contracted version of strain shown in
(1.6). Note that the six components of stress σα with α = 1 . . . 6 can be arranged
into a column array, denoted by curly brackets { } as in (1.10), but {σ} is not
a vector, but just a convenient way to arrange the six unique components of a
symmetric second-order tensor.
1.5.1 Alternate Contracted Notation

Some finite element analysis (FEA) packages use different contracted notations, as
shown in Table 1.1. For example, to transform stresses or strains from standard
notation to Abaqus TM notation, a transformation matrix can be used as follows
{σA } = [T ]{σ} (1.10)

where the subscript ()A denotes a quantity in Abaqus notation. Also note that { }
denotes a column array, in this case of six elements, and [ ] denotes a matrix, in
this case the 6×6 rotation matrix given by
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Table 1.1: Contracted notation convention used by various FEA software packages
Standard LS-DYNA and
convention Abaqus/Standard Abaqus/Explicit ANSYS/Mechanical
11 −→ 1 11 −→ 1 11 −→ 1 11 −→ 1
22 −→ 2 22 −→ 2 22 −→ 2 22 −→ 2
33 −→ 3 33 −→ 3 33 −→ 3 33 −→ 3
23 −→ 4 12 −→ 4 12 −→ 4 12 −→ 4
13 −→ 5 13 −→ 5 23 −→ 5 23 −→ 5
12 −→ 6 23 −→ 6 13 −→ 6 13 −→ 6
 
1 0 0 0 0 0

 0 1 0 0 0 0 

 0 0 1 0 0 0 
[T ] =   (1.11)

 0 0 0 0 0 1 

 0 0 0 0 1 0 
0 0 0 1 0 0
The stiffness matrix transforms as follows
[CA ] = [T ]T [C][T ] (1.12)
For LS-DYNA
R
and ANSYS
R
, the transformation matrix is
 
1 0 0 0 0 0

 0 1 0 0 0 0 

 0 0 1 0 0 0 
[T ] =   (1.13)

 0 0 0 0 0 1 

 0 0 0 1 0 0 
0 0 0 0 1 0
1.6 Equilibrium and Virtual Work

The three equations of equilibrium at every point in a body are written in tensor
notation as
σij,j + fi = 0 (1.14)
∂
where fi is the body force per unit volume and ( ),j = . When body forces are
∂xj
negligible, the expanded form of the equilibrium equations, written in the laminate
coordinate system x-y-z, is
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
∂σxx ∂σxy ∂σxz

+ + =0
∂x ∂y ∂z
∂σxy ∂σyy ∂σyz
+ + =0
∂x ∂y ∂z
∂σxz ∂σyz ∂σzz
+ + =0 (1.15)
∂x ∂y ∂z
The principle of virtual work (PVW) provides an alternative to the equations

of equilibrium [4]. Since the PVW is an integral expression, it is more convenient
than (1.14) for finite element formulation. The PVW reads
Z Z Z
σij δij dV − ti δui dS − fi δui dV = 0 (1.16)
V S V
where ti are the surface tractions per unit area acting on the surface S. The negative
sign means that work is done by external forces (ti , fi ) on the body. The forces and
the displacements follow the same sign convention; that is, a component is positive
when it points in the positive direction of the respective axis. The first term in (1.16)
is the virtual work performed by the internal stresses and it is positive following the
same sign convention.
Example 1.1 Find the displacement function u(x) for a slender rod of cross-sectional area
A, length L, modulus E and density ρ, hanging from the top end and subjected to its own
weight. Use a coordinate x pointing downward with the origin at the top end.
Solution to Example 1.1 We assume a quadratic displacement function
u(x) = C0 + C1 x + C2 x2
Using the boundary condition (BC) at the top yields C0 = 0. The PVW (1.16) simplifies
because the only nonzero strain is x and there is no surface traction. Using Hooke’s law
Z L Z L
Ex δx Adx − ρgδuAdx = 0
0 0
From the assumed displacement
δu = xδC1 + x2 δC2
du
x = = C1 + 2xC2
dx
δx = δC1 + 2xδC2
Substituting
Z L Z L
EA (C1 + 2xC2 )(δC1 + 2xδC2 )dx − ρgA (xδC1 + x2 δC2 )dx = 0
o 0
Integrating and collecting terms in δC1 and δC2 separately
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
ρgL2 4 ρgL3
(EC2 L2 + EC1 L − )δC1 + ( EC2 L3 + EC1 L2 − )δC2 = 0
2 3 3
Since δC1 and δC2 have arbitrary (virtual) values, two equations in two unknowns are
obtained, one inside each parenthesis. Solving them we get
Lρg ρg
C1 = ; C2 = −
E 2E
Substituting back into u(x)
ρg
u(x) =(2L − x)x
2E
which coincides with the exact solution from mechanics of materials.
1.7 Boundary Conditions

1.7.1 Traction Boundary Conditions
The solution of problems in solid mechanics requires that boundary conditions be
specified. The boundary conditions may be specified in terms of components of
displacement, stress, or a combination of both. For any point on an arbitrary
surface, the traction Ti is defined as the vector consisting of the three components
of stress acting on the surface at the point of interest. As indicated in Figure
1.4 the traction vector consists of one component of normal stress, σnn , and two
components of shear stress, σnt and σns . The traction vector can be written using
Cauchy’s law
3
X
Ti = σji nj = σji nj (1.17)
j
where nj is the unit normal to the surface at the point under consideration.2 For a
plane perpendicular to the x1 axis ni = (1, 0, 0) and the components of the traction
are T1 = σ11 , T2 = σ12 , and T3 = σ13 .
1.7.2 Free Surface Boundary Conditions

The condition that a surface be free of stress is equivalent to all components of
traction being zero, i.e., Tn = σnn = 0, Tt = σnt = 0, and Ts = σns = 0. It is
possible that only selected components of the traction be zero while others are not
zero. For example, pure pressure loading corresponds to nonzero normal stress and
zero shear stresses.
2
Einstein’s summation convention can be introduced with (1.17) as an example. Any pair of
repeated indices implies a summation over all the values of the index in question. Furthermore,
each pair of repeated indices represents a contraction. That is, the order of resulting tensor, in this
case order one for Ti , is two less than the sum of the orders of the tensors involved in the operation.
The resulting tensor keeps only the free indices that are not involved in the contraction–in this
case only i remains.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 1.5: Traction continuity across an interface.
1.8 Continuity Conditions
1.8.1 Traction Continuity
Equilibrium (action and reaction) requires that the traction components Ti must be
continuous across any surface. Mathematically this is stated as Ti+ −Ti− = 0. Using
− −
(1.17), Ti+ = σji
+
nj . Since n+ +
j = −nj , we have σji = σji . In terms of individual
stress components, σnn + = σ − , σ + = σ − , and σ + = σ − (Figure 1.5). Thus, the
nn nt nt ns ns
normal and shear components of stress acting on a surface must be continuous across
that surface. There are no continuity requirements on the other three components
+ − + 6= σ − , and σ + 6= σ − . Lack of
of stress. That is, it is possible that σtt 6= σtt , σss ss ts ts
continuity of the two normal and one shear components of stress is very common
because the material properties are discontinuous across lamina boundaries.
1.8.2 Displacement Continuity
Certain conditions on displacements must be satisfied along any surface in a per-

fectly bonded continuum. Consider for example buckling of a cylinder under ex-
ternal pressure (Figure 1.6). The displacements associated with the material from
−
either side of the line A-A must be identical u+ i = ui . The continuity conditions
must be satisfied at every point in a perfectly bonded continuum. However, con-
tinuity is not required in the presence of debonding or sliding between regions or
phases of a material. For the example shown, continuity of slope must be satisfied
∂w+ ∂w−
also ( = ), where w is the radial displacement.
∂θ ∂θ
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 1.6: Buckling of an encased cylindrical pipe under external pressure.
1.9 Compatibility
The strain displacement equations (1.5) provide six equations for only three un-
known displacements ui . Thus, integration of equations (1.5) to determine the
unknown displacements will not have a single-valued solution unless the strains εij
satisfy certain conditions. Arbitrary specification of the εij could result in discon-
tinuities in the material, including gaps and/or overlapping regions.
The necessary conditions for single-valued displacements are the compatibility
conditions. Although these six equations are available [2], they are not used here
because the displacement method, which is used throughout this book, does not
require them. That is, in solving problems, the form of displacements ui is always
assumed a priori. Then, the strains are computed with (1.5), and the stress with
(1.46). Finally, equilibrium is enforced by using the PVW (1.16).
1.10 Coordinate Transformations

The coordinates of point P in the prime coordinate system can be found from its
coordinates in the unprimed system. From Figure 1.7, the coordinates of point P
are
x01 = x1 cos θ + x2 sin θ

x02 = −x1 sin θ + x2 cos θ
x03 = x3 (1.18)
or
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
x2 P
si nq
x2
i na
l a m
x2
co
x’1
sq
x’2
s q
ers co
x1
fib x1
sin
laminate
q
x1
Figure 1.7: Coordinate transformation.
x0i = aij xj (1.19)
or in matrix notation
0
x = [a] {x} (1.20)
where aij are the components of the unit vectors of the primed system e0i on the
unprimed system ej , by rows [2]
e1 e2 e3
e01 a11 a12 a13
aij = cos(e0i , ej ) = (1.21)
e02 a21 a22 a23
e03 a31 a32 a33
If primed coordinates denote the lamina coordinates and unprimed denote the
laminate coordinates, then (1.19) transforms vectors from laminate to lamina coor-
dinates. The inverse transformation simply uses the transpose matrix
{x} = [a]T x0

(1.22)
Example 1.2 A composite lamina has fiber orientation θ = 30◦ . Construct the [a] matrix
by calculating the direction cosines of the lamina system, i.e., the components of the unit
vectors of the lamina system (x0i ) on the laminate system (xj ).
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 1.8: Coordinate transformation for axial-symmetric analysis.
Solution to Example 1.2 From Figure 1.7 and (1.19) we have

√
3
a11 = cos θ =
2
1
a12 = sin θ =
2
a13 = 0
1
a21 = − sin θ = −
√ 2
3
a22 = cos θ =
2
a23 = 0
a31 = 0
a32 = 0
a33 = 1
Example 1.3 A fiber reinforced composite tube is wound in the hoop direction (1-direction).
Formulas for the stiffness values (E1 , E2 , etc.) are given in that system. However, when
analyzing the cross-section of this material with generalized plane strain elements (CAX4 in
Abaqus), the model is typically constructed in the structural X, Y, Z system. It is therefore
necessary to provide the stiffness values in the structural system as Ex , Ey , etc. Con-
struct the transformation matrix [a]T to go from lamina coordinates (1-2-3) to structural
coordinates in Figure 1.8.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Solution to Example 1.3 First, construct [a] using the definition (1.21). Taking each
unit vector (1-2-3) at a time we construct the matrix [a] by rows. The i-th row contains the
components of (i = 1, 2, 3) along (X-Y-Z).
[a] X Y Z
1 0 0 1
2 0 −1 0
3 1 0 0
The required transformation is just the transpose of the matrix above.
1.10.1 Stress Transformation

A second-order tensor σpq can be thought as the (uncontracted) outer product3 of
two vectors Vp and Vq
σpq = Vp ⊗ Vq (1.23)
each of which transforms as (1.19)
0
σij = aip Vp ⊗ ajq Vq (1.24)
Therefore,
0
σij = aip ajq σpq (1.25)
or, in matrix notation
{σ 0 } = [a]{σ}[a]T (1.26)
0 in contracted notation
For example, expand σ11
σ10 = a211 σ1 + a212 σ2 + a213 σ3 + 2a11 a12 σ6 + 2a11 a13 σ5 + 2a12 a13 σ4 (1.27)
0 in contracted notation yields
Expanding σ12
σ60 = a11 a21 σ1 + a12 a22 σ2 + a13 a23 σ3 + (a11 a22 + a12 a21 )σ6 (1.28)
+ (a11 a23 + a13 a21 )σ5 + (a12 a23 + a13 a22 )σ4
The following algorithm is used to obtain a 6 × 6 coordinate transformation

matrix [T] such that (1.25) is rewritten in contracted notation as
σα0 = Tαβ σβ (1.29)

If α ≤ 3 and β ≤ 3 then i = j and p = q, so
Tαβ = aip aip = a2ip no sum on i, p (1.30)

If α ≤ 3 and β > 3 then i = j but p 6= q, and taking into account that switching
p by q yields the same value of β = 9 − p − q as per (1.9) we have
3
The outer product preserves all indices of the entities involved, thus creating a tensor of order
equal to the sum of the order of the entities involved.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Tαβ = aip aiq + aiq aip = 2aip aiq no sum on i, p (1.31)

If α > 3, then i 6= j, but we want only one stress, say σij , not σji because they
are numerically equal. In fact σα = σij = σji with α = 9 − i − j. If in addition β ≤ 3
then p = q and we get
Tαβ = aip ajp no sum on i, p (1.32)

When α > 3 and β > 3, i 6= j and p 6= q so we get
Tαβ = aip ajq + aiq ajp (1.33)

which completes the derivation of Tαβ . Expanding (1.30–1.33) and using (1.21) we
get
a211 a212 a213

 
2 a12 a13 2 a11 a13 2 a11 a12

 a221 a22 2 a223 2 a22 a23 2 a21 a23 2 a21 a22 

 a231 a232 a233 2 a32 a33 2 a31 a33 2 a31 a32 
[T ] =  

 a21 a31 a22 a32 a23 a33 a22 a33 + a23 a32 a21 a33 + a23 a31 a21 a32 + a22 a31 
 a11 a31 a12 a32 a13 a33 a12 a33 + a13 a32 a11 a33 + a13 a31 a11 a32 + a12 a31 
a11 a21 a12 a22 a13 a23 a12 a23 + a13 a22 a11 a23 + a13 a21 a11 a22 + a12 a21
(1.34)
A MATLAB R
program that can be used to generate (1.34) is shown next (also
available in [5]).
% Derivation of the transformation matrix [T]

clear all;
syms T alpha R
syms a a11 a12 a13 a21 a22 a23 a31 a32 a33
a = [a11,a12,a13;
a21,a22,a23;
a31,a32,a33];
T(1:6,1:6) = 0;
for i=1:1:3
for j=1:1:3
if i==j; alpha = j; else alpha = 9-i-j; end
for p=1:1:3
for q=1:1:3
if p==q beta = p; else beta = 9-p-q; end
T(alpha,beta) = 0;
if alpha<=3 & beta<= 3; T(alpha,beta)=a(i,p)*a(i,p); end
if alpha> 3 & beta<= 3; T(alpha,beta)=a(i,p)*a(j,p); end
if alpha<=3 & beta>3; T(alpha,beta)=a(i,q)*a(i,p)+a(i,p)*a(i,q);end
if alpha>3 & beta>3; T(alpha,beta)=a(i,p)*a(j,q)+a(i,q)*a(j,p);end
end
end
end
end
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
T
R = eye(6,6); R(4,4)=2; R(5,5)=2; R(6,6)=2; % Reuter matrix
Tbar = R*T*R^(-1)
1.10.2 Strain Transformation

The tensor components of strain εij transform in the same way as the stress com-
ponents
ε0ij = aip ajq εpq (1.35)

or
ε0α = Tαβ εβ (1.36)

with Tαβ given by (1.34). However, the three engineering shear strains γxz , γyz , γxy
are normally used instead of tensor shear strains εxz , εyz , εxy . The engineering
strains ( instead of ε) are defined in (1.5). They can be obtained from the tensor
components by the following relationship
δ = Rδγ εγ (1.37)
with the Reuter matrix given by
 
1 0 0 0 0 0

 0 1 0 0 0 0 

 0 0 1 0 0 0 
[R] =   (1.38)

 0 0 0 2 0 0 

 0 0 0 0 2 0 
0 0 0 0 0 2
Then, the coordinate transformation of engineering strain results from (1.36)
and (1.37) as
0α = T αβ β (1.39)
with
= [R][T ][R]−1

T (1.40)
used only to transform engineering strains. Explicitly we have

T =
a211 a212 a213
 
a12 a13 a11 a13 a11 a12

 a221 a222 a223 a22 a23 a21 a23 a21 a22 


 a231 a232 a233 a32 a33 a31 a33 a31 a32 


 2 a21 a31 2 a22 a32 2 a23 a33 a22 a33 + a23 a32 a21 a33 + a23 a31 a21 a32 + a22 a31 

 2 a11 a31 2 a12 a32 2 a13 a33 a12 a33 + a13 a32 a11 a33 + a13 a31 a11 a32 + a12 a31 
2 a11 a21 2 a12 a22 2 a13 a23 a12 a23 + a13 a22 a11 a23 + a13 a21 a11 a22 + a12 a21
(1.41)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
1.11 Transformation of Constitutive Equations

The constitutive equations that relate stress σ to strain ε are defined using tensor
strains (ε, not ), as
σ 0 = C0 : ε0
0 0
σij = Cijkl ε0kl (1.42)
where both tensor and index notations have been used.4

For simplicity consider an orthotropic material (Section 1.12.3). Then, it is
0 , and σ 0 as
possible to write σ11 12
0 0
σ11 = C1111 ε011 + C1122
0
ε022 + C1133
0
ε033
0 0
σ12 = C1212 ε012 + C1221
0
ε021 = 2C1212
0
ε012 (1.43)
Rewriting (1.43) in contracted notation, it is clear that in contracted notation

all the shear strains appear twice, as follows
σ10 = C11
0 0 0 0
ε1 + C12 0 0
ε2 + C13 ε3 (1.44)
σ60 = 0 0
2C66 ε6
The factor 2 in front of the tensor shear strains is caused by two facts, the minor
symmetry of the tensors C and ε (see (1.5,1.55,1.56) and the contraction of the last
two indices of Cijkl with the strain εkl in (1.43). Therefore, any double contraction
of tensors with minor symmetry needs to be corrected by a Reuter matrix (1.38)
when written in the contracted notation. Next, (1.42) can be written as
σα0 = Cαβ
0
Rβδ ε0δ (1.45)
Note that the Reuter matrix in (1.45) can be combined with the tensor strains
using (1.37), to write
σα0 = Cαβ
0
0β (1.46)
in terms of engineering strains. To obtain the stiffness matrix [C] in the laminate
coordinate system, introduce (1.29) and (1.39) into (1.46) so that
0
Tαδ σδ = Cαβ T βγ γ (1.47)
It can be shown that

[T ]−1 = [T ]T (1.48)
Therefore
{σ} = [C]{} (1.49)
4
A double contraction involves contraction of two indices, in this case k and l, and it is denoted
by : in tensor notation. Also note the use of boldface to indicate tensors in tensor notation.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
with
[C] = [T ]T [C 0 ][T ] (1.50)

and
[C 0 ] = [T ]−T [C][T ]−1 = [T ][C][T ]T (1.51)

The compliance matrix is the inverse of the stiffness matrix, not the inverse of
the fourth-order tensor Cijkl . Therefore,
[S 0 ] = [C 0 ]−1 (1.52)
Taking into account (1.48) and (1.50), the compliance matrix transforms as
[S] = [T ]T [S 0 ][T ] (1.53)

[S 0 ] = [T ]−T [S][T ]−1 = [T ][S][T ]T (1.54)
1.12 3D Constitutive Equations

Hooke’s law in three dimensions (3D) takes the form of (1.42). The 3D stiffness
tensor Cijkl is a fourth-order tensor with 81 components. For anisotropic materials
only 21 components are independent. That is, the remaining 60 components can
be written in terms of the other 21. The one dimensional case (1D), studied in
mechanics of materials, is recovered when all the stress components are zero except
σ11 . Only for the 1D case, σ11 = σ,ε11 = ,C1111 = E, and σ = E. All the
derivations in this section are carried out in lamina coordinates but for simplicity
the prime symbol (0 ) is omitted, in this section only.
In (1.42), exchanging the dummy indexes i by j, and k by l we have
σji = Cjilk εlk (1.55)

Since the stress and strain tensors are symmetric, i.e., σij = σji and εkl = εkl ,
it follows that
Cijkl = Cjikl = Cijlk = Cjilk (1.56)

which effectively reduces the number of independent components from 81 to 36. For
example, C1213 = C2131 and so on. Then, the 36 independent components can be
written as a 6×6 matrix.
Furthermore, an elastic material does not dissipate energy. All elastic energy
stored during loading is recovered during unloading. Therefore, the elastic energy
at any point on the stress-strain curve is independent on the path that was followed
to arrive at that point. A path independent function is called a potential function.
In this case, the potential is the strain energy density u
e(εij ). Expanding the strain
energy density in a Taylor power series
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
1 ∂2u

∂eu e
u
e=u
e0 + εij + εij εkl + ... (1.57)
∂εij 0 2 ∂εij ∂εkl 0
Now take a derivative with respect to εij
∂eu 1
= 0 + βij + (αijkl εkl + αklij εij ) (1.58)
∂εij 2
where βij and αijkl are constants. From here, one can write
0
σij − σij = Cijkl εkl (1.59)
0 = β is the residual stress and α
where σij ij ijkl = 1/2(Cijkl + Cklik ) = Cijkl is the
symmetric stiffness tensor (see (1.56)). Equation (1.59) is a generalization of (1.55)
including residual stresses.
Using contracted notation, the generalized Hooke’s law becomes
     

 σ 1 
 C 11 C 12 C 13 C 14 C 15 C 16 
 1 




 σ 2




 C12 C22 C23 C24 C25 C26  
 


 2




σ3 C C C C C C
    
13 23 33 34 35 36 3

=  (1.60)

 σ4   C14 C24 C34 C44 C45 C46  

 γ4 
σ    C15 C25 C35 C45 C55 C56   γ 
   
 5   5 

  
 
σ6

C16 C26 C36 C46 C56 C66 γ6

Once again, the 1D case is covered when σα = 0 if α 6= 1. Then, σ1 = σ, 1 =

, C11 = E.
1.12.1 Anisotropic Material

Equation (1.60) represents a fully anisotropic material. Such a material has prop-
erties that change with the orientation. For example, the material body depicted in
Figure 1.9 deforms differently in the directions P, T, and Q, even if the forces ap-
plied along the directions P, T, and Q are equal. The number of constants required
to describe anisotropic materials is 21.
The inverse of the stiffness matrix is the compliance matrix [S] = [C]−1 . The
constitutive equation (3D Hooke’s law) is written in terms of compliances as follows
     

 1 
 S 11 S12 S 13 S14 S 15 S16 
 σ1 




 2




 S12 S22 S23 S24 S25 S26  
 


 σ2




3 S S S S S S σ
    
13 23 33 34 35 36 3

=  (1.61)

 γ4 
  S14 S24 S34 S44 S45 S46  

 σ4 

γ   S15 S25 S35 S45 S55 S56   σ 
   
 5   5 

  
 
γ6

S16 S26 S36 S46 S56 S66 σ6

The [S] matrix is also symmetric and it has 21 independent constants. For the
1D case, σ = 0 if p 6= 1. Then, σ1 = σ, 1 = , S11 = 1/E.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 1.9: Anisotropic material.
Figure 1.10: Monoclinic material.
1.12.2 Monoclinic Material

If a material has one plane of symmetry (Figure 1.10) it is called monoclinic and
13 constants are required to describe it. One plane of symmetry means that the
properties are the same at symmetric points (z and −z as in Figure 1.10).
When the material is symmetric about the 1-2 plane, the material properties
are identical upon reflection with respect to the 1-2 plane. For such reflection the
a-matrix (1.21) is
x1 x2 x3
00
e1 (1.62)
 
1 0 0
e002  0 1 0 
e003 0 0 −1
where 00 has been used to avoid confusion with the lamina coordinate system that
is denoted without 0 in this section but with 0 elsewhere in this book. From (1.40)
we get
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
 
1 0 0 0 0 0

 0 1 0 0 0 0 

 0 0 1 0 0 0 
T =  (1.63)

 0 0 0 −1 0 0 

 0 0 0 0 −1 0 
0 0 0 0 0 1

The effect of T is to multiply rows and columns 4 and 5 in [C] by −1. The
diagonal terms C44 and C55 remain positive because they are multiplied twice.
Therefore, Ci400 = −C with i 6= 4, 5, C 00 = −C with i 6= 4, 5, with everything else
i4 i5 i5
unchanged. Since the material properties in a monoclinic material cannot change
by a reflection, it must be C4i = Ci4 = 0 with i 6= 4, 5, C5i = Ci5 = 0 with i 6= 4, 5.
That is, 3D Hooke’s law reduces to
     

 σ1  C 11 C 12 C 13 0 0 C 16  1 
 σ    C C C 0 0 C

  


 2 

 12 22 23 26  
 2 

    
σ3  C13 C23 C33 0 0 C
  
36  3

= (1.64)

 σ4 
  0 0 0 C44 C45 0   
 γ4 

σ   0 0 0 C45 C55 0   γ 
   
 5   5 

  
 
σ6

C16 C26 C36 0 0 C66 γ6

and in terms of the compliances to

     

 1  S11 S12 S13 0 0 S16 
 σ1 




 2 



 S12 S22 S23 0 0 S26  

  σ2 


   
3 S13 S23 S33 0 0 S36 σ3
    
=  (1.65)

 γ 4 

 0
 0 0 S44 S45 0  
  σ4 

γ 0 0 0 S45 S55 0 σ5
    
 5 

 
  
 

  
γ6

S16 S26 S36 0 0 S66

σ6

1.12.3 Orthotropic Material

An orthotropic material has three planes of symmetry that coincide with the co-
ordinate planes. It can be shown that if two orthogonal planes of symmetry exist,
there is always a third orthogonal plane of symmetry. Nine constants are required
to describe this type of material.
The symmetry planes can be Cartesian, as depicted in Figure 1.11, or they may
correspond to any other coordinate representation (cylindrical, spherical, etc.). For
example, the trunk of a tree has cylindrical orthotropy because of the growth rings.
However, most practical materials exhibit Cartesian orthotropy. A unidirectional
fiber reinforced composite may be considered to be orthotropic. One plane of sym-
metry is perpendicular to the fiber direction, and the other two are parallel to the
fiber direction and orthogonal among themselves.
In addition to the reflection about the 1-2 plane discussed in Section 1.12.2,
a second reflection about the 1-3 plane should not affect the properties of the or-
thotropic materials. In this case the a-matrix is
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 1.11: Orthotropic material.
 
1 0 0
[a] =  0 −1 0  (1.66)
0 0 1
The T -matrix from (1.40) is

 
1 0 0 0 0 0

 0 1 0 0 0 0 

 0 0 1 0 0 0 
T =  (1.67)

 0 0 0 −1 0 0 

 0 0 0 0 1 0 
0 0 0 0 0 −1
This will make Ci6 = −Ci6 , i 6= 4, 6 and Ci4 = −Ci4 , i = 6 4, 6. Since the
material has symmetry about the 1-3 plane, this means that Ci6 = C6i = 0 , i 6= 6.
In this case, 3D Hooke’s law reduces to
     

 σ1 
 C11 C12 C13 0 0 0 
 1 




 σ2 



 C12 C22 C23 0 0 0  

  2 


   
σ3 C13 C23 C33 0 0 0 3
    
=  (1.68)

 σ4 


 0 0 0 C44 0 0  
  γ4 

σ5  0 0 0 0 C55 0 γ5

    
 


 
 
 


σ6

0 0 0 0 0 C66

γ6


     

 1 
 S11 S12 S13 0 0 0 
 σ 1 




 2




 S12 S22 S23 0 0 0  

 σ 2



    
3 S13 S23 S33 0 0 0 σ3
    
=  (1.69)
 γ4  0 0 0 S44 0 0  σ4 
   
   
γ   0 0 0 0 S55 0 σ5 
   
 5 

   
 
  
γ6

0 0 0 0 0 S66

σ6

i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 1.12: Randomly distributed E-glass fibers with 200X magnification.
Note that if the material has two planes of symmetry, it automatically has three
because applying the procedure once more for a third plane (the 2-3 plane) will not
change (1.68–1.69).
1.12.4 Transversely Isotropic Material

A transversely isotropic material has one axis of symmetry. For example, the fiber
direction of a unidirectional fiber reinforced composite can be considered an axis of
symmetry if the fibers are randomly distributed in the cross-section (Figure 1.12).
In this case, any plane containing the fiber direction is a plane of symmetry. A
transversely isotropic material is described by five constants. When the axis of
symmetry is the fiber direction (1-direction), 3D Hooke’s law reduces to
     

 σ1 
 C11 C12 C12 0 0 0 
 1 




 σ2




 C12 C22 C23 0 0 0  

  2 


   
σ3 C12 C23 C22 0 0 0 3
    
=  (1.70)
 σ4 
   0 0 0 (C22 − C23 )/2 0 0  
  γ4 

σ 0 0 0 0 C66 0 γ5
    
5

 
  
 


   
σ6

0 0 0 0 0 C66

γ6

     

 1 
 S11 S12 S12 0 0 0 
 σ 1 




 2




 S12 S22 S23 0 0 0  

 σ 2



    
3 S12 S23 S22 0 0 0 σ3
    
=  (1.71)
 γ4 
  
 0 0 0 2(S22 − S23 ) 0 0 
  σ4 
 
γ   0 0 0 0 S66 0 σ5 
   
 5 

   
 
  
γ6

0 0 0 0 0 S66

σ6

Note the equations would be different if the axis of symmetry is not the 1-
direction. In terms of engineering properties (Section 1.13), and taking into account
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
that the directions 2 and 3 are indistinguishable, the following relations apply for a
transversely isotropic material:
E2 = E3
ν12 = ν13 (1.72)
G12 = G13
In addition, any two perpendicular directions on the plane 2-3 can be taken as
axes. In other words, the plane 2-3 is isotropic. Therefore, the following holds in
the 2-3 plane
E2
G23 = (1.73)
2(1 + ν23 )
just as it holds for isotropic materials (see Problem 1.14).
1.12.5 Isotropic Material

The most common materials of industrial use are isotropic, like aluminium, steel,
etc. Isotropic materials have an infinite number of planes of symmetry, meaning
that the properties are independent of the orientation. Only two constants are
needed to represent the elastic properties. These two properties can be the Young’s
modulus E and the Poisson’s ratio ν, but several other pairs of constants are used
whenever it is convenient. However, any pair of properties has to be related to any
other pair. For example, you could describe isotropic materials by E and G, but
the shear modulus of isotropic materials is related to E and ν by
E
G= (1.74)
2(1 + ν)
Also, the Lamé constants are sometimes used for convenience, in this case the
two constants are
Eν
λ= (1.75)
(1 + ν)(1 − 2ν)
µ=G
To form yet another pair, any of the above properties could be substituted by
the bulk modulus K, as follows
E
K= (1.76)
3(1 − 2ν)
which relates the hydrostatic pressure p to the volumetric strain as
p = K (1+ 2+ 3 ) (1.77)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
For isotropic materials, the 3D Hooke’s law is written in terms of only two
constants C11 and C12 as
   C11 C12 C12 0 0 0

  
 σ 1 
  1 
C12 C11 C12 0 0 0
   


 σ 2






 
 2 


C12 C12 C11 0 0 0
   
σ3 3
     
=
 
(C11 −C12 )
 σ 4 
  0 0 0 2 0 0 
  γ4 
(C11 −C12 )
   
σ   γ5
    
0 0 0 0 0
 5 
    
2

 
 

σ6 (C11 −C12 )
 
γ6

0 0 0 0 0 2
(1.78)
In terms of compliances, once again, two constants are used, S11 and S12 as
follows
     

 1 
 S11 S12 S12 0 0 0 
 σ1 




 2 

  S12 S11 S12 0 0 0
 
 
 
 σ2 



3  S12 S12 S11 0 0 0 σ3
    
=   (1.79)
 γ4 
  0 0 0 2s 0 0   σ4 
    
γ  0 0 0 0 2s 0 σ5
    
5

 
 
 


   
γ6

0 0 0 0 0 2s

σ6

s = S11 − S12
Not only are the various constants related in pairs, but also certain restrictions
apply on the values that these constants may have for real materials. Since the
Young and shear moduli must always be positive, the Poisson’s ratio must be ν >
−1. Furthermore, since the bulk modulus must be positive, we have ν < 12 . Finally,
the Poisson’s ratio of isotropic materials is constrained by −1 < ν < 12 .
1.13 Engineering Constants

Please note from here forward 0 denotes the lamina coordinate system. Our next
task is to write the components of the stiffness and compliance matrices in terms of
engineering constants for orthotropic materials. For this purpose it is easier to work
with the compliance matrix, which is defined as the inverse of the stiffness matrix.
In lamina coordinates [S 0 ] = [C 0 ]−1 . The compliance matrix is used to write the
relationship between strains and stresses (1.69) for an orthotropic material. Let’s
rewrite the first of (1.69), which corresponds to the strain in the 1-direction (fiber
direction)
01 = S11
0
σ10 + S12
0
σ20 + S13
0
σ30 (1.80)
and let’s perform a thought experiment. Note that [S 0 ] is used to emphasize the fact
that we are working in the lamina coordinate system. First, apply a tensile stress
along the 1-direction (fiber direction) as in Figure 1.13, with all the other stresses
equal to zero, and compute the strain produced in the 1-direction, which is
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
2 Poisson Effect
σ Matrix
σ
Fiber 1
Matrix
L ΔL
Figure 1.13: Longitudinal loading.
σ1 0
01 = (1.81)
E1
Then, apply a stress in the 2-direction only, and compute the strain in the
1-direction using the appropriate Poisson’s ratio [1]
σ20
01 = −ν21 (1.82)
E2
Now, apply a stress in the 3-direction only, and compute the strain in the 1-
direction using the appropriate Poisson’s ratio,
σ30
01 = −ν31 (1.83)
E3
The total strain 01 is the sum of equations (1.81), (1.82), and (1.83)
1 0 ν21 0 ν31 0
01 = σ1 − σ2 − σ (1.84)
E1 E2 E3 3
Comparing (1.84) with (1.80) we conclude that
0 1 ν21 0 ν31
S11 = ; S0 = − ;S = − (1.85)
E1 12 E2 13 E3
Repeat the same procedure for the equations corresponding to 02 and 03 to
obtain the coefficients in the second and third rows of the compliance matrix (1.69).
For the shear terms use the 4th, 5th, and 6th rows of the compliance matrix
(1.69). For example, from Figure 1.14 we write
σ60 = 06 G12 = 2ε06 G12 (1.86)

which compared to the 6th row of (1.69) leads to S66 = 1/G12 .
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
3
2
6 4
2
1
6
6 Fiber 4
6
Fiber
(a) In-plane shear 6 (b) Interlaminar shear 4
Figure 1.14: Shear loading.
For an orthotropic material, the compliance matrix [S 0 ] is defined in the lamina

coordinate system as
 1 −ν21 −ν31 
0 0 0
 E1 E2 E3 
 −ν12 1 −ν32 
 0 0 0 
 E1 E2 E3 
−ν13 −ν23 1
 
 
 0 0 0 
[S 0 ] = 
 E1 E2 E3 
(1.87)
1 
 0 0 0 0 0 
G23
 
 
 1 
 0 0 0 0 0 
 G13 
 1 
0 0 0 0 0
G12
where Ei , Gij , and νij , are the elastic moduli, shear moduli, and Poisson’s ratios,
respectively. Furthermore, the subscripts indicate lamina coordinates, i.e.,
νij = νx0i x0j and Eii = Ex0i (1.88)
Since [S 0 ] is symmetric, the following must be satisfied

νij νji
= , i, j = 1..3 (1.89)
Eii Ejj
Furthermore, Poisson’s ratios are defined so that the lateral strain is given by
νj = −νij i (1.90)
In ANSYS, the Poisson’s ratios are defined differently that in this textbook.
In fact, νxy , νxz , νyz are denoted PRXY, PRXZ, and PRYZ, while νyx , νzx , νzy are
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
denoted by NUXY, NUXZ, and NUYZ. On the contrary, Abaqus uses the standard
notation also used in this textbook. That is, the symbols NU12, NU13, NU23,
follow the convention described by (1.90).
After computing Sij , the components of stress are obtained by using (1.46)
or (1.49). This formulation predicts realistic behavior for finite displacement and
rotations as long as the strains are small. This formulation is expensive to use
since it needs 18 state variables: 12 components of the strain displacement matrix
computed in the initial configuration (ui,j and ur,i ur,j ) plus 6 direction cosines [a]
to account for finite rotations.
However, in (1.87) only nine constants are independent because the matrix [S 0 ]
must be symmetric (see 1.93), so
 1 ν12 ν13 
− − 0 0 0
 E1 E1 E1 
 ν12 1 ν23 
 − − 0 0 0 
 E1 E2 E2 
ν13 ν23 1
 
− −
 
 0 0 0 
0
[S ] = 
 E1 E2 E3 
(1.91)
1 
 0 0 0 0 0 
G23
 
 
 1 
 0 0 0 0 0 
 G13 
 1 
0 0 0 0 0
G12
The stiffness matrix can be computed also in terms of engineering constants by
inverting the above equation so that [C 0 ] = [S 0 ]−1 , with components given in terms
of engineering constants as
0 1 − ν23 ν32
C11 =
E2 E3 ∆
0 ν21 + ν31 ν23 ν12 + ν32 ν13
C12 = =
E2 E3 ∆ E1 E3 ∆
0 ν31 + ν21 ν32 ν13 + ν12 ν23
C13 = =
E2 E3 ∆ E1 E2 ∆
0 1 − ν13 ν31
C22 =
E1 E3 ∆
0 ν32 + ν12 ν31 ν23 + ν21 ν13
C23 = =
E1 E3 ∆ E1 E2 ∆
0 1 − ν12 ν21
C33 =
E1 E2 ∆
0
C44 = G23
0
C55 = G13
0
C66 = G12 (1.92)
1 − ν12 ν21 − ν23 ν32 − ν31 ν13 − 2ν21 ν32 ν13
∆=
E1 E2 E3
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
So far both [S 0 ] and [C 0 ] are 6×6 matrices with 9 independent constants for
the case of orthotropic materials. If the material is transversely isotropic G13 =
G12 , ν13 = ν12 , E3 = E2 .
1.13.1 Restrictions on Engineering Constants

It is important to note that because of the symmetry of the compliance matrix
(1.91), the following restrictions on engineering constants apply
νij νji
= ; i, j = 1..3; i 6= j (1.93)
Ei Ej
Further restrictions on the values of the elastic constants can be derived from
the fact that all diagonal terms in both the compliance and stiffness matrices
must be positive. Since all the engineering elastic constants must be positive
(E1 , E2 , E3 , G12 , G23 , G31 > 0), all the diagonal terms of the stiffness matrix (1.92)
will be positive if the following two conditions are met. The first condition is that
(1 − νij νji ) > 0 for i, j = 1..3 and i 6= j, which leads to the following restriction on
the values of the engineering constants
s
Ei
0 < νij < ; i, j = 1..3; i 6= j (1.94)
Ej
The second condition is that
∆ = 1 − ν12 ν21 − ν23 ν32 − ν31 ν13 − 2ν21 ν32 ν13 > 0 (1.95)
These restrictions can be used to check experimental data. For example, consider
an experimental program in which if E1 and ν12 are measured in a longitudinal test
(fibers in the direction of loading) by using two strain gauges, one longitudinal and
one transverse, and E2 and ν21 are measured in the transverse tensile tests (fibers
perpendicular to loading). For the test procedure to be valid, all the four data
values, E1, E2 , ν12 and ν21 must conform to (1.93–1.95) within the margin allowed
by experimental errors.
Example 1.4 Sonti et al. [6] performed a series of tests on pultruded glass-fiber reinforced
composites. From tensile tests along the longitudinal axis, the average of eight tests gives
E1 = 19.981 GP a and ν12 = 0.274. The average of eight tests in the transverse direction
gives E2 = 11.389 GP a and ν21 = 0.192. Does this data fall within the constraints on
elastic constants?
Solution to Example 1.4 First compute both sides of (1.93) for i, j = 1, 2 as
E1 19.981
= = 72.9 GP a
ν12 0.274
E2 11.389
= = 59.3 GP a
ν21 0.192
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
The transverse result is 23% lower than expected. Either E2 measured is too low or ν21
measured is 23% higher than what it should be. In any case a 23% difference deserves some
scrutiny.
Next check (1.94)
r
E1
abs(ν12 ) <
E2
0.274 < 1.32
r
E2
abs(ν21 ) <
E1
0.192 < 0.75
Finally, there is insufficient data to evaluate the last of the restrictions on elastic con-
stants from (1.95).
1.14 From 3D to Plane Stress Equations

Setting σ3 = 0 in the compliance equations (1.69) of an orthotropic material implies
that the third row and column of the compliance matrix are not used
 0   0 0 0 σ10
  
 1  S11 S12 S13 0 0 0  
02  0 0 0 σ20
   


 
  S12 S22 S23 0 0 0 
 
 



 0  0 0 0 σ30 = 0
    
3 S13 S23 S33 0 0 0
   
0 = 0   σ0
 (1.96)

 γ4 

 0 0 0 S44 0 0   4 



 0
γ5 

  0 0 0 0
0 S55 0  

 σ50




 0    
γ6 0 0 0 0 0
0 S66

σ60

so, the first two equations plus the last one can be written separately of the re-
maining, in terms of a 3×3 reduced compliance matrix [S] and using γ = 2, we
have  0   0 0
 0 
 1  S11 S12 0  σ1 
02 0
=  S12 0
S22 0  σ0 (1.97)
 0  0  20 
γ6 0 0 S66 σ6
The third equation is seldom used
03 = S13
0
σ10 + S23
0
σ20 (1.98)
and the remaining two equations can be written separately as

0 0 0
γ4 S44 0 σ4
0 = 0 (1.99)
γ5 0 S55 σ50
To compute stress components from strains, (1.97) can be inverted to get {σ} =
[Q]{} or  0   0
Q11 Q012
 0 
 σ1  0  1 
σ20 =  Q012 Q022 0  0 (1.100)
 0  0  20 
σ6 0 0 Q66 γ6
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
where the matrix [Q0 ] = [S3×3

0 ]−1 is the reduced stiffness matrix for plane stress.
Note that while the components of the reduced compliance matrix [S3×3 0 ] are nu-
merically identical to the corresponding entries in the 6×6 compliance matrix, the
components of the reduced stiffness matrix [Q0 ] are not numerically equal to the cor-
responding entries on the 6×6 stiffness matrix [C 0 ], thus the change in name. This
is because the inverse of a 3×3 matrix produces different values than the inverse of
a 6×6 matrix. The set of equations is completed by writing
0 0 0
σ4 C44 0 γ4
0 = 0 (1.101)
σ5 0 C55 γ60
0 and C 0 are numerically equal to the corresponding entries
where the coefficient C44 55
in the 6×6 stiffness matrix because the 2×2 matrix in (1.101) is diagonal.
Example 1.5 Show that the change in the thickness t3 of a plate is negligible when com-
pared to the in-plane elongations a1 and b2 . Use the data from a composite plate with thick-
ness t = 0.635 mm, and dimensions a = 279 mm and b = 203 mm. Take E1 = 19.981 GP a,
E2 = 11.389 GP a, ν12 = 0.274.
Solution to Example 1.5 Assuming that the 0.635 mm thick glass-reinforced Polyester
plate is transversely isotropic, take E3 = E2 = 11.389 GP a, ν13 = ν12 = 0.274, G31 = G12 .
Sonti et al. [6] report the average of eight torsion tests as G12 = 3.789 GP a. Lacking
experimental data, assume ν23 ≈ νm = 0.3, G23 ≈ Gm = 0.385 GP a, with the properties
of the Polyester matrix taken from [1]. The remaining properties in (1.91) can be obtained,
using (1.93), as

E2 11.389
ν21 = ν12 = 0.274 = 0.156
E1 19.981

E3 11.389
ν31 = ν13 = 0.274 = 0.156
E1 19.981

E3 11.389
ν32 = ν23 = 0.3 = 0.3
E2 11.389
Because transverse isotropy G13 = G12 = 3.789 GP a. Now, assume a state of stress
σ10 = σ20 = 0.1 GP a, σ40 = σ50 = σ60 = 0 and σ30 = 0 because of the assumption of plane stress.
Using (1.97) we get
0.1 0.1(0.156)
01 = S11
0
σ10 + S12
0
σ20 = − = 3.635 10−3
19.981 11.389
0.1(0.156) 0.1
02 = S12
0
σ10 + S22
0
σ20 = − + = 7.411 10−3
11.389 11.389
0.274(0.1) 0.3(0.1)
03 = S13
0
σ10 + S23
0
σ20 = − − = −4.005 10−3
19.981 11.389
Finally
t03 = −0.635(4.005 10−3 ) = −2.543 10−3 mm
a01 = 279(3.635 10−3 ) = 1.014 mm
b02 = 203(7.411 10−3 ) = 1.504 mm
Since the elongation in the transverse direction is so small, it is neglected in the deriva-
tion of the plate equations in [1, Section 6.1].
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
1.15 Apparent Laminate Properties

The stiffness matrix [C] of a balanced, symmetric laminate with N laminas is built
by adding the lamina matrices in laminate coordinate system multiplied by the
thickness ratio tk /t of each lamina, where t is the laminate thickness and tk denotes
the thickness of the k -th lamina
N
X tk
[C] = [Ck ] (1.102)
t
k=1
Note that compliances cannot be added nor averaged. The laminate compliance
is obtained inverting the 6×6 stiffness matrix, as
[S] = [C]−1 (1.103)
A laminate is called balanced if the total thickness of laminas oriented with

respect to the laminate direction at +θ and −θ are the same. Such a laminate has
orthotropic stiffness [C] and compliance [S]. In terms of the apparent engineering
properties of the laminate, the compliance is
 1 νyx νzx 
− − 0 0 0
 Ex Ey Ez 
 νxy 1 νzy 
 − − 0 0 0 

 Ex Ey Ez 

 νxz νyz 1 
 − − 0 0 0 

[S] =  Ex Ey Ez 
 (1.104)
 1 
 0 0 0 0 0 
 Gyz 
1
 
 
 0 0 0 0 0 
 Gxz 
 1 
0 0 0 0 0
Gxy
Since the compliance must be symmetric, it must satisfy (1.93) with i, j =

x, y, z. Therefore, it is possible to compute the apparent engineering properties of
a laminate in terms of the laminate compliance, as follows
Ex = 1/S11 νxy = −S21 /S11

Ey = 1/S22 νxz = −S31 /S11
Ez = 1/S33 νyz = −S32 /S22
Gyz = 1/S44
Gxz = 1/S55
Gxy = 1/S66 (1.105)
Example 1.6 Compute the laminate properties of [0/90/ ± 30]S with tk = 1.5 mm, Ef =
241 GPa, νf = 0.2, Em = 3.12 GPa, νm = 0.38, fiber volume fraction Vf = 0.6, where f,m,
denote fiber and matrix, respectively.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Solution to Example 1.6 First use periodic microstructure micromechanics (6.8) to ob-
tain the lamina properties (in MPa).
E1 = 145, 880 G12 = 4, 386 ν12 = ν13 = 0.263

E2 = 11, 590 G23 = 3, 767 ν23 = 0.538
Then, compute the compliance matrix [S 0 ] using (1.91), the rotation matrix [T ] using
(1.34), the compliance [S] in laminate coordinate system using (1.53), and the stiffness
[C] = [S]−1 in the laminate coordinate system for each lamina. Then, average them using
(1.102), invert the average, and finally using (1.105) get
Ex = 75, 924 Gxy = 38, 048 νxy = 0.370

Ey = 44, 913 Gyz = 3, 999 νyz = 0.394
Ez = 15, 119 Gxz = 4, 154 νxz = 0.276
This example can be solved with the following MATLAB code:
function Ex106
% Example 1.6 laminate stiffness as per section 1.15
clc
% Lamina properties calculated with PMM in cadec-online
% Cprime (1.92) calculated with
% http://www.cadec-online.com/Chapters/Chapter5/3DConstitutiveEquations
% /StiffnessMatrix.aspx
Cprime = [
149395.9113 6754.740801 6754.740801 0 0 0;
6754.740801 16627.00749 9093.208855 0 0 0;
6754.740801 9093.208855 16627.00749 0 0 0;
0 0 0 3766.899318 0 0;
0 0 0 0 4385.907534 0;
0 0 0 0 0 4385.907534;
]
theta = [0,90,30,-30];
thickness = [1,1,1,1]*1.5; % mm
laminateThickness = sum(thickness);
C = zeros(6);
for i=1:length(theta)
[T,Tbar] = RotationMatrix3D(theta(i));
C = C + Tbar*Cprime*Tbar*thickness(i)/laminateThickness; % (1.102)
end
S = C^-1; % (1.103)
display(S);
Ex =1/S(1,1) % (1.105)
Ey =1/S(2,2)
Ez =1/S(3,3)
Gxy =1/S(6,6)
Gyz =1/S(4,4)
Gxz =1/S(5,5)
PRxy=-S(2,1)/S(1,1)
PRyz=-S(3,2)/S(2,2)
PRxz=-S(3,1)/S(1,1)
end
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
which uses the following code for coordinate transformation in 3D:
function [T,Tbar] = RotationMatrix3D(t);

R = eye(6,6); R(4,4)=2; R(5,5)=2; R(6,6)=2; % Reuter matrix
a = [cosd(t), sind(t), 0;sind(t), cosd(t), 0; 0, 0, 1];
T(1:6,1:6) = 0;
for i=1:1:3
for j=1:1:3
if i==j; alpha = j; else alpha = 9-i-j; end
for p=1:1:3
for q=1:1:3
if p==q beta = p; else beta = 9-p-q; end
T(alpha,beta) = 0;
if alpha<=3 & beta<= 3; T(alpha,beta)=a(i,p)*a(i,p); end
if alpha> 3 & beta<= 3; T(alpha,beta)=a(i,p)*a(j,p); end
if alpha<=3 & beta>3; T(alpha,beta)=a(i,q)*a(i,p)+a(i,p)*a(i,q);end
if alpha>3 & beta>3; T(alpha,beta)=a(i,p)*a(j,q)+a(i,q)*a(j,p);end
end
end
end
end
Tbar = R*T*R^(-1);
end
Suggested Problems
Problem 1.1 Using the principle of virtual work (PVW), find a quadratic displacement
function u(x) in 0 < x < L of a tapered slender rod of length L, fixed at the origin and
loaded axially in tension at the free end. The cross-section area changes linearly and the
areas are A1 > A2 at the fixed and free ends, respectively. The material is homogeneous
and isotropic with modulus E.
Problem 1.2 Using the principle of virtual work (PVW), find a quadratic rotation angle
function θ(x) in 0 < x < L of a tapered slender shaft of circular cross-section and length L,
fixed at the origin and loaded by a torque T at the free end. The cross-section area changes
linearly and the areas are A1 > A2 at the fixed and free ends, respectively. The material is
homogeneous and isotropic with shear modulus G.
Problem 1.3 Construct a rotation matrix [a] resulting from three consecutive reflections
about (a) the x-y plane, (b) the x-z plane, (c) the y-z plane. The resulting system does not
follow the right-hand rule.
Problem 1.4 Construct three rotation matrices [a] for rotations θ = π about (a) the x-axis,
(b) the y-axis, (c) the z-axis.
Problem 1.5 Using  

10 2 1
σ =  2 5 1
1 1 3
and [a] of Example 1.2, verify that (1.29) yields the same result as (1.26).
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Problem 1.6 Write a computer program to evaluate the compliance and stiffness matrices
in terms of engineering properties. Take the input from a file and the output to another file.
Validate the program with your own examples. You may use material properties from [1,
Tables 1.3–1.4] and assume the material is transversely isotropic as per Section 1.12.4.
Show all work in a report.
Problem 1.7 Write a computer program to transform the stiffness and compliance matrix
from lamina coordinates C 0 , S 0 , to another coordinate system C, S, by a rotation −θ around
the z-axis (Figure 1.7). The data C 0 , S 0 , θ, should be read from a file. The output C, S
should be written to another file. Validate your program with your own examples. You may
use material properties from [1, Tables 1.3–1.4] and assume the material is transversely
isotropic as per Section 1.12.4. Show all work in a report.
Problem 1.8 Verify numerically (1.92) against [S]−1 for the material of your choice. You
may use material properties from [1, Tables 1.3–1.4] and assume the material is transversely
isotropic as per Section 1.12.4.
Problem 1.9 The following data has been obtained experimentally for a composite based
on a unidirectional carbon-epoxy prepreg (MR50 carbon fiber at 63% by volume in LTM25
Epoxy). Determine if the restrictions on elastic constants are satisfied.
E1 = 156.403 GP a, E2 = 7.786 GP a
ν12 = 0.352, ν21 = 0.016
G12 = 3.762 GP a
u u
σ1t = 1.826 GP a, σ1c = 1.134 GP a
u u
σ2t = 19 M P a, σ2c = 131 M P a
σ6u = 75 M P a
u1t = 11, 900 10−6 , u1c = 8, 180 10−6
u2t = 2, 480 10−6 , u2c = 22, 100 10−6
u
γ12 = 20, 000 10−6
Problem 1.10 Explain contracted notation for stresses and strains.
Problem 1.11 What is an orthotropic material and how many constants are needed to
describe it?
Problem 1.12 What is a transversely isotropic material and how many constants are
needed to describe it?
Problem 1.13 Use the three rotations matrices in Problem 1.4 to verify (1.48) numeri-
cally.
Problem 1.14 Prove (1.73) using (1.71) and (1.91).
Problem 1.15 Demonstrate that a material having two perpendicular planes of symmetry
also has a third. Apply a reflection about the 2-3 plane to (1.68) using the procedure in
Section 1.12.3.
Problem 1.16 What is a plane stress assumption?
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Problem 1.17 Write a computer program to evaluate the laminate engineering properties
for symmetric balanced laminates. All laminas are of the same material. Input data consists
of all the engineering constants for a transversely isotropic material, number of laminas N ,
thickness and angle for all the laminas tk , θk with k = 1...N . Use Sections 1.15, 1.12.4, and
1.13.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Chapter 2
Introduction to Finite Element

Analysis
In this textbook, the finite element method (FEM) is used as a tool to solve practical
problems. For the most part, commercial packages, mainly ANSYS R
, are used in
the examples. Computer programming is limited to implementing material models
and post-processing algorithms. When commercial codes lack needed features, other
codes are used, which are provided in [5]. A basic understanding of the finite element
method is necessary for effective use of any finite element software. Therefore, this
chapter contains a brief introduction intended for those readers who have not had a
formal course or prior knowledge about the finite element method. Furthermore, an
introduction to ANSYS Mechanical APDL is presented to familiarize the reader with
the typical procedures used for finite element modeling using commercial software.
2.1 Basic FEM Procedure

Consider the axial deformation of a rod. The ordinary differential equation (ODE)
describing the deformation of the rod is

d du
− EA −f =0 ; 0≤x≤L (2.1)
dx dx
where E, A are the modulus and cross-section area of the rod, respectively, and
f is the distributed force. The boundary conditions for the case illustrated in Figure
2.1 are
u(0) = 0

du
EA =P (2.2)
dx x=L
As it is customary in mechanics of materials textbooks, the real rod shown in

Figure 2.1(a) is mathematically modeled as a line in Figure 2.1(b). The rod occupies
the domain [0, L] along the real axis x.
37
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 2.1: Physical and mathematical (idealization) model.
2.1.1 Discretization
The next step is to divide the domain into discrete elements, as shown in Figure
2.2.
2.1.2 Element Equations

To derive the element equations, an integral form of the ordinary differential equa-
tion (ODE) is used, which is obtained by integrating the product of the ODE times
a weight function v as follows
Z xB
d du
0= v − EA − f dx (2.3)
xA dx dx
This is called a weak form because the solution u(x) does not have to satisfy the
ODE (2.1) for all and every one of the infinite values of x in [0, L], in a strong sense.
Instead, the solution u(x) only has to satisfy the ODE in (2.3) in a weighted average
sense. It is therefore easier to find a weak solution than a strong one. Although for
the case of the rod, the strong (exact) solution is known, most problems of composite
mechanics do not have an exact solution. The governing equation is obtained by
integrating (2.3) by parts as follows
xB xB
du xB
Z Z
dv du
0= EA dx − vf dx − v EA (2.4)
xA dx dx xA dx xA
where v(x) is a weight function, which is usually set equal to the primary variable
u(x). From the boundary term, it is concluded that
– specifying v(x) at xA or xB is an essential boundary condition
– specifying EA du

dx at either end is the natural boundary condition
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Introduction to Finite Element Analysis 39
Figure 2.2: Discretization into elements.
While u(x) is the primary variable, EA du

dx = EAx = Aσx is the secondary
variable. Let
u(xA ) = ue1
u(xB ) = ue2

du
− EA = P1e
dx xA

du
EA = P2e (2.5)
dx xB
Then, the governing equation becomes
Z xB
dv du
0= EA − vf dx − P1e v(xA ) − P2e v(xB ) = B(v, u) − l(v) (2.6)
xA dx dx
with
Z xB
dv du
B(u, v) = EA dx
xA dx dx
Z xB
l(v) = vf dx + P1e v(xA ) + P2e v(xB ) (2.7)
xA
2.1.3 Approximation over an Element

Now, the unknown u(x) is approximated as a linear combination (series expansion)
of known functions Nie (x) and unknown coefficients aej , as
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
n
X
ue (x) = aej Nje (x)
j=1
where aej are the coefficients to be found and Nje (x) are the interpolation functions.
For the weight function v(x), the Ritz method can be used [4], in which v(x) =
Nje (x). Substituting in the governing equation (2.6) we get
n Z xB
dNie dNje
X Z xB
e
EA dx aj = Nie f dx + P1e Nie (xA ) + P2e Nie (xB ) (2.8)
xA dx dx xA
j=1
which can be written as

n
X
e e
Kij aj = Fie (2.9)
j=1
or in matrix form
[K e ]{ae } = {F e } (2.10)
where [K e ] is the element stiffness matrix, {F e } is the element vector equivalent
force, and {ae } are the element unknown parameters.
2.1.4 Interpolation Functions

Although any complete set of linearly independent functions could be used as in-
terpolation functions, it is convenient to choose the function in such a way that the
unknown coefficients represent the nodal displacements, that is ai = ui . For a two-
node element spanning the interval xe ≤ x ≤ xe+1 , the following linear interpolation
functions (Figure 2.3) can be used
xe+1 − x
N1e =
he
e x − xe
N2 = (2.11)
he
where he = xe+1 − xe is the element length. These interpolation functions satisfy

the following conditions

0 if i 6= j
Nie (xj ) = (2.12)
1 if i=j
2
X
Nie (x) = 1 (2.13)
i=1
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 2.3: Linear interpolation functions for a two-node element rod.
Figure 2.4: Two-dimensional interpolation functions.
which guarantees that the unknown coefficients represent the nodal displacements,
i.e., ai = ui .
Many other interpolation functions can be used, each one with some advan-
tages and disadvantages. The interpolation functions are intimately related to the
number of nodes of the element. Figure 2.4 illustrates the shape of the interpola-
tion functions N1 and N5 (corresponding to nodes 1 and 5) in an eight-node shell
element.
Broadly speaking, more nodes per element imply more accuracy and less need
for a fine mesh, but also imply higher cost in terms of computer time. Figure 2.5
illustrates how the approximate solution converges to the exact one as the number
of elements increases from 2 to 4 or as the number of nodes in the element increases
from 2 for the linear element to 3 for the quadratic element.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 2.5: Discretization error.
2.1.5 Element Equations for a Specific Problem

With interpolation functions that satisfy the conditions in (2.12–2.13), it is possible
to rewrite (2.10) as
[K e ]{ue } = {F e } (2.14)
where {ue } are the nodal displacements, [K e ] is the element stiffness matrix given
by
dN1e dN1e dN1e dN2e

 
R xB R xB
 xA EA dx dx dx xA EA dx dx dx 
[K e ] =   (2.15)
 Rx
B
dN2e dN1e R xB dN2e dN2e 
xA EA dx dx dx xA EA dx dx dx
and {F e } is the element force vector

( R xB
N1e f dx + P1e
)
xA
{Fie } = R xB (2.16)
xA N2e f dx + P2e
For a two-node rod element number e, the constant cross-section area Ae , the
element length he , and the modulus E are fixed. These values define the tensile-
compression element stiffness as
EAe
ke = (2.17)
he
The external loads on the element are the distributed force fe , the force at
end number 1, P1e , and the force at end number 2, P2e . Using these values, the
linear interpolation functions (2.11), as well as (2.15) and (2.16), the element matrix
stiffness and the equivalent nodal forces become
e
−k e

e k EAe 1 −1
[K ] = = (2.18)
−k e k e he −1 1
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 2.6: Connectivity between three two-node elements.
P1e

fe he 1
{F e } = + (2.19)
2 1 P2e
2.1.6 Assembly of Element Equations

The element unknown parameters correspond to displacements at the element nodes.
Since a node must have the same displacement on both adjacent elements, the
value is unique. For example, using the connectivity of elements shown in Figure
2.6, unique labels are assigned to the displacements, using capital letters. While a
superscript denotes an element number, a subscript indicates a nodal number, as
follows
u11 = U1
u12 = U2 = u21
u21 = U3 = u31
u32 = U4 (2.20)
Now, the element equations can be assembled into the global system. First, the
contribution of element #1 is
k 1 −k 1
      1 
0 0  U1 f1 h1 /2 P
 11
    
 −k 1 k 1
     
0 0  U2 f1 h1 /2 P2
    

 0
 = + (2.21)
0 0 0  U 0  0
 3
 
    
  

0 0 0 0 U4 0 0
    
Add the contribution of element #2, as follows
k1 −k 1 P11
     
0 0  U1   f1 h1 /2  
 −k k + k 2 −k 2
1 1 P21 + P12
    
0  U2 f1 h1 /2 + f2 h2 /2
    
  = +
 0 −k 2 k2 0  U f2 h2 /2 P22
 3
     
 
  
  
0 0 0 0 U4 0 0
  
(2.22)
Finally, add element #3 to obtain the fully assembled system, as follows
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
k1 −k 1 P11
      
0 0  U1
   f1 h1 /2   
 −k 1 k 1 + k 2 2   1
P2 + P12
    
−k 0   U2 f1 h1 /2 + f2 h2 /2
  
  = +
 0 −k 2 k 2 + k 3 −k 3   U  f2 h2 /2 + f3 h3 /2  P22 + P13 
 3
 
 
  
  
0 0 −k 3 k3 U4 f3 h3 /2 P23
  
(2.23)
2.1.7 Boundary Conditions

By equilibrium (see Figure 2.2), the internal loads cancel whenever two elements
share a node, or
P21 + P12 = 0
P22 + P13 = 0 (2.24)
The remaining P11 and P32 are the forces at the end of the rod. If either end of
the rod is fixed, then the displacement must be set to zero at that end. Say the end
at x = 0 is fixed, then U1 = 0. If the end at x = L is free, then P32 must be specified,
since U4 6= 0. If it is not specified, then it is assumed that the force is zero.
2.1.8 Solution of the Equations

Since U1 = 0, eliminating the first row and column of the stiffness matrix, a 3×3
system of algebraic equations is obtained, and solved for 3 unknowns: U2 , U3 , U4 .
Once a solution for U2 is found, the reaction P11 is computed from the first equation
of (2.23), as follows
f1 h1
−k 1 U2 = + P11 (2.25)
2
2.1.9 Solution Inside the Elements

Now, the solution Ui at 4 points along the rod is available. Next, the solution at
any location x can be computed by interpolating with the interpolation functions,
as follows
2
X
U e (x) = Uje Nje (x) (2.26)
j=1
or
 U1 N11 (x) + U2 N21 (x)


if 0 ≤ x ≤ h1
u(x) = U2 N12 (x) + U3 N22 (x) if h1 ≤ x ≤ h1 + h2 (2.27)
U3 N13 (x) + U4 N23 (x) if h1 + h2 ≤ x ≤ h1 + h2 + h3

i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
2.1.10 Derived Results

Strains
Strains are computed using (1.5) directly from the known displacements inside the
element. For example,
2
du X e dNje
x = = Uj (2.28)
dx dx
j=1
Note that if Nje (x) are linear functions, the strains are constant over the element.
In general the quality of the strains is one order of magnitude poorer than the
primary variable (displacements).
Stresses
Stress values are usually computed from strains through the constitutive equations.
In this example, with one-dimensional stress-strain behavior
σx = E x (2.29)
Note that the quality of stresses is the same as that of the strains.
2.2 General Finite Element Procedure

The derivation of the element equations, assembly, and solution for any type of
elements is similar to that of the one-dimensional rod element described in Section
2.1, with the exception that the principle of virtual work (PVW, 1.16) is used instead
of the governing equation (2.1). The PVW provides a weak form similar to that in
(2.4). Expanding (1.16) for a full 3D state of deformation, the internal virtual work
is
Z
δWI = (σxx δxx + σyy δyy + σzz δzz + σyz δγyz + σxz δγxz + σxy δγxy ) dV
Z
= σ T δ dV (2.30)
V
where
σ T = [σxx , σyy , σzz , σyz , σxz , σxy ]

δT = [δxx , δyy , δzz , δγyz , δγxz , δγxy ] (2.31)
Next, the external work is

Z Z
T
δWE = f δu dV + tT δu dS (2.32)
V S
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
where the volume forces per unit volume and surface forces per unit area are
f T = [fx , fy , fz ]
tT = [tx , ty, tz ] (2.33)
Here, underline ( ) denotes a one-dimensional array, not necessarily a vector.

For example, u is a vector but σ are the six components of stress arranged in
a six-element array. The virtual strains are strains that would be produced by
virtual displacements δu (x). Therefore, virtual strains are computed from virtual
displacements using the strain-displacement equations (1.5). In matrix notation
=∂u
δ = ∂ δu (2.34)
where
∂ ∂ ∂
 
0 0 0
 ∂x ∂y ∂z 
 ∂ ∂ ∂ 
∂= 0 0 0  (2.35)
 
 ∂y ∂x ∂z 
 ∂ ∂ ∂ 
0 0 0
∂z ∂y ∂x
Then, the PVW is written in matrix notation as
Z Z Z
σ T ∂ δu dV = f T δu dV + tT δu dS (2.36)
V V S
The integrals over the volume V and surface S of the body can be broken element
by element over m elements, as
m Z
X m Z
X Z
T T T
σ ∂ δu dV = f δu dV + t δu dS (2.37)
e=1 Ve e=1 Ve Se
Whenever two elements share a surface, the contributions of the second inte-
gral cancel out, just as the internal loads canceled in Section 2.1.7. The stress
components are given by the constitutive equations. For a linear material
σ=C (2.38)
with C given by (1.68). The internal virtual work over each element becomes
Z Z
e T
δWI = σ δ dV = T C δ dV (2.39)
Ve Ve
The expansion of the displacements can be written in matrix form as
u=N a (2.40)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
where N contains the element interpolation functions and a the nodal displacements
of the element, just as in Section 2.1.4. Therefore, the strains are
=∂u=∂N a=Ba (2.41)

where B = ∂ N is the strain-displacement matrix. Now, the discretized form of the
internal virtual work over an element can be computed as
Z Z
δWIe = T T
a B C B δa dV = a T
B T C B dV δa = aT K e δa (2.42)
Ve Ve
where the element stiffness matrix K e is

Z
e
K = B T C B dV (2.43)
Ve
The external virtual work becomes
Z Z
δWEe = T
f δu dV + tT δu dS
Ve Se
Z Z
= T
f N dV + t N dS δa = (P e )T δa
T
(2.44)
Ve Se
where the element force vector is

Z Z
e T
P = N f dV + N T t dS (2.45)
Ve Se
The integrals over the element volume Ve and element surface Se are usually
evaluated numerically by the Gauss integration procedure. For the volume integral,
such a procedure needs evaluation of the integrand at a few points inside the volume.
Such points, which are called Gauss points, are important for two reasons. First,
the constitutive matrix C is evaluated at those locations. Second, the most accurate
values of strains (and stresses) are obtained at those locations too.
The assembly of the element equations δWIe and δWEe into the PVW for the
whole body is done similarly to the process in Section 2.1.6. Obviously the process
is more complicated than for rod elements. The details of such process and its
computer programming are part of finite element technology, which is outside the
scope of this textbook. Eventually all the element stiffness matrices K e and element
force vectors P e are assembled into a global system for the whole body
Ka=P (2.46)
Next, boundary conditions are applied on the system (2.46) in a systematic way
resembling the procedure in Section 2.1.6. Next, the algebraic system of equations
(2.46) is solved to find the nodal displacement array a over the whole body. Since
the nodal displacements results for every element can be found somewhere in a, it
is possible to go back to (2.34) and to (2.38) to compute the strains and stresses
anywhere inside the elements.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Example 2.1 Compute the element stiffness matrix (2.43) and the equivalent force vector
(2.45) of a rod discretized with one element. Use linear interpolation functions such as
(2.11). Compare the result with (2.18-2.19).
Solution to Example 2.1 Let Ae be the transverse area of the rod and he the element
length, with xe = 0 and xe+1 = he . Substituting these values in the linear interpolation
functions from equation (2.11), the interpolation functions arrays are obtained as follows
xe+1 − x
 
" e # " #
N 1 1 − x/h e
NT = = x− he
xe  =
 
N2e x/he
he
The strain-displacement array is obtained as
∂N1e /∂x

T T −1/he
B =∂ N = =
∂N2e /∂x 1/he
The rod element has a one-dimensional strain-stress state with linear elastic behavior.
Therefore
C=E
Then, using equation (2.43) we can write
Z Z he
e T
−1/he
K = B C B dV = −1/he 1/he E Ae dx
Ve 0 1/he
The element stiffness matrix is obtained by integration

e EAe 1 −1
[K ] =
he −1 1
To calculate the equivalent vector force, fe is defined as the distributed force on element,
P1e is the force at end x = 0, and P2e is the force at end x = he . Substituting into equation
(2.45) we obtain
he
P1e
Z Z Z
e T T 1 − x/he
P = N f dV + N t dS = fe dx +
Ve Se 0 x/he P2e
The element equivalent force vector is obtained by integration
e
e fe he 1 P1
P = +
2 1 P2e
Using MATLAB
R
% FEACM solution Example 2.1

clear all
syms x x1 x2 h E A f real
% shape functions (linear Lagrange polynomial)
N(1)=(x-x2)/(x1-x2)
N(2)=(x-x1)/(x2-x1)
N=subs(N,{x1,x2},{0,h})
% Strain-displacement matrix B
B=diff(N,x)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
% Stiffness matrix
K_e = int(B’*E*B,x,0,h)*A
% Equivalent nodal force vector
P_e=int(f*N,0,h)
2.3 Solid Modeling, Analysis, and Visualization

Many commercial programs exist with finite element analysis capabilities for dif-
ferent engineering disciplines. They help solve a variety of problems from simple
linear static analysis to nonlinear transient analysis. A few of these codes, such as
ANSYS R
and Abaqus TM , have special capabilities to analyze composite materials
and they accept custom, user-programmed constitutive equations and element for-
mulations. Since these software packages not only provide analysis tools, geometric
modeling, and visualization of results, but also they can be integrated in the larger
design, production, and product life-cycle process, they are often called complete
analysis environments or computer aided engineering (CAE) systems.
Modern finite element analysis (FEA) software are commonly organized into
three blocks: the pre-processor, the processor, and the post-processor. In the pre-
processor, the model is built defining the geometry, material properties, and element
type. Also, loads and boundary conditions are entered in the pre-processor, but they
may also be entered during the solution phase. With this information, the processor
can compute the stiffness matrix and the force vector. Next, the algebraic equations
(2.46) are solved and the solution is obtained in the form of displacement values. In
the last block–the post-processor–derived results, such as stress, strain, and failure
ratios, are computed. The solution can be reviewed using graphic tools.
In the remainder of this chapter, a general description of the procedures and the
specific steps for a basic FEA are presented using examples executed with ANSYS.
Although the emphasis of the textbook is on the mechanics of composite materials,
concepts are illustrated with examples that are solved using ANSYS. Solutions to
similar examples using Abaqus are available in [7].
The first requirement of the model is the geometry. Then, material properties are
given for the various parts that make up the geometry. Next, loads and boundary
conditions are applied on the geometry. Next, the geometry is discretized into
elements, which are defined in terms of the nodes and element connectivity. The
element type is chosen to represent the type of problem to be solved. Next, the
model is solved. Finally, derived results are computed and visualized.
2.3.1 Model Geometry

The model geometry is obtained specifying all nodes, their position, and the element
connectivity. The connectivity information allows the program to assemble the
element stiffness matrix and the element equivalent force vector to obtain the global
equilibrium equations, as shown in Section 2.1.6.
There are two ways to generate the model. The first is to manually create a
mesh. The second is to use solid modeling, and then mesh the solid to get the node
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
and element distribution.
Manual Meshing
Manual mesh generation was the only method before solid modeling became wide-
spread among commercial packages. It is still the only option with some older
and custom software, although in those cases it is always possible to use a general
purpose, solid modeling pre-processor to generate the mesh. With manual meshing,
the user creates nodes, then connects the nodes into elements. Afterward, the user
applies boundary conditions and loads directly on nodes and/or elements. Manual
meshing is used in Example 2.2.
Example 2.2 Use ANSYS to model the curved beam shown in Figure 2.7. Use manual
meshing for generating the mesh geometry. Since the thickness is small and constant, use
planar solid elements with plane stress analysis. Add the boundary conditions and loads
shown in the figure. Solve the problem and visualize a contour plot of von Mises stress
on the deformed shape. The thickness of the part is 4.0 mm. The material properties are
E = 195, 000 M P a, ν = 0.3.
Solution to Example 2.2 The commands listed below, which are available on the Web
site [5], define the model geometry by using manual meshing. The characters after (!) are
comments. These commands can be typed one line at a time in the ANSYS command
window (see Appendix C). Alternatively, in the ANSYS command window, read the text
file by entering /input,file,ext, where file is the name of the file, and ext is the file
extension.
/TITLE, Ex. 2.2 Bending curved beam (manual meshing)

/PREP7 ! Start pre-processor module
ET,1,PLANE182 ! Element type #1: PLANE182
KEYOPT,1,3,3 ! Key option #3 = 3, plane stress with thickness input
R,1,4 ! Real constant #1: Th = 4 mm
MP,EX,1,195000 ! Material #1: E=195000 MPa
MP,NUXY,1,0.3 ! Material #1: Poisson coefficient 0.3
! Nodes and elements
CSYS,1 ! Activate polar coordinate system
N,1,20,180 ! Define node #1: radius=20mm, angle=180
N,10,20,90 ! Define node #10: radius=20mm, angle=90
FILL,1,10 ! Fill nodes between node 1 and 10
NGEN,9,20,1,10,,2.5 ! Generate new node rows increasing radius 2.5 mm
CSYS,0 ! Activate Cartesian coordinate system
N,15,20,20 ! Define node #15: x=20mm, y=20mm
FILL,10,15 ! Fill nodes between node 10 and 15
NGEN,9,20,11,15,,,2.5 ! Generate new node rows increasing y 2.5 mm
E,1,2,22,21 ! Define element #1,joining nodes 1,2,22,21
EGEN,8,20,1 ! Generate a row of elements
EGEN,14,1,ALL ! Generate the rest of elements
FINISH ! Exit pre-processor module
/SOLU ! Solution module
ANTYPE,STATIC ! Set static analysis
D,1,ALL,0,0,161,20 ! Impose Clamped BC
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 2.7: Curved beam.
ESEL,S,ELEM,,105,112 ! Select a subset of elements

SFE,ALL,2,PRES,0,-9.5 ! Apply pressure on subset
ESEL,ALL ! Reselect all elements
SOLVE ! Solve the current load state
FINISH ! Exit solution module
/POST1 ! Start Post-processor module
PLNSOL,S,EQV,2,1 ! Contour plot Von Mises equivalent stress
FINISH ! Exit post-processor module
The maximum value of von Mises stress is 108.124 M P a, but this value is sensitive to
mesh density, so the value will vary if you use a different mesh.
Solid Modeling
With solid modeling, the user creates a geometric representation of the geometry
using solid model constructs, such as volumes, areas, lines, and points. Boundary
conditions, loads, and material properties can be assigned to parts of the solid model
before meshing. In this way, re-meshing can be done without losing, or having to
remove, the loads and boundary conditions. The models are meshed just prior to
the solution. Solid modeling is used in Example 2.3.
Example 2.3 Use ANSYS to generate the same model in Example 2.2, but this time use
Solid Modeling commands for generating the mesh geometry.
Solution to Example 2.3 The commands listed below generate the geometry using Solid
Modeling [5].
/TITLE, Ex. 2.3 Bending curved beam (Solid Modeling)

ET,1,PLANE182 ! Element type #1: PLANE182
KEYOPT,1,3,3 ! Key option #3 = 3, plane stress
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
R,1,4 ! Real constant #1: Th = 4 mm

MP,EX,1,195000 ! Material #1: E=195000 MPa
MP,NUXY,1,0.3 ! Material #1: Poisson coefficient 0.3
! Geometry
CYL4,0,0,40,90,20,180 ! Generate curved area
BLC4,0,20,20,20 ! Generate rectangular area
AGLUE,all ! Glue both areas
LESIZE,2,,,8 ! Define divisions of elements by lines
LESIZE,4,,,8
LESIZE,6,,,8
LESIZE,1,,,10
LESIZE,3,,,10
LESIZE,9,,,5
LESIZE,10,,,5
MSHKEY,1 ! Force meshing by quadrilateral elements
AMESH,all ! Mesh all areas
DL,2,1,all,0 ! Impose Clamped BC
SFL,6,PRES,-9.5 ! Apply pressure
The maximum value of von Mises stress 109.238 M P a, but this value is sensitive to
mesh density, so the value will vary if you use a different mesh.
2.3.2 Material and Section Properties

Parts must be associated to materials. Depending on the analysis, material prop-
erties can be linear (linear elastic analysis) or nonlinear (e.g., damage mechanics
analysis), isotropic or orthotropic, constant or temperature-dependent. Entering
the correct materials properties is one of the most important aspects of a successful
analysis of composite materials. A great deal of attention is devoted to material
properties in the rest of the textbook. For now it will suffice to illustrate the process
using a linear elastic, isotropic material. For structural analysis, elastic properties
must be defined according to Section (1.12). Other mechanical properties, such as
strength, density, and thermal expansion coefficients are optional and their defini-
tion depends on the objectives of the analysis.
All elements need material properties, but structural elements need additional
parameters that vary with the type of element. These parameters result from an-
alytical integration of the 3D governing equations while formulating the element.
For example, the cross-section area A appears in (2.1) because the 3D partial dif-
ferential equations have been integrated over the cross-section of the rod to arrive
at the ordinary differential equation (2.1). Beam elements require the cross-section
area and the moment of inertia to be specified. Laminated shell elements require
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Table 2.1: Some of the elements available in Abaqus and ANSYS

Abaqus ANSYS nodes DOF Element Description
Structural elements
T2D2 - 2 uX uY line bar/truss, 2D space
T3D2 LINK180 2 uX uY uZ line bar/truss, 3D space
- COMBIN14 2 uX uY uZ spring/damper, 3D space
B21 - 2 uX uY line beam in 2D space
θX θY
B31 BEAM188 2 uX uY uZ line beam in 3D space
θX θY θ Z
CPE4R PLANE182 4 uX uY solid quadrilateral
in 2D space
CPE8R PLANE183 8 uX uY solid quadrilateral
in 2D space
S4R SHELL181 4 uX uY uZ shell quadrilateral
θX θY θZ in 3D space (conventional)
S8R SHELL281 8 uX uY uZ shell quadrilateral
θX θY θZ in 3D space (conventional)
S8R5 - 4 uX uY uZ thin shell quadrilateral
θX θY in 3D space (conventional)
SAX1 SHELL208 2 uX uY θZ axisymmetric
SAX2 SHELL209 3 uX uY θZ axisymmetric
Continuum elements1
C3D8 SOLID185 8 uX uY uZ solid hexahedra
in 3D space
C3D20 SOLID186 20 uX uY uZ solid hexahedra
in 3D space
SC8R SOLSH190 8 uX uY uZ shell hexahedra
in 3D space (continuum)
the laminate stacking sequence (LSS). Continuum elements (see Table 2.1) do not
require additional parameters, only material properties, because the geometry is
fully described by the mesh. However, continuum elements representing laminated
composites still require the LSS.
2.3.3 Assembly
If more than one part exists, assembly is necessary to put the parts together into
what is called an assembly, which represents the physical object you are trying to
1
Continuum means that the 3D volume is discretized with and no section properties are required.
Structural means that the volume is discretized as 2D or 1D and section properties, such as shell
thickness, are required to complete the description of the volume. 3D solid elements are typical
continuum elements. Conventional shells and beams are typical structural elements. Continuum
shells are continuum elements with kinematic constrains to represent shell behavior.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
analyze.
2.3.4 Solution Steps

Next, the analysis process is normally broken down into several steps, each repre-
senting different loading and constraint conditions. The minimum number of steps
is two: an initial step and at least one additional step. No loads can be applied on
the initial step, only boundary conditions.
2.3.5 Loads
In structural analysis, loads are defined by forces, pressures, inertial forces (as grav-
ity), and specified displacements, all applied to the model. Specification of different
kinds of loads for the finite element model are explained in the following sections.
The reactions obtained by fixing a nodal degree of freedom (displacements and
rotations) are discussed in Section 2.3.6.
Loads can be applied on nodes by means of concentrated forces and moments, as
shown in Example 2.4. Also, loads can be distributed over the elements as: surface
loads, body loads, inertia loads, or other coupled-field loads (for example, thermal
strains). Surface loads are used in Example 2.5.
A surface load is a distributed load applied over a surface, for example a pressure.
A body load is a volumetric load, for example expansion of material by temperature
increase in structural analysis. Inertia loads are those attributable to the inertia of
a body, such as gravitational acceleration, angular velocity, and acceleration.
A concentrated load applied on a node is directly added to the force vector.
However, the element interpolation functions are used to compute the equivalent
forces vector due to distributed loads.
2.3.6 Boundary Conditions

The boundary conditions are the known values of the degrees of freedom (DOF)
on the boundary. In structural analysis, the DOF are displacements and rotations.
With this information, the software knows which values of a in (2.46) are known or
unknown.
Constrained Displacements and Rotations
In general, a node can have more than one degree of freedom. For example, if the
finite element model uses beam elements in 2D space, there are three DOF: the
horizontal displacement, the vertical displacement, and the rotation around an axis
perpendicular to the plane. Constraining different sets of DOF results in different
boundary conditions being applied. In the 2D beam element case, constraining only
the horizontal and vertical displacements results in a simple support, but constrain-
ing all the DOF results in a clamped condition.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 2.8: Convention for rotations of a plate or shell.
Symmetry Conditions
Symmetry conditions can be applied to reduce the size of the model without loss of
accuracy. Four types of symmetry must exist concurrently: symmetry of geometry,
boundary conditions, material, and loads. Under these conditions the solution will
also be symmetric. For example, symmetry with respect to the y − z plane means
that the nodes on the symmetry plane have the following constraints
ux = 0 ; θy = 0 ; θz = 0 (2.47)
where ux is the displacement along the x-direction, θy and θz are the rotations
around the y and z axis, respectively (Figure 2.8). Note that the definition of rota-
tions used in shell theory (φi , see Section 3.1) is different than the usual definition
of rotations θi that follows the right-hand rule. Rotations in ANSYS are described
using right-hand-rule rotations θi . Symmetry boundary conditions on nodes in the
symmetry plane involve the restriction of DOF translations out-of-plane with re-
spect to the symmetry plane and restriction of the DOF rotations in-plane with
respect to the symmetry plane. Symmetry boundary conditions are used in Exam-
ple 2.5.
Antisymmetry Conditions
Antisymmetry conditions are similar to the symmetry conditions. They can be
applied when the model exhibits antisymmetry of loads but otherwise the model
exhibits symmetry of geometry, symmetry of boundary conditions, and symmetry of
material. Antisymmetry boundary conditions on nodes in the antisymmetry plane
involve restriction of DOF translations in the antisymmetry plane and restriction
of DOF rotations out-of-plane with respect to the antisymmetry plane.
Periodicity Conditions
When the material, load, boundary conditions, and geometry are periodic with pe-
riod (x, y, z) = (2ai , 2bi , 2ci ), only a portion of the structure needs to be modeled,
with dimensions (2ai , 2bi , 2ci ). The fact that the structure repeats itself periodi-
cally means that the solution will also be periodic. Periodicity conditions can be
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
imposed by different means. One possibility involves using constrained equations

(CE) between DOF (see Section 6.2) or using Lagrange multipliers.
2.3.7 Meshing and Element Type

Next, the assembly needs to be meshed. Before meshing, the element type needs to
be chosen. Finite Element Analysis (FEA) programs have an element library that
contains many different element types. The element type determines the element
formulation used. For example, the degree of freedom set, the interpolation func-
tions, whether the element is for 2D or 3D space, etc. The element type identifies
the element category: bar/rod tensile-compression, beam bending, solid, shell, lam-
inate shell, etc. Each commercial code identifies element formulations with different
labels. Identification labels and basic characteristics of a few element formulations
are shown in Table 2.1. Also, each element type has different options. For example,
on a planar solid element, an option allows one to choose between plane strain and
plane stress analysis.
2.3.8 Solution Phase

In the solution phase of the analysis, the solver subroutine included in the finite
element program solves the simultaneous set of equations (2.46) that the finite
element method generates. Usually, the primary solution is obtained by solving for
the nodal degree of freedom values, i.e., displacements and rotations. Then, derived
results, such as stresses and strains, are calculated at the integration points. In
ANSYS, primary results are called nodal solutions and derived results are called
element solutions.
Several methods of solving the system of simultaneous equations are available.
Some methods are better for larger models, others are faster for nonlinear analysis,
others allow one to distribute the solution by parallel computation. Commercial
finite element programs solve these equations in batch mode. The frontal direct
solution method is commonly used because it is rather efficient for FEA. When the
analysis is nonlinear, the equations must be solved repeatedly, thus increasing the
computational time significantly.
2.3.9 Post-Processing and Visualization

Once the solution has been calculated, the post-processor can be used to review and
to analyze the results. Results can be reviewed graphically or by listing the values
numerically. Since a model usually contains a considerable amount of results, it
may be better to use graphical tools. Post-processors of commercial codes produce
contour plots of stress and strain distributions, deformed shapes, etc. The software
usually includes derived calculations such as error estimation, load case combina-
tions, or path operations. Examples 2.4 and 2.5 include commands to review the
results by listing and by graphic output, respectively.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Example 2.4 Use ANSYS to find the axial displacement at the axially loaded end of a bar
clamped at the other end. The bar is made of steel E = 200, 000 M P a, diameter d = 9 mm,
length L = 750 mm, and load P = 100, 000 N. Also find the stress and strain. Use three
two-node (linear) link elements.
Solution to Example 2.4 The ANSYS command sequence for this example is listed be-
low. You can either type these commands on the command window, or you can type them
on a text file [5], then, on the command window enter /input,file,ext, where file is the
name of the file, and ext is the file extension (see Appendix C).
/TITLE, Ex. 2.4 Axially loaded bar

ET,1,LINK180 ! Define element type #1 LINK180 - 3D bar
R,1,63.6173 ! Define real constant A=63.6173 mm^2
MP,EX,1,200E3 ! Define elastic modulus E=200,000 MPa
N,1 ! Define node 1, coordinates=0,0,0
NGEN,4,1,1,,,250 ! Generate 3 additional nodes
! distance between adjacent nodes 250 mm
E,1,2 ! Generate element 1 by node 1 to 2
EGEN,3,1,1 ! Generate element 2,3
/SOLU ! Start Solution module
OUTPR,ALL,LAST,
D,1,all ! Define b.c. in node 1, totally fixed
F,4,FX,100E3 ! Define horizontal force in node 4
/STAT,SOLU ! Provides a solution status summary
To see the printout you need to execute the following commands manually by typing
them in the ANSYS command window or using the graphical user interface (GUI):

PRNSOL,U,X ! Print in a list the horizonal disp.
PRESOL,ELEM ! Print all line element results
PRRSOL,FX ! Print horizontal reactions
A convenient combination of units for this case is N ewton, mm, and M P a. The
analysis results can be easily verified by strength of material calculations, as follows
PL (750)(100000)
Ux = = = 5.894 mm
AE (63.617)(200000)
P 100000
σ= = = 1571.9 MPa
A 63.617
σ
= = 7.859 · 10−3
E
Example 2.5 Use ANSYS to find the stress concentration factor of a rectangular notched
strap. The dimensions and the load state are defined in Figure 2.9. Use eight-node (quadratic)
quadrilateral plane stress elements.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 2.9: Rectangular notched strap analyzed in Example 2.5.
Solution to Example 2.5 The ANSYS command sequence for this example is listed be-
low. These commands can be typed on the command window or in a text file [5], then, on
the command window enter /input,file,ext, where file is the name of the file, and ext
is the file extension (see Appendix C).
/TITLE, Ex. 2.5 Stress concentration

ET,1,PLANE183 ! Define element type #1 PLANE183 8-noded 2-D
KEYOPT,1,3,3 ! Key option #3 = 3, plane stress
R,1,4 ! Define real constant th=4 mm
MP,EX,1,190E3 ! Define elastic modulus E=190,000 MPa
MP,PRXY,1,0.3 ! Define Poisson coefficient 0.3
BLC4,0,0,50,18.5 ! Define squared area 18.5x50 mm
CYL4,0,20,7.5 ! Define circular area radius 7,5 mm
ASBA,1,2 ! Subtract previous areas
ESIZE,1.5,0, ! Define element size
MSHKEY,0 ! Free mesh
AMESH,all ! Mesh
DL, 1, ,SYMM ! Define symmetry b. conditions in line 1
DL, 9, ,SYMM ! Define symmetry b. conditions in line 9
SFL,2,PRES,-10 ! Apply pressure on line 2
PLNSOL,S,X,2,1 ! Contour plot horizontal stress
PLVECT,S ! Vector plot principal stress
The stress in the net area without stress concentration is

P 10 · 37 · 4
σo = = = 14.8 MPa
A 25 · 4
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
The maximum horizontal stress close to the notch is 28 MPa obtained from the finite
element model (PLNSOL,S,X,2,1). Therefore, the concentration factor is
σmax
k= = 1.89
σo
Example 2.6 Using ANSYS generate a model for a dome (Figure 2.10) with different types
of elements (shell and beam elements), using two materials, and different section data. Use
solid modeling to generate the mesh geometry. Report the minimum and maximum values
of von Mises stress, and the maximum displacement.
Solution to Example 2.6 The element types in ANSYS are defined by the ET command
[8]. The element types can be defined by their library names (see Table 2.1) and given
reference numbers to be used later. For example, the commands shown below define two
element types, BEAM188 and SHELL181, and assign them type reference numbers 1 and
2, respectively.
ET,1,BEAM188 ! Define element type #1 BEAM188

ET,2,SHELL181 ! Define element type #2 SHELL181
For material definition, MP can be used along with the appropriate property label; e.g.,
EX for Young’s modulus, NUXY for Poisson’s ratio, etc. For isotropic material, only the
X-direction properties need to be defined. The remaining properties in the other directions
default to the X-direction values. Also a reference number is used for each material. For
example, the following code defines two materials
MP,EX,1,200E3 ! Define material #1 E=200000MPa

MP,NUXY,1,0.29 ! Define material #1 Poisson ratio
MP,EX,2,190E3 ! Define material #2 E=190000MPa
Section properties are required for all elements where one or more dimensions have been
integrated a priori, e.g., beam and shell elements. That is, beam elements require the area,
and shell elements the thickness. In previous versions of ANSYS, section properties were
introduced via REAL constants but that is changing to using SECTION data. For this
example,
SECTYPE,1,BEAM,RECT ! Define SECTION 1

SECDATA,10.,10. ! Assign cross-section 10x10 squared
SECTYPE,2,SHELL ! Define SECTION 2
SECDATA,6. ! Assign Th = 6 mm
The commands shown above define a database with a table of elements type, another
with materials, and lastly another with section data. The reference number of each table
can be selected by using the commands TYPE, MAT, and SECNUM before defining the
mesh, as shown in the ANSYS command sequence listed below. These commands can be
typed on the command window, or in a text file [5], then, on the command window enter
/input,file,ext, where file is the name of the file, and ext is the file extension (see
Appendix C).
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
/TITLE, Ex. 2.6 Dome

! Define element types
ET,1,BEAM188 ! Define element type #1 BEAM188
ET,2,SHELL181 ! Define element type #2 SHELL181
! Define materials
MP,EX,1,200E3 ! Define material #1 E=200,000 MPa
MP,EX,2,190E3 ! Define material #2 E=190,000 MPa
! Define sections
SECTYPE,1,BEAM,RECT ! Define SECTION 1
SECDATA,10.,10. ! Assign cross-section 10x10 squared
! Create geometry by solid modelling
SPH4, , ,500 ! Define sphere radius 500 mm
BLOCK,-600,600,-600,600,-600,0 ! Define blocks for substract ...
BLOCK, 300,600,-600,600,0,600 ! ...to sphere
BLOCK,-300,-600,-600,600,0,600
BLOCK,-600,600,300,600,0,600
BLOCK,-600,600,-300,-600,0,600
VADD,2,3,4,5,6 ! Add all blocks
VSBV,1,7 ! Substract blocks to the shephere
WPAVE,0,0,200 ! Offset working plane z=+200 mm
VSBW,ALL ! Divide Volume by working plane
! Mesh geometry
ESIZE,20 ! Define element size
TYPE,2 ! Assign SHELL to elements defined next
MAT,2 ! Assign material #2 to elements defined next
SECNUM,2 ! Assign section #2 to elements defined next
AMESH,8,9 ! Mesh areas 8 and 9 (top surface dome).
AMESH,12,15 ! Mesh areas 12,13,14 and 15 (side surface dome)
TYPE,1 ! Assign BEAM to elements defined next
MAT,1 ! Assign material #1 to elements defined next
LMESH,1,2 ! Mesh lines 1 and 2 (columns)
LMESH,4,5 ! Mesh lines 4 and 5 (columns)
D,2007,ALL,0,0,2037,10 ! Impose Clamped BC-lower-end of the columns
LSEL,S,LINE,,13,16,1 ! Select a subset of lines
DL,ALL,,UX,0,0 ! Impose BC-vertical edges of the side walls
DL,ALL,,UY,0,0 ! Impose BC-vertical edges of the side walls
LSEL,ALL ! Reselect all lines
ESEL,S,SEC,,2 ! Select a subset of elements
SFE,ALL,2,PRES,0,-100 ! Apply pressure on subset
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 2.10: Mesh obtained by the command sequence used to generate a dome.
ESEL,ALL ! Reselect all elements

!FINISH ! Exit post-processor module
Using PLNSOL,S,EQV,2,1, the maximum value of von Mises stress 83, 513 M P a at the
attachment to the columns. The minimum value of von Mises stress 255 M P a on the shell
body. The maximum displacement is 155 mm.
Suggested Problems
Problem 2.1 Solve Example 2.4 explicitly as it is done in Section 2.1, using only two
elements. Show all work.
Problem 2.2 From the solution of Problem 2.1, compute the axial displacement at (a)
x = 500 mm, (b) x = 700 mm.
Problem 2.3 Using the same procedure in Example 2.1 calculate the element stiffness ma-
trix and the equivalent force vector of a three-node element rod with quadratic interpolation
functions. The interpolation functions are
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
x − x2 x − x3 x − x3 x − x1 x − x1 x − x2
N1e = N2e = N3e =
x1 − x2 x1 − x3 x2 − x3 x2 − x1 x3 − x1 x3 − x2
where x1 , x2 and x3 are the coordinate positions of node 1, 2, and 3 respectively. Use
x1 = 0, x2 = h/2 and x3 = h, where h is the element length. Show all work.
Problem 2.4 Program a finite element code using the element formulation obtained in
Example 2.1 and the assembly procedure shown in Section 2.1.6. With this code, solve
Example 2.4. Show all work in a report.
Problem 2.5 Program a finite element code using the element formulation obtained in
Problem 2.3 and the assembly procedure shown in Section 2.1.6. With this code, solve
Example 2.4. Show all work in a report.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Chapter 3
Elasticity and Strength of

Laminates
Most composite structures are built as assemblies of plates and shells. This is
because the structure is more efficient when it carries membrane loads. Another
important reason is that thick laminates are difficult to produce.
For example, consider a beam made of an homogeneous material with tensile
and compressive strength σu subjected to bending moment M . Further, consider
a solid beam of square cross-section (Figure 3.1), equal width and depth 2c, with
area A, inertia I, and section modulus S given by
A = 4c2
4
I = c4
3
I 4
S = = c3 (3.1)
c 3
When the stress on the surface of the beam reaches the failure stress σu , the
bending moment per unit area is
Mu Sσu 1
mu = = = cσu (3.2)
A A 3
Now consider a square hollow tube (Figure 3.1) of dimensions 2c × 2c and wall
thickness t, with 2c >> t, so that the following approximations are valid
A = 4(2c)t = 8ct
t(2c)3

16
I=2 + c (2ct) = tc3
2
12 3
I 16 2
S = = tc (3.3)
c 3
Then
16 2
Mu Sσu 3 tc σu 2
mu = = = = cσu (3.4)
A A 8ct 3
63
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 3.1: Solid section (a) and hollow square tube (b).
The failure moment per unit area mu is twice as large for a hollow square tube
with thin walls than for a solid section.
Of course, the failure moment is limited by buckling of the thin walls (see Chap-
ter 4). This is the reason buckling analysis is so important for composites. Most
composite structures are designed under buckling constraints because the thick-
nesses are small and the material is very strong, so normally one does not encounter
material failure as in metallic structures (e.g., yield stress) but structural failure such
as buckling.
Plates are a particular case of shells, having no initial curvature. Therefore,
only shells will be mentioned in the sequel. Shells are modeled as two-dimensional
structures because two dimensions (length and width) are much larger than thick-
ness. The thickness coordinate is eliminated from the governing equations so that
the 3D problem simplifies to 2D. In the process, the thickness becomes a parameter
that is known and supplied to the analysis model.
Modeling of laminated composites differs from modeling conventional materials
in three aspects. First, the constitutive equations of each lamina are orthotropic
(Section 1.12.3). Second, the constitutive equations of the element depend on the
kinematic assumptions of the shell theory used and their implementation into the el-
ement. Finally, material symmetry is as important as geometric and load symmetry
when trying to use symmetry conditions in the models.
3.1 Kinematics of Shells

Shell elements are based on various shell theories which in turn are based on kine-
matic assumptions. That is, there are some underlying assumptions about the likely
type of deformation of the material through the thickness of the shell. These as-
sumptions are needed to reduce the 3D governing equations to 2D. Such assumptions
are more or less appropriate for various situations, as discussed next.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Elasticity and Strength of Laminates 65
Figure 3.2: Assumed deformation in FSDT.1
3.1.1 First-Order Shear Deformation Theory

The most popular composite shell theory is the first-order shear deformation theory
(FSDT). It is based on the following assumptions:
i. A straight line drawn through the thickness of the shell in the undeformed
configuration may rotate but it will remain straight when the shell deforms.
The angles it forms (if any) with the normal to the undeformed mid-surface
are denoted by φx and φy when measured in the x − z and y − z planes,
respectively (Figures 2.8 and 3.2).
ii. As the shell deforms, the change in thickness of the shell is negligible.
These assumptions are verified by experimental observation in most laminated

shells when the following are true:
– The aspect ratio r = a/t, defined as the ratio between the shortest surface
dimension a and the thickness t, is larger than 10.
– The stiffness of the laminas in shell coordinates (x, y, z) do not differ by more
than two orders of magnitude. This restriction effectively rules out sandwich
shells, where the core is much softer than the faces.
Based on the assumptions above, the displacement of a generic point B anywhere

in the shell can be written in terms of the displacement and rotations at the mid-
surface C as
1
Reprinted from Introduction to Composite Materials Design, E. J. Barbero, Figure 6.2, copy-
right (1999), with permission from Taylor & Francis.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
u(x, y, z) = u0 (x, y) − zφx (x, y)

v(x, y, z) = v0 (x, y) − zφy (x, y) (3.5)
w(x, y, z) = w0 (x, y)
The mid-surface variables on the right-hand side of (3.5) are functions of only
two coordinates (x and y), thus the shell theory is 2D. On the left-hand side, the
displacements are functions of three coordinates, and thus correspond to the 3D
representation of the material. At the 3D level, we use the 3D constitutive equations
(1.68) and the 3D strain-displacement equations (1.5), which now can be written in
terms of 2D quantities as
∂u0 ∂φx
x (x, y, z) = −z = 0x + zκx
∂x ∂x
∂v0 ∂φy
y (x, y, z) = −z = 0y + zκy
∂y ∂y

∂u0 ∂v0 ∂φx ∂φy 0
γxy (x, y, z) = + −z + = γxy + zκxy
∂y ∂x ∂y ∂x
∂w0
γyz (x, y) = −φy +
∂y
∂w0
γxz (x, y) = −φx +
∂x
z = 0 (3.6)
where
0 , also called membrane strains, represent
– The mid-surface strains 0x , 0y , γxy
stretching and in-plane shear of the mid-surface.
– The change in curvature κx , κy , κxy , which are close but not exactly the same
as the geometric curvatures of the mid-surface. They are exactly that for the
Kirchhoff theory discussed in Section 3.1.2.
– The intralaminar shear strains γxz , γyz , which are through-the-thickness shear
deformations. These are small but not negligible for laminated composites be-
cause the intralaminar shear moduli G23 , G13 are small when compared with
the in-plane modulus E1 . Metals are relatively stiff in shear (G = E/2(1+ν)),
and thus the intralaminar strains are negligible. In addition, the intralami-
nar shear strength of composites F4 , F5 are relatively small when compared
to the in-plane strength values F1t , F1c , thus making evaluation of intralami-
nar strains (and possibly stresses) a necessity. On the other hand, the shear
strength of metals is comparable to their tensile strength, and since the in-
tralaminar stress is always smaller than the in-plane stress, it is not necessary
to check for intralaminar failure of metallic homogeneous shells. That is not
the case for laminated metallic shells since the adhesive is not quite strong
and it may fail by intralaminar shear.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
(c)
Figure 3.3: Stress resultants acting on a plate or shell element: (a) forces per unit
length, and (b) moments per unit length, and (c) definition of shell theory rotations
φ compared to mathematical angles θ.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
While the 3D constitutive equations relate strains to stress, the laminate con-
stitutive equations relate mid-surface strains and curvatures. The laminate consti-
tutive equations are obtained by using the definition of stress resultants. While in
3D elasticity every material point is under stress, a shell is loaded by stress resul-
tants (Figure 3.3), which are simply integrals of the stress components through the
thickness of the shell, as follows
   k
 Nx  N Z
X  σx zk 
Ny = σy dz

Nxy

k=1 z k−1

σxy

X N Z zk k
Vy σyz
= dz
Vx σxz
k=1 zk−1
   k
 Mx  X N Z zk  σ x 
My = σy z dz (3.7)
Mxy k=1 zk−1 σxy
   
where N is the number of laminas, zk−1 and zk are the coordinates at the bottom
and top surfaces of the k-th lamina, respectively. Replacing the plane stress version
of the 3D constitutive equations in shell local coordinates (1.100–1.101) at each
lamina and performing the integration we get
0x
    
 Nx  A11 A12 A16 B11 B12 B16  
0y
   


 Ny 

  A12 A22 A26 B12 B22 B26 

 


0
    
Nxy A16 A26 A66 B16 B26 B66 γxy
    
=  (3.8)

 Mx 


 B11 B12 B16 D11 D12 D16 
 κx 

 My B12 B22 B26 D12 D22 D26 κy

 
   

 
 
 


Mxy

B16 B26 B66 D16 D26 D66
 
κxy

Vy H44 H45 γyz
=
Vx H45 H55 γxz
where
N
X
Aij = Qij t ;
k k
i, j = 1, 2, 6
k=1
XN

Bij = Qij t z̄ ;
k k k
i, j = 1, 2, 6
k=1
N
t3
X
tk z̄k2 + k

Dij = Qij k ; i, j = 1, 2, 6
12
k=1
N
t3k

5 X ∗ 4
Hij = Qij tk − tk z 2k + ; i, j = 4, 5 (3.9)
4 k t2 12
k=1
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i

where Qij k are the coefficients in laminate coordinates of the plane-stress stiffness
matrix for lamina number k, tk is the thickness of lamina k, and z̄k is the coordinate
of the middle surface of the k-th lamina. For an in-depth discussion of the meaning
of various terms see [1]. In summary, the Aij coefficients represent in-plane stiffness
of the laminate, the Dij coefficients represent bending stiffness, the Bij represent
bending-extension coupling, and the Hij represent intralaminar shear stiffness. All
these coefficients can be calculated by (3.9) and are implemented in widely available
software packages such as CADEC [9].
When membrane and bending deformations are uncoupled (e.g., symmetric lam-
inates), the governing equations of FSDT involve three variables for solving the
bending problem (w0 , φx , φy ) and two to solve the membrane problem (u0 , v 0 ).
Bending-extension coupling means that all five variables will have to be found si-
multaneously, which is what finite element analysis (FEA) software codes do for
every case, whether the problem is coupled or not.
The equilibrium equations of plates can be derived by using the principle of vir-
tual work (PVW, see (1.16)). Furthermore, the governing equations can be derived
by substituting the constitutive equations (3.8) into the equilibrium equations.
Simply Supported Boundary Conditions in Plates

Composite plates with coupling effects may have bending, shear, and membrane
deformations coupled even if loaded by pure bending, pure shear, or pure in-plane
loads (see p. 141 in [1]). While the term simply supported always means to restrict
the transverse deflection w(x, y), it does not uniquely define the boundary conditions
on the in-plane displacements un and us , normal and tangent to the boundary,
respectively. In the context of analytical solutions, it is customary to restrict either
un or us . Therefore, the following possibilities exist
– SS-1: w = us = φs = 0; Nn = N
bn ; Mn = M
cn
– SS-2: w = un = φs = 0; Nns = N
bns ; Mn = M
dn
In type SS-1, a normal force and a moment are specified. In SS-2, a shear
force and a moment are specified. If the laminate does not have bending-extension
coupling, and the analysis is geometrically linear, transverse loads will not induce
un . The naming convention for the rotations is the same as that used for moment
resultants in Figure 3.3, where a subscript n indicates the direction normal to the
edge of the shell, and a subscript s indicates the direction tangent to the edge (see
also [10, Figure 6.2.1]). Furthermore b represents a fixed known value that may or
may not be zero.
3.1.2 Kirchhoff Theory

Historically, Kirchhoff theory was preferred because the governing equations can be
written in terms of only one variable, the transverse deflection of the shell w0 . In
the pre-information age, it was easier to obtain analytical solutions in terms of only
one variable rather than in terms of the three variables needed in FSDT. This means
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
that a wealth of closed form design equations and approximate solutions exist in
engineering design manuals which are based on Kirchhoff theory [11]. Such simple
design formulas can still be used for preliminary design of composite shells if we
are careful and we understand their limitations. Metallic shells were and still are
commonly modeled with Kirchhoff theory. The FSDT governing equations can be
reduced to Kirchhoff governing equations, and closed form solutions can be found,
as shown in [10].
In Kirchhoff theory the intralaminar shear strain is assumed to be zero. From
the last two equations in (3.6) we get
∂w0
φx =
∂x
∂w0
φy = (3.10)
∂y
and introducing them into the first three equations in (3.6) we get
∂u0 ∂ 2 w0
x (x, y, z) = −z = 0x + zκx
∂x ∂x2
∂v0 ∂ 2 w0
y (x, y, z) = −z = 0y + zκy
∂y ∂y 2
∂u0 ∂v0 ∂ 2 w0 0
γxy (x, y, z) = + − 2z = γxy + zκxy (3.11)
∂y ∂x ∂x∂y
Notice that the variables φx , φy have been eliminated and Kirchhoff theory only
uses three variables u0 (x, y), v0 (x, y), and w0 (x, y). This makes analytical solutions
easier to find, but numerically Kirchhoff theory is more complex to implement.
Since second derivatives of w0 are needed to write the strains, the weak form (2.30)
will have second derivatives of w0 . This will require that the interpolation functions
(see Section 2.1.4) have C 1 continuity. That is, the interpolation functions must
be such that not only the displacements but also the slopes be continuous across
element boundaries. In other words, both the displacement w0 and the slopes
∂w0 /∂x, ∂w0 /∂y will have to be identical at the boundary between elements when
calculated from either element sharing the boundary. This is difficult to implement.
Consider the case of beam bending. The ordinary differential equation (ODE)
with an applied distributed load qb(x) is
d4 w0
EI = qb(x) (3.12)
dx4
The weak form is obtained as in (2.3)

xB
d4 w0
Z
0= v −EI + qb(x) dx (3.13)
xA dx4
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 3.4: (a) Micromechanics, (b) lamina level, and (c) laminate level approach.
Integrating by parts twice

Z xB 2 Z xB
d v d2 w0 d3 w0 dv d 2 w0
0= 2
EI dx + vEI − EI − vb
q (x)dx
xA dx dx2 dx3 dx dx2 xA
xB Z xB
xB dv
0 = B(v, w0 ) + [vQx ]xA − Mx − vM
c(x)dx
dx xA x A
0 = B(v, w0 ) + L(v) (3.14)
When the elements are assembled as in Section 2.1.6, it turns out that adjacent
elements i and i + 1 that share a node have identical deflection but opposite shear
force Qx and bending moment Mx at their common node, as follows
wi = wi+1
Qi = −Qi+1
M i = −M i+1 (3.15)
For the shear forces to cancel as in (2.24), it is only required to have v i = v i+1 ,
which is satisfied by C 0 continuity elements having wi = wi+1 at the common
node. For the bending moments to cancel as in (2.24), it is required that dwi /dx =
dwi+1 /dx. This can only be done if the elements have C 1 continuity. That is, the
slopes dwi /dx = dwi+1 /dx must be identical at the common node. Such elements
are difficult to work with ( [12, page 276]).
In FSDT theory, only first derivatives are used in the strains (3.6). So, the weak
form (2.30) has only first derivatives and, like (2.24), all the internal generalized
forces cancel at common nodes with only C 0 element continuity.
3.2 Finite Element Analysis of Laminates

Deformation and stress analysis of laminated composites can be done at different
levels (Figure 3.4). The level of detail necessary for description of the material
depends on the level of post-processing desired.
When a great level of detail is necessary (Figure 3.4.a), the strain and stress
are computed at the constituent level, i.e., fiber and matrix. In this case, it is
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
necessary to describe the microstructure, including the fiber shape and geometrical
distribution, and the material properties of the constituents. More details are given
in Chapter 6 where micromechanical modeling is used to generate properties for any
combination of fibers and matrix. Also, when the composite material is a woven
fabric, or the laminate is very thick, or when studying localized phenomena such as
free edges effects (Chapter 5), the composite should be analyzed as solid, as shown
in Chapter 5. However, it must be noted that most of the laminated structures can
be analyzed using the plates and shell simplifications explained in Section 3.1.
At the other end of the spectrum (Figure 3.4.c), the composite material can
be considered as a homogeneous equivalent material. In this case, its structural
behavior can be analyzed by using orthotropic properties shown in Chapter 1. If
the whole laminate is analyzed as a homogeneous equivalent shell, using the macro-
scale level approach (Figure 3.4.c), the stress distribution in the laminate cannot
be obtained. However, this very simple description of the laminate is sufficient
when only displacements, buckling loads and modes, or vibration frequencies and
modes are required. In these cases, only the laminate stiffness (3.8) is needed (see
Section 3.2.5). In certain cases, even a simpler material description will suffice. For
example, when the laminate is only unidirectional, or if the laminate is balanced
and symmetric (see [1, Section 6.3]), the laminate can be modeled as a single lamina
of orthotropic material (Section 3.2.6).
In most cases, stress and strains need to be calculated for every lamina in the
laminate. Then, the actual laminate stacking sequence (LSS) must be input to the
program (Section 3.2.7). In this case, the elastic properties of each lamina, as well
as thickness and fiber orientation of every lamina must be given. This method is
usually called the mesoscale level approach (Figure 3.4.b).
A unidirectional lamina can be satisfactorily approximated as being transversely
isotropic. Then, it suffices to use E3 = E2 , and G23 = E3 /2(1 + ν23 ) in the equa-
tions for an orthotropic material. The elastic properties of a unidirectional lamina
can be computed using micromechanics (Chapter 6) or with experimental data of
unidirectional laminates. Material properties of some unidirectional composites are
shown in Table 3.1.
In the analysis of most composite structures, it is usual to avoid the microme-
chanics approach and to obtain experimentally the properties of the unidirectional
lamina, or even the whole laminate. However, the experimental approach is not
ideal because a change of constituents or fiber volume fraction during the design
process invalidates all the material data and requires a new experimental program
for the new material. It is better to calculate the elastic properties of the lamina
using micromechanics formulas, using software such as [9] (see also Section 6.1).
Unfortunately, micromechanics formulas are not accurate to predict strength, so
experimental work cannot be ruled out completely.
In summary, laminate properties can be specified in two ways:
– by the constitutive matrices A, B, D, and H, or
– by specifying the laminate stacking sequence (LSS) and properties for every
lamina.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Table 3.1: Material properties of unidirectional carbon/epoxy composites

Property2 Unit AS4D/9310 T300/5208
E1 [GPa] 133.86 136.00
E2 = E3 [GPa] 7.706 9.80
G12 = G13 [GPa] 4.306 4.70
G23 [GPa] 2.76 4.261
ν12 = ν13 0.301 0.280
ν23 0.396 0.150
Vf 0.55
ρ [g/cm3 ] 1.52 1.54
α1 [10−6 /◦ C] 0.32
α2 [10−6 /◦ C] 25.89
F1t [MPa] 1830 1550
F1c [MPa] 1096 1090
F2t = F3t [MPa] 57 59
F2c = F3c [MPa] 228 207
F4 [MPa] 141 128
F6 [MPa] 71 75
When the constitutive matrices A, B, D, H, of the laminate are used to define

the laminate, the shell element cannot distinguish between different laminas. It can
only relate generalized forces and moments to generalized strains and curvatures.
On the other hand, laminated shell elements have the capability to compute the
laminate properties using the Laminate Stacking Sequence (LSS) and the laminas
properties.
3.2.1 Element Types

In commercial finite element codes, various plates and shells theories are imple-
mented and differentiated by element types, called shell elements [13, 14]. The
main shell elements available in ANSYS R
for the analysis of composite materi-
als/structures are shown in Table 2.1.
Shell elements allow one to model thin to moderately thick shells, down to a
side-to-thickness ratio of 10. While some of them have 3 or 4 nodes, others have 8
nodes, thus using interpolation functions of higher degree. Shell elements are defined
in 3D space and have 5 or 6 degrees of freedom (DOF) at each node (translations
in the nodal x, y, and z directions and rotations about the nodal x, y, and z axis).
The 6th DOF (rotation about the z axis) is included in the shell formulation to
allow modeling of folded plates, but it would not be necessary if the shell surface is
smooth.
Modeling of different types of laminates with various levels of detail is explained
in the next few sections.
2
F4 was calculated with [1, Eq. 4.109] assuming the angle of the fracture plane to be α0 = 54◦ .
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
3.2.2 Sandwich Shells

For a sandwich shell, the core is much softer that the faces, and the transverse shear
deformations are significant regardless of the total thickness of the shell. Conven-
tional shell elements may not have the shear flexibility required. In ANSYS, ele-
ments SHELL181 and SHELL281 model the transverse shear deflection using an
energy equivalence method, which is recommended for sandwich shells. See Exam-
ple 3.7.
3.2.3 Nodes and Curvature

Typically, four node elements are flat, but since R14.0, ANSYS SHELL181 is capable
of recovering the initial shell curvature from the mesh. For curved shells, it is better
to use eight node elements such as SHELL281. The degenerate triangular option
of SHELL181 is not recommended but can be used if needed as filler elements if
the mesh generation algorithm requires it. However, SHELL281 is acceptable when
used as a triangle.
3.2.4 Drilling Rotation

Conventional shell elements are based on shell theory, which constrains the 3D con-
tinuum deformations according to some kinematic assumptions, such as Kirchhoff,
FSDT, or one of many others. In continuum theory, the deformation of a point in
3D can be described in terms of the relative displacements of two points, thus re-
quiring six degrees of freedom (three per point). Both Kirchhoff and FSDT theories
formally reduce the requirement to three displacements and two in-plane rotations
at a single point. The two rotations φx , φy , are rotations of the normal to the ref-
erence surface. These are called in-plane rotations because the rotation vectors lie
on the surface of the shell. For a smoothly curved or flat shell, there is no need
of tracking the rotation of the normal around itself φz . This last rotation is called
drilling rotation. However, if the shell has a fold, an in-plane rotation on one side
of the fold corresponds to a drilling rotation on the other side (Figure 3.5). Thus,
to enforce compatibility of displacements, the drilling rotations become necessary.
Since elements having 5 DOF do not have drilling rotation, they are suited to model
smoothly curved shells but not folded shells.
3.2.5 A-B-D-H Input Data for Laminate FEA

As previously mentioned, macroscale level (laminate level) analysis is adequate if
only deflections, modal analysis, or buckling analysis are to be performed, with no
requirement for detailed stress analysis. Then it is not necessary to specify the
laminate stacking sequence (LSS), the thickness, and the elastic properties of each
lamina of the laminate. Only the elastic laminate properties (A, B, D, H, matrices)
defined in (3.9) are required. This is convenient because it allows one to input
the aggregate composite material behavior with few parameters. The reduction of
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
(1)
fz(1)
(1)
B y
(2)
x
1
fy(1)
fx(1) A
fx(2) fz(2)
(1)
x
(2)
2
fy
(2)
y
(1)
Figure 3.5: Exploded view of a plate folded along AB. Drilling rotations φz and
(2)
φz associated to elements 1 and 2, respectively.
the complexity of the input data allows modeling of laminates with an unlimited
number of laminas, using only four matrices.
When the A, B, D, H, matrices are used to define the finite element analysis,
the computer model knows the correct stiffness but it does not know the LSS.
Therefore, the software can compute the deformation response (including buckling
and vibrations) and even the strain distribution through the thickness of the shell,
but it cannot compute the stress components because it does not know where the
lamina material properties change from lamina to lamina.
The A, B, D, H input data can be found by using (3.9) or [9]. Then, these are
input into the finite element software, as illustrated in Example 3.1.
Example 3.1 Consider a simply supported square plate ax = ay = 2000 mm, thickness
t = 10 mm laminated with AS4D/9310 (Table 3.1) in a [0/90]n configuration. Tabulate
the center deflection perpendicular to the plate surface when the number of laminas is n =
1, 5, 10, 15, 20. The plate is loaded in compression with and edge load Nx = −1 N/mm and
(Ny = Nxy = Mx = My = Mxy = 0). Use symmetry to model 1/4 of the plate. Generate
the A, B, D, and H matrices and enter them into ANSYS.
Solution to Example 3.1 Due to the symmetry of the plate, only 1/4 of it will be mod-
eled. The matrices A, B, D, H, are calculated using (3.9), which are implemented in CADEC
[9]. In ANSYS, SHELL181 and SHELL281 allow data input in the form of A, B, D, H, ma-
trices using preintegrated general shell section via SECTYPE,1,GENS. Then, the laminate
stiffness matrices must be placed in the input file using four matrices called A, B, D, E, as
follows
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
SSPA,711563,23328.8,0,711563,0,43600 ! A11,A21,A31,A22,A32,A33
SSPB,-1.58515e+006,0,0,1.58515e+006,0,0 ! B11,B21,B31,B22,B32,B33
SSPD,5.92969e+006,194407,0,5.92969e+006,0,363333 ! D11,D21,D31,D22,D32,D33
SSPE,29666.7,0,29666.7 ! H11,H21,H22
SLIST ! List A,B,D,H submatrices
In this case the input data does not need material properties. The complete input file,
which is available on the Web site [5], is listed below for n = 1. See Appendix C for an
introduction to the software interface.
Note that in ANSYS a pressure applied on the boundary acts on the element; that is,
opposite to the outside normal to the boundary and thus a compressive edge load in this case
is applied as SFL,2,PRES,1.0.
/TITLE,Simply Supported [0/90] Plate Nx=1 N/mm - SHELL281

! Material is AS4D/9310 - laminate [0/90]n, n=1
/UNITS,MPA ! Units are in mm, MPa, and Newtons
/PREP7 ! Pre-processor module
! This input data does not need material properties
ET,1,SHELL281 ! Chooses SHELL281 element for analysis

SECTYPE,1,GENS ! Use preintegrated general shell section (ABDH matrix)
! ABDH matrix definition

SSPA,711563,23328.8,0,711563,0,43600 ! A11,A21,A31,A22,A32,A33
SSPB,-1.58515e+006,0,0,1.58515e+006,0,0 ! B11,B21,B31,B22,B32,B33
SSPD,5.92969e+006,194407,0,5.92969e+006,0,363333 ! D11,D21,D31,D22,D32,D33
SSPE,29666.7,0,29666.7 ! H11,H21,H22
SLIST ! List A,B,D,H submatrices
! Geometry and mesh

RECTNG,0,1000,0,1000 ! Creates a rectangle with x=1 m and y=1 m
ESIZE,100 ! Element size 100 mm
AMESH,all ! Mesh the area

DL,2,1,uz,0 ! Impose Simple Supported BC
DL,3,1,uz,0
DL,1,1,symm ! Impose Symmetry BC
DL,4,1,symm
!d,all,rotz ! Constraint rotations about z axes (optional)
SFL,2,PRES,1.0 ! Apply uniform pressure in N/mm
SOLVE ! Solve current load state

/POST1 ! Post-processor module

PLDISP ! Display deformed shape
PLNSOL,u,z ! Display contour of displacements z
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
!plesol,s,x ! Notice this cannot be done, try it

The solution is tabulated in Table 3.2, using PLDISP. Bending extension coupling pro-
duces a lateral deflection, which diminishes as the number of laminas grow.
Table 3.2: Lateral deflection vs. number of laminas in Example 3.1

n δ [mm]
1 0.2191
5 0.0211
10 0.0104
15 0.0069
20 0.0052
3.2.6 Equivalent Orthotropic Input for Laminate FEA

Some FEA codes do not have laminated elements and do not accept the A, B, D,
and H matrices as explained in Section 3.2.5. However, if they have orthotropic
elements, it is still possible to perform deformation, vibration, and buckling analysis
for laminated composites, as it is shown in this section.
Unidirectional Laminate FEA

Standard shell elements, even if they are not laminated, can be used to model a
unidirectional laminate and still obtain correct results of displacements, strains, and
stress. The geometry of shells is a surface that represents the mid-surface of the
real shell, located halfway through the thickness. The positive thickness coordinate
points along a normal to the shell mid-surface (local z -direction which coincides
with the 3-direction). This is the normal definition of shells and it is used in shell
elements, as shown in Example 3.2.
Example 3.2 Use ANSYS to model a simply supported rectangular plate with dimensions
ax = 4000 mm, ay = 2000 mm, thickness t = 10 mm. Apply a uniform transverse load
q0 = 0.12 × 10−3 M P a. The material is a unidirectional laminate AS4D/9310 (Table 3.1),
with the fibers oriented in the x-direction. Determine the deflection of the center point of
the plate. This example is continued in Example 3.8.
Solution to Example 3.2 The thickness coordinate is eliminated from the governing equa-
tions so that the 3D problem simplifies to 2D. In the process, the thickness becomes a pa-
rameter, which is known and supplied to the modeling software. Most software packages
differentiate between material properties and parameters but both are supplied as known in-
put data. For example, the shell thickness is supplied to ANSYS as section data (SECDATA),
while material properties are entered separately with UIMP.
SHELL281, an 8-node shell element, is used in this example. Symmetry with respect to
the x-z and y-z planes is used to model 1/4 of the plate. The APDL file commands, also
available in [5, FEAcomp Ex302.inp], are shown below
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
/TITLE,Simply Supported Plate under q=0.12e-3 MPa - SHELL281

! Material is UD AS4D/3100 Carbon/Epoxy - 8 laminas 1.25 mm thick

! Material properties FOR AS4D/9110 orthotropic laminate
UIMP,1,EX,EY,EZ,133.86E3,7.706E3,7.706E3
UIMP,1,GXY,GYZ,GXZ,4.306E3,2.76E3,4.306E3
UIMP,1,PRXY,PRYZ,PRXZ,0.301,0.396,0.301

SECTYPE,1,SHELL
SECDATA,10,1,0 ! Thickness=10mm, Material 1, Layer 0 deg
! Geometry and mesh

AMESH,ALL ! Mesh the area

DL,3,1,uz,0
DL,4,1,symm
SFA,ALL,2,PRES,1.2E-4, ! Apply uniform pressure in MPa


PLDISP,1 ! plots displaced plate
PLNSOL,u,z ! contour plot of z direction displacements
PLESOL,s,x ! contour plot of x direction stress
The maximum deflection is 17.47 mm at the center of the plate, using PLDISP,1.
Symmetric Laminate FEA

If a multidirectional laminate is balanced and symmetric, the apparent laminate
orthotropic properties can be found as explained in Section 1.15. The apparent
laminate properties represent the stiffness of an equivalent (fictitious) orthotropic
plate that behaves like the actual laminate under in-plane loads. These apparent
properties should not be used to predict bending response. When the only important
response is bending, e.g., a thick cantilever plate under bending, the formulation
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
shown in [1, (6.33)] should be used to obtain the apparent laminate properties.
However, in most structural designs using composite shells, the laminate works
under in-plane loads and the formulation in Section 1.15 should be used.
If the laminate is symmetric but not balanced, the axes of orthotropy are rotated
with respect to the laminate coordinate system, but still the laminate is equivalent to
an orthotropic material as per Section 1.15. For example, a unidirectional laminate
oriented at an angle θ with respect to global axes, should be modeled on a coordinate
system oriented along the fiber direction (see Section 3.2.10).
Example 3.3 Use ANSYS to model a simply supported rectangular plate with dimensions
ax = 2, 000 mm, ay = 2, 000 mm, for a laminate [±45/0]S . Apply a tensile edge load
Nx = 200 N/mm. Find the maximum horizontal displacement. Each lamina is 1 mm thick
with the following properties
E1 = 37.88 GP a G12 = 3.405 GP a ν12 = 0.299

E2 = 9.407 GP a G23 = 3.308 GP a ν23 = 0.422
Solution to Example 3.3 Since the laminate is balanced symmetric, compute the aver-
aged laminate properties Ex , Ey , and so on using Section 1.15, for example using the fol-
lowing MATLAB R
code (available in [5, Ex 3 3.m])
function Ex33
% Example 3.3 laminate stiffness as per section 1.15
clc
% G23 = E2/2/(1+nu23)
% Cprime (1.92) calculated with
% http://www.cadec-online.com/Chapters/Chapter5/3DConstitutiveEquations
% /StiffnessMatrix.aspx
E1 = 37.88E3;
E2 = 9.407E3; E3 = E2;
G12 = 3.405E3; G13 = G12;
nu12= 0.299; nu13= nu12;
nu23= 0.422;
G23 = E2/2/(1+nu23);
% (1.104)
Sprime = [
1/E1, -nu12/E1, -nu13/E1,0,0,0;
-nu12/E1, 1/E2, -nu23/E2,0,0,0;
-nu13/E1, -nu23/E2, 1/E3,0,0,0;
0,0,0,1/G23,0,0;
0,0,0,0,1/G13,0;
0,0,0,0,0,1/G12];
Cprime = Sprime^-1;
theta = [45,-45,0];
thickness = [1,1,1];
laminateThickness = sum(thickness);
C = zeros(6);
for i=1:length(theta);
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
[T,Tbar] = RotationMatrix3D(theta(i));
C = C + Tbar*Cprime*Tbar*thickness(i)/laminateThickness; % (1.102)
end
S = C^-1; % (1.103)
Ex =1/S(1,1) % (1.105)
Ey =1/S(2,2)
Ez =1/S(3,3)
Gxy =1/S(6,6)
Gyz =1/S(4,4)
Gxz =1/S(5,5)
PRxy=-S(2,1)/S(1,1)
PRyz=-S(3,2)/S(2,2)
PRxz=-S(3,1)/S(1,1)
end
The function RotationMatrix3D is given in Example 1.6. Then, the MATLAB code
yields
Ex = 13.825 GP a Gxy = 24.637 GP a νxy = 0.832

Ey = 8.007 GP a Gyz = 3.340 GP a νyz = 0.235
Ez = 10.220 GP a Gxz = 3.372 GP a νxz = 0.069
The APDL command file is shown below and also available in [5, FEAcomp Ex303].
Note that in ANSYS a pressure applied on the boundary acts on the element; that is, opposite
to the outside normal to the boundary and thus a tensile load in this case is applied as
SFL,2,PRES,-200.0.
/TITLE,Simply Supported Plate Nx=200 N/mm - equivalent [45/-45/0]s

! [45/-45/0]s laminate of E-Glass/Vinyl, vf=0.5 and th=1*6 mm
! quarter plate with symmetry conditions

! Equivalent orthotropic material properties for the LAMINATE
UIMP,1, EX, EY, EZ, 13.825E3, 8.007E3, 10.220E3
UIMP,1, GXY, GYZ, GXZ, 24.636E3, 3.340E3, 3.372E3
UIMP,1, PRXY, PRYZ, PRXZ, 0.832, 0.232, 0.069
ET,1,SHELL281 ! Chooses Shell 281 element for analysis

SECTYPE,1,SHELL ! Section shell set #1
SECDATA,1,1,45 ! 1st layer: mat. #1, 45 deg, Th=1.0 mm
SECDATA,1,1,-45 ! 2nd layer: mat. #1, -45 deg, Th=1.0 mm
SECDATA,1,1,0 ! 3rd layer: mat. #1, 0 deg, Th=1.0 mm
SECDATA,1,1,0 ! Same layers in symmetrical order
SECDATA,1,1,-45
SECDATA,1,1,45
! Geometry and mesh

/PNUM,LINE,1
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
LPLOT

DL,3,1,uz,0
DL,4,1,symm
!D,ALL,ROTZ ! Constraint rotations about z axes (optional)
SFL,2,PRES,-200 ! Apply uniform linear load in N/mm


/VIEW,,1,1,1
!PLDISP,2 ! Plots displaced plate
PLNSOL,u,x ! Contour plot of x direction displacements
!plesol,s,x ! Notice the stress results are incorrect
The resulting maximum horizontal displacement on a quarter-plate model is 0.996 mm

at the edge of the plate. The planes x = 0 and y = 0 are not symmetry planes for a
[±45/0]S but once the laminate is represented by equivalent orthotropic properties, as is
done in this example, the lack of symmetry at the lamina level is lost and it does not have
any effect on the mid-surface displacements. Therefore, one-quarter of the plate represents
well the entire plate as long as no stress analysis is performed. Furthermore, displacement
and mid-surface strain analysis can be done with the laminate replaced by an equivalent
orthotropic material. However, even if the full plate were to be modeled, the stress values in
the equivalent orthotropic material are not the actual stress values of the laminate. While
the material analyzed in this example is not homogeneous, but laminated, the material in
Example 3.2 is a homogeneous unidirectional material. Therefore, the stress values are not
correct in this example but they are correct in Example 3.2.
Asymmetric Laminate FEA
If the laminate is not symmetric, bending-extension coupling must be considered.

Strictly speaking, such material is not orthotropic and should not be modeled with
an equivalent laminate material. Even then, if only orthotropic shell elements are
available and the bending-extension coupling effects are not severe, the material
could be approximated by an orthotropic material by neglecting the matrices B
and D. The ratios defined in [1, (6.37)–(6.38)] can be used to assess the quality of
the approximation obtained using apparent elastic properties. Care must be taken
for unbalanced laminates that the A and H matrix are formulated in a coordinate
system coinciding with the axes of orthotropy of the laminate.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
3.2.7 LSS for Multidirectional Laminate FEA

For computation of strain and stress at the mesoscale (lamina level), it is necessary
to know the description of the laminate and the properties of each lamina. The
description of the multidirectional laminate includes the LSS, which specifies the
angle of each lamina with respect to the x -axis of the laminate, the thickness, and
the elastic material properties of each lamina. Then, the software computes the
matrices A, B, D, and H internally. In this way, the software can compute the
stress components in each lamina. This approach is illustrated in the following
example.
Example 3.4 Consider a simply supported square plate ax = ay = 2000 mm, t = 10 mm

thick, laminated with AS4D/9310 (Table 3.1) in a [0/90/±45]S symmetric laminate config-
uration. The plate is loaded with a tensile load Nx = 100 N/mm and (Ny = Nxy = Mx =
My = Mxy = 0). Compute maximum displacement at the edge of the plate.
Solution to Example 3.4 The problem can be solved using shell elements SHELL181 or
SHELL281. Other than the element selection, the models are identical. Note the LSS is
given starting at lamina #1 at the bottom. Although a quarter model with symmetry could
be used, a full plate is modeled to show how to constrain the model from experiencing rigid
body motion.
/TITLE,Simply Supported [0/90/45/-45]s - uniform load - SHELL181

! Material is AS4D/9310 - [0/90/45/-45]s, Th=1.25 mm per lamina

UIMP,1,EX,EY,EZ,133.86E3,7.706E3,7.706E3

KEYOPT,1,3,2 ! Set KEYOPT(3)=2, Full integration
! (recommended for SHELL 181/composites)
KEYOPT,1,8,1 ! Set KEYOPT(8)=1, Store data for all laminas
SECTYPE,1,SHELL,,La1 ! Section shell set #1, [0/90/45/-45]s, label=La1
SECDATA, 1.25,1,0.0,3 ! 1st lamina: mat. #1, 0 deg, Th=1.25 mm
SECDATA, 1.25,1,90,3 ! 2nd lamina: mat. #1, 90 deg, Th=1.25 mm
SECDATA, 1.25,1,45,3 ! 3nd lamina: mat. #1, +45 deg, Th=1.25 mm
SECDATA, 1.25,1,-45,3 ! 4rt lamina: mat. #1, -45 deg, Th=1.25 mm
SECDATA, 1.25,1,-45,3 ! Same laminas in symmetrical order
SECDATA, 1.25,1,45,3
SECDATA, 1.25,1,90,3
SECDATA, 1.25,1,0.0,3
SECOFFSET,MID ! Nodes on the laminate middle thickness
! Geometry and mesh

RECTNG,-1000,1000,-1000,1000 ! Creates a rectangle with x=2 m and y=2 m
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i


DL,3,1,uz,0
DL,1,1,uz,0
DL,4,1,uz,0
CEN_NODE=NODE(0,0,0) ! Center node
D,CEN_NODE,UX ! Constraint nodes to avoid rigid body motion
LEFT_NODE=NODE(-1000,0,0) ! Middle node in left edge
D,LEFT_NODE,UY
RIGH_NODE=NODE(1000,0,0) ! Middle node in right edge
D,RIGH_NODE,UY
SFL,2,PRES,-100 ! Apply uniform edge pressure in N/mm

SFL,4,PRES,-100

PLDISP,1 ! Display the deformed plate
The maximum displacement at the edge of the plate is 0.206 mm.

Note that SECOFFSET,MID is used to place the nodes on the mid-surface. The deforma-
tion of the lamina is in-plane only because the laminate is symmetric, balanced, and in-plane
loaded. The distributed edge load is applied at the mid-surface. However, the same model is
re-executed with SECOFFSET,BOT (nodes on bottom face) or with SECOFFSET,TOP (nodes on
top face), the edge load will be applied on the bottom edge or on the top edge of the laminate.
In these cases, out-plane deformations will appear.
Other model definition aspects that can be controlled include: the position of the bottom
and top surfaces of the laminate (i.e., the direction of the vector normal to the surface of
the shell), the relative position of the shell surface through the real laminate thickness (at
the bottom, at the middle, or at the top), the orientation of laminate reference axis (as is
shown in Section 3.2.10), and so on.
3.2.8 FEA of Ply Drop-Off Laminates

Sometimes it is convenient to set the reference surface at the bottom (or top) of the
shell. One such case is when the laminate has ply drop-offs, as shown in Figure 3.6.
When the design calls for a reduction of laminate thickness, plies can be gradually
terminated from the thick to the thin part of the shell. As a rule of thumb, ply
drop-off should be limited to a 1:16 to 1:20 ratio (T h : L ratio in Figure 3.6) unless
detailed analysis and/or testing supports a steeper drop-off ratio. For this case, it
is convenient to specify the geometry of the smooth surface, or tool surface.
Then, every time a ply or set of plies is dropped, the material and thickness for
those elements is changed. This is illustrated in the next examples. Not all software
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 3.6: Ply drop-off of length (L) and thickness (T h) and finite element model
simplifications.
has this capability and it may be necessary to assume that the mid-surface is smooth
while in reality only the tool surface is smooth. As long as the thickness is small
compared to the other two dimensions of the structure, such assumption is unlikely
to have a dramatic effect in the results of a global analysis, such as deformation,
buckling, and even membrane stress analysis. The exact description of the thickness
geometry begins to play a role when detailed 3D stress analysis of the ply drop-off
region is required, but at that point, a 3D local model is more adequate.
Example 3.5 A ply drop-off is defined between the laminate A, [90/0]S , and the laminate
B, [90/0]. The ply drop-off ratio is 1:20 The lamina thickness is 0.75 mm. Consider a
composite strip 120 mm long and 100 mm wide under tension N = 10 N/mm applied to
the bottom edges on the strip. Use symmetry to model 1/2 of the tape. Visualize and report
the maximum transverse deflection.
Solution to Example 3.5 Using shell elements SHELL181, three different sections are
defined, one for A, one for B, and one section to model the ply drop-off between them.
The thickness of the drop-off is 0.75 × 2 = 1.5 mm. With a ply drop-off ratio 1:20, the
length of the ply drop-off is 1.5 × 20 = 30 mm. Every 30 mm there is a section change.
The bottom lamina is designated as lamina #1, and additional laminas are stacked from
bottom to top in the positive normal direction of the element coordinate system. The APDL
code is shown below and also available in [5, FEAcomp Ex305.inp].
/TITLE,Tape with Ply Drop-off between [90/0]s and [90/0]

! Material is AS4D/9310 - Th=0.75 mm per lamina - SHELL181

UIMP,1,EX,EY,EZ,133.86E3,7.706E3,7.706E3
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i

! (recommended for SHELL181/composites)
SECTYPE,1,SHELL,,A ! Section shell set #1, [90/0]s, A
SECDATA, 0.75,1,90,3 ! 1st lamina: mat. #1, 90 deg, Th=0.75 mm
SECDATA, 0.75,1,0,3 ! 3rd lamina: mat. #1, 0 deg, Th=0.75 mm
SECDATA, 0.75,1,90,3 ! 4th lamina: mat. #1, 90 deg, Th=0.75 mm
SECOFFSET,BOT ! Nodes on the laminate BOTTOM thickness
SECTYPE,2,SHELL,,DROP ! Section shell set #2, [90/0/0], DROP

SECTYPE,3,SHELL,,B ! Section shell set #2, [90/0], B

! Geometry and mesh

RECTNG,0,60,0,50 ! Laminate A x=60 mm and y=50 mm
RECTNG,60,(60+30),0,50 ! Laminate Drop x=15 mm and y=50 mm
RECTNG,(60+30),(120),0,50 ! Laminate B x=60 m and y=50 mm
AGLUE,all ! Glue all areas
SECNUM,1
AMESH,1 ! Mesh the area number 1
SECNUM,2
SECNUM,3

DL,4,1,all,0 ! Impose clamped BC
DL,13,4,symm
DL,15,5,symm
SFL,10,PRES,-10 ! Apply uniform pressure in N/mm

i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 3.7: Laminate with dropped laminas.
The maximum transverse deflection is 2.579 mm.

When an inner lamina extends over only part of the geometry, it is convenient
that the remaining laminas maintain their numbering through the entire model.
Otherwise, if the continuity in numbering of laminas is lost, the post-processing
and result visualization will be extremely difficult. A model with a few laminas,
some of which are dropped over part of the laminate, is shown in Figure 3.7.
Example 3.6 Define the three different sections of the laminate shown in Figure 3.7. The
laminate in section A is a [+45/ − 45/0/90/0]. The thickness of each lamina is 1.2 mm.
Solution to Example 3.6 Using shell elements SHELL181, the different sections are de-
fined as shown in the APDL code shown below [5, FEAcomp Ex306]. The bottom lamina is
designated as lamina #1, and additional laminas are stacked from bottom to top in the pos-
itive normal direction of the element coordinate system. The dropped laminas are modeled
using zero thickness in order to maintain continuous numbering of the remaining laminas.
! Material is AS4D/9310 - [+45/-45/0/90/0]s, Th=1.2 mm per lamina

/UNITS,MPa ! Units are in mm, MPa, and Newtons

UIMP,1,EX,EY,EZ,133.86E3,7.706E3,7.706E3

! (recommended for SHELL181/composites)
SECTYPE,1,SHELL,,A ! Section shell set #1, section A
SECDATA, 1.2,1,45,3 ! 1st lamina: mat. #1, +45 deg, Th=1.2 mm
SECDATA, 1.2,1,-45,3 ! 2nd lamina: mat. #1, -45 deg, Th=1.2 mm
SECOFFSET,BOT ! Nodes on the laminate bottom thickness
SECTYPE,1,SHELL,,B ! Section shell set #1, section B

SECDATA, 0 ,1,0,3 ! 3rd lamina: Thk=0 mm, do not compute
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i

SECTYPE,1,SHELL,,C ! Section shell set #1, section C

SECDATA, 0 ,1,0,3 ! 3rd lamina: Thk=0 mm, do not compute
SECDATA, 0 ,1,90,3 ! 4th lamina: Thk=0 mm, do not compute
3.2.9 FEA of Sandwich Shells

Some laminates can be considered sandwich when specifically designed for sandwich
construction with thin faceplates and a thick, relatively weak, core. The faceplates
are intended to carry all, or almost all, of the bending and in-plane normal load.
Conversely, the core is assumed to carry all of the transverse shear. Example 3.7
shows how to define and calculate a sandwich cantilever beam.
The following assumptions are customarily made for a sandwich shell:
– The terms Hij in (3.9) depend only on the middle lamina (core) and they can
be calculated as
∗
Hij = Qij tcore ; i, j = 4, 5 (3.16)
core
– The transverse shear moduli (G23 and G13 ) are set to zero for the top and
bottom laminas (face laminas).
– The transverse shear strains and stresses in the face laminas are neglected or
assumed to be zero.
– The transverse shear strains and shear stresses in the core are assumed con-
stant through the thickness.
Example 3.7 Calculate the maximum deflection of a sandwich cantilever beam subject to
an end load Fz = −100. The beam is made of a sandwich of two outer aluminum plates
(with thickness 1 mm each, E = 69 GP a, ν = 0.3) and an inner core of foam (with thickness
50 mm, E = 300 M P a, ν = 0.1).
Solution to Example 3.7 ANSYS element SHELL281 is used in this example. Another
element with capabilities to analyze sandwich structures is SHELL181. Both elements model
the transverse shear deflection using an energy equivalence method, which is recommended
for sandwich shells. The APDL code is shown below and in [5, FEAcomp Ex307].
/TITLE, Cantilever Beam with Sandwich [Al/foam/Al] material

! Material is Aluminium (Th=1mm) and FOAM (Th=50 mm) - SHELL281

! Material definition
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
MP,EX,1,69e3 ! Aluminium Young’s modulus

MP,NUXY,1,0.3 ! Aluminium Poisson ratio
MP,EX,2,300 ! Foam Young’s modulus
MP,NUXY,2,0.1 ! Foam Poisson ratio

KEYOPT,1,8,1 ! Set KEYOPT(8)=1, Store data for all laminas
SECDATA,1,1,0 ! 1st lamina: mat. #1, 0 deg, Th=1.0 mm
SECDATA,50,2,0 ! 2nd lamina: mat. #2, 0 deg, Th=50.0 mm
SECDATA,1,1,0 ! 3rd lamina: mat. #1, 0 deg, Th=1.0 mm
! Geometry and mesh

RECTNG,0,3e3,0,600 ! Creates a rectangle with x=3 m and y=600 mm

DL,4,1,all,0 ! Impose Clamped BC
NSEL,S,LOC,x,3e3
CP,1,UZ,ALL ! Coupling DOF set, vertical displacement
NSEL,R,LOC,y,300
F,all,FZ,-100 ! Apply force in a end line node
NSEL,all

PLDISP,1
The maximum displacement is 16.0 mm.
3.2.10 Element Coordinate System

In the pre-processor, during the definition of the laminate, it is very important to
know the orientation of the laminate coordinate system. Material properties, the
relative lamina orientation with respect to the laminate axis, and other parameters
and properties are defined in the laminate coordinate system, unless specified oth-
erwise. Also, it can be used to obtain the derived results (strains and stress) in
these directions. In FEA, the laminate coordinate system is associated to the ele-
ment coordinate system, with a unique right-handed orthogonal system associated
to each element.
The element coordinate system orientation is associated with the element type.
For bar or beam elements the orientation of the x-axis is generally along the line
defined by the end nodes of the element. For solid elements in two and three
dimensions, the orientation is typically defined parallel to the global coordinate
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
system. For shell elements this is not useful. Axes x and y need to be defined
on the element surface, with the z-axis always normal to the surface. The default
orientation of x and y axes depends on the commercial code and the element type.
There are various ways to define the default orientation of x and y in shell
elements (see Figure 3.8).
In Figure 3.8.(a) the x-axis is aligned with the edge defined by the first and
second nodes of each element, the z-axis normal to the shell surface (with the
outward direction determined by the right-hand rule), and the y-axis perpendicular
to the x- and z-axis.
MSC-MARCTM calculates the orientation of the x-axis from the lines defined
by the middle points of the edges as shown in Figure 3.8(b).
In ANSYS, Figure 3.8.(b), the orientation of the xe axis is aligned with the shell
surface coordinate at the center of the element, connecting the mid-sides of edges
i − l and j − k, and formally defined as
∂{x}/∂s
xe = (3.17)
|∂{x}/∂s|
Xn
{x} = hi (s, r){x}i
i=1
where {x} are the nodal coordinates in the global coordinate system; s, r, are the
coordinates of the isoparametric master element, and hi are the shape functions [12].
In Abaqus TM , Figure 3.8.(c), the local x-direction is calculated projecting the
global X -direction onto the surface of the element, i.e.,
î = (Iˆ − î∗ )/kIˆ − î∗ k

k̂ = n̂
ĵ = k̂ × î (3.18)
with î∗ = (Iˆ · n̂)n̂ being the projection of the global x -direction, I,
ˆ along the normal
ˆ ◦
of the element, n̂. If the global X -direction, I, is within 0.1 of the normal to the
element, n̂, i.e., they are almost parallel, the local x -direction is calculated using the
global z -direction, K̂, instead of I. ˆ The local coordinate system can be redefined
using the procedures described in the Abaqus documentation [14, Section 2.2.5].
If no additional rotation is specified, the laminate coordinate system coincides
with the element coordinate system. Additional rotations are specified by ESYS in
ANSYS and CSYS in Abaqus, as shown in Examples 3.8 and 3.9.
In Examples 3.2–3.7, only rectangular plates with rectangular elements are ana-
lyzed. All of them have the first and the second node aligned with the global X-axis.
Therefore, the material axes have been chosen parallel to the global axis. But this
doesn’t need to be the case. Most commercial codes have utilities to change the
element coordinate system. Example 3.8 illustrates how to change the element co-
ordinate system orientation in a plate. Example 3.9 illustrates how it can be done
in a shell with curvature. Example 3.10 illustrates how different orientations can
be used in different locations of the structure.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Z ze ye
node l Z
ze node k node l node k
X Y
X Y
ye xe
node i xe node j node i node j

(a) ^
n (b)
^
K
node l ^
^
J k
^
I ^
^
j
node k
i
node i ^
I
node j (c)
Figure 3.8: Default orientations of element coordinate systems in shell elements:

(a) Along side 1, (b) ANSYS, (c) Abaqus.
Example 3.8 Use a local coordinate system to model the plate of Example 3.2 if the or-
thotropic material is rotated +30 degrees with respect the x-direction. This example contin-
ues in Example 3.14.
Solution to Example 3.8 This example illustrates the use of local coordinate system in a
rectangular plate. In ANSYS a local coordinate system is defined using LOCAL commands,
which can be Cartesian, cylindrical, or spherical. Then, each element is linked to a pre-
viously defined local coordinate system using the element property ESYS. The objective is
to orient the x-axis element coordinate system parallel to the x-axis laminate coordinate
system. Also, it is possible to define element coordinate system orientations by user written
subroutines [15]. Element coordinate systems may be displayed as a triad with the /PSYMB
command or as an ESYS number (if specified) with the /PNUM command.
Although this example is a continuation of Example 3.2, the material is no longer sym-
metric so we cannot reuse FEAcomp Ex302.inp and we must construct a new full model.
The APDL file commands, also available in [5, FEAcomp Ex308 full plate.inp], are shown
below
/TITLE,SIMPLY SUPPORTED PLATE UNDER UNIFORM LOAD Q=1.2E-4 MPA - SHELL281

! FULL PLATE
/UNITS,MPA ! UNITS ARE IN MM, MPA, AND NEWTONS
/PREP7 ! PRE-PROCESSOR MODULE
! MATERIAL PROPERTIES FOR AS4D/9310 ORTHOTROPIC LAMINATE
UIMP,1,EX,EY,EZ,133.86E3,7.706E3,7.706E3
ET,1,SHELL281 ! CHOOSES SHELL281 ELEMENT FOR ANALYSIS
SECTYPE,1,SHELL ! SECTION SHELL SET #1
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
SECDATA,10,1,0 ! THICKNESS=10MM, MATERIAL 1, 0-DEG LAYER
! GEOMETRY AND MESH

LOCAL,11,0,,,,30,0,0 ! DEFINE LOCAL COORD. SYSTEM, XYROT=30 DEG
ESYS,11 ! SET COORD. SYSTEM FOR ELEMENTS MESHED
RECTNG,-2000,2000,-1000,1000 ! RECTANGLE WITH X=4 M AND Y=2 M
ESIZE,250 ! ELEMENT SIZE 250 MM
AMESH,ALL ! MESH THE AREA
CSYS,0 ! GO BACK TO DEFAULT COORD. SYSTEM
/PSYMB,ESYS,1 ! SET ON DISPLAY LAMINATE ORIENTATION
EPLOT ! DISPLAY ELEMENTS
FINISH ! EXIT PRE-PROCESSOR MODULE
/SOLU ! SOLUTION MODULE

DL,2,1,UZ,0 ! IMPOSE SIMPLE SUPPORTED BC, SS1
DL,2,1,UY,0
DL,2,1,ROTX,0
DL,4,1,UY,0
DL,4,1,ROTX,0
DL,1,1,UX,0
DL,1,1,ROTY,0
DL,3,1,UX,0
DL,3,1,ROTY,0
D,ALL,ROTZ ! CONSTRAINT ROTATIONS ABOUT Z AXES (OPTIONAL)
SFA,ALL,2,PRES,1.2E-4 ! APPLY UNIFORM PRESSURE IN MPA
SOLVE ! SOLVE CURRENT LOAD STATE
FINISH ! EXIT SOLUTION MODULE
/POST1 ! POST-PROCESSOR MODULE

PLDISP,1 ! PLOTS DEFORMED PLATE
RSYS,SOLU ! ACTIVATE RESULTS IN SOLUTION COOD. SYSTEM
PLESOL,S,X ! CONTOUR PLOT STRESS 1(FIBER), LAYER2, TOP FACE
FINISH ! EXIT POST-PROCESSOR MODULE
The maximum deflection of the model is 8.448 mm.
Example 3.9 Align the laminate coordinate system with the global Y-axis of a 3D curved
shell.
Solution to Example 3.9 For the analysis of laminated composite shells, it is very im-
portant to define clearly a reference direction with respect to which one can specify the fiber
direction of each lamina. One way of doing it, is to force the laminate coordinate system
(c.s.) to be aligned with the projection of one of the global axes on the surface of the shell.
For shells defined in 3D, the ESYS orientation uses the projection of the local system on
the shell surface. The element x-axis is determined from the projection of the local x-axis
on the shell surface. The z-axis is determined normal to the shell surface (with the outward
direction determined by the right-hand rule), and the y-axis perpendicular to the x- and
z-axis. For elements without mid-side nodes (linear interpolation functions), the projection
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
is evaluated at the element centroid and it is assumed constant in direction throughout

the element. For elements with mid-side nodes (quadratic interpolation functions), the
projection is evaluated at each integration point and may vary in direction throughout the
element. See the APDL file listed below (available in [5, FEAcomp Ex309]) to align the
element x-axis with the global Y-axis. See Figure 3.9.
/TITLE,Curved surface - SHELL281

!Material is AS4D/9310 Carbon/Epoxy [90/45/-45]T - Th=1.05mm per lamina
!units are in mm, MPa, and Newtons

UIMP,1,EX,EY,EZ,133.86E3,7.706E3,7.706E3

SECDATA, 1.05,1,90 ! 1st lamina: mat. #1, 90 deg, Th=1.05 mm
SECDATA, 1.05,1,45 ! 2nd lamina: mat. #1, 45 deg, Th=1.05 mm
SECDATA, 1.05,1,-45 ! 3nd lamina: mat. #1, -45 deg, Th=1.05 mm
! Create geometry by solid modeling

K,1,300,0,135
K,2,0,0,235
K,3,-300,0,135
K,4,200,200,0
K,5,0,200,135
K,6,-200,200,0
L,1,4
L,3,6
BSPLIN,1,2,3
BSPLIN,4,5,6
AL,ALL
! Mesh geometry
LOCAL,11,0,,,,90,0,0 ! Define local coord. system, XYrot=90 deg
ESYS,11 ! Set coord. system for elements meshed
AMESH,1 ! Mesh the area
CSYS,0 ! Go back to default coord. system
/PSYMB,esys,1 ! Set on display laminate orientation
/TYPE,1,0 ! Not hidden surfaces
EPLOT ! Display elements
Example 3.10 Model in ANSYS a flanged tube with axial and radial laminate orientation.
In the cylindrical part, the reference axis will be in the longitudinal direction. In the flange,
the reference axis will be radial (see Figure 3.10).
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 3.9: Curved shell in Example 3.9.
Figure 3.10: Reference axis in a flange tube.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Solution to Example 3.10 Different orientation systems are needed for different model
locations. Therefore, two local reference axes are defined and activated using the ESYS
command. The APDL file listed below aligns the elements on the cylinder in the axial
direction and aligns the elements on the flange in the radial direction [5, FEAcomp Ex310].
See Figure 3.11.
/TITLE,Flange tube with axial and radial laminate orientation - SHELL 281
! Material is AS4D/3100 Carbon/Epoxy [0/45/45] - Th=1.05 mm per lamina
! Units are in mm, MPa, and Newtons

UIMP,1,EX,EY,EZ,133.86E3,7.706E3,7.706E3

SECDATA,1.05,1,0 ! 1st lamina: mat. #1, 0 deg, Th=1.05 mm
SECDATA,1.05,1,45 ! 2nd lamina: mat. #1, 45 deg, Th=1.05 mm
SECDATA,1.05,1,-45 ! 3rd lamina: mat. #1, -45 deg, Th=1.05 mm
! Create geometry by solid modeling

CYL4,0,0,350,,,,300
CYL4,0,0,350,,550
AGLUE,3,4,5 ! Glue areas, area 5 become area 6
! Mesh geometry
LOCAL,11,0,,,,0,0,90 ! Define rotation=90 deg around Y (cylinder)
LOCAL,12,1,,,,0,0,0 ! Define polar coordinate system (flange)
AMESH,3,4 ! Mesh the cylindrical areas (areas 3 and 4)
AMESH,6 ! Mesh the flange area (area 6)
CSYS,0 ! Go back to default coord. system
/PSYMB,esys,1 ! Set on display laminate orientation
/TYPE,1,0 ! Not hidden surfaces
EPLOT ! Display elements
3.2.11 Constraints
Constraints are used to restrict the motion of portions of a model in a variety of
ways, as summarized below:
Tie Constraint tie two regions regardless of the meshes on the two regions being
identical or not. See Example 6.3.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 3.11: Flange in Example 3.10.
Rigid Body Constraint couples the motion of regions in an assembly to the mo-
tion of a reference point. Relative positions of regions that make up the rigid
body will remain unchanged during the analysis. With a rigid body con-
straint, pin nodes makes all displacements equal to those of the reference
point, and tie nodes makes all displacements and rotations equal to those
of the reference point. Rigid body constraints are very useful to force pinned
boundary conditions as in Example 3.11.
Coupling Constraint couples the motion of a surface to the motion of a single

point. Although a mesh of 3D solids has only 3 DOF per node, the reference
point will have 6 DOF. Therefore, it will be necessary to constraint the rota-
tions of the reference point (RP) if the intent is to have the coupled surface
translate but not rotate as a rigid surface.
Multipoint Constraint (MPC) couples the motion of selected slave nodes to

the motion of a single point.
Shell-to-Solid Coupling Constraint couples the motion of a shell edge to the

motion of an adjacent solid face. This is useful for detailed modeling of small
regions using solid elements while large portions of the model are idealized
with shell elements.
Embedded Region Constraint embeds a (small) region of a model within a host

region of a (larger) model. This is useful for global-local analysis, multiscale
analysis, and multiphysics analysis.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Example 3.11 Create a finite element model for a pultruded composite column under axial
compression load P = 11, 452 N [16] and calculate the end axial displacement u(L/2), where
x = 0 is located at the half-length of the column. The column is simply supported (pinned)
at both ends x = (−L/2, L2). Its length is L = 1.816 m. The cross-section of the column
is that of a wide-flange I-beam (also called H-beam) with equal outside height and width,
H = W = 304.8 mm. The thickness of both the flange and the web is tf = tw = 12.7 mm.
The material properties are given by the A, B, D, and H matrices, with units [mm M P a],
[mm2 M P a], [mm3 M P a], and [mm M P a], respectively. For the flange:
   
335,053 47,658 0 -29,251 -1,154 0
[A] =  47,658 146,155 0  ; [B] =  -1,154 -5,262 0 
0 0 49,984 0 0 -2,274
 
4,261,183 686,071 0
34,216 0
[D] =  686,071 2,023,742 0  ; [H] =
0 31,190
0 0 677,544
For the web:

   
338,016 44,127 0 -6,088 -14,698 0
[A] =  44,127 143,646 0  ; [B] =  -14,698 -6,088 0 
0 0 49,997 0 0 0
 
4,769,538 650,127 0
34,654 0
[D] =  650,127 2,155,470 0  ; [H] =
0 31,623
0 0 739,467
This example continues in Example 4.4.
Solution to Example 3.11 The APDL file is shown next [5, FEAcomp Ex311.inp]. A
number of modeling techniques are illustrated, which are very useful for FEA of composite
structures. Solid modeling is based on areas, and the use of lines to effectively impose
boundary conditions and to control the mesh refinement is illustrated. Only one-half of the
length of the column is modeled using symmetry boundary conditions. The loaded end is
constrained to move as a rigid body using CERIG so that the pinned boundary condition is
properly simulated.
The model is set up parametrically so that all the geometric parameters of the column,
such as the length, as well as mesh refinement can be easily changed. Only displacements
can be displayed because the model is set up with A-B-D-H matrices, but the .inp file can be
easily modified to enter the LSS along with lamina material properties as in Example 3.4.
LOCAL and ESYS coordinate systems are used so that all the local and the element coordinate
systems are oriented similarly; this is necessary to facilitate the specification of directionally
dependent materials properties and also interpretation of stress and strain results.
/TITLE, H-COLUMN, Ref: CST 58 (1998) 1335-1341 SHELL181 - ANSYS R14

!Ref: COMPOSITE SCIENCE AND TECHNOLOGY 58 (1998) 1335-1341
/UNITS,MPA !UNITS ARE mm, MPa, Newtons
WPSTYLE,,,,,,,,1 !WORKPLANE VISIBLE
/VSCALE,1,2.5,0 !2.5 LARGER ARROWS
!ARROW & TRIAD COLORS: WHITE=X, GREEN=Y, BLUE=Z
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
/TRIAD,LBOT !MOVE COORDINATE LABELS TO LEFT-BOTTOM

/VIEW,,1,2,3 !OBLIQUE VIEW
/PNUM,KP,0 !THESE ARE ALL THE NUMBERING OPTIONS
/PNUM,LINE,0 !ENTITY NUMBERING OFF=0, ON=1
/PNUM,AREA,1
/PNUM,VOLU,0
/PNUM,NODE,0
/PNUM,TABN,0
/PNUM,SVAL,0
/NUMBER,0
/PBC,ALL,,1 !THESE ARE ALL THE BC DISPLAY OPTIONS

/PBC,NFOR,,0 !DISPLAY ALL APPLIED BC, OFF=0, ON=1
/PBC,NMOM,,0
/PBC,RFOR,,0
/PBC,RMOM,,0
/PBC,PATH,,0
/PSYMB,CS,1 !THESE ARE ALL THE SYMBOL DISPLAY OPTIONS

/PSYMB,NDIR,0
/PSYMB,ESYS,0
/PSYMB,LDIV,1 !SHOW LESIZE ON LINES
/PSYMB,LDIR,1 !SHOW LDIR TO DECIDE ON LESIZE BIAS
/PSYMB,ADIR,0
/PSYMB,ECON,0
/PSYMB,XNODE,0
/PSYMB,DOT,1
/PSYMB,PCONV,
/PSYMB,LAYR,0
/PSYMB,FBCS,0
/PREP7 !ENTER THE PREPROCESSOR

!DEFINE PARAMETRIC VALUES===================
LOAD=11452 !APPLIED LOAD [N]
L2=1816/2 !COLUMN HALF LENGTH [mm]
WW=304.8 !OUTER WEB WIDHT [mm]
FW=WW !OUTER FLANGE WIDTH
WT=12.7 !WEB THICKNESS [mm]
FT=WT !FLANGE THICKNESS
WW1=WW-FT !MID-PLANE WEB WIDTH
FW1=FW !MID-PLANE FLANGE WIDTH
WW2=WW1/2
FW2=FW1/2
CSAREA=2*FW*FT+(WW-FT)*WT !CROSS-SECTION AREA
NELEN=10 !NUMBER OF ELEMENTS ALONG THE LENGTH
NEWEB=4 !NUMBER OF ELEMENTS ON THE WEB
NEFLA=2 !NUMBER OF ELEMENTS ON 1/2 FLANGE
!
! (recommended in SHELL181/composites)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
!====================================================================
! Commands for Preintegrated Shell Section Data (A-B-D-H MATRICES)
! SSPA,a11,A21,A31,A22,A32,A33,Temperature Membrane stiffness
! SSPB,B11,B21,B31,B22,B32,B33,Temperature Coupling stiffness
! SSPD,D11,D21,D31,D22,D32,D33,Temperature Bending stiffness
! SSPE,H11,H21,H22,Temperature Transverse shear stiffness
!
!FLANGE [mm,Mpa,Newton]
SECTYPE,1,GENS ! Define SECTION 1 (FLANGE)
SSPA,335053,47658,0,146155,0,49984 ! A11,A21,A31,A22,A32,A33
SSPB,-29251,-1154,0,-5262,0,-2274 ! B11,B21,B31,B22,B32,B33
SSPD,4261183,686071,0,2023742,0,677544 ! D11,D21,D31,D22,D32,D33
SSPE,34216,0,31190 ! H11,H21,H22
!
! WEB [mm, Mpa, Newton]
SECTYPE,2,GENS ! Define SECTION 2 (WEB)
SSPA,338016,44127,0,143646,0,49997 ! A11,A21,A31,A22,A32,A33
SSPB,-6088,-14698,0,-6088,0,0 ! B11,B21,B31,B22,B32,B33
SSPD,4769538,650127,0,21554700,739467 ! D11,D21,D31,D22,D32,D33
SSPE,34654,0,31623 ! H11,H21,H22
!
!DEFINE SOLID MODEL USING AREAS======================
RECTNG,0,L2,-WW2,WW2, !WEB
WPAVE,0,WW2,0 !MOVE WORKPLANE TO TOP FLANGE
WPRO,,-90.000000, !ROTATE WORKPLANE AS FLANGE
RECTNG,0,L2,-FW2,FW2, !TOP FLANGE
WPAVE,0,-WW2,0
RECTNG,0,L2,-FW2,FW2, !BOTTOM FLANGE
AOVLAP,all !JOINS AREAS CREATING INTERSECTIONS IF NEEDED
NUMCMP,AREA !COMPRESS AREA NUMBERS
/REPLOT
WPSTYLE,,,,,,,,0 !HIDE WORKPLANE
!
LPLOT !PLOT LINES
LSEL,S,LOC,X,0 !SELECT SYMMETRY END
/REPLOT
/PBC,ALL,,1 !DISPLAY ALL APPLIED BC, OFF=0, ON=1
DL,ALL, ,SYMM !APPLY SYMMETRY BC
!NOTE SYMM DISPLAYED AS S ON LINES, BUT WILL NOT SHOW ON NODES LATER
ALLSEL,ALL !RESELECT ALL ENTITIES
/REPLOT
!
!MESHING============================================================
LOCAL,11,0,,,, , ,0 !DEFINE LOCAL COORD SYS TO ALIGN W/MAT PROPS
ESYS,11 !USE IT FOR ALL ELEMS
/VSCALE,1,2.5,0 !2.5 LARGER ARROWS
/PSYMB,ESYS,1 !DISPLAY IT
/PSYMB,ESYS,0 !DO NOT DISPLAY IT
/VSCALE,1,1.0,0 !RESET TO DEFAULT ARROW LENGTH
MSHAPE,0,2D !QUADRILATERAL 0, MESHING 2D
MSHKEY,1 !MAPPED MESHING 1 (FREE WOULD BE 0)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
!
LESIZE,ALL,,,NEFLA !ALL LINES DIVIDED IN NEFLA ELEMENTS
LESIZE, 2,,,NEWEB,0 !LINE 2, DIV NEWEB, NO BIAS, OVERRIDES PREVIOUS
LESIZE, 4,,,NEWEB
LESIZE, 1,,,NELEN,1/2 !LINE 1, DIV 10, BIAS 1/2 TOWARDS LINE END
LESIZE, 3,,,NELEN,2 !LINE 3, DIV 10, BIAS 2 TOWARDS LINE ORIGIN
LESIZE, 5,,,NELEN,1/2
LESIZE, 7,,,NELEN,2
LESIZE, 9,,,NELEN,1/2
LESIZE,11,,,NELEN,2
/PNUM,REAL,1 !COLOR AND NUMBER ELEMENTS BY REAL SET
ASEL,ALL !SELECT ALL AREAS
ASEL,S,,,1 !SELECT AREA 1 (WEB)
AATT,,2, !USE SECTION 2 FOR THE WEB
AMESH,ALL !MESH ALL AREAS CURRENTLY SELECTED (I.E., WEB)
ASEL,S,,,ALL !SELECT ALL
ASEL,U,,,1 !UNSELECT AREA 1 TO KEEP THE FLANGES
AATT,,1 !USE SECTION 1 FOR THE FLANGE
AMESH,ALL !MESH ALL AREAS CURRENTLY SELECTED (I.E., FLANGE)
ASEL,ALL
/PNUM,REAL,0 !SUPRESS NUMBERING
CSYS,0 !RETURN TO GLOBAL COORD SYSTEM TO DISPLAY STRESSES
!
NSEL,S,LOC,X,0 !PREVENT RIGID BODY TRANSLATION
NSEL,R,LOC,Y,0
D,ALL,UY,0
D,ALL,UZ,0
D,ALL,ROTX,0 !PREVENT RIGIT BODY TWIST
NSEL,ALL
!
NSEL,S,LOC,X,L2 !SELECT LOAD END
NSEL,R,LOC,Y,0 !SELECT CENTER NODE ONLY
/PNUM,NODE,1
NPLOT
*GET,MYNODE,NODE,,NUM,MIN !GET LABEL OF CENTER NODE
NSEL,S,LOC,X,L2 !SELECT LOAD END AGAIN
/PNUM,NODE,0 !TURN OFF NODE NUMBER DISPLAY
/REPLOT
! APPLY RIGID BC AT LOADED END
CERIG,MYNODE,ALL,UXYZ, , , , !MYNODE MASTER, ALL OTHER SLAVES
F,MYNODE,FX,-LOAD !APPLY COMPRESSION LOAD
!
ALLSEL,ALL !RESELECT EVERYTHING
FINISH !EXIT PREPROCESSOR
!
/SOLU !ENTER SOLUTION MODULE
ANTYPE,STATIC !STATIC ANALYSIS
RESCONTROL,LINEAR !CONTROL FILE WRITING FOR MULTIFRAME RESTARTS
SOLVE
FINISH !EXIT SOLUTION MODULE
!
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 3.12: Deformed and undeformed shapes of the H column in Example 3.11.
/POST1 !ENTER POSTPROCESSOR MODULE

/VIEW,,1,1,1 !ISOMETRIC VIEW
PLDISP,2 !PLOT DEFORMED SHAPE AND UNDEF OUTLINE
!FINISH
The axial displacement at the loaded end of the column is 0.036 mm. See Figure 3.12.
This example continues in Example 4.4.
3.3 Failure Criteria

Failure criteria are curve fits of experimental data that attempt to predict failure un-
der multiaxial stress based on experimental data obtained under uniaxial stress. All
failure criteria described in this section predict the first occurrence of failure in one
of the laminas but are unable to track failure propagation until complete laminate
failure. Damage mechanics is used in Chapters 8 and 9 to track damage evolution
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
up to laminate failure. The truncated-maximum-strain criterion estimates laminate

failure without tracking damage evolution by making certain approximations and
assumptions about the behavior of the laminate [1].
In this section, failure criteria are presented using the notion of failure index,
which is used for several FEA packages, and it is defined as
stress
IF = (3.19)
strength
Failure is predicted when IF ≥ 1. The strength ratio [1, Section 7.1.1] is the
inverse of the failure index
1 strength
R= = (3.20)
IF stress
Failure is predicted when R ≤ 1.
3.3.1 2D Failure Criteria

Strength-based failure criteria are commonly used in FEA to predict failure events
in composite structures. Numerous criteria exist for unidirectional (UD) laminas
subjected to a state of plane stress (σ3 = 0). The most commonly used are described
in [1]. They are:
– Maximum stress criterion

– Maximum strain criterion
– Truncated maximum strain criterion, and
– Interacting failure criterion
A few additional criteria are presented in this section.
Hashin Failure Criterion
The Hashin failure criterion (HFC) proposes four separate modes of failure:
– Fiber tension
– Fiber compression
– Matrix tension
– Matrix compression
that are predicted by four separate equations, as follows3 :
3
The Hashin failure equations on the right-hand side of (3.21–3.24) yield squares of failure
indexes IF . Compare to the Maximum Stress Criterion (i.e., IF f t = σ1 /F1T ) and so on.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
2 2
σ1 σ6
IF2 f t = +α if σ1 ≥ 0 (3.21)
F1t F6
2
σ1
IF2 f c = if σ1 < 0 (3.22)
F1c
2
σ6 2

σ2
IF2 mt = + if σ2 ≥ 0 (3.23)
F2t F6
" #
σ2 2 F2c 2
2
σ2 σ6
IF2 mc = + −1 + if σ2 < 0 (3.24)
2F4 2F4 F2c F6
where α is a weight factor to give more or less emphasis to the influence of shear on
fiber failure. With α = 0, Hashin failure criterion (FC) and Maximum Stress FC
would predict longitudinal tensile failure of the unidirectional lamina at the same
stress σ1 . Note (3.22) predicts longitudinal compressive failure without influence
of the shear stress, although it is well known that the state of shear has a strong
influence on the longitudinal compression failure [17].
Equations (3.21–3.24) define the square of failure indexes according to Hashin
FC. The values of F1c and F2c are considered positive throughout this textbook and
most of the literature. The Tsai-Hill, Azzi-Tsai-Hill, and Tsai-Wu failure criteria
are not recommended because they over emphasize the interaction between fiber
(σ1 ) and transverse matrix (σ2 ) damage modes.
Puck Failure Criterion

The Puck failure criterion [18] distinguishes between fiber failure (FF) and matrix
failure (MF). In the case of plane stress, the MF criteria discriminates three different
modes. Mode A is when transverse cracks appear in the lamina under transverse
tensile stress with or without in-plane shear stress. Mode B also denotes transverse
cracks, but in this case they appear under in-plane shear stress with small transverse
compression stress. Mode C indicates the onset of oblique cracks (typically with
an angle of 53◦ in carbon epoxy laminates) when the material is under significant
transverse compression.
The FF and the three MF modes yield separate failure indexes. The Puck
criterion assumes that FF only depends on longitudinal tension. Therefore, the
failure index for FF is defined as

σ1 /F1t if σ1 > 0
IF F = (3.25)
−σ1 /F1c if σ1 < 0
The MF failure indexes have different expressions depending on the mode that
becomes active. With positive transverse stress, mode A is active. In this case, the
failure index for matrix dominated tensile failure (mode A) is
s
σ6 2 F2t 2 σ2 2

σ2
IM F,A = + 1 − p6t + p6t if σ2 ≥ 0 (3.26)
F6 F6 F2t F6
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
where p6t is a fitting parameter. Lacking experimental values, it is assumed that

p6t = 0.3 [18].
Under negative transverse stress, either mode B or mode C is active, depending
on the relationship between in-plane shear stress and transverse shear stress. The
limit between mode B and C is defined by the relation F2A /F6A , where
"r #
F6 F2c
F2A = 1 + 2p6c −1 (3.27)
2p6c F6
p
F6A = F6 1 + 2p2c (3.28)
and p2c is defined as
F2A
p2c = p6c (3.29)
F6
and p6c is another fitting parameter. Lacking experimental values, it is assumed
that p6c = 0.2 [18].
Finally, the failure index for matrix dominated shear (mode B) is
(
σ2 < 0
q
1 2
IM F,B = σ62 + (p6c σ2 ) + p6c σ2 if σ2 F2A (3.30)
F6 σ6 ≤ F6A
and for matrix dominated compression (mode C) is

" 2 # (
F2c σ6 σ2 2 σ2 < 0
IM F,C = − + if σ2 F2A (3.31)
σ2 2(1 + p2c )F6 F2c σ6 ≥ F6A
3.3.2 3D Failure Criteria

Failure criteria presented here are 3D generalizations of the ones presented in [1,
Section 7.1]. The user of FEA packages should be careful because some packages
use only the in-plane stress components for the computation of the failure index
(e.g., Abaqus), even though all six stress components may be available from the
analysis. In those cases the intralaminar and thickness components of stress should
be evaluated separately to see if they lead to failure.
In this section, the numerical subscript denotes the directions of (1) fiber, (2)
in-plane transverse to the fibers, and (3) through the thickness of the lamina. The
letter subscript denotes (t) tensile and (c) compressive. Contracted notation is used
for the shear components as described in Section 1.5.
Maximum Strain Criterion

The failure index is defined as


 1 /1t if 1 > 0 or −1 /1c if 1 < 0



 2 /2t if 2 > 0 or −2 /2c if 2 < 0
3 /3t if 3 > 0 or −3 /3c if 3 < 0

IF = max (3.32)

 abs(γ4 )/γ4u
abs(γ5 )/γ5u





abs(γ6 )/γ6u
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
The quantities in the denominator are the ultimate strains of the unidirectional
lamina. Note that compression ultimate strains in (3.32) are positive numbers.
Maximum Stress Criterion

The failure index is defined as

 σ1 /F1t
 if σ1 > 0 or −σ1 /F1c if σ1 < 0



 σ2 /F2t if σ2 > 0 or −σ2 /F2c if σ2 < 0
σ3 /F3t if σ3 > 0 or −σ3 /F3c if σ3 < 0

IF = max (3.33)

 abs(σ4 )/F4
abs(σ5 )/F5





abs(σ6 )/F6
The letter F is used here to denote a strength value for a unidirectional lamina
as in [19]. Note that compression strength in (3.33) are positive numbers.
Tsai-Wu Criterion
Using the Tsai-Wu criterion the failure index is defined as
 s −1
2
1 B B 1
IF = = − + +  (3.34)
R 2A 2A A
with
σ12 σ22 σ32 σ2 σ2 σ2
A= + + + 42 + 52 + 62
F1t F1c F2t F2c F3t F3c F4 F5 F6
σ2 σ3 σ1 σ3 σ1 σ2
+ c4 √ + c5 √ + c6 √ (3.35)
F2t F2c F3t F3c F1t F1c F3t F3c F1t F1c F2t F2c
and
−1 −1 −1 −1 −1 −1

B = F1t − F1c σ1 + F2t − F2c σ2 + F3t − F3c σ3 (3.36)
where ci , i = 4..6, are the Tsai-Wu coupling coefficients, that by default are taken
to be −1. Note that compression strength in (3.35) and (3.36) are here positive
numbers.
The through-the-thickness strength values F3t and F3c are seldom available in
the open literature, so it is common practice to use the corresponding in-plane
transverse values of strength. Also, the intralaminar strength F5 is commonly as-
sumed to be equal to the in-plane shear strength. Lacking experimental data for
the remaining intralaminar strength F4 , it can be estimated as the shear strength
of the matrix.
Example 3.12 Compute the failure index IF in each lamina of Example 3.4 using the
maximum stress failure criterion and the Tsai-Wu criterion. The lamina strength values
are given in Table 3.1, p. 73. Determine the strength ratio of the laminate using both
criteria.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Solution to Example 3.12 Once the model is solved, inside /POST1 module, the FC
commands (FC, FCDELE, FCLIST, etc.) can be used to define the failure criteria param-
eters. Before defining the failure criteria parameters, ANSYS must show the solution in
“results” coordinate system using RSYS, SOLU command. The LAYER command is used
to select the lamina where the failure criterion is to be calculated.
After Example 3.4, include the commands below to compute the IF of each lamina.
Note that in ANSYS compression strength must be introduced using negative numbers. If
the compression strength value is not given, ANSYS takes the compression strength equal to
the negative value of the tensile strength. Also note that ANSYS uses x, y, z to denote the
lamina coordinates that are denoted as 1, 2, 3, in this textbook.
/TITLE,Simply Supported [0/90/45/-45]s - uniform load - SHELL281

! Material is AS4D/9310 - [0/90/45/-45]s, Th=1.25 mm per lamina

uimp,1,ex,ey,ez,133.86E3,7.706E3,7.706E3
uimp,1,gxy,gyz,gxz,4.306E3,2.76E3,4.306E3
uimp,1,prxy,pryz,prxz,0.301,0.396,0.301

KEYOPT,1,8,1 ! Set KEYOPT(8)=1, Store data for all layers
SECDATA, 1.25,1,0 ! 1st layer: mat. #1, 0 deg, Th=1.25 mm
SECDATA, 1.25,1,90 ! 2nd layer: mat. #1, 90 deg, Th=1.25 mm
SECDATA, 1.25,1,45 ! 3nd layer: mat. #1, +45 deg, Th=1.25 mm
SECDATA, 1.25,1,-45 ! 4rt layer: mat. #1, -45 deg, Th=1.25 mm
SECDATA, 1.25,1,-45 ! Same layers in symmetrical order
SECDATA, 1.25,1,45
SECDATA, 1.25,1,90
SECDATA, 1.25,1,0
SECOFFSET,MID ! NODES ON THE LAMINATE MIDLLE THICKNESS
! Geometry and mesh


DL,3,1,uz,0
DL,4,1,symm
SFL,2,PRES,-100 ! Apply uniform pressure in N/mm
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
/post1 ! Post-processor module

RSYS,SOLU ! Activate the ESYS
!Failure criteria definition

FC,1,s,xten, 1830 ! F1t strength
FC,1,s,xcmp,-1096 ! F1c strength
FC,1,s,yten, 57 ! F2t strength
FC,1,s,ycmp,-228 ! F2c strength
FC,1,s,zten, 1e6 ! F3c=-F3t strength (large value so it does not compute)
FC,1,s,xy,71 ! F6 strength
FC,1,s,yz,1e6 ! F4 strength (large value, not compute)
FC,1,s,xz,1e6 ! F5 strength (large value, not compute)
FC,1,s,XYCP,-1 ! c6 coefficient. Defaults to -1.0
FC,1,s,YZCP,-1 ! c4 coefficient. Defaults to -1.0
FC,1,s,XZCP,-1 ! c5 coefficient. Defaults to -1.0
LAYER, 1 ! Select layer #1

PRNSOL,S,FAIL ! Print table with FAIL index, where:
! MAXF is index failure for Maximum Stress
! TWSI is index failure for Tsai-Wu criteria
! Repeat this with the others layers

!LAYER, 2 ! Select layer #2
!PRNSOL,S,FAIL ! Print table with FAIL index
!FINISH ! End Post-process module
The MAXF and the TWSR are the failure index defined in Eqs. (3.33) and (3.34) respec-
tively. The TWSI, called Tsai-Wu “strength index,” is the addition of the value A in Eq.
(3.35) and value B in Eq. (3.36), i.e., T W SI = A + B. This “index” does not have
engineering interpretation and we recommend not to use it.
The solution is tabulated in Table 3.3, showing the failure indexes and the strength ratios
obtained for maximum stress criterion and Tsai-Wu criterion in each lamina.
Table 3.3: Failure indexes and strength ratios for each lamina in Example 3.12
Layer Maximum Stress Tsai-Wu
IF R IF R
#1, 0o 0.0144 69.34 0.0144 69.38
#2, 90o 0.0243 41.16 0.0294 34.04
#3, +45o 0.0157 63.84 0.0199 50.18
#4, −45 o 0.0157 63.84 0.0199 50.18
Example 3.13 Compute the 2D Tsai-Wu failure index IF in each lamina of Example 3.4
using APDL language. The lamina strength values are given in Table 3.1.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Solution to Example 3.13 First, run FEAcomp_Ex304_shell181.inp to calculate dis-

placements, strain, and stress in all laminas. Then, the computation of Tsai-Wu failure
index IF is achieved by running TSAIWU2D.mac. Both are avaliable in [5, FEAcomp Ex313].
Using the APDL scripting language of ANSYS [20], it is possible to automate common
tasks (macros) or build parametric models. To create macros, a set of commands can be
saved in a text file using extension .mac in the ANSYS working directory. Then, these
commands can be executed by using the name of the .mac file.
After executing FEAcomp_Ex304_shell181.inp, you should manually enter the follow-
ing APDL commands at the ANSYS Command Prompt located at the top of the graphical
user interface (GUI), immediately below the Menu Bar (File Select List...).

RSYS,SOLU ! activate the solution reference axes
! REPEAT FOR ALL LAYERS
LAYER,1 ! select the lamina
TSAIWU2D ! execute the macro
PRNSOL,EPSW ! PRINT FAILURE INDEX IN A LIST
When you are done, exit the post-processor:
FINISH
In this example, the commands to compute, print, and plot the 2D Tsai-Wu failure
index are saved in a file named TSAIWU2D.mac, reproduced below:
! Macro: Tsai-Wu failure criterion using APDL macro language

! Tested with SHELL181 (2013)
! Define parameters
F1t= 1830 ! F1t strength
F1c= 1096 ! F1c strength
F2t= 57 ! F2t strength
F2c= 228 ! F2c strength
F6 = 71 ! F6 strength
c6 = -1 ! Tsai-Wu coefficient
! initialize arrays
*get,nelem,elem,,num,max ! get number of elements
*get,nnode,node,,num,max ! get number of nodes
*set,I_F, ! delete array if already used

*set,sel, ! delete array if already used
*dim,I_F,,nnode ! set up array for element nodes
*dim,sel,,nnode ! set up array for select vector
nsle,s,corner ! select only nodes at element corners

*vget,sel(1),node,1,nsel ! mask for compute only corners
! compute Tsai-Wu failure criterion

*do,in,1,nnode
*if,sel(in),gt,0,then ! read only selected nodes
*get,s_1,node,in,s,x ! get stress each node
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
*get,s_2,node,in,s,y
*get,s_6,node,in,s,xy
A1= s_1**2/(F1t*F1c)
A2= s_2**2/(F2t*F2c)
A6= s_6**2/(F6)**2
A12= c6*s_1*s_2/(F1t*F1c*F2t*F2c)**0.5
A = A1+A2+A6+A12
B= (1/F1t-1/F1c)*s_1+(1/F2t-1/F2c)*s_2
R_tw=-B/(2*A)+((B/2/A)**2+1/A)**0.5
If_tw=1/R_tw
I_F(in)=If_tw
*endif
*enddo
*vput,I_F,node,1,epsw, ! write failure index in results database

! end macro
The solution is tabulated in Table 3.4, showing the failure indexes obtained by using the
Tsai-Wu criterion in each lamina.
Table 3.4: Failure indexes and strength ratios for each lamina in Example 3.13
Layer IF R
#1, 0 o 0.0144 69.44
#2, 90o 0.0294 34.01
#3, +45 o 0.0200 50.00
#4, −45o 0.0200 50.00
Example 3.14 Compute the Tsai-Wu failure index IF on each lamina of a quasi-isotropic
laminate [0/90/ ± 45]S , otherwise identical to Example 3.8, using a USERMAT subroutine
(usermatps 314.f90 for shell elements). The lamina strength values are given in Table
3.1, p. 73.
Solution to Example 3.14 See user material subroutine usermatps 314.f90 and model
file FEAcomp Ex314.inp on the Web site [5]. Refer to Appendix C for program compilation
and execution details.
First, follow the instructions in Appendix C.1.1 to create a USERMATLib.DLL. For this
particular example, copy usermatps 314.f90 to c: \Ansys\ User\ usermatps. f90 . Note
the change of file name. At this stage, probably you are overwriting a usermatps.f90 that
was in your work directory. If so, make sure you keep a backup.
Next, double click on AnsUserMatEjb.bat to create the DLL.
Next, open ANSYS/Mechanical, then File>Read Input from>FEAcomp\ Ex314. inp
available in [5] and shown next.
/TITLE, SS FULL PLATE, UNIFORM LOAD Q=1.2E-4 MPA, SHELL181/281

! FULL PLATE
!FROM EX.3.8: MATERIAL PROPERTIES FOR AS4D/9310 ORTHOTROPIC LAMINATE
!UIMP,1,EX,EY,EZ,133.86E3,7.706E3,7.706E3
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
!UIMP,1,GXY,GYZ,GXZ,4.306E3,2.76E3,4.306E3
!UIMP,1,PRXY,PRYZ,PRXZ,0.301,0.396,0.301
! 12 PROPERTIES AS FOLLOWS
! E1 E2 NU12 NU23 G12 G23
! F1T F1C F2T F2C F6 C6
TBDELETE,USER,1,,, ! DELETE USER PROPERTIES IN CASE WERE DEFINED BEFORE
! A TB,USER, CARD IMMEDIATELY FORCES ANSYS TO LOOK FOR A USERMAT
TB,USER,1,1,12, ! 12 PROPERTIES FOR MATERIAL 1, 1 TEMPERATURE
TBTEMP,0. ! ZERO TEMPERATURE
TBDATA,,133.86E3,7.706E3,0.301,0.396,4.306E3,2.76E3
TBDATA,,1830,1096,57,228,71,-1
TB,STAT,1,,2, ! 2 STATE VARIABLES FOR MATERIAL 1
TBDATA,1,0.,0.,,,, ! INITIALIZE THE STATE VARIABLES TO ZERO
ET,1,SHELL181,,,2 ! FOR SHELL 181 SET KEYOPT(3)=2, FULL INTEGRATION

!CHOICE !ET,1,SHELL281,,, ! FOR SHELL 281 DEFAULT REDUCED INTEGRATION
KEYOPT,1,8,1 ! SET KEYOPT(8)=1, STORAGE DATA: ALL LAYERS
SECTYPE,1,SHELL ! SECTION SHELL SET #1
SECDATA, 1.25,1,0.0,3 ! 1ST LAYER: MAT. #1, 0 DEG, TH=1.25 MM
SECDATA, 1.25,1,90,3 ! 2ND LAYER: MAT. #1, 90 DEG, TH=1.25 MM
SECDATA, 1.25,1,45,3 ! 3ND LAYER: MAT. #1, +45 DEG, TH=1.25 MM
SECDATA, 1.25,1,-45,3 ! 4RT LAYER: MAT. #1, -45 DEG, TH=1.25 MM
SECDATA, 1.25,1,-45,3 ! SAME LAYERS IN SYMMETRICAL ORDER
SECDATA, 1.25,1,45,3
SECDATA, 1.25,1,90,3
SECDATA, 1.25,1,0.0,3
SECOFFSET,MID ! NODES ON THE LAMINATE MIDDLE THICKNESS
! GEOMETRY AND MESH

LOCAL,11,0,,,,30,0,0 ! DEFINE LOCAL COORD. SYSTEM, XYROT=30 DEG
ESYS,11 ! SET COORD. SYSTEM FOR ELEMENTS MESHED
RECTNG,-2000,2000,-1000,1000 ! RECTANGLE WITH X=4 M AND Y=2 M
ESIZE,250 ! ELEMENT SIZE 250 MM
CSYS,0 ! GO BACK TO DEFAULT COORD. SYSTEM
/PSYMB,ESYS,1 ! SET ON DISPLAY LAMINATE ORIENTATION
EPLOT ! DISPLAY ELEMENTS

DL,2,1,UY,0
DL,2,1,ROTX,0
DL,4,1,UY,0
DL,4,1,ROTX,0
DL,1,1,UX,0
DL,1,1,ROTY,0
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
DL,3,1,UX,0
DL,3,1,ROTY,0
D,ALL,ROTZ ! CONSTRAINT ROTATIONS ABOUT Z AXES (OPTIONAL)
SFA,ALL,2,PRES,1.2E-4 ! APPLY UNIFORM PRESSURE IN MPA
ANTYPE,STATIC ! SET STATIC ANALYSIS

OUTRES,SVAR,1 ! STORE STATE VARIABLES

PLDISP,1 ! PLOTS DISPLACED PLATE
/GRA,FULL
RSYS,SOLU ! ACTIVATE RESULTS IN SOLUTION COOD. SYSTEM
LAYER,2
PLESOL,SVAR,1 ! CONTOUR PLOT STRESS STATE VARIABLE 1
Suggested Problems
Problem 3.1 Compute the maximum bending moment per unit cross-sectional area mu
that can be applied to a beam of circular hollow cross-section of outside radius ro and
inner radius ri . The loading is pure bending, no shear. The material is homogeneous and
failure occurs when the maximum stress reaches the strength σu of the material. The hollow
section is filled with foam to prevent buckling. Derive an expression for the efficiency of the
cross-section as the ratio of mu of the hollow beam by mu of a solid rod of same outside
radius. Faced with the problem of using a strong and relatively expensive material, would
you recommend a small or large radius?
Problem 3.2 Compute the maximum outside radius for a cantilever beam of length L,
loaded by a tip load P, otherwise similar to the beam in Problem 3.1 but subjected to pure
shear loading. The shear strength is τu = σu /2. Consider only shear. Buckling of the thin
wall is likely to limit further the practical thickness of the wall.
Problem 3.3 Compute the maximum deflection per unit volume δV that can be applied to
a beam of circular hollow cross-section of outside radius ro and inner radius ri . This is
a cantilever beam of length L, loaded by a tip load P . The hollow section is made of an
homogeneous material with moduli E and G = E/2.5, filled with foam to prevent buckling.
Derive an expression for the efficiency of the cross-section as the ratio of δV between the
hollow cross-section and a solid rod of the same outside radius. Faced with the problem of
using a relatively expensive and not quite stiff material, would you recommend a small or
large radius?
Problem 3.4 Write a computer program to evaluate (3.9). The program data input is the
LSS, the thickness of the laminas, and the material elastic properties. The output should be
written in a file. Show all work in a report.
Problem 3.5 Using the program of Problem 3.4 compute the A, B, D, and H matrices for
the following laminates. The material is AS4D/9310 and all laminas are 0.85 mm thick.
Comment on the coupling of the constitutive equations for each case: (a) one lamina [0], (b)
one lamina [30], (c) [0/90]2 , (d) [0/90]s , (e) [0/90]8 , (f ) [±45]2 = [+45/ − 45/ + 45/ − 45],
(g) [±45]s = [+45/ − 45/ − 45/ + 45], (h) [±45/0/90/ ± 30]. Show all work in a report.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Problem 3.6 (FEA) Compute the value and location of the absolute maximum transverse
shear strain γ23 in Example 3.2. At that location plot the distribution of γ23 through the
thickness of the plate. Is that distribution a reasonable answer?
Problem 3.7 (FEA) Recompute Example 3.2 with a doubly sinusoidal load
q(x, y) = q0 sin(πx/2a) sin(πx/2b)
where 2a, 2b are the plate dimensions in x and y, respectively. Compare the result with the
exact solution at the center of the plate, that is
w0 = 16q0 b4 /[π 4 (D11 s4 + 2(D12 + 2D66 )s2 + D22 )]
where s = b/a [10, (5.2.8–5.2.10)].
Problem 3.8 (FEA) Calculate the first vibration frequency $11 of the plate with the ana-
2
lytical solution $mn = π 4 [D11 m4 s4 + 2(D12 + 2D66 )m2 n2 s2 + D22 n4 ]/(16ρhb4 ), where ρ, h
are the density and thickness of the plate, respectively ( [10, (5.7.8)]).
Problem 3.9 (FEA) Using ANSYS finite element code, generate a rectangular plate with
ax = 1000 mm and by = 100 mm. The laminas are made of AS4D/9310 (Table 3.1) 1.2 mm
thick. Look up four different LSS laminates where appear: (a) bending extension coupling
effect, (b) thermal expansion coupling effect, (c) torsion extension coupling effect, and (d)
shear extension (these coupling effects are shown in [1, Figure 6.7]). Model (i) one half of
the plate, 500 × 100 mm, and (ii) one quarter of the plate, 500 × 50 mm, applying symmetry
conditions and report when it is correct or not to use each of these reduced models. Show
all work in a report.
Problem 3.10 Using a program (e.g., MATLAB) to plot the failure limits (with If = 1)
of maximum stress, Tsai-Wu, and Puck failure criteria in the plane σ1 − σ2 , and in the
plane σ2 − σ6 .
Problem 3.11 (FEA) Compute the failure index IF at the center point on each lamina of
Example 3.12 using the maximum stress failure criterion and the Tsai-Wu failure criterion.
The lamina strength values are given in Table 3.1. (a) Calculate the failure indexes using
the FC commands in ANSYS, (b) Write the nodal stress results at the top and bottom of
each lamina in a file. Then, using an external program (e.g., MATLAB) compute the same
failure indexes as in part (a) and compare them at the center of the plate. Show all work in
a report.
Problem 3.12 Compute the failure index IF on each lamina of Example 3.12 using the
Puck failure criterion. The lamina strength values are given in Table 3.1. Calculate the
failure indexes using: (a) APDL script in ANSYS and (b) a USERMAT subroutine (for
shell elements usermatps.f). Show all work in a report.
Problem 3.13 Visualize and report the maximum value of transverse deflection U3 for the
ply drop-off Example 3.6 with a ply drop-off ratio 1:10. The strip is 120 mm long and
100 mm, loaded by tension Nx = 10 N/mm applied to the bottom edges on the strip. Use
symmetry to model 1/2 of the tape. The material is AS4D/9310 (Table 3.1).
Problem 3.14 Perform a modal analysis using ANSYS for a 1×1 m plate with all edges
fixed using the same laminate layup properties of Example 3.3. Compare the 10 lower
eigenvalues obtained with “layered shell” (Example 3.4) and “equivalent orthotropic” (Ex-
ample 3.3) approaches.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Chapter 4
Buckling
Most composite structures are thin walled. This is a natural consequence of the
following facts:
– Composites are stronger than conventional materials. Then, it is possible to

carry very high loads with a small area, and thus small thickness in most
components.
– Composites are expensive when compared to conventional materials. There-

fore, there is a strong motivation to reduce the volume, and thus the thickness
as much as possible.
– The cost of polymer matrix composites increases with their stiffness. The
stiffness in the fiber direction can be estimated by using the fiber-dominated
rule of mixtures, E1 = Ef Vf . For example, when glass fibers are combined
with a polymer matrix, the resulting composite stiffness is lower than that of
aluminum. Using Aramid yields a stiffness comparable to aluminum. Carbon
fibers yield composite stiffness lower than steel. Therefore, there is strong
motivation to increase the moment of inertia of beams and stiffeners without
increasing the cross-sectional area. The best option is to increase the moment
of inertia by enlarging the cross-section dimensions and reducing the thickness.
All the above factors often lead to the design of composite structures with larger,
thin-walled cross-sections, with modes of failure likely to be controlled by buckling.
4.1 Eigenvalue Buckling Analysis

Buckling is loss of stability due to geometric effects rather than material failure.
But it can lead to material failure and collapse if the ensuing deformations are
not restrained. Most structures can operate in a linear elastic range. That is,
they return to the undeformed configuration upon removal of the load. Permanent
deformations result if the elastic range is exceeded, as when matrix cracking occurs
in a composite.
113
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
L
secondary path P
bifurcation
Lcr
primary path
w
imperfect path
w
Figure 4.1: Equilibrium paths for the perfect column.
Consider a simply supported column of area A, length L, and moment of inertia

I, made of homogeneous material with modulus E and strength F along the length
of the column. The column is loaded by a compressive load P acting on the centroid
of the cross-section [3]. If the column geometry, loading, and material have no
imperfections, the axial deformation is
u = P L/EA (4.1)
with no lateral deformation w = 0. The deformation of the structure (u, v, w) before
buckling occurs is called the primary path (Figure 4.1). The slightest imperfection
will make the column buckle when
Pcr = π 2 (EI)/L2 (4.2)
The load capacity for long slender columns will be controlled by buckling, as
opposed to the crushing strength of the material. What happens after the column
reaches its critical load depends largely on the support conditions. For the simply
supported column, the lateral deflection1
w = A sin(πx/L) (4.3)
will grow indefinitely (A → ∞) when the load just barely exceeds PCR . Such large
lateral deflections will cause the material to fail and the column will collapse. The
behavior of the structure after buckling has occurred is called post-buckling.
The simply supported column in Figure 4.1 experiences no deformations in the
shape of the buckling mode (4.3) before buckling actually happens. In this case, it is
said that the structure has a trivial primary path. This is a consequence of having
a perfect structure with perfectly aligned loading. For these type of structures,
1
With x measured from one end of the column.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Buckling 115
buckling occurs at a bifurcation point. A bifurcation point is the intersection of the

primary path with the secondary path, i.e., the post-buckling path [21].
The bifurcation loads, one for every possible mode of buckling, are fairly easy to
obtain using commercial software. The geometry of the structure is that of the per-
fect undeformed configuration, loaded with the nominal loads, and the material is
elastic. Such analysis requires a minimum of effort on the part of the analyst. Com-
mercial programs refer to this analysis as an eigenvalue buckling analysis because
the critical loads are the eigenvalues λi of the discretized system of equations
([K] − λ[Ks ]){v} = 0 (4.4)

where K and Ks are the stiffness and stress stiffness matrix, respectively, and v is
the column of eigenvectors (buckling modes) [21].
Example 4.1 Consider a simple supported plate, with side dimensions ax = 1000 mm,
ay = 500 mm, edgewise loaded in compression with Nx = Ny = 1 N/mm. The plate
is made of [(0/90)3 ]S , AS4/9310 (Table 4.1) composite with fiber volume fraction 0.6 and
total thickness tT = 10.2 mm. Compute the critical load of the lowest four modes using
eigenvalue analysis. Visualize the lower four modes.
Table 4.1: Lamina elastic properties. AS4/9310, Vf = 0.6

Young’s Moduli Shear Moduli Poisson’s Ratio
E1 = 145880 MPa G12 = G13 = 4386 MPa ν12 = ν13 = 0.263
E2 = E3 = 13312 MPa G23 = 4529 MPa ν23 = 0.470
Solution to Example 4.1 Since the laminate is symmetric, and stress computation lam-
ina by lamina is not required, the critical loads can be obtained using three different ap-
proaches. This example continues in Example 4.3.
First approach: Equivalent Laminate Moduli. The equivalent laminate moduli

are calculated and used along with an orthotropic shell element. In this case, laminate
moduli represent the stiffness of an equivalent orthotropic plate that behaves likes the actual
laminate under in-plane loads, neglecting the bending loads (see Section 3.2.6). Laminate
moduli can be found as explained in Section 1.15. Introduce the lamina properties (Table
4.1) into (1.91), rotate each lamina (1.53), add then according to (1.102) get the laminate
moduli (1.105) listed in Table 4.2. Note that in some cases bending laminate moduli Exb ,
etc. [1, (6.36)] may give more accurate results than in-plane laminate moduli Ex , etc. [1,
(6.35)].
The portion of the ANSYS R
input file used to enter the laminate moduli is listed below
and available on the Web site [5, FEAComp Ex401 orthotropic]. Element type SHELL281 is
used.
Table 4.2: Equivalent laminate moduli for [0/903 ]S

Ex = 79985 MPa Gxy = 4386 MPa νxy = 0.044
Ey = 79985 MPa Gyz = 4458 MPa νyz = 0.415
Ez = 16128 MPa Gxz = 4458 MPa νxz = 0.415
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
/TITLE, Bifurcation, Orthotropic, FEACM with ANSYS (c) Barbero (2012)


! Equivalent Laminate Properties
UIMP,1,EX,EY,EZ,79985,79985,16128
UIMP,1,GXY,GYZ,GXZ,4386,4458,4458

SECDATA,10.2,1 ! Laminate section: Th=10.2 mm, Mat. #1
! Geometry and mesh

ESIZE,,25 ! 25 divisions for edge
/SOLU ! Solution module, (i) STATIC ANALYSIS

PSTRESS,ON ! Calculate the stress stiffness matrix
DL,3,1,uz,0
DL,4,1,symm
!Load application
SFL,2,PRES,1 ! Apply uniform pressure in x=500 mm
SFL,3,PRES,1 ! Apply uniform pressure in y=250 mm
/SOLU ! Solution module, (ii) find the BIFURCATION LOADS

ANTYPE,BUCK
BUCOPT,SUBSP,10 ! Find the first 10 bifurcations loads
SOLVE ! Solve
/SOLU ! Solution module, (ii) find the BUCKLING MODES

EXPASS
MXPAND
SOLVE ! Solve

SET,LIST ! List the critical loads
SET,1,2 ! Set mode number 2 shape
PLDISP,1 ! Display the mode 2 shape displacements
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Buckling 117
The buckling loads are shown in Table 4.3.
Second approach: Using A, B, D, H, matrices. To calculate the A, B, D, H,

matrices, introduce the lamina properties (Table 4.1) into (3.9). The resulting laminate
matrices are
 
817036 35937.6 0 0 0 0


 35937.6 817036 0 0 0 0 

A B  0 0 44737.2 0 0 0 
= 
B D 
 0 0 0 8.55845 106 311579 0 

 0 0 0 311579 5.60896 106 0 
0 0 0 0 0 387872

37812.8 0
[H] =
0 37964.7
The ANSYS input file used to define the laminate using SHELL281 elements and A, B, C, H,
matrices is listed below.
/TITLE, Bifurcation, ABDH input, FEACM with ANSYS (c) Barbero (2012)

! This example does not need material properties

SECTYPE,1,GENS ! Use preintegrated general shell section (ABDH matrix)
! ABDH matrix definition

SSPA,817036,35937.6,0,817036,0,44737.2 ! A11,A21,A31,A22,A32,A33
SSPB,0,0,0,0,0,0 ! B11,B21,B31,B22,B32,B33
SSPD,8.55845e+006,311579,0,5.60896e+006,0,387872 ! D11,D21,D31,D22,D32,D33
SSPE,37812.8,0,37964.7 ! H11,H21,H22
! Geometry and mesh


DL,3,1,uz,0
DL,4,1,symm
!D,ALL,ROTZ ! Constrain rotations about z axes (optional)
!Load application
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i


ANTYPE,BUCK
SOLVE ! Solve

EXPASS
MXPAND
SOLVE ! Solve

The buckling loads are shown in Table 4.3.
Third approach: Using LLS. The laminate stacking sequence (LSS) and the lamina
properties (Table 4.1) are entered. The ANSYS input file commands to define the laminate
are listed below. Element type SHELL281 is used.
/TITLE, Bifurcation, LSS input, FEACM with ANSYS (c) Barbero (2012)

! Material properties for a lamina
UIMP,1,EX,EY,EZ,145880,13312,13312
UIMP,1,GXY,GYZ,GXZ,4386,4529,4386
ET,1,SHELL281 ! Chooses Shell281 element for analysis

SECDATA,0.85,1,0 ! Repeat the pattern
SECDATA,0.85,1,90
SECDATA,0.85,1,0
SECDATA,0.85,1,90 ! Same laminas in symmetrical order
SECDATA,0.85,1,90
SECDATA,0.85,1,0
SECDATA,0.85,1,90
SECDATA,0.85,1,0
SECDATA,0.85,1,90
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Buckling 119
SECDATA,0.85,1,0
! Geometry and mesh


DL,2,1,UZ,0 ! Impose Simple Supported BC
DL,3,1,UZ,0
DL,1,1,SYMM ! Impose Symmetry BC
DL,4,1,SYMM
!Load application

ANTYPE,BUCK
SOLVE ! Solve

EXPASS
MXPAND
SOLVE ! Solve

The procedure for obtaining the solution of “Eigenvalue Buckling Analysis” in ANSYS
has three steps: (i) solve the static solution using the PSTRESS,ON command to obtain the
stress stiffness matrix, (ii) obtain the bifurcation loads using the eigenvalue buckling solution,
and (iii) expand the solution if the buckled mode shapes are needed. By running the code
listed below, the critical load and buckling mode shape for every mode are obtained.

DL,2,1,UZ,0 ! Impose Simple Supported BC
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
DL,3,1,UZ,0
DL,1,1,SYMM ! Impose Symmetry BC
DL,4,1,SYMM
!Load application

ANTYPE,BUCK
SOLVE ! Solve

EXPASS
MXPAND
SOLVE ! Solve
Using the command SET,LIST in the post-processor, a list with the critical buckling
loads is obtained. With SET,1,n, where n is the mode number, it is possible to select differ-
ent solutions corresponding to different mode shapes, which can be plotted using PLDISP,1
command, as indicated in the listing below.

The results of the first five modes are summarized in Table 4.3 for the equivalent lamina,
A, B, D, H, matrix input, as well as for LSS. Values are shown for only five modes because
lack of accuracy of results for modes above 1/2 the number of iteration vectors used in the
subspace method.
Table 4.3: Bifurcation loads [N/mm]

Mode 1 2 3 4 5
Orthotropic Equivalent 252.70 570.55 1547.40 2150.80 2318.20
ABDH Input 209.53 639.98 1802.80 1822.50 1863.40
LSS Input 209.42 640.43 1794.93 1826.86 1854.23
4.1.1 Imperfection Sensitivity

To illustrate the influence of imperfections in buckling, let us consider the solid
lines in Figure 4.1. The lateral deflection is zero for any load below the bifurcation
load PCR , that is on the primary path of the perfect structure. The primary path
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Buckling 121
intersects the secondary path at the bifurcation point, for which the load is PCR .
The post-critical behavior of the column is indifferent and slightly stable. Indifferent
means that the column can deflect right or left. Stable post-critical path means that
the column can take a slightly higher load once it has buckled. For a column, this
stiffening behavior is so small that one cannot rely upon it to carry any load beyond
PCR . In fact, the column will deform laterally so much that the material will fail
and the system will collapse. Unlike columns, simply supported plates experience
significant stiffening on the secondary path.
4.1.2 Asymmetric Bifurcation

Consider the frame illustrated in Figure 4.2. An eigenvalue analysis using one finite
element per bar [21, Sections 5.9 and 7.8] reveals the bifurcation load
PCR = 8.932(10−6 )AE (4.5)
but gives no indication about the nature of the critical state: whether it is stable
or not, whether the post-critical path is symmetric or not, and so on. We shall see
later on that the frame has an asymmetric, and thus unstable post-critical path, as
represented in Figure 4.2. That is, the post-critical path has a slope
P (1) = 18.73(10−9 )AE (1/rad) (4.6)
in the force-rotation diagram in Figure 4.2, where θ is the rotation of the joint at
the load point.
In general, the problem with eigenvalue analysis is that it provides no indication
as to the nature of the post-critical path. If the post-critical path is stiffening and
symmetric as in Figure 4.1, the real structure may have a load capacity close to the
bifurcation load. But if the post-critical path is unstable and/or asymmetric, as in
Figure 4.2 or if there is mode interaction [16, 22–26], the real structure may have a
load capacity much smaller than the bifurcation load. In order to use the informa-
tion provided by eigenvalue analysis, it is necessary to understand and quantify the
post-buckling behavior.
4.1.3 Post-Critical Path

One way to investigate the post-buckling behavior is to perform a continuation
analysis of the imperfect structure, as presented in Section 4.2. This is perfectly
possible, but complicated and time consuming, as it will be seen later in this chap-
ter. A more expedient solution can be obtained using software capable of predicting
the nature of the post-critical path, including symmetry, curvature, and mode in-
teraction. If the secondary path is stable and symmetric, the bifurcation load can
be used as a good estimate of the load capacity of the structure. The curvature of
the post-critical path gives a good indication of the post-buckling stiffening and it
can be used to a certain extent to predict post-buckling deformations.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
. A
B
P
. B
q
L A
Y
. q
X
Figure 4.2: Two-bar frame.
The bifurcation load, slope, and curvature of the post-critical path emerging
from the bifurcation (4.1) can be computed with BMI3 [23–25] available in [5]. The
post-buckling behavior is represented by the following formula
1 (2) 2
Λ = Λ(cr) + Λ(1) s + Λ s + ... (4.7)
2
where s is the perturbation parameter, which is chosen as one component of the
displacement of one node, Λ(cr) is the bifurcation multiplier, Λ(1) is the slope, and
Λ(2) is the curvature of the post-critical path [22, (43)], [16, 23–26]. When the slope
is zero, the post-critical path is symmetric. Therefore, buckling is indifferent, and
the real structure will buckle to either side. There is no way to predict which way it
is going to buckle, unless of course one knows the shape of the imperfections on the
real plate, which is seldom the case. A positive curvature denotes stiffening during
post-buckling, and a negative one indicates that the stiffness decreases.
Example 4.2 Consider the simple supported plate of Example 4.1. Compute the bifurcation
multiplier ΛCR , the critical load NCR , the slope Λ(1) , and the curvature Λ(2) of the post-
critical path. Estimate the load when the maximum lateral deflection is equal to the thickness
of the plate. As perturbation parameter, use the largest displacement component of the
buckling mode with lowest buckling load.
Solution to Example 4.2 The program BMI3 [22], available in [5], is used in this case to
compute the bifurcation multiplier ΛCR , the slope Λ(1) , and the curvature Λ(2) of the post-
critical path. Refer to Appendix C for a description of the software interface and operation
procedure. BMI3 c
is used from within the ANSYS graphical user interface (GUI) in this
example. Since BMI3 requires the A-B-D-H matrices, the ANSYS input file is a slightly
modified version of that used for the second approach in Example 4.1, as follows [5,
FEAcomp Ex402 ABDH.inp]
/TITLE, ORTHOTROPIC PLATE WITH EDGE LOAD, BIFURCATION ANALYSIS

i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Buckling 123

! THIS INPUT DATA DOES NOT NEED MATERIAL PROPERTIES
ET,1,SHELL99,,2 ! CHOOSES SHELL99 ELEMENT FOR ANALYSIS

! SET KEYOPT(2)=2, THEN SUPPLY ABDH MATRIX
KEYOPT,1,10,2 ! SET KEYOPT(10)=2, PRINT ABDH MATRIX FILE.ABD
! REAL CONSTANT SET #1, ABDH MATRIX DEFINITION

R,1,817036,35937.6,0,0,0,0 ! REAL SET #1, A11,A12,0,A16,0,0
RMODIF,1,7,817036,0,0,0,0 ! REAL SET #1, A22,0,A26,0,0
RMODIF,1,16,44737.2,0,0 ! REAL SET #1, A66,0,0
RMODIF,1,19,37812.8,0 ! REAL SET #1, H44,H45
RMODIF,1,21,37964.7 ! REAL SET #1, H55
RMODIF,1,43,8.55845E+006,311579,0,0,0,0 ! REAL SET #1, D11,D12,0,D16,0,0
RMODIF,1,49,5.60896E+006,0,0,0,0 ! REAL SET #1, D22,0,D26,0,0
RMODIF,1,58,387872,0,0 ! REAL SET #1, D66,0,0
RMODIF,1,76,,10.2 ! REAL SET #1, AVERAGE DENSITY AND THICKNESS
! GEOMETRY AND MESH

RECTNG,0,500,0,250 ! CREATES A RECTANGLE WITH X=1 M AND Y=1 M
ESIZE,50 ! 50 SIZE ELEMENT EDGE
/SOLU ! SOLUTION MODULE, (I) STATIC ANALYSIS

PSTRESS,ON ! CALCULATE THE STRESS STIFFNESS MATRIX
DL,2,1,UZ,0 ! IMPOSE SIMPLE SUPPORTED BC
DL,3,1,UZ,0
DL,1,1,SYMM ! IMPOSE SYMMETRY BC
DL,4,1,SYMM
!D,ALL,ROTZ ! CONSTRAINT ROTATIONS ABOUT Z AXES (OPTIONAL)
!LOAD APPLICATION
!SFL,2,PRES,1 ! APPLY UNIFORM PRESSURE IN X=500 MM
LSEL,S,LOC,X,500
NSLL,S,1
*GET,NNOD,NODE,,COUNT
F,ALL,FX,-(1*250)/NNOD
!SFL,3,PRES,1 ! APPLY UNIFORM PRESSURE IN Y=250 MM

LSEL,S,LOC,Y,250
NSLL,S,1
*GET,NNOD,NODE,,COUNT
F,ALL,FY,-(1*500)/NNOD
ALLSEL

i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
/SOLU ! SOLUTION MODULE, (II) FIND THE BIFURCATION LOADS

ANTYPE,BUCK
BUCOPT,SUBSP,10 ! FIND THE FIRST 10 BIFURCATIONS LOADS
SOLVE ! SOLVE
/SOLU ! SOLUTION MODULE, (II) FIND THE BUCKLING MODES

EXPASS
MXPAND
SOLVE ! SOLVE

SET,LIST ! LIST THE CRITICAL LOADS
SET,1,1 ! SET MODE NUMBER 1 SHAPE
/VIEW,,1,1,1
PLDISP,2 ! DISPLAY THE MODE SHAPE
The model in ANSYS and solved by using the “Eigenvalue buckling analysis” procedure
for obtaining the bifurcation loads Λ(cr) . The critical lowest load is displayed on the ANSYS
GUI as FREQ=210.2 for STEP=1, SUB=1. Since the buckling mode is scaled to a maximum
amplitude of 1.0, we get DMX=1. A list of buckling loads can be recalled by the command
SET,LIST.
Next, details about how to run BMI3 from within ANSYS are given in Appendix C.2.2.
Within the ANSYS GUI, Run the APDL macro ans2i (available in [5]) simply by entering
ans2i in the ANSYS command line to calculate parameters of the post-critical path. BMI3
will be executed.
Find the ANSYS Output Window minimized in your Taskbar and open it up. Then,
manually introduce the following responses to the prompts:
– ...sort (0 = none, 1 = x, 2 = y, 3 = z): 1. To minimize the bandwidth, usually it is

best to sort along the direction that has more elements and/or nodes.
– ...for perturbation analysis (y/n)?: n. In this way, BMI3 chooses as perturbation pa-
rameter the largest displacement component of the buckling mode with lowest buckling
load. In this case, that corresponds to the first buckling mode, the node in the middle
of the plate, and the deflection direction δ.
The following results are obtained:
Λ(cr) = −209.0418; Λ(1) ≈ 0; Λ(2) = −0.2308
Since BMI3 solves the problem using reversed loads (see Appendix C), then (4.7) becomes
−N = Λ(cr) + Λ(1) s + 21 Λ(2) s2

N = −Λ(cr) − Λ(1) s − 21 Λ(2) s2
and, in this case the perturbation direction is s = −δ, so
N = −Λ(cr) − Λ(1) (−δ) − 21 Λ(2) (−δ)2

N = −Λ(cr) + Λ(1) δ − 21 Λ(2) δ 2
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Buckling 125
Figure 4.3: Equilibrium paths for a perfect plate.
Therefore, using the results from BMI3, the secondary path of bifurcation analysis is
N = −(−209.0418) + (0) δ − (−0.1154) δ 2 = 209.0418 + 0.1154 δ 2
Since the slope Λ(1) is zero, the post-critical path is symmetric. The post-buckling load
when the lateral deflection (w) is equal to the thickness (s = T h = 10.2 mm) is equal to
221 N/mm, as shown in Figure 4.3.
Figure 4.3 can be drawn based on the results from BMI3 by using the MATLAB
R
code shown next.
% plot Figure~4.3, FEA of Comp Mater Using ANSYS--Second Ed.

% Ever J. Barbero (c) 2007, 2013
clear all; close all;
Th=0.85*12 %10.8
lambda0= +(-209.04) %216.66% 210.7879
lambda1= -(-0) %-0.24389% 2.063125
lambda2= +(-0.2308) %0.3602 %0.36291% 0.368516
s_ini=-0.5*Th;
s_fin=1.5*Th;
s_inc = [s_ini:0.1:s_fin];
P_inc = lambda0 + lambda1 .* s_inc + lambda2.*s_inc.^2;
s = 1*Th;
P = lambda0 + lambda1*s + lambda2*s^2;
Po= lambda0;
s1_inc = [0,0]; %[0.2,0.2];
P1_inc = [0,Po];
so_inc = [0:1:Th/2];
Po_inc = lambda0 .* so_inc./so_inc;
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
% Ansys results OPEN I/0 FILES

n_file = ’FEAcomp_Ex402_conti_225’ % series name
fidinp = fopen([n_file,’.txt’],’r’);
results = (fscanf(fidinp,’%g %g’,[2 inf]));
fclose(fidinp);
results(1,:)=(results(1,:)-1)*(225)
linetxt1 = [’N_{s=Th}=’ num2str(-int32(P)) ’ N/mm \rightarrow ’]
linetxt2 = [’ \leftarrow N_{cr}=’ num2str(-int32(Po)) ’ N/mm’]
linewth = 2.8
figure1 = figure
plot(s_inc,-P_inc,’k’,’LineWidth’,linewth)
xlabel(’s [mm]’,’Fontsize’,16);ylabel(’N [N/mm]’,’Fontsize’,16);
hold on
plot(s1_inc,-P1_inc,’k’,’LineWidth’,linewth)
plot(so_inc,-Po_inc,’k-.’,’LineWidth’,1)
text(s,-P,linetxt1,’HorizontalAlignment’,’right’,’Fontsize’,16)
text(Th/2,-Po,linetxt2,’HorizontalAlignment’,’left’,’Fontsize’,16)
axis([s_ini s_fin 0 -min(P_inc)])
set(gca,’Fontsize’,16);
plot(s,-P,’ks’,’MarkerSize’,10,’LineWidth’,2)
plot(0,-Po,’ks’,’MarkerSize’,10,’LineWidth’,2)
hold off
saveas(figure1,[’Fig4_3’],’pdf’);
4.2 Continuation Methods

The strain to failure of polymer matrix composites (PMC) is high. Compare 1.29%
for AS4/3501 and 2.9% for S-glass/epoxy with only 0.2% for steel and 0.4% for
aluminum. That means that buckling deformations can go into the post-buckling
regime while the material remains elastic. However, great care must be taken that
no matrix dominated degradation mode takes place, in which case the material will
not remain elastic (see Chapter 8). Eigenvalue buckling analysis is relatively simple
as long as the material remains elastic because classical theory of elastic stability
can be used, as was done in Section 4.1. Material nonlinearity is one reason that
motivates an incremental analysis. Another reason is to evaluate the magnitude of
the buckling load for an imperfection sensitive structure.
In an incremental analysis, also called continuation analysis, the load is increased
gradually step by step. At each step, the deformation, and possibly the changing
material properties, are evaluated. Incremental analysis must include some type of
imperfection, in the geometry, material, or alignment of loads. Lacking any imper-
fection, incremental analysis will track the linear solution, revealing no bifurcations
or limit points.
Continuation methods are a form of geometrically nonlinear analysis. The sys-
tem must have a nontrivial fundamental path, such as a flat plate with asymmetric
laminate stacking sequence (LSS) under edge loads.
If the system has a trivial fundamental path, such a flat plate with symmetric
LSS under edge load, the nontrivial fundamental path can be forced by introducing
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Buckling 127
an imperfection. Several types of imperfections are possible, including material im-

perfections (e.g., unsymmetrical LSS), geometric imperfections, or load eccentricity
are used.
Since the real geometric imperfections are seldom known, the preferred artificial
geometric imperfection is in the form of the bifurcation mode having the lowest
bifurcation load. This is true in most cases; however, in some cases, a second mode
that is associated to imperfections that are more damaging to the structure should
be used [27]. Also, if the structure has an asymmetric post-buckling path, as the
two-bar example in Figure 4.2, care must be taken not to force the structure along
the stiffening path.
Finite element analysis (FEA) codes allow the user to modify a mesh by su-
perposing an imperfection in the shape of any mode from a previous bifurcation
analysis onto the perfect geometry (see Example 4.3).
Example 4.3 Using ANSYS, apply a geometric imperfection wp (x, y) = δ0 φ(x, y) to Ex-
ample 4.1 and plot the load-multiplier vs. maximum lateral deflection for an imperfection
magnitude δ0 = T h/10 and δ0 = T h/100, where T h is the total laminate thickness, and
φ(x, y) is the buckling mode corresponding to the lowest bifurcation load found in Example
4.1.
Solution to Example 4.3 First the buckling modes are found using the bifurcation method
(execute the commands shown in Example 4.1).
Then the nodal positions are updated using the UPGEOM command. Using this command
the displacements from a previous analysis can be added in order to update the geometry of
the finite element model to that of the deformed configuration.
Since the displacements have been obtained from a mode shape, the maximum displace-
ment in the results file is 1.0. The UPGEOM allows the user to define a multiplier for dis-
placements being added to the nodal coordinates. In this case, the multiplier factors chosen
are δ0 = T h/10 and δ0 = T h/100. Therefore, an initial deflection equal to the first mode of
buckling with a central deflection δ0 is forced on the structure.
Using a continuation method with this imperfect geometry, the continuation equilibrium
paths shown in Figure 4.4 are obtained. It can be seen that eventually the continuation
solution approaches the secondary path of the perfect structure, shown by dashed lines in
Figure 4.4. For smaller imperfections, the continuation solution follows more closely the
primary path, then the secondary path. A structure with large imperfections deviates more
from the behavior of the perfect structure, as show by the solution corresponding to an
imperfection δ0 = T h/10.
/TITLE, Geometric imperfection, FEACM with ANSYS (c) Barbero (2012)


! Material properties for lamina
UIMP,1,EX,EY,EZ,145880,13312,13312
UIMP,1,GXY,GYZ,GXZ,4386,4529,4386
ET,1,SHELL281 ! Chooses Shell281 element for analysis

i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i

SECDATA,0.85,1,0 ! Repeat the pattern
SECDATA,0.85,1,90
SECDATA,0.85,1,0
SECDATA,0.85,1,90 ! Same laminas in symmetrical order
SECDATA,0.85,1,90
SECDATA,0.85,1,0
SECDATA,0.85,1,90
SECDATA,0.85,1,0
SECDATA,0.85,1,90
SECDATA,0.85,1,0
! Geometry and mesh

ESIZE,50 ! 50 size element edge

DL,3,1,uz,0
DL,4,1,symm
!Load application

ANTYPE,BUCK
SOLVE ! Solve

EXPASS
MXPAND
SOLVE ! Solve

ftr=(10.8/10) ! Multiplicator shape factor (Th/10)
UPGEOM,ftr,1,1,file,rst
! ftr: Multiplier for displacements added to coordinates
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Buckling 129
200
← δ0= Th/100
150
N [N/mm]
← δ0= Th/10
100
50
0
-5 0 5 10 15
s [mm]
Figure 4.4: Equilibrium paths for a [(0/90)3 ]S plate, with δ0 = T h/10 and δ0 =
T h/100.
! 1,1 : Load step 1, substep=1, equivalent to mode =1

! file,rst: results file to obtain displacements
/SOLU ! Solution module, Continuation loads

NLGEOM,1 ! Use large displacements analysis
OUTRES,ALL,ALL ! Keep results of each substep
mult=225 ! Apply loads until N = 225 N/mm

SFL,2,PRES,1*mult ! Apply uniform pressure in x=500 mm
SFL,3,PRES,1*mult ! Apply uniform pressure in y=250 mm
ARCLEN,1,10,0.1 ! Use ARCLENG method to obtain solution

NSUBST,50,0,0 ! #Substeps

LINES,1000 ! List without breaks between pages
NSOL,2,1,U,Z,UZ_node1 ! Load deflexion in central plate node
PLVAR,2 ! DISPLAY VARIABLES evolution
PRVAR,2 ! PRINT VARIABLES evolution
FINISH
Example 4.4 Find the buckling load multiplier and the first mode shape for Example 3.11.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Solution to Example 4.4 The solution is found by first executing the APDL code for
Example 3.11, available in [5, FEAComp Ex 311.inp]. Then, execute the APDL shown
next.
!Buckling analysis for H-COLUMN CST 58 (1998) 1335-1341 - ANSYS R14

/SOLU !ENTER SOLUTION MODULE
!FIRST PHASE
ANTYPE,STATIC,RESTART,,,PERTURB !RESTART THE LINEAR BASE ANALYSIS
PERTURB,BUCKLE,,,ALLKEEP !LINEAR PERTURBATION EIGENVALUE BUCKLING
!KEEP ALL THE BOUNDARY CONDITIONS
SOLVE,ELFORM !REGENERATE ELEMENT MATRICES
!SECOND PHASE
BUCOPT,SUBSP,1 !SUBSPACE ITERATION EIGENSOLVER
!EXTRACT 1 BUCKLING MODE
MXPAND,1 !NUMBER OF MODES TO EXPAND AND WRITE
SOLVE
FINISH !EXIT SOLUTION MODULE
/POST1 !POSTPROCESSOR MODULE

FILE,,RSTP !*.RSTP FILE TO REVIEW RESULTS FROM LINEAR PERTURBATION
SET,1,1 !SELECT 1ST LOAD CASE, 1ST EIGENVALUE
*GET,LCR,TIME !GET THE EIGENVALUE IN USER DEFINED VARIABLE LCR
PLDISP,2 !PLOT MODE SHAPE AND OUTLINE OF UNDEFORMED SHAPE
The load multiplier Λcr can be read from the GUI as FREQ=54.288 on the same
screen that shows the mode shape for mode one (STEP=1, SUB=1). The buckling
load is simply the product of the load multiplier by the applied load P cr = 54.142 ×
11, 452 = 597, 130 N . The buckling mode shape can be seen in Figure 4.5. The value
of DMX is irrelevant because in eigenvalue analysis the amplitude of the deformed
shape of the buckling modes is undetermined.
Suggested Problems
Problem 4.1 Compute the bifurcation load P c of the two-bar frame in Figure 4.2 using
one quadratic beam element per bar. Each bar has length L = 580 mm, area A = 41 mm2 ,
inertia I = 8.5 mm4 , height H = 10 mm, and modulus E = 200 GP a. The connection
between the two bars is rigid.
Problem 4.2 Perform a convergence study on the bifurcation load P c of the two-bar frame
in Problem 4.1 by increasing the number of elements per bar N until the bifurcation load
converges within 2%. Plot P c vs. N .
Problem 4.3 Recalculate Example 4.2 when the LSS changes to [(0/90)6 ]T , thus becoming
asymmetric. Do not introduce any imperfection but rather analyze the perfect system, which
in this case is asymmetric.
Problem 4.4 Recalculate Example 4.2 with [(0/90)6 ]T , and Nx = 1, Ny = Nxy = 0. Do

not introduce any imperfection but rather analyze the perfect system, which in this case is
asymmetric.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Buckling 131
Figure 4.5: Buckling mode shape, Example 4.4.
Problem 4.5 Using a FEA code, plot the continuation solution for δ0 = T h/100 as in
Figure 4.4, for a cylindrical shell with distributed axial compression on the edges. The
cylinder has a length of L = 965 mm and a mid-surface radius of a = 242 mm. The LSS
is [(0/90)6 ]S , with lamina thickness t = 0.127 mm. The laminas are of E-glass/epoxy with
E1 =54 GPa, E2 =18 GPa, G12 = 9 GPa, ν12 = 0.25, and ν23 = 0.38.
Problem 4.6 Compute the maximum stress failure index If of Problem 4.5 at P = Λ(cr) .
The strength values are F1t = 1034 MPa, F1c = 1034 MPa, F2t = 31 MPa, F2c = 138 MPa,
and F6t = 41 MPa.
Problem 4.7 Plot the imperfection sensitivity of the cylindrical shell of Problem 4.5, for
imperfections in the range (T h/200) < s < T h.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Chapter 5
Free Edge Stresses
In-plane loading Nx , Ny , Nxy , of symmetric laminates induces only in-plane stress

σx , σy , σxy , in the interior of the laminate. Near the free edges, interlaminar stresses
σz , σyz , σxz , are induced due to the imbalance of the in-plane stress components at
the free edge.
For illustration, consider a long laminated strip of length 2L, width 2b << 2L,
and thickness 2H < 2b (Figure 5.1). The strip is loaded by an axial force Nx only.
For a balanced, symmetric laminate the mid-plane strains and curvatures (see (3.6))
are uniform over the entire cross-section and given by
0x = α11 Nx
0y = α12 Nx
0
γxy =0
kx = ky = kxy = 0 (5.1)
where α11 , α12 , are in-plane laminate compliances, which are obtained by inverting
(3.8); see also [1, (6.21)]. From the constitutive equation [1, (6.24)] for lamina k,
we get

k k
σxk = Q11 α11 + Q12 α12 Nx

k k
σyk = Q12 α11 + Q22 α12 Nx

k k k
σxy = Q16 α11 + Q26 α12 Nx
σzk = σxz
k k
= σyz =0 (5.2)
A piece of laminate taken out of the interior of the laminate will have balanced
σy and σxy on opposite faces; the free body diagram (FBD) is in equilibrium without
the need for any additional forces. In this case we say the stress components are
self-equilibrating. At the free edge in Figure 5.1, σy =σxy =σyz =0. If σy and σxy are
not zero in the interior of the laminate, but are zero at the free edge, then some
other stresses must equilibrate them.
133
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 5.1: Tensile coupon.1
5.1 Poisson’s Mismatch

A lamina subjected to tensile loading in one direction will contract in the direction
perpendicular to the load. If two or more laminas with different Poisson’s ratios are
bonded together, interlaminar stress will be induced to force all laminas to deform
equally at the interfaces (Figure 5.2). Over the entire laminate thickness, these
stresses add up to zero since there is no transverse loading Ny applied. In other
words, they are self-equilibrating in such a way that
Z zN
σy dz = 0 (5.3)
z0
where z0 and zN are the coordinates of the bottom and top surfaces, respectively.
5.1.1 Interlaminar Force

As noted in (5.3), the in-plane stress σy calculated with classical lamination theory
(CLT) is self-equilibrating when added through the whole thickness of the laminate
[1, Chapter 6]. But on a portion of the laminate (above zk in Figure 5.3), the stresses
σy may not be self-equilibrating. Therefore, the contraction or expansion of one or
more laminas must be equilibrated by interlaminar shear stress σyz . Since there is
no shear loading on the laminate, the integral of σyz over the entire width of the
sample must vanish. Over half the width of the laminate, however, an interlaminar
shear force exists if the stress σy above or below the surface is not self-equilibrating.
The magnitude of these per unit length forces can be estimated by integrating the
interlaminar shear stress σyz over half the width of the laminate (0 < y < b). By
1
Reprinted from Mechanics of Fibrous Composites, C. T. Herakovich, Figure 8.1, copyright
(1998), with permission from John Wiley & Sons, Inc.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Free Edge Stresses 135
Figure 5.2: Poisson’s effect.2
equilibrium
Z b Z zN
Fyz (zk ) = σyz(z=z ) dy = − σy dz (5.4)
k
0 zk
The interlaminar shear stress σyz is not available from classical lamination theory
but the transverse stress σy is. Therefore, the magnitude of the interlaminar shear
force can be computed anywhere through the thickness of a laminate in terms of
the known transverse stress distribution σy .
The in-plane stress σy in a balanced, symmetric laminate under tensile load is
constant in each lamina. Therefore, when the interlaminar force is evaluated at an
interface (located at z = zk ), the integration above reduces to
N
X
Fyz (zk ) = − σyi ti (5.5)
i=k
where ti are the thicknesses of the laminas.

The magnitude of the interlaminar shear force Fyz can be used to compare
different stacking sequences in an effort to minimize the free-edge interlaminar shear
stress σyz . However, the force does not indicate how large the actual stress is.
Therefore, it can be used to compare different laminate stacking sequence (LSS)
but not as a failure criterion.
5.1.2 Interlaminar Moment

The interlaminar shear stress σyz produces shear strain γyz , which must vanish at
the center line of the sample because of symmetry. Therefore, σyz = 0 at the center
line. Also, at the free edge, σyz must vanish because σzy vanishes there. But for
any position zk above which σy is not self-equilibrating, σyz must be different from
zero somewhere between the edge and the center line. A numerical solution of σyz
2
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 5.3: Free body diagram of sublaminate used for computation of Poisson-
induced forces Fyz and moments Mz .
is plotted in Figure 5.5 in terms of the distance y/b from the free edge. It reveals
that σyz grows rapidly near the free edge and then tapers out at the interior of the
laminate.
A not self-equilibrating distribution of stress yields both a force Fyz (5.5) and a
moment. To compute the moment Mz , take moments of the stress σy with respect
to point A in Figure 5.3. A non-vanishing moment produced by σy can only be
equilibrated by a moment produced by transverse stress σz . Therefore, the moment
Mz is defined as
Z b Z zN
Mz (zk ) = σz(z=z ) ydy = (z − zk )σy dz (5.6)
k
0 zk
where zk is the coordinate of the top surface of lamina k, and zN is the coordinate of
the top surface of the laminate (see [1, Figure 6.6] for the definition of the coordinate
system through the thickness of the laminate).
The existence of σz is corroborated by free-edge delamination during a tensile
test, at a much lower load than the failure load of the laminate. The magnitude
of the moment can be used to compare different stacking sequences in an effort to
minimize the thickness stress σz . However, the moment does not indicate how large
the actual stress is, and thus it cannot be used as a failure criterion.
The in-plane stress σy in a balanced laminate under tensile load is constant in
each lamina. Therefore, when the interlaminar moment is evaluated at an interface
(located at z = zk ), the integration above reduces to
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
N
X t2i
Mz (zk ) = σyi (zi ti + − zk t i ) (5.7)
2
i=k
Since σz is a by-product of σyz , which vanishes at y = 0 due to symmetry, then
σz must vanish at the center line of the specimen (y = 0) but it is large near the
edge. Since no vertical load is applied, the integral of σz must be zero. Therefore, it
must be tensile (positive) on some regions and compressive (negative) at others. A
numerical solution reveals that σz grows rapidly near the free edge, dips to negative
values, and then tapers out at the interior of the laminate. A numerical solution of
σz is plotted in Figure 5.5 in terms of the distance y/b from the free edge. However,
σz → ∞ as y → b. This is a singularity that is not handled well by finite element
analysis (FEA). Therefore the results, even for y < b, will be very dependent on
the mesh refinement. Furthermore, since σz → ∞, the results cannot be used in a
failure criterion without further consideration.
Example 5.1 Compute Fyz and Mz at all interfaces of a balanced [02 /902 ]s symmetric
laminate (Figure 5.1) loaded with Nx = 175 KN/m. Use unidirectional lamina carbon/epoxy
properties E1 =139 GPa, E2 =14.5 GPa, G12 = G13 = 5.86 GPa, G23 = 5.25 GPa, ν12 =
ν13 = 0.21, ν23 = 0.38. The lamina thickness is tk = 0.127 mm.
Solution to Example 5.1 The in-plane stress distribution σy through the thickness can be
obtained by the procedure described in [1, Section 6.2], which is implemented in CADEC [9].
The stress values are shown in Table 5.1.
To calculate Fyz , compute the contribution of all laminas above a given interface using
(5.5). The in-plane stress σy in a balanced laminate under in-plane load is constant in each
lamina, so (5.5) applies. For other cases, (5.4) can be integrated exactly since σy is linear
in z, or Fyz can be approximated by (5.5) using the average σy in each lamina.
Since the laminate is balanced and loaded with in-plane loads only, Mz can be computed
using (5.7). Otherwise, use (5.6) or approximate Mz by using the average σz in each lamina
into (5.7).
The results are shown Table 5.1 and Figure 5.4.
Example 5.2 Plot σyz and σz vs y for 0 < y < b at the 90/0 interface above the middle
surface of a [0/90]s laminate with properties E1 = 139 GPa, E2 = 14.5 GPa, G12 = G13 =
5.86 GPa, G23 = 5.25 GPa, ν12 = ν13 = 0.21, ν23 = 0.38. Take 2b = 20 mm, length of
the sample 2L = 80 mm, thickness of each lamina tk = 1.25 mm. Load the sample with a
uniform strain x = 0.01 by applying a uniform displacement at x = L. Use orthotropic solid
elements on each lamina. Refine the mesh towards the free edge. Use at least two quadratic
elements through the thickness of each lamina and an element aspect ratio approximately
one near the free edge.
Solution to Example 5.2 Note that it is not necessary to model the whole geometry.
Symmetry can be used to model only the quadrant with x > 0, y > 0, z > 0; that is one-
eighth of the plate, as shown in Figure 5.3. Since any cross-section y − z has the same
behavior, only a short segment between x = 0 and x = L? needs to be modeled. Since free
edge effects also occur at x = 0 and x = L?, take L? = 8h and plot the results at x = L ? /2
to avoid free edge effects at the two loaded ends of the model. The solution is shown in
Figure 5.5.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
1
[0 /90 ]
2 2s
0.8
0.6
0.4
0.2
z/H
-0.2
-0.4
-0.6
Fyz/Fyz max
-0.8
Mz/Mz max
-1
-1 -0.5 0 0.5 1
Fyz/Fyz max; Mz/Mz max
Figure 5.4: Interlaminar force Fyz and moment Mz due to Poisson’s effect.
20
15
10
5
σ [MPa]
-5
-10
σyz
-15
σz
-20
0 0.2 0.4 0.6 0.8 1
y/b
Figure 5.5: Interlaminar stress σyz and σz at the 90/0 interface of a carbon/epoxy
[0/90]S laminate (FEA).
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Table 5.1: Poisson’s interlaminar force Fyz

k Pos σy tk z Fyz Mz
[MPa] [mm] [mm] [kN/m] [N m/m]
8 TOP 5.55 10−3 0.508 0.000
8 BOT 5.55 10−3 0.127 0.381 -0.705 0.045
7 TOP 5.55 10−3 0.381 -0.705
7 BOT 5.55 10−3 0.127 0.254 -1.410 0.179
6 TOP -5.55 10−3 0.254 -1.410
6 BOT -5.55 10−3 0.127 0.127 -0.705 0.313
5 TOP -5.55 10−3 0.127 -0.705
5 BOT -5.55 10−3 0.127 0.000 0.000 0.358
4 TOP -5.55 10−3 0.000 0.000
4 BOT -5.55 10−3 0.127 -0.127 0.705 0.313
3 TOP -5.55 10−3 -0.127 0.705
3 BOT -5.55 10−3 0.127 -0.254 1.410 0.179
2 TOP 5.55 10−3 -0.254 1.410
2 BOT 5.55 10−3 0.127 -0.381 0.705 0.045
1 TOP 5.55 10−3 -0.381 0.705
1 BOT 5.55 10−3 0.127 -0.508 0.000 0.000
See the command input file below. The PATH commands define, plot, and print the stress
values shown in Figure 5.5.
/TITLE,Free Edge Analysis [0/90]s laminate
*SET,THZ,1.25 ! thickness of lamina in mm
*SET,LX,8*THZ ! 1/2 length of laminate in mm
*SET,BY,10.0 ! 1/2 width of laminate in mm
*SET,NEX,8 ! number of elements in x/z direction
*SET,NEY,14 ! number of elements in y direction
*SET,EPSX,0.01 ! uniform strain in x direction
! Equivalent Material properties

UIMP,1,EX,EY,EZ,139E3,14.5E3,14.5E3
ET,1,SOLID186 ! Chooses SOLID186 element for analysis

KEYOPT,1,2,1 ! KEYOPT(2) = 1 Enables Full Integration
! Define material orientation by local Coordinate

LOCAL,11,,0,0,0,90 ! defines 90 degree local cs
CSYS,0 ! set active cs to cart. system
! Generate Geometry
BLOCK,0,LX,0,BY,0,THZ ! 90 degrees lamina
BLOCK,0,LX,0,BY,THZ,2*THZ ! 0 degress lamina
VGLUE,ALL ! Glue volumes
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
! Mesh Control and Mesh

LESIZE,ALL,,,NEX ! line number divisions = nex
LSEL,S,LOC,Z,0 ! selects lines z=0
LSEL,A,LOC,Z,THZ ! add lines z=thz to selection
LSEL,A,LOC,Z,2*THZ ! add lines z=2thz to selection
LSEL,R,LOC,X,0 ! reselects lines x=0
LESIZE,ALL,,,NEY,15,1,,,1 ! define element size in selected lines
LSEL,R,LOC,X,LX ! reselects lines x=lx
LESIZE,ALL,,,NEY,(1/15),1,,,1 ! define ele. size selected lines
LSEL,ALL ! select all lines
MSHKEY,1 ! Specifies mapped meshing
ESYS,11 ! Selects 90 degrees material orientation
VMESH,1 ! Meshes 90 degrees lamina
VMESH,3 ! Meshes 0 degree lamina
/SOLU ! Solution module,

ASEL,S,LOC,X,0
ASEL,A,LOC,Y,0
ASEL,A,LOC,Z,0
DA,ALL,SYMM ! Impose Symmetry BC
ASEL,S,LOC,X,LX
DA,ALL,UX,(EPSX*LX) ! Impose displacement on the end
ALLSEL,ALL ! Selects all areas

RSYS,0 ! Set results in global coordinates system
PATH,INTERFACE,2,,100 ! Define a path between two points, 100 values
PPATH,1,0,0,0,THZ,0 ! 1st point of the path location
PPATH,2,0,0,bY,THZ,0 ! 2nd point of the path location
PDEF,zero,EPSW,,AVG ! Compute zero axis (optional)
PLPATH,SZ,SXZ,SYZ,ZERO ! Plot Sz,Sxz,Syz
/PAGE,1000,,1000 ! Define print list without skips between pages

PATH,INTERFACE,2,,100 ! path between two points, compute 100 values
PPATH,1,0,LX/2,0,THZ,0 ! 1st point of the path location
PPATH,2,0,LX/2,BY,THZ,0 ! 2nd point of the path location
PDEF,Sz ,S,Z,AVG ! Compute Sz
PDEF,Syz,S,YZ,AVG ! Compute Syz
PDEF,zero,EPSW,,!AVG ! Compute zero axis (optional)
PLPATH,SZ,SYZ,ZERO ! Plot Sz,Syz
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i

PRPATH,SZ,SYZ ! Print Sz,Syz
5.2 Coefficient of Mutual Influence

In classical lamination theory, it is assumed that the portion of the laminate being
analyzed is far from the edges of the laminate. Stress resultants N and M are then
applied to a portion of the laminate and these induce in-plane stress σx , σy , σxy ,
on each lamina. In the interior of the laminate, interlaminar stress σxz , σyz , are
induced only if shear forces are applied.
For uniaxial loading Nx , the transverse stresses generated in each lamina as a
result of Poisson’s effect must cancel out to yield a null laminate force Ny . Also, the
in-plane shear stress on off-axis laminas must cancel out with those of other laminas
to yield zero shear force Nxy for the laminate. The situation is more complex
near the edges as the various components of in-plane stress do not cancel each out
other across the lamina interfaces. For the time being, let us revisit the concept
of laminate engineering properties. In material axes, the plane stress compliance
equations are
    
 1  S11 S12 0  σ1 
2 =  S12 S22 0  σ2 (5.8)
γ6 0 0 S66 σ6
   
It is also known that the compliance coefficients can be written in terms of

engineering properties as
 
1/E1 −ν12 /E1 0
[S] =  −ν12 /E1 1/E2 0  (5.9)
0 0 1/G12
For an off-axis lamina (oriented arbitrarily with respect to the global axes), we
have
    
 x  S 11 S 12 S 16  σx 
y =  S 12 S 22 S 26  σy (5.10)
γxy S 16 S 26 S 66 σxy
   
Here it can be seen that uniaxial load (σy = σxy = 0) yields shear strain as a
result of the shear-extension coupling
γxy = S 16 σx (5.11)
where
S 16 = (2S11 − 2S12 − S66 ) sin θ cos3 θ (5.12)

3
− (2S22 − 2S12 − S66 ) sin θ cos θ
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 5.6: Deformation caused by mutual influence.3

Now, the coefficients of S can be defined in term of the engineering properties
for the off-axis lamina as
S 11 = 1/Ex ; S 12 = −νxy /Ex = −νyx /Ey (5.13)

S 22 = 1/Ey ; S 66 = 1/Gxy

To complete the definition of S in (5.10), two new engineering properties
describing shear-extension coupling, ηxy,x and ηxy,y , are defined as
ηxy,x ηxy,y
S 16 = ; S 26 = (5.14)
Ex Ex
The engineering properties ηxy,x and ηxy,y are called coefficients of mutual in-
fluence and they represent the shear caused by stretching. Their formal definition
is obtained by imposing an axial stress and measuring the resulting shear strain
γij
ηij,i = (5.15)
i
Alternatively, two other coefficients of mutual influence could be defined to

represent the stretching caused by shear
ηx,xy ηy,xy
S 16 = ; S 26 = (5.16)
Gxy Gxy
These are defined by imposing a shear stress and measuring the axial strain
i
ηi,ij = (5.17)
γij
3
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 5.7: Free body diagram of sublaminate to compute the interlaminar force
Fxz due to mutual influence.
5.2.1 Interlaminar Stress due to Mutual Influence

Off-axis laminas induce in-plane shear stress when subject to axial loading because
the natural shear deformations that would occur on an isolated lamina (Figure 5.6)
are constrained by the other laminas. Through the whole thickness of the laminate,
these stresses cancel out, but over unbalanced sublaminates (e.g., the top lamina in
Figure 5.6), they amount to a net shear.
That shear can only be balanced by interlaminar stress σzx at the bottom of the
sublaminate (Figure 5.7). Then, summation of forces along x leads to a net force
Z b Z zN
Fxz (zk ) = σzx(z=z ) dy = − σxy dz (5.18)
k
0 zk
Once again, the in-plane shear stress calculated with classical lamination theory
(CLT) [1, Chapter 6] can be used to compute the interlaminar force per unit length
Fxz . For in-plane loading, CLT yields constant shear stress in each lamina. When
the interlaminar force is evaluated at an interface (located at z = zk ), the integration
above reduces to
N
X
i
Fxz (zk ) = − σxy ti (5.19)
i=k
The force Fxz , as well as the values of the coefficients of mutual influence can be
used to qualitatively select the LSS with the least interlaminar stress. Actual values
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
of interlaminar stresses can be found by numerical analysis. However, σz → ∞ as

y → b. This is a singularity that is not handled well by FEA. Therefore the results,
even for y < b, are very dependent on mesh refinement. Furthermore, since σz → ∞,
the results cannot be used in a failure criterion without further consideration. A
numerical approximation of σxz for a [±45]S laminate is plotted in Figure 5.9 in
terms of the distance y’ from the free edge.
Example 5.3 Compute Fxz at all interfaces of a [302 /−302 ]s balanced symmetric laminate
(Figure 5.1) loaded with Nx = 175 KN/m. The material properties are given in Example
5.1. The lamina thickness is tk = 0.127 mm.
Solution to Example 5.3 In-plane shear stress σxy through the thickness of the laminate
can be obtained following the same procedure used to obtain σy in Example 5.1.
For a symmetric balanced laminate under in-plane loads, use (5.19). For a general
laminate under general load, use (5.18) or approximate Fxz by (5.19) taking the average of
σxy in each lamina.
The results are obtained with a spreadsheet and shown in Table 5.2 and Figure 5.8.
Table 5.2: Interlaminar force Fxz due to mutual influence

k Pos σxy tk z Fxz
[MPa] [mm] [mm] [kN/m]
8 TOP 78.6 10−3 0.508 0.000
8 BOT 78.6 10−3 0.127 0.381 -9.982
7 TOP 78.6 10−3 0.381 -9.982
7 BOT 78.6 10−3 0.127 0.254 -19.964
6 TOP -78.6 10−3 0.254 -19.964
6 BOT -78.6 10−3 0.127 0.127 -9.982
5 TOP -78.6 10−3 0.127 -9.982
5 BOT -78.6 10−3 0.127 0.000 0.000
4 TOP -78.6 10−3 0.000 0.000
4 BOT -78.6 10−3 0.127 -0.127 9.982
3 TOP -78.6 10−3 -0.127 9.982
3 BOT -78.6 10−3 0.127 -0.254 19.964
2 TOP 78.6 10−3 -0.254 19.964
2 BOT 78.6 10−3 0.127 -0.381 9.982
1 TOP 78.6 10−3 -0.381 9.982
1 BOT 78.6 10−3 0.127 -0.508 0.000
Example 5.4 Plot σxz at the interface above the middle surface of a [±45]S laminate using
the material properties, geometry, and loading of Example 5.2.
Solution to Example 5.4 Note that in this case it is not possible to use the same sym-
metry conditions used in Example 5.2. Since the LSS is symmetric, it is possible to model
half of the laminate (z > 0). Since the LSS contains laminas at angles other than 0 and
90, the plane x = 0 is not a symmetry plane, but rather a plane with x = 0. Also, the edge
effects at the ends of the model in x = 0 and in x = L? are now important, so the results
must be plotted at x = L ? /2 to avoid free edge effects at the loaded ends. The solution is
shown in Figure 5.9.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
1
[30 /-30 ]
2 2s
0.8
0.6
0.4
0.2
z/H
-0.2
-0.4
-0.6
-0.8
Fxz/Fxz max
-1
-1 -0.5 0 0.5 1
Fxz/Fxz max
Figure 5.8: Interlaminar shear force due to mutual influence Fxz .
40
30
20
10
0
σ [MPa]
-10
-20
-30
-40
-50
σxz
-60
0 0.2 0.4 0.6 0.8 1
y/b
Figure 5.9: Interlaminar shear stress σxz at the interface above the middle-surface
of a carbon/epoxy [±45]S laminate (FEA).
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
/TITLE,Free Edge Analysis [45/-45]s laminate

*SET,THZ,1.25 ! thickness of lamina in mm
*SET,LX,8*THZ ! 1/2 length of laminate in mm
*SET,BY,10.0 ! 1/2 width of laminate in mm
*SET,NEX,8 ! number of elements in x/z direction
*SET,NEY,14 ! number of elements in y direction
*SET,EPSX,0.01 ! uniform strain in x direction

UIMP,1,EX,EY,EZ,139E3,14.5E3,14.5E3

KEYOPT,1,2,1 ! KEYOPT(2) = 1 Enables Full Integration

LOCAL,11,,0,0,0,-45 ! defines -45 degree local cs
LOCAL,12,,0,0,0,45 ! defines +45 degree local cs
! Generate Geometry
BLOCK,0,LX,0,BY,0,THZ ! -45 degrees lamina
BLOCK,0,LX,0,BY,THZ,2*THZ ! +45 degress lamina
VSYMM,Y,ALL

LSEL,R,LOC,Y,0,2*BY
LESIZE,ALL,,,NEY,20,1,,,1 ! define element size in selected lines
LSEL,R,LOC,Y,-2*BY,0
LESIZE,ALL,,,NEY,(1/20),1,,,1 ! define element size in selected lines
LSEL,R,LOC,X,LX ! reselects lines x=lx
LESIZE,ALL,,,NEY,(1/20),1,,,1 ! define ele. size in selected lines
LSEL,ALL ! select all lines
ESYS,11 ! Selects -45 degrees material orientation
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
VMESH,1 ! Meshes -45 degrees lamina

VMESH,6 ! Meshes -45 degrees lamina

ASEL,S,LOC,X,0
ASEL,A,LOC,Z,0
ASEL,S,LOC,X,LX
DA,ALL,UX,(EPSX*LX) ! Impose displacement on the end = epsX*LX
DK,2,ALL

PATH,INTERFACE,2,,100 ! Define a path between two points, compute 100 values
PPATH,1,0,LX/2,0,THZ,0 ! 1st point of the path location
PPATH,2,0,LX/2,BY,THZ,0 ! 2nd point of the path location
PDEF,SXZ,S,XZ,!AVG ! Compute Sxz
PLPATH,SXZ,ZERO ! Plot Sxz
PRPATH,SXZ ! Print Sxz
Suggested Problems
Problem 5.1 Write a computer program to use tabulated data of σy and σxy (at the top
and bottom of every lamina) to compute Fyz , Fxz , and Mz , for all locations through the
thickness of a laminate with any number of laminas. Using the program, plot Fyz , Fxz , and
Mz , through the thickness −4t < z < 4t of a [±45/0/90]s laminate with lamina thickness
t = 0.125 mm, loaded with Nx = 100 kN/m. Use carbon/epoxy properties E1 =139 GPa,
E2 =14.5 GPa, G12 = G13 = 5.86 GPa, G23 = 5.25 GPa, ν12 = ν13 = 0.21, ν23 = 0.38.
Submit a report including the source code of the program.
Problem 5.2 Repeat Problem 5.1 for Mx = 1 Nm/m. Submit a report including the source
code of the program.
Problem 5.3 Plot σz /σx0 and σyz /σx0 vs. y/b (0 < y/b < 1) at x = L/2, at the first
interface above the mid-surface for laminates, [0/90]s and with lamina thickness t = 0.512
mm, loaded with x = 0.01. Compute the far-field uniform stress σx0 in terms of the applied
strain. Use quadratic solid elements and a mesh biased toward the free edge (bias 0.1) to
model 1/8 of a tensile specimen (see Example 5.2), of width 2b = 25.4 mm and length 2L =
20 mm. Use carbon/epoxy properties E1 = 139 GPa, E2 = 14.5 GPa, G12 = G13 = 5.86
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
GPa, G23 = 5.25 GPa, ν12 = ν13 = 0.21, ν23 = 0.38. Attempt to keep the aspect ratio of
the elements near the free edge close to one. Submit the input command file to obtain the
solution and the plot. In addition, submit the plot.
Problem 5.4 For the laminate and loading described in Problem 5.3, plot σz /σx0 and
σyz /σx0 versus z/tk (0 < z/tk < 2) above the mid-surface, at a distance 0.1tk from the
free edge and x = L/2. Study the effect of mesh refinement by providing four curves with
different number of divisions along the z-direction. Attempt to keep the aspect ratio of the
elements near the free edge close to one. Submit the input command file to obtain the
solution and the plot. In addition, submit the plot.
Problem 5.5 Plot σxz /σx0 as in Problem 5.3 for all the interfaces above the middle surface
of a [±102 ]S laminate.
Problem 5.6 Plot σxz /σx0 as in Problem 5.4 for a [±102 ]S laminate.
Problem 5.7 Use solid elements and a biased mesh to model 1/8 of a tensile specimen (see
Example 5.2), of width 2b = 24 mm and length 2L = 20 mm. The laminate is [±45/0/90]s
with lamina thickness t = 0.125 mm, loaded with Nx = 175 KN/m. Use carbon/epoxy
properties E1 = 139 GPa, E2 = 14.5 GPa, G12 = G13 = 5.25 GPa, G23 = 5.86 GPa,
ν12 = ν13 = 0.21, ν23 = 0.38. Plot the three interlaminar stress components, from the edge
to the center line of the specimen, at the mid-surface of each lamina. Lump all four plots
of the same stress into a single plot. Submit the input command file to obtain the solution
and the three plots. In addition, submit the three plots.
Problem 5.8 Plot Ex /E2 , Gxy /G12 , 10νxy , −ηxy,x and −ηx,xy in the same plot vs θ in
the range −π/2 < θ < π/2 for a unidirectional single lamina oriented at an angle θ. The
material is S-glass/epoxy [1, Tables 1.3–1.4].
Problem 5.9 Using the plot from Problem 5.8 and considering a [θ1 /θ2 ]S laminate, what
are the worst combinations of values θ1 , θ2 for (a) Poisson’s mismatch and (b) shear mis-
match.
Problem 5.10 In a single plot, compare -ηxy,x of E-glass/epoxy, Kevlar49/epoxy, and

T800/3900-2 in the range -π/2 < θ < π/2 ( [1, Tables 1.3–1.4]).
Problem 5.11 Obtain contour plots of the three deformations ux , uy , uz (independently)

on the top surface of a [±45]s laminate. Use dimensions, load, and material properties of
Problem 5.7. Explain your findings.
Problem 5.12 Repeat Problem 5.11 for a [0/90]s laminate. Explain your findings.
Problem 5.13 Use solid elements and a biased mesh to model 1/4 of a tensile specimen
(Figure 5.1) of a total width 2b = 12 mm and length 2L = 24 mm. Compare in the same
plot σz vs z/H for [±15/ ± 45]s and [±(15/45)] of SCS-6/aluminum with 50% fiber volume.
Use micromechanics (6.8) to predict the unidirectional composite properties. The lamina
thickness tk = 0.25 mm. The laminate is loaded with x = 0.01.
Al-2014-T6 SCS-6
( [28, App. B]) ( [1, Tables 2.1–2.2])
E [GPa] 75.0 427.0
G [GPa] 27.0 177.9
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Problem 5.14 Use a FEA model similar to Problem 5.13 to plot σxz /σzxmax vs θ (0 < θ <
π/2) for a [±θ]s SCS-6/ Al laminate with x = 0.01.
Problem 5.15 Use the FEA model of Problem 5.13 to plot σz vs y/b (0 < y < 0.95b)
at the mid-surface of the [±15/ ± 45]s laminate. Note σz → ∞ near y = b, so the actual
value from FEA at y = b is mesh dependent. Investigate mesh dependency at y = 0.95b by
tabulating the result using different mesh densities.
Problem 5.16 Use an FEA model similar to Problem 5.13 to plot σx , σxy and σxz vs y/b
(0 < y < b) when a [±θ]s SCS-6/Al laminate is subjected to 1% axial strain (x = 0.01).
Problem 5.17 A [0/90]s laminate with properties E1 = 139 GPa, E2 = 14.5 GPa, G12 =
G13 = 5.86 GPa, G23 = 5.25 GPa, ν12 = ν13 = 0.21, ν23 = 0.38 is shown in Figure 5.1. The
strength properties of the lamina are F1t = 1550 MPa, F1c = 1090 MPa, F2t = F2c = 59
MPa, and F6 = 75 MPa. Take 2b = 20 mm, length of the sample 2L = 200 mm, thickness
of each lamina tk = 1.25 mm. Load the sample with a uniform strain x = 0.01 by applying
a uniform displacement. Use symmetry to model only the quadrant with x > 0, y > 0,
z > 0. Use orthotropic solid elements on each lamina, with at least two quadratic elements
through the thickness of each lamina. Compute the 3D Tsai-Wu failure index IF using a
USERMAT subroutine for solid elements. Obtain the contour plot of IF in each lamina (do
not use results averaging). Show all work in a report.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Chapter 6
Computational Micromechanics
In Chapter 1, the elastic properties of composite materials were assumed to be

available in the form of elastic modulus E, shear modulus G, Poisson’s ratio ν, and
so on. For heterogeneous materials such as composites, a large number of material
properties are needed, and experimental determination of these many properties is
a tedious and expensive process. Furthermore, the values of these properties change
as a function of the volume fraction of reinforcement and so on. An alternative, or
at least a complement to experimentation, is to use homogenization techniques to
predict the elastic properties of the composite in terms of the elastic properties of the
constituents (matrix and reinforcements). Since homogenization models are based
on more or less accurate modeling of the microstructure, these models are also called
micromechanics models, and the techniques used to obtain approximate values of
the composite’s properties are called micromechanics methods or techniques [1].
Micromechanics models can be classified into empirical, semiempirical, analytical,
and numerical. Accurate semiempirical models are described in [1].
This book deals only with strictly analytical or numerical models that do not
require empirical adjusting factors, so that no experimentation is required. Since
most of this book deals with 3D analysis, emphasis is placed on micromechanics
models that can estimate the whole set of elastic properties using a single model,
rather than using a disjoint collection of models based on different assumptions to
assemble the set of properties needed. Many analytical techniques of homogeniza-
tion are based on the equivalent eigenstrain method [29, 30], which considers the
problem of a single ellipsoidal inclusion embedded in an infinite elastic medium.
The Eshelby solution is used in [31] to develop a method that takes into account,
approximately, the interactions among the inclusions. One of the more used homog-
enization techniques is the self-consistent method [32], which considers a random
distribution of inclusions in an infinite medium. The infinite medium is assumed
to have properties equal to the unknown properties sought. Therefore, an iterative
procedure is used to obtain the overall moduli. Homogenization of composites with
periodic microstructure has been accomplished by using various techniques including
an extension of the Eshelby inclusion problem [29, 30], the Fourier series technique
(see Section 6.1.3 and [33, 34]), and variational principles. The periodic eigenstrain
method was further developed to determine the overall relaxation moduli of linear
151
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
viscoelastic composite materials (see Section 7.6 and [35,36]). A particular case, the
cell method for periodic media, considers a unit cell with a square inclusion [37].
The analytical procedures mentioned so far yield approximate estimates of the
exact solution of the micromechanics problem. These estimates must lie between
lower and upper bounds for the solution. Several variational principles were devel-
oped to evaluate bounds on the homogenized elastic properties of macroscopically
isotropic heterogeneous materials [38]. Those bounds depend only on the volume
fractions and the physical properties of the constituents.
In order to study the nonlinear material behavior of composites with periodic mi-
crostructure, numerical methods, mainly the finite element method, are employed.
Nonlinear finite element analysis of metal matrix composites has been studied by
looking at the behavior of the microstructure subjected to an assigned load his-
tory [39]. Bounds on overall instantaneous elastoplastic properties of composites
have been derived by using the finite element method [40].
6.1 Analytical Homogenization

As discussed in the introduction, estimates of the average properties of heteroge-
neous media can be obtained by various analytical methods. Detailed derivations
of the equations fall outside the scope of this book.
Available analytical models vary greatly in complexity and accuracy. Simple
analytical models yield formulas for the stiffness C and compliance S tensors of the
composite [37, (2.9) and (2.12)], such as
X X
C= Vi Ci Ai ; Vi Ai = I
X X
S= Vi Si Bi ; Vi Bi = I (6.1)
where Vi , Ci , Si , are the volume fraction, stiffness, and compliance tensors (in con-
tracted notation)1 of the i-th phase in the composite, respectively, and I is the 6×6
identity matrix. Furthermore, Ai , Bi , are the strain and stress concentration ten-
sors (in contracted notation) of the i-th phase [37]. For fiber reinforced composites,
i = f, m, represent the fiber and matrix phases, respectively.
6.1.1 Reuss Model

The Reuss model (also called rule of mixtures), assumes that the strain tensors2
in the fiber, matrix, and composite are the same ε = εf = εm , so, the strain
concentration tensors are all equal to the 6×6 identity matrix Ai = I. The rule of
mixtures (ROM) formulas for E1 and ν12 are derived and computed in this way.
1
Fourth-order tensors with minor symmetry are represented by a 6×6 matrix taking advantage
of contracted notation.
2
Tensors are indicated by boldface type, or by their components using index notation.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Computational Micromechanics 153
RVE 2a3
2a2
Figure 6.1: Three possible representative volume elements (RVE) for a composite
material with a periodic, square fiber array.
6.1.2 Voigt Model

The Voigt model (also called inverse rule of mixtures), assumes that the stress
tensors in the fiber, matrix, and composite are the same σ = σ f = σ m , so, the
stress concentration tensors are all equal to the 6×6 identity matrix Bi = I. The
inverse rule of mixtures (IROM) formulas for E2 and G12 are derived and computed
in this way. More realistic concentration tensors are given in [41, Appendix B].
6.1.3 Periodic Microstructure Model

If the composite has a periodic microstructure, or if it can be approximated as
having such a microstructure (see Section 6.1.4), then the Fourier series can be
used to estimate all the components of the stiffness tensor of a composite. Explicit
formulas for a composite reinforced by isotropic, circular-cylindrical fibers, which
are periodically arranged in a square array (Figure 6.1), were developed by [34]
and they are presented here. The fibers are aligned with the x1 axis, and they are
equally spaced (2a2 = 2a3 ). If the fibers are randomly distributed in the cross-
section, the resulting composite has transversely isotropic properties, as explained
in Section 6.1.4. The case of a composite reinforced with transversely isotropic
fibers is presented in [36], and the resulting equations are implemented in [9].
Because the microstructure has a square symmetry, the stiffness tensor has six
unique coefficients given by
Vf S32 S62 − S72 aS6 + bS7 a2 − b2

∗ 2S6 S3 aS3
C11 = λm + 2 µm − − − + + +
D µ2m µ2m g µm c µ2m g 2 µm gc 4 c2

∗ Vf S3 S6 − S7 a + b
C12= λm + b − −
D 2cµm 2cµm g 4 c2
ba + b2

∗ Vf aS7
C23 = λm + −
D 2 µm gc 4 c2
a2 − b2

∗ Vf aS3 aS6
C22 = λm + 2 µm − − + +
D 2 µm c 2 µm gc 4 c2
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
−1
∗ 2 S3 −1 4 S7
C44 = µm − Vf − + (µm − µf ) +
µm µm (2 − 2νm )
−1
S3
∗
C66 = µm − Vf − + (µm − µf )−1 (6.2)
µm
where
aS32 aS6 S3 a(S62 − S72 )
D= − + + (6.3)
2 µ2m c µ2m gc 2µ2m g 2 c
S3 (b2 − a2 ) S6 (a2 − b2 ) + S7 (ab + b2 ) (a3 − 2b3 − 3 ab2 )
+ + +
2 µm c2 2 µm gc2 8 c3
and
a = µf − µm − 2 µf νm + 2 µm νf
b = −µm νm + µf νf + 2 µm νm νf − 2 µ1 νm νf
c = (µm − µf )(µf − µm + µf νf − µm νm + 2 µm νf − 2 µf νm +
+ 2 µm νm νf − 2 µf νm νf )
g = (2 − 2νm ) (6.4)
The subscripts ()m , ()f refer to matrix and fiber, respectively. Assuming the
fiber and matrix are both isotropic (Section 1.12.5), Lamé constants of both mate-
rials are obtained by using (1.75) in terms of the Young’s modulus E, the Poisson’s
ratio ν, and the shear modulus G = µ.
For a composite reinforced by long circular cylindrical fibers, periodically ar-
ranged in a square array (Figure 6.1), aligned with x1 -axis, with a2 = a3 , the
constants S3 , S6 , S7 are given as follows [34]
S3 = 0.49247 − 0.47603Vf − 0.02748Vf2

S6 = 0.36844 − 0.14944Vf − 0.27152Vf2 (6.5)
S7 = 0.12346 − 0.32035Vf + 0.23517Vf2
The resulting tensor C∗ has a square symmetry (not transverse isotropy) due to
the microstructural periodic arrangement in the form of a square array. The tensor
C∗ is therefore described by six constants. However, most composites have random
arrangement of the fibers (see Figure 1.12), resulting in a transversely isotropic
stiffness tensor, with only five independent constants. Therefore, the tensor C for a
transversely isotropic material is derived from the tensor C∗ in Section 6.1.4, next.
6.1.4 Transversely Isotropic Averaging

In order to obtain a transversely isotropic stiffness tensor (Section 1.12.4), equiva-
lent in the average sense to the stiffness tensor with square symmetry, the following
averaging procedure is used. A rotation θ of the tensor C∗ about the x1 -axis pro-
duces
T
B(θ) = T (θ)C∗ T (θ) (6.6)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
where T (θ) is the coordinate transformation matrix (see (1.50)). Then the equiva-
lent transversely isotropic tensor is obtained by averaging as follows
Zπ
1
B= B(θ)dθ (6.7)
π
0
Then, using the relations between the engineering constants and the components
of the B tensor, the following expressions are obtained explicitly in terms of the
coefficients (6.2–6.5) of the tensor C∗
∗2
2 C12
∗
E1 = C11 − ∗ + C∗
C22 23
∗ C ∗ + 2 C ∗ C ∗ − 4 C ∗2 (C ∗ − C ∗ + 2 C ∗ )

2 C11 22 11 23 12 22 23 44
E2 = ∗ C ∗ + C ∗ C ∗ + 2 C ∗ C ∗ − 4 C ∗2
3 C11 22 11 23 11 44 12
∗
G12 = G13 = C66
C∗
ν12 = ν13 = ∗ 12 ∗
C22 + C23
∗ ∗
C C + 3 C11 ∗ C ∗ − 2 C ∗ C ∗ − 4 C ∗2
ν23 = 11∗ 22∗ 23 11 44 12
∗ C ∗ + 2 C ∗ C ∗ − 4 C ∗2 (6.8)
3 C11 C22 + C11 23 11 44 12
Note that the transverse shear modulus G23 can be written in terms of the other
engineering constants as
∗
C22 C∗ C∗ E2
G23 = − 23 + 44 =
4 4 2 2(1 + ν23 )
or directly in terms of µm , µf as
f aS3 a(S7 + S6 ) ba + 2b2 − a2

G23 = µm − [(− + − )
4D 2 µm c 2 µm gc 4 c2
(6.9)
2 S3 4 S7
+2(− + (µm − µf )−1 + )−1 ]
µm µm (2 − 2νm )
where D is given by (6.3), a, b, c and g are given by (6.4) and S3 , S6 , and S7 can be
evaluated by (6.5). These equations are implemented in PMMIE.m and PMMIE.xls,
available in [5, /Examples/Ch6Ex/]. For the case of transversely isotropic fibers,
they are implemented in [9].
Example 6.1 Compute the elastic properties of a composite material reinforced with par-
allel cylindrical fibers randomly distributed in the cross-section. The constituent properties
are Ef = 241 GP a, νf = 0.2, Em = 3.12 GP a, νm = 0.38, fiber volume fraction Vf = 0.4.
Solution to Example 6.1 The results shown in Table 6.1 are obtained using [9], which
implements the periodic microstructure model (PMM) equations for the case of transversely
isotropic fibers.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Table 6.1: Lamina elastic properties for Vf = 0.4

Young’s Moduli Poisson’s Ratio Shear Moduli
E1 = 98, 306 M P a ν12 = ν13 = 0.298 G12 = G13 = 2, 594 M P a
E2 = E3 = 6, 552 M P a ν23 = 0.600
Figure 6.2: Composite material with hexagonal array.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 6.3: Cross-section of the composite material.
6.2 Numerical Homogenization
The composite material considered in this section has cylindrical fibers of infinite
length, embedded in an elastic matrix, as shown in Figure 6.2. The cross-section
of the composite obtained by intersecting with a plane orthogonal to the fiber axis
is shown in Figure 6.3, which clearly shows a periodic microstructure. Because of
the periodicity, the three-dimensional representative volume element (RVE) shown
in Figure 6.4 can be used for finite element analysis.
In general, composites reinforced with parallel fibers display orthotropic mate-

rial properties (Section 1.12.3) at the mesoscale (lamina level). In special cases,
such as the hexagonal array shown in Figures 6.2 and 6.3, the properties become
transversely isotropic (Section 1.12.4). In most commercially fabricated compos-
ites, it is impossible to control the placement of the fibers so precisely and most
of the time the resulting microstructure is random, as shown in Figure 1.12. A
random microstructure results in transversely isotropic properties at the mesoscale.
The analysis of composites with random microstructure still can be done using a
fictitious periodic microstructure, such as that shown in Figure 6.1, then averag-
ing the stiffness tensor C as in Section 6.1.4 to obtain the stiffness tensor of a
transversely isotropic material. A simpler alternative is to assume that the random
microstructure is well approximated by the hexagonal microstructure displayed in
Figure 6.3. Analysis of such microstructure directly yields a transversely isotropic
stiffness tensor, represented by (1.70), which is reproduced here for convenience
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 6.4: Representative volume element (RVE).
     

 σ 1 
 C11 C12 C12 0 0 0 
 1 




 σ 2




 C12 C22 C23 0 0 0 
 
  2 


   
σ3 C12 C23 C22 0 0 0 3
    
= 1
 (6.10)
 σ4  0 0 0 2 (C22 − C23 ) 0 0 γ4
    
    
σ   0 0 0 0 C66 0 γ5
    
 5 

  
 

  
σ6

0 0 0 0 0 C66

γ6

where the 1-axis aligned with the fiber direction and an over-bar indicates the aver-
age computed over the volume of the RVE. Once the components of the transversely
isotropic tensor C are known, the five elastic properties of the homogenized material
can be computed by (6.11), i.e., the longitudinal and transversal Young’s moduli
E1 and E2 , the longitudinal and transversal Poisson’s ratios ν12 and ν23 , and the
longitudinal shear modulus G12 , as follows
E1 = C11 − 2C12 2 /(C + C )

22 23
ν12 =C 12 /(C22 + C 23 )
2 (C − C )/ C C − C 2

E2 = C11 (C22 + C23 ) − 2C12 22 23 11 22 12 (6.11)
2 / C C − C2

ν23 = C11 C23 − C12 11 22 12
G12 = C66
The shear modulus G23 in the transversal plane can be obtained by the classical
relation (1.74) or directly as follows
1 E2
G23 = C44 = (C22 − C23 ) = (6.12)
2 2 (1 + ν23 )
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
In order to evaluate the overall elastic matrix C of the composite, the RVE is
subjected to an average strain β [42]. The six components of strain ε0ij are applied
by enforcing the following boundary conditions on the displacement components
−a2 ≤ x2 ≤ a2
ui (a1 , x2 , x3 ) − ui (−a1 , x2 , x3 ) = 2a1 ε0i1 (6.13)
−a3 ≤ x3 ≤ a3
−a1 ≤ x1 ≤ a1
ui (x1 , a2 , x3 ) − ui (x1 , −a2 , x3 ) = 2a2 ε0i2 (6.14)
−a3 ≤ x3 ≤ a3
−a1 ≤ x1 ≤ a1
ui (x1 , x2 , a3 ) − ui (x1 , x2 , −a3 ) = 2a3 ε0i3 (6.15)
−a2 ≤ x2 ≤ a2
Note that tensor components of strain, defined in (1.5) are used in (6.13-6.15).
Also, note that a superscript ()0 indicates an applied strain, while a bar indicates
a volume average. Furthermore, 2aj ε0ij is the displacement necessary to enforce a
strain ε0ij over a distance 2aj (Figure 6.4).
The strain ε0ij applied on the boundary by using (6.13–6.15) results in a complex
state of strain inside the RVE. However, the volume average of the strain in the
RVE equals the applied strain,3 i.e.,
Z
1
εij = εij dV = ε0ij (6.16)
V V
For the homogeneous composite material, the relationship between average stress
and strain is
σ α = Cαβ β (6.17)
where the relationship between i, j = 1..3 and β = 1..6 is given by the definition of
contracted notation in (1.9). Thus, the components of the tensor C are determined
solving six elastic models of the RVE subjected to the boundary conditions (6.13–
6.15), where only one component of the strain 0β is different from zero for each of
the six problems.
By choosing a unit value of applied strain, and once the problem defined by the
boundary conditions (6.13–6.15) is solved, it is possible to compute the stress field
σα , whose average gives the required components of the elastic matrix, one column
at a time, as
Z
1
Cαβ = σ α = σα (x1 , x2 , x3 ) dV with 0β = 1 (6.18)
V V
where α, β = 1 . . . 6 (see Section 1.5). The integrals (6.18) are evaluated within each
finite element using the Gauss-Legendre quadrature. Commercial programs, such as
ANSYS R
, have the capability to compute the average stress and volume, element
by element. Therefore, computation of the integral (6.18) is a trivial matter. For
more details see Example 6.2.
The coefficients in C are found by setting a different problem for each column
in (6.10), as follows.
3
As long as there are no discontinuities, such as voids or cracks, inside the RVE.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
First Column of C
In order to determine the components Ci1 , with i = 1, 2, 3, the following strain is
applied to stretch the RVE in the fiber direction (x1 -direction)
o1 = 1 o2 = o3 = γ4o = γ5o = γ6o = 0 (6.19)
Thus, the displacement boundary conditions (6.13–6.15) for the RVE in Figure
6.4 become
u1 (+a1 , x2 , x3 ) − u1 (−a1 , x2 , x3 ) = 2a1

−a2 ≤ x2 ≤ a2
u2 (+a1 , x2 , x3 ) − u2 (−a1 , x2 , x3 ) = 0
−a3 ≤ x3 ≤ a3
u3 (+a1 , x2 , x3 ) − u3 (−a1 , x2 , x3 ) = 0
−a1 ≤ x1 ≤ a1 (6.20)
ui (x1 , +a2 , x3 ) − ui (x1 , −a2 , x3 ) = 0
−a3 ≤ x3 ≤ a3
−a1 ≤ x1 ≤ a1
ui (x1 , x2 , +a3 ) − ui (x1 , x2 , −a3 ) = 0
−a2 ≤ x2 ≤ a2
The conditions (6.20) are constraints on the relative displacements between op-
posite faces of the RVE. Because of the symmetries of the RVE and symmetry of
the constraints (6.20), only one-eighth of the RVE needs to be modeled in finite ele-
ment analysis (FEA). Assuming the top-right-front portion is modeled (Figure 6.5),
the following equivalent external boundary conditions, i.e., boundary conditions on
components of displacements and stresses, can be used
u1 (a1 , x2 , x3 ) = a1
u1 (0, x2 , x3 ) = 0
σ12 (a1 , x2 , x3 ) = 0 0 ≤ x2 ≤ a2
σ12 (0, x2 , x3 ) = 0 0 ≤ x3 ≤ a3
σ13 (a1 , x2 , x3 ) = 0
σ13 (0, x2 , x3 ) = 0
u2 (x1 , a2 , x3 ) = 0
u2 (x1 , 0, x3 ) = 0
σ21 (x1 , a2 , x3 ) = 0 0 ≤ x1 ≤ a1
(6.21)
σ21 (x1 , 0, x3 ) = 0 0 ≤ x3 ≤ a3
σ23 (x1 , a2 , x3 ) = 0
σ23 (x1 , 0, x3 ) = 0
u3 (x1 , x2 , a3 ) = 0
u3 (x1 , x2 , 0) = 0
σ31 (x1 , x2 , a3 ) = 0 0 ≤ x1 ≤ a1
σ31 (x1 , x2 , 0) = 0 0 ≤ x2 ≤ a2
σ32 (x1 , x2 , a3 ) = 0
σ32 (x1 , x2 , 0) = 0
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 6.5: One-eighth model of the RVE. Note that the model is set up with the
fiber along the z-axis, which corresponds to the x1 -direction in the equations.
These boundary conditions are very easy to apply. Symmetry boundary con-
ditions are applied on the planes x1 = 0, x2 = 0, x3 = 0. Then, a uniform
displacement is applied on the plane x1 = a1 . The stress boundary conditions do
not need to be applied explicitly in a displacement-based formulation. The displace-
ment components in (6.21) represent strains that are not zero along the x1 -direction
and zero along the other two directions. The stress boundary conditions listed in
(6.21) reflect the fact that, in the coordinate system used, the composite material
is macroscopically orthotropic and that the constituent materials are orthotropic
too. Therefore, there is no coupling between extension and shear strains. This is
evidenced by the zero coefficients above the diagonal in columns 4 to 6 in (6.10).
The coefficients in column one of (6.10) are found by using (6.18), as follows
Z
1
Cα1 = σα = σα (x1 , x2 , x3 ) dV (6.22)
V V
Second Column of C
The components Cα2 , with α = 1, 2, 3, are determined by setting
o2 = 1 o1 = o3 = γ4o = γ5o = γ6o = 0 (6.23)
Thus, the following boundary conditions on displacements can be used
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
u1 (a1 , x2 , x3 ) = 0
u1 (0, x2 , x3 ) = 0
u2 (x1 , a2 , x3 ) = a2
(6.24)
u2 (x1 , 0, x3 ) = 0
u3 (x1 , x2 , a3 ) = 0
u3 (x1 , x2 , 0) = 0
The trivial stress boundary conditions have not been listed because they are
automatically enforced by the displacement-based FEA formulation. Using (6.18),
the stiffness terms in the second column of C are computed as
Z
1
Cα2 = σ α = σα2 (x1 , x2 , x3 ) dV (6.25)
V V
Third Column of C
Because of the transverse isotropy of the material (6.10), the components of the
third column of the matrix C can be determined from the first and the second
column, so no further computation is required. However, if desired, the components
Cα3 , with α = 1, 2, 3, can be found by applying the following strain
o3 = 1 o1 = o2 = γ4o = γ4o = γ5o = 0 (6.26)

Thus, the following boundary conditions on displacement can be used
u1 (a1 , x2 , x3 ) = 0
u1 (0, x2 , x3 ) = 0
u2 (x1 , a2 , x3 ) = 0
(6.27)
u2 (x1 , 0, x3 ) = 0
u3 (x1 , x2 , a3 ) = a3
u3 (x1 , x2 , 0) = 0
The required components of C are determined by averaging the stress field as
in (6.18).
Example 6.2 Compute E1 , E2 , ν12 , and ν23 for a unidirectional composite with isotropic
fibers Ef = 241 GPa, νf = 0.2, and isotropic matrix Em = 3.12 GPa, νm = 0.38 with fiber
volume fraction Vf = 0.4. The fiber diameter is df = 7 µm, placed in a hexagonal array as
Solution to Example 6.2 The dimensions a2 and a3 of the RVE, as shown in Figure 6.4,
are chosen to obtain Vf = 0.4 with a hexagonal array microstructure. The fiber volume and
the total volume of the RVE are
2
df
vf = 4a1 π ; vt = 2a1 2a2 2a3
2
The ratio between both is the volume fraction. Therefore,
2
(df /2)
Vf = π = 0.4
2 a2 a3
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Additionally, the relation between a2 and a3 is established by the hexagonal array pattern
a3 = a2 tan(60◦ )
These two relations yield a2 and a3 , while the a1 dimension can be chosen arbitrarily.
In this case, the RVE dimensions are
a1 = a2 /4 ; a2 = 5.2701 µm ; a3 = 9.1281 µm
Since this RVE is symmetric, it is possible to model one-eighth of the RVE, as shown
in Figure 6.5. The ANSYS R
command list below is used to model one-eighth of the RVE.
/TITLE,ONE-EIGHT Symmetric Model of RVE hexagonal array

!Define variables for parametric modeling
rf=3.5 ! Radius fiber in microns
a2=5.2701 ! x2 length in microns
a1=a2/4 ! x1 length in microns

MP,EX,1,0.241 ! Fiber material properties in TeraPascals [TPa]
MP,PRXY,1,0.2
MP,EX,2,3.12e-3 ! Matrix material properties in TeraPascals [TPa]
MP,PRXY,2,0.38
ET,1,SOLID186 ! Choose SOLID186 element type
BLOCK,0,a2,0,a3,0,a1 ! Geometry definition

CYLIND,rf,,0,a1,0,90
WPOFF,a2,a3
CYLIND,rf,,0,a1,180,270
VOVLAP,all ! Overlap volumes
NUMCMP,all ! Renumbering volumes
/DEVICE,VECTOR,1
/VIEW,1,1,2,3
/ANG,1
/PNUM,VOLU,1
/PNUM,MAT,1
/REPLOT
LSEL,U,LOC,Z,A1 ! MESHING CONTROL

LSEL,U,LOC,Z,0
LESIZE,ALL,,,2 ! NUMBER OF DIVISIONS TROUGTH THE THICKNESS
VSEL,S,,,1,2
ASLV,S
LSLA,S
LESIZE,ALL,,,6 ! NUMBER OF DIVISIONS ON THE FIBER
LSEL,S,LOC,Y,A3
LSEL,A,LOC,Y,0
LESIZE,ALL,,,3 ! NUMBER OF DIVISIONS ON THE MATRIX
ALLSEL,ALL
LESIZE,ALL,,,8 ! NUMBER OF DIVISIONS ON THE MATRIX
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
MAT,1 ! ASSOCIATE MATERIAL #1 WITH VOLUMES 1 AND 2

VMESH,1,2 ! MESH VOLUME 1 AND 2
MAT,2 ! ASSOCIATE MATERIAL #2 WITH VOLUME 3
VSWEEP,3 ! MESH BY SWEEP PROCEDURE VOLUME 3
EPLOT
The boundary conditions are defined in three load steps, which are then used to obtain
the coefficients Cαβ in columns one, two, and three. A unit strain is applied along each
direction, each time. Equation (6.18) is then used to obtain the stiffness coefficients.

LSCLEAR,ALL ! BOUNDARY CONDITIONS COLUMN 1

ASEL,S,LOC,X,0 ! MODEL X DIRECTION = 2 MATERIAL DIRECTION
ASEL,A,LOC,X,A2
DA,ALL,UX,0
ASEL,S,LOC,Y,0 ! MODEL Y DIRECTION = 3 MATERIAL DIRECTION
ASEL,A,LOC,Y,A3
DA,ALL,UY,0
ASEL,S,LOC,Z,0 ! MODEL Z DIRECTION = 1 MATERIAL DIRECTION
DA,ALL,UZ,0
ASEL,S,LOC,Z,A1
DA,ALL,UZ,A1
ASEL,ALL
LSWRITE,1
LSCLEAR,ALL ! boundary conditions column 2

ASEL,S,LOC,X,0 ! model x direction = 2 material direction
DA,ALL,UX,0
ASEL,S,LOC,X,A2
DA,ALL,UX,A2
ASEL,S,LOC,Y,0 ! model y direction = 3 material direction
ASEL,A,LOC,Y,A3
DA,ALL,UY,0
ASEL,S,LOC,Z,0 ! model z direction = 1 material direction
ASEL,A,LOC,Z,A1
DA,ALL,UZ,0
ASEL,ALL
LSWRITE,2
LSCLEAR,ALL ! boundary conditions column 3

ASEL,S,LOC,X,0 ! model x direction = 2 material direction
ASEL,A,LOC,X,A2
DA,ALL,UX,0
ASEL,S,LOC,Y,0 ! model y direction = 3 material direction
DA,ALL,UY,0
ASEL,S,LOC,Y,A3
DA,ALL,UY,A3
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
ASEL,S,LOC,Z,0 ! model z direction = 1 material direction

ASEL,A,LOC,Z,A1
DA,ALL,UZ,0
ASEL,ALL
LSWRITE,3
LSSOLVE,1,3 ! Solve all load sets

The APDL language macro srecover, shown below, is defined in order to compute the
average stress in the RVE.
*CREATE,SRECOVER !,mac ! macro to calculate average stress

/NOPR
ETABLE, ,VOLU, ! Get element volume
ETABLE, ,S,X ! Get element stress
ETABLE, ,S,Y
ETABLE, ,S,Z
ETABLE, ,S,XY
ETABLE, ,S,XZ
ETABLE, ,S,YZ
ETABLE, ,S,YZ
SMULT,SXV,VOLU,SX,1,1, ! Stress by element volume
SMULT,SYV,VOLU,SY,1,1,
SMULT,SZV,VOLU,SZ,1,1,
SMULT,SXYV,VOLU,SXY,1,1,
SMULT,SXZV,VOLU,SXZ,1,1,
SMULT,SYZV,VOLU,SYZ,1,1,
SSUM
*GET,TOTVOL,SSUM,,ITEM,VOLU ! integrate stress
*GET,TOTSX ,SSUM,,ITEM,SXV
*GET,TOTSY ,SSUM,,ITEM,SYV
*GET,TOTSZ ,SSUM,,ITEM,SZV
*GET,TOTSXY ,SSUM,,ITEM,SXYV
*GET,TOTSXZ ,SSUM,,ITEM,SXZV
*GET,TOTSYZ ,SSUM,,ITEM,SYZV
SXX0 = TOTSX/TOTVOL ! compute average stress

SYY0 = TOTSY/TOTVOL
SZZ0 = TOTSZ/TOTVOL
SXY0 = TOTSXY/TOTVOL
SXZ0 = TOTSXZ/TOTVOL
SYZ0 = TOTSYZ/TOTVOL
/GOPR
*END !SRECOVER
The coefficients Cαβ and the equivalent engineering elastic constants are computed using
the previous macro, as follows.

/DEVICE,VECTOR,0
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Table 6.2: Calculated elastic properties of the unidirectional lamina.

Property PMM FEA
E1 [MPa] 98,306 98,302
E2 [MPa] 6,552 7,479
ν12 = ν13 0.298 0.298
ν23 0.6 0.540
G12 2,594 (*)
(*) Not possible with the boundary conditions used in this example.
PLESOL,S,Z,1
SET,1 ! First column coefficients

*USE,SRECOVER
C11 = Szz0
C21 = Sxx0
C31 = Syy0
SET,2 ! Second column coefficients

*USE,SRECOVER
C12 = Szz0
C22 = Sxx0
C32 = Syy0
SET,3 ! Third column coefficients

*USE,SRECOVER
C13 = Szz0
C23 = Sxx0
C33 = Syy0
EL=C11-2*C12*C21/(C22+C23) ! Longitudinal E1 modulus

nuL=C12/(C22+C23) ! 12 Poisson coefficient
ET=(C11*(C22+C23)-2*C12*C12)*(C22-C23)/(C11*C22-C12*C21)
! Transversal E2 modulus
nuT=(C11*C23-C12*C21)/(C11*C22-C12*C21) ! 23 Poisson coefficient
GT=(C22-C23)/2 ! or GT=ET/2/(1+nuT) ! 23 Shear stiffness
You need to look for the results in the ANSYS output window, which is min-
imized in the Windows Taskbar. The results are in the same units as the elastic
properties, i.e., TPa; then converted to MPa to be displayed in Table 6.2.
Fourth Column of C
If the material is orthotropic, a procedure similar to that used for column number
six must be used. But for a transversally isotropic material, only the term C44 is
nonzero in column 4 of (6.10) and it can be determined as a function of the other
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
components as
1
C44 = (C22 − C23 ) (6.28)
2
Fifth Column of C
If the material is orthotropic, a procedure similar to that used for column number
six must be used. But for a transversally isotropic material, only the term C55 = C66
is nonzero in column 5 of (6.10) and it can be found from column number six.
Sixth Column of C
Because of the lack of symmetry of the loads, in this case it is not possible to use
boundary conditions as was done for the first three columns. Thus, the boundary
conditions must be enforced by using coupling constraint equations (called CE in
most FEA commercial packages).
According to (6.10), only the term C66 is different from zero. The components
Cα6 are determined by setting
γ60 = ε012 + ε021 = 1.0 01 = 02 = 03 = γ40 = γ50 = 0 (6.29)
Note that ε012 = 1/2 is applied between x1 = ±a1 and another one-half is applied
between x2 = ±a2 . In this case, the CE applied between two periodic faces (except
points in the edges and vertices) are given as a particular case of (6.13–6.15) as
follows
u1 (a1 , x2 , x3 ) − u1 (−a1 , x2 , x3 ) = 0
−a2 < x2 < a2
u2 (a1 , x2 , x3 ) − u2 (−a1 , x2 , x3 ) = a1
−a3 < x3 < a3
u3 (a1 , x2 , x3 ) − u3 (−a1 , x2 , x3 ) = 0
u1 (x1 , a2 , x3 ) − u1 (x1 , −a2 , x3 ) = a2
−a1 < x1 < a1
u2 (x1 , a2 , x3 ) − u2 (x1 , −a2 , x3 ) = 0 (6.30)
−a3 < x3 < a3
u3 (x1 , a2 , x3 ) − u3 (x1 , −a2 , x3 ) = 0
u1 (x1 , x2 , a3 ) − u1 (x1 , x2 , −a3 ) = 0
−a1 < x1 < a1
u2 (x1 , x2 , a3 ) − u2 (x1 , x2 , −a3 ) = 0
−a2 < x2 < a2
u3 (x1 , x2 , a3 ) − u3 (x1 , x2 , −a3 ) = 0
Note that (6.30) are applied between opposite points on the faces of the RVE
but not on edges and vertices. In FEA, CE are applied between degrees of freedom
(DOF). Once a DOF has been used in a CE, it cannot be used in another CE. For
example, the first of (6.30) for x2 = a2 becomes
u1 (a1 , a2 , x3 ) − u1 (−a1 , a2 , x3 ) = 0 (6.31)
The DOF associated to u1 (a1 , a2 , x3 ) (for all −a3 < x3 < a3 ) are eliminated
because they are identical to u1 (−a1 , a2 , x3 ), as required by (6.31) and enforced by
a CE based on the same. Once the DOF are eliminated, they cannot be used in
another CE. For example, the fourth of (6.30) at x1 = a1 is
u1 (a1 , a2 , x3 ) − u1 (a1 , −a2 , x3 ) = 0 (6.32)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 6.6: Top view of the RVE showing that two displacements (vertical and
horizontal) must be applied at edges to impose shear strain (shown as points A, B, C,
and D in the figure).
but this CE cannot be enforced because the DOF associated to u1 (a1 , a2 , x3 ) have
been eliminated by the CE associated to (6.31). As a corollary, constraint equa-
tions on the edges and vertices of the RVE must be written separately from (6.30).
Furthermore, only three equations, one for each component of displacement ui can
be written between a pair of edges or pair of vertices. Simply put, there are only
three displacements that can be used to enforce periodicity conditions.
For pairs of edges, the task at hand is to reduce the first six equations of (6.30)
to three equations that can be applied between pairs of edges for the interval −a3 <
x3 < a3 . Note that the new equations will not be applied at x3 = ±a3 because
those are vertices, which will be dealt with separately. Therefore, the last three
equations of (6.30) are inconsequential at this point.
The only way to reduce six equations to three, in terms of six unique DOF, is
to add the equations for diagonally opposite edges. Figure 6.6 is a top view of the
RVE looking from the positive x3 axis. Point A in Figure 6.6 represents the edge
formed by the planes x1 = a1 and x2 = a2 . This location is constrained by the first
of (6.30) at that location, which is precisely (6.31). Point C in Figure 6.6 represents
the edge formed by the planes x1 = −a1 and x2 = −a2 . This location is constrained
by the fourth of (6.30), which at that location reduces to
u1 (−a1 , a2 , x3 ) − u1 (−a1 , −a2 , x3 ) = a2 (6.33)
Adding (6.31) and (6.33) yields a single equation as follows
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
u1 (a1 , a2 , x3 ) − u1 (−a1 , −a2 , x3 ) = a2 (6.34)
Repeating the procedure for the components u2 and u3 , and grouping the re-
sulting equations with (6.34) results in
u1 (a1 , a2 , x3 ) − u1 (−a1 , −a2 , x3 ) = a2

u2 (a1 , a2 , x3 ) − u2 (−a1 , −a2 , x3 ) = a1 −a3 < x3 < a3 (6.35)
u3 (a1 , a2 , x3 ) − u3 (−a1 , −a2 , x3 ) = 0
Considering (6.30) between edges B and D in Figure 6.6 results in
u1 (a1 , −a2 , x3 ) − u1 (−a1 , a2 , x3 ) = −a2

u2 (a1 , −a2 , x3 ) − u2 (−a1 , a2 , x3 ) = a1 −a3 < x3 < a3 (6.36)
u3 (a1 , −a2 , x3 ) − u3 (−a1 , a2 , x3 ) = 0
The planes x1 = ±a1 and x3 = ±a3 define two pairs of edges restrained by the
following six CE
u1 (+a1 , x2 , +a3 ) − u1 (−a1 , x2 , −a3 ) = 0

u2 (+a1 , x2 , +a3 ) − u2 (−a1 , x2 , −a3 ) = a1 −a2 < x2 < a2
u3 (+a1 , x2 , +a3 ) − u3 (−a1 , x2 , −a3 ) = 0
(6.37)
u1 (+a1 , x2 , −a3 ) − u1 (−a1 , x2 , +a3 ) = 0
u2 (+a1 , x2 , −a3 ) − u2 (−a1 , x2 , +a3 ) = a1 −a2 < x2 < a2
u3 (+a1 , x2 , −a3 ) − u3 (−a1 , x2 , +a3 ) = 0
The six CE for the two pairs of edges defined by the planes x2 = ±a2 and
x3 = ±a3 are
u1 (x1 , +a2 , +a3 ) − u1 (x1 , −a2 , −a3 ) = a2

u2 (x1 , +a2 , +a3 ) − u2 (x1 , −a2 , −a3 ) = 0 −a1 < x1 < a1
u3 (x1 , +a2 , +a3 ) − u3 (x1 , −a2 , −a3 ) = 0
(6.38)
u1 (x1 , +a2 , −a3 ) − u1 (x1 , −a2 , +a3 ) = a2
u2 (x1 , +a2 , −a3 ) − u2 (x1 , −a2 , +a3 ) = 0 −a1 < x1 < a1
u3 (x1 , +a2 , −a3 ) − u3 (x1 , −a2 , +a3 ) = 0
Note that (6.35–6.38) are not applied at the vertices because redundant CE
would appear among pairs of vertices that are located symmetrically with respect
to the center of the RVE’s volume. Therefore, each of the four pairs of vertices need
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
to be constrained one at a time. The resulting CE are as follows
u1 (+a1 , +a2 , +a3 ) − u1 (−a1 , −a2 , −a3 ) = a2

u2 (+a1 , +a2 , +a3 ) − u2 (−a1 , −a2 , −a3 ) = a1
u3 (+a1 , +a2 , +a3 ) − u3 (−a1 , −a2 , −a3 ) = 0
u1 (+a1 , +a2 , −a3 ) − u1 (−a1 , −a2 , +a3 ) = a2
u2 (+a1 , +a2 , −a3 ) − u2 (−a1 , −a2 , +a3 ) = a1
u3 (+a1 , +a2 , −a3 ) − u3 (−a1 , −a2 , +a3 ) = 0
(6.39)
u1 (−a1 , +a2 , +a3 ) − u1 (+a1 , −a2 , −a3 ) = a2
u2 (−a1 , +a2 , +a3 ) − u2 (+a1 , −a2 , −a3 ) = −a1
u3 (−a1 , +a2 , +a3 ) − u3 (+a1 , −a2 , −a3 ) = 0
u1 (+a1 , −a2 , +a3 ) − u1 (−a1 , +a2 , −a3 ) = −a2
u2 (+a1 , −a2 , +a3 ) − u2 (−a1 , +a2 , −a3 ) = a1
u3 (+a1 , −a2 , +a3 ) − u3 (−a1 , +a2 , −a3 ) = 0
Equations (6.30–6.39) constrain the volume of the RVE with a unit strain given
by (6.29). The FEA of this model yields all the component of stress. As discussed
previously, element by element averages of these components of stress are available
from the FEA (see macro srecover in Example 6.1) or they can be easily computed
by post-processing. Therefore, the coefficient C66 , for this case is found using (6.18)
written as Z
1
C66 = σ 6 = σ6 (x1 , x2 , x3 ) dV with γ60 = 1 (6.40)
V V
Finally, the elastic properties of the composite are determined using (6.11).
Example 6.3 Compute G12 for the composite in Example 6.2.
Solution to Example 6.3 To compute G12 = C66 , the RVE shown in Figure 6.4 must be
used along with the CE explained in (6.30–6.39). The dimensions to define the RVE are
the same used in Example 6.1. Therefore, the fiber diameter is df = 7 µm and the RVE
dimensions are
a1 = a2 /4 ; a2 = 5.2701 µm ; a3 = 9.1281 µm
See the ANSYS command list below to model the whole RVE.
/TITLE, Full Model of RVE, hexagonal array
rf=3.5 ! Radius fiber in microns

a1=a2/4 ! x1 length in microns

MP,EX,1,0.241 ! Fiber material properties in TeraPascals [TPa]
MP,PRXY,1,0.2
MP,EX,2,3.12e-3 ! Matrix material properties in TeraPascals [TPa]
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
MP,PRXY,2,0.38
ET,1,SOLID186 ! Choose SOLID186 element type
BLOCK,-A2,A2,-A3,A3,-A1,A1,
CYLIND,RF, ,-A1,A1, 0, 90,
CYLIND,RF, ,-A1,A1, 90,180,
CYLIND,RF, ,-A1,A1,180,270,
CYLIND,RF, ,-A1,A1,270,360,
CYLIND,RF, ,-A1,A1, 0, 90,
CYLIND,RF, ,-A1,A1, 90,180,
CYLIND,RF, ,-A1,A1,180,270,
CYLIND,RF, ,-A1,A1,270,360,
VGEN,1,6,,,-A2,-A3,,,,1
VGEN,1,7,,, A2,-A3,,,,1
VGEN,1,8,,, A2, A3,,,,1
VGEN,1,9,,,-A2, A3,,,,1
ALLSEL,ALL
VOVLAP,all ! Overlap volumes
NUMCMP,all ! Renumbering all volumes, volume 9 is the matrix
/DEVICE,VECTOR,1
/VIEW,1,1,2,3
/ANG,1
/PNUM,VOLU,1
/PNUM,MAT,1
/REPLOT
LSEL,U,LOC,Z,A1 ! meshing control

LSEL,U,LOC,Z,-A1
LESIZE,ALL,,,4 ! number of divisions trougth the thickness
VSEL,S,,,1,8
ASLV,S
LSLA,S
LESIZE,ALL,,,6 ! number of divisions on the fiber
LSEL,S,LOC,Y,A3
LSEL,A,LOC,Y,-A3
LESIZE,ALL,,,6 ! number of divisions on the matrix
ALLSEL,ALL
LESIZE,ALL,,,16 ! number of divisions on the matrix
MAT,1 ! Associate material #1 with volumes 1 and 2

VMESH,1,8 ! Mesh volume 1 and 2
MAT,2 ! Associate material #2 with volume 3
VSWEEP,9 ! Mesh by sweep procedure volume 3
EPLOT
The APDL macro ceRVE.mac, available in [5], is used to define the CE and to implement
(6.30–6.39). The macro is made available on the Web site because it is too long to be printed
here. The RVE dimensions and the applied strain are input arguments to the macro. In
this example, only a strain γ6 = 1.0 is applied.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Table 6.3: In-plane shear modulus of the unidirectional lamina

Property PMM FEA
G12 [M P a] 2,594 2,583

! ceRVe arguments:
! a1,a2,a3,eps1,eps2,eps3,eps4,eps5,eps6
*use,ceRVE,a1,a2,a3,0,0,0,0,0,0.5
SOLVE ! Solve analysis
To compute the average stress in the RVE, it is possible to use the macro srecover,
shown in Example 6.1. On account of the applied strain being equal to unity, the computed
average stress is equal to C66 . Therefore, G12 = C66 = 2, 583 MPa (Table 6.3).
*use,srecover
C66 = Sxz0
6.3 Global-Local Analysis

In global-local analysis (Figure 6.7), an RVE is used to to perform a refined com-
putation at each Gauss integration point of the global model. The global model is
used to compute the displacements and resulting strains, assuming that the ma-
terial is homogeneous. The local model takes the inhomogeneities into account by
modeling them with an RVE and thus providing a better computation of stress,
state variables, as well as secant and tangent constitutive tensors. In a way, the
local analysis is a surrogate for a constitutive equation that might be unknown due
to the complexity of the material behavior inside the RVE. Also, the computational
cost may be too high to model the entire structure with the refinement that can be
afforded inside the RVE.
Equations (6.13–6.15) are used in Section 6.2 to enforce one component of strain
at a time, with the objective of finding the equivalent elastic properties of the
material. Equations (6.13–6.15) are still valid for a general state of strain applied
to the RVE but care must be taken with the specification of periodic boundary
conditions at the edges and vertices, as discussed on page 167. Equations (6.13–
6.15) are nine constraint equations that can be imposed between all the pairs of
periodic points on the faces of the RVE except on the edges and vertices.
On the faces x1 = ±a1 , u1 is used to impose ε011 , u2 is used to impose ε021 = γ6 /2,
and u3 is used to impose ε031 = γ5 /2. To achieve this, (6.13) is expanded into its
three components, using tensor notation for strains, as follows
u1 (a1 , x2 , x3 ) − u1 (−a1 , x2 , x3 ) = 2a1 ε011
u2 (a1 , x2 , x3 ) − u2 (−a1 , x2 , x3 ) = 2a1 ε021
u3 (a1 , x2 , x3 ) − u3 (−a1 , x2 , x3 ) = 2a1 ε031 (6.41)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 6.7: Global-local analysis using RVE.
On the faces x2 = ±a2 , u1 is used to impose ε012 = γ6 /2, u2 is used to impose

ε022 ,
and u3 is used to impose ε032 = γ4 /2. Therefore, (6.14) is expanded into its
u1 (x1 , a2 , x3 ) − u1 (x1 , −a2 , x3 ) = 2a2 ε12

u2 (x1 , a2 , x3 ) − u2 (x1 , −a2 , x3 ) = 2a2 ε22
u3 (x1 , a2 , x3 ) − u3 (x1 , −a2 , x3 ) = 2a2 ε32 (6.42)
On the faces x3 = ±a3 , u1 is used to impose ε013 = γ5 /2, u2 is used to impose

ε023= γ4 /2, and u3 is used to impose ε033 . Therefore, (6.15) is expanded into its
u1 (x1 , x2 , a3 ) − u1 (x1 , x2 , −a3 ) = 2a3 ε13

u2 (x1 , x2 , a3 ) − u2 (x1 , x2 , −a3 ) = 2a3 ε23
u3 (x1 , x2 , a3 ) − u3 (x1 , x2 , −a3 ) = 2a3 ε33 (6.43)
Since each edge belongs to two faces, on every edge, it would seem that each
component of displacement would be used to impose two CE, one from each face, as
given by (6.41–6.43). However, as discussed on page 167, only one CE can be written
for each component of displacement. Therefore, edges must be dealt with separately.
Similarly, since three faces converge at a vertex, three periodic CE, one from each
face, need to be imposed using a single component of displacement. Following a
derivation similar to that presented on page 167, the following is obtained.
The planes x1 = ±a1 and x2 = ±a2 define two pairs of edges, for which (6.41–
6.43) reduce to the following six equations (with i = 1, 2, 3), as follows
ui (+a1 , +a2 , x3 ) − ui (−a1 , −a2 , x3 ) − 2a1 εi1 − 2a2 εi2 = 0

ui (+a1 , −a2 , x3 ) − ui (−a1 , +a2 , x3 ) − 2a1 εi1 + 2a2 εi2 = 0 (6.44)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
ui (+a1 , x2 , +a3 ) − ui (−a1 , x2 , −a3 ) − 2a1 εi1 − 2a3 εi3 = 0

ui (+a1 , x2 , −a3 ) − ui (−a1 , x2 , +a3 ) − 2a1 εi1 + 2a3 εi3 = 0 (6.45)
ui (x1 , +a2 , +a3 ) − ui (x1 , −a2 , −a3 ) − 2a2 εi2 − 2a3 εi3 = 0
ui (x1 , +a2 , −a3 ) − ui (x1 , −a2 , +a3 ) − 2a2 εi2 + 2a3 εi3 = 0 (6.46)
Four pairs of corners need to be analyzed one at a time. For each pair, the
corners are located symmetrically with respect to the center of the RVE located at
coordinates (0, 0, 0). The resulting CE are as follows
ui (+a1 , +a2 , +a3 ) − ui (−a1 , −a2 , −a3 ) − 2a1 εi1 − 2a2 εi2 − 2a3 εi3 = 0
ui (+a1 , +a2 , −a3 ) − ui (−a1 , −a2 , +a3 ) − 2a1 εi1 − 2a2 εi2 + 2a3 εi3 = 0
ui (−a1 , +a2 , +a3 ) − ui (+a1 , −a2 , −a3 ) + 2a1 εi1 − 2a2 εi2 − 2a3 εi3 = 0
ui (+a1 , −a2 , +a3 ) − ui (−a1 , +a2 , −a3 ) − 2a1 εi1 + 2a2 εi2 − 2a3 εi3 = 0
(6.47)
Example 6.4 Apply 02 = 0.2% and γ40 = 0.1% simultaneously to the composite in Example
6.2. Compute the average σ 2 and σ 12 in the RVE and the maximum stress σ2 and σ12
anywhere in the RVE.
Solution to Example 6.4 The same procedure used in Example 6.3 is used to define the
model. The APDL macro ceRVE.mac available in [5] is used to define the CE. The macro
needs the RVE dimensions and the applied strain as input arguments. In this example,
components of strain 2 = 0.2% and γ4 = 0.1% are applied, as follows

! units: TeraPascals, and microns, eps non-dimensional
! ceRVe arguments:
*use,ceRVE,a1,a2,a3,0,2e-3,0,1e-3,0,0
The macro srecover is used to compute the average stress of the RVE. The maximum
stress in the RVE can be computed using the commands PLESOL,S,1 or PRESOL,S,PRIN.

*use,srecover ! Compute average properties
S_1 = Szz0
S_2 = Sxx0
S_3 = Syy0
S_4 = Sxy0
S_5 = Syz0
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 6.8: Laminated RVE.
S_6 = Sxz0
VSEL,s,,,9
ESLV,S
plesol,s,x,1 ! Contour plot of S2 on matrix
plesol,s,xy,1 ! Contour plot of S23 on matrix
Taking into account the relation between the lamina coordinate system (c.s. 1,2,3) and
ANSYS global c.s. X,Y,Z on which the mesh is defined, the results are transformed and
shown in Table 6.4.
Table 6.4: Maximum stress on the matrix and average stress in the RVE
Average Results Maximum on the Matrix
σ2 = 10.0 MPa σ 2 = 29.5 MPa
σ12 = 2.42 MPa σ 12 = 6.07 MPa
6.4 Laminated RVE

A similar procedure to that used to obtain the RVE at the microscale can be used
to analyze laminates on the mesoscale. In this case the RVE represents a laminate.
Therefore, the through-thickness direction should remain free to expand along the
thickness. For example, with laminas parallel to the x-y plane, then σz = 0 and
(6.15) is not enforced, so that the thickness coordinate is free to contract (see
Figure 6.8). In general, the RVE must include the whole thickness. For symmetrical
laminates subjected to in-plane loads, the RVE can be defined with half the thickness
using symmetry boundary conditions (see Example 6.5).
The CE for a laminated RVE are simpler. Only (6.13) and (6.14) must be
enforced. In an hexahedral RVE, such as shown in Figure 6.8, only four faces
(x1 = ±a1 and x2 = ±a2 ) and the four edges defined by these faces need to be
considered.
Therefore, in a laminated RVE the constraint equations (6.13) and (6.14) become
the following. On the periodic pair of faces x1 = ±a1 , the CE are derived from (6.13)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
as
u1 (a1 , x2 , x3 ) − u1 (−a1 , x2 , x3 ) − 2a1 ε11 = 0
u2 (a1 , x2 , x3 ) − u2 (−a1 , x2 , x3 ) − 2a1 ε21 = 0 (6.48)
u3 (a1 , x2 , x3 ) − u3 (−a1 , x2 , x3 ) − 2a1 ε31 = 0
and on the pair of faces x2 = ±a2 , the CE are are derived from (6.14)
u1 (x1 , a2 , x3 ) − u1 (x1 , −a2 , x3 ) − 2a2 ε12 = 0

u2 (x1 , a2 , x3 ) − u2 (x1 , −a2 , x3 ) − 2a2 ε22 = 0 (6.49)
u3 (x1 , a2 , x3 ) − u3 (x1 , −a2 , x3 ) − 2a2 ε32 = 0
The planes x1 = ±a1 and x2 = ±a2 define two pairs of periodic edges, for which
(6.13–6.14) reduce to the following equations
u1 (+a1 , +a2 , x3 ) − u1 (−a1 , −a2 , x3 ) − 2a1 ε11 − 2a2 ε12 = 0

u2 (+a1 , +a2 , x3 ) − u2 (−a1 , −a2 , x3 ) − 2a1 ε21 − 2a2 ε22 = 0 (6.50)
u3 (+a1 , +a2 , x3 ) − u3 (−a1 , −a2 , x3 ) − 2a3 ε31 = 0
and
u1 (+a1 , −a2 , x3 ) − u1 (−a1 , +a2 , x3 ) − 2a1 ε11 + 2a2 ε12 = 0

u2 (+a1 , −a2 , x3 ) − u2 (−a1 , +a2 , x3 ) − 2a1 ε21 + 2a2 ε22 = 0 (6.51)
u3 (+a1 , −a2 , x3 ) − u3 (−a1 , +a2 , x3 ) − 2a3 ε32 = 0
For in-plane analysis, ε31 = ε32 = 0 and the third equation in (6.48)–(6.51) are
automatically satisfied.
Example 6.5 Compute Gxy for a [0/90/−45/45]S laminate with properties E1 = 139 GPa,
E2 = 14.5 GPa, G12 = G13 = 5.86 GPa, G23 = 2.93 GPa, ν12 = ν13 = 0.21, ν23 = 0.38
and lamina thickness tk = 1.25 mm.
0
Solution to Example 6.5 A shear strain γxy = 1 is applied to the RVE. The laminate
shear stiffness Gxy is obtained directly by computing the average stress in the RVE. As a
result of laminate symmetry and in-plane load, an RVE of half thickness with symmetry
boundary conditions in z = 0 can be used. The following APDL commands define the model
and the laminate.
/TITLE,RVE of [0/90/-45/45]s laminate

TH =1.25 ! Thickness of lamina in mm

A1 =1 ! Half length of RVE in x direction
A2 =1 ! Half length of RVE in y direction

UIMP,1,EX,EY,EZ,139E3,14.5E3,14.5E3
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i

LOCAL,12,,0,0,0,-45 ! defines -45 degree local cs
! Generate Geometry
BLOCK,-A1,A1,-A2,A2,0,TH ! 45 degrees layer
BLOCK,-A1,A1,-A2,A2,1*TH,2*TH ! -45 degrees layer
BLOCK,-A1,A1,-A2,A2,2*TH,3*TH ! 90 degrees layer
BLOCK,-A1,A1,-A2,A2,3*TH,4*TH ! 0 degrees layer

NUMCMP,ALL
LESIZE,ALL,,,2
VMESH,1 ! Meshes 45 degrees layer
ESYS,12 ! Selects -45 degrees material orientation
VMESH,2 ! Meshes -45 degree layer
VMESH,3 ! Meshes 0 degree layer
The APDL macro ceRVElaminate.mac available in [5] is used to define the CE, thus
implementing (6.48)–(6.51). The macro needs the RVE dimensions and the applied strain
0
as input arguments. In this example, only a strain γxy = 1.0 is applied.

NSEL,S,LOC,Z,0
D,ALL,UZ ! Symmetry z=0
NSEL,R,LOC,Y,0
NSEL,R,LOC,X,0
D,ALL,ALL
NSEL,ALL
! ceRVElaminate arguments:
! a1,a2,epsX,epsY,epsXY
*use,ceRVElaminate,a1,a2,0,0,1
To compute the average stress along the RVE, it is possible to use the macro srecover,
used in Example 6.1.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i

/DEVICE,VECTOR,0
/PNUM,MAT,0
*USE,SRECOVER
G_XY = SXY0
PLESOL,S,XY,1
The result G XY is shown in the ANSYS output window, which is minimized in

the Taskbar. On account of the applied strain being equal to unity, the computed
average stress is equal to C66 . Therefore, G12 = 21, 441 MPa.
Suggested Problems
Problem 6.1 Consider a unidirectional composite with isotropic fibers Ef = 241 GP a,
νf = 0.2, and isotropic matrix Em = 3.12 GP a, νm = 0.38 with fiber volume fraction
Vf = 0.4. The fiber diameter is df = 7 µm, placed in a square array as shown in Figure 6.1.
Choose an RVE including one full fiber in the center, with vertical faces spaced 2 a2 and
horizontal faces spaced 2 a3 .
i. compute the first 3 columns of the stiffness matrix in (6.10).
ii. compute C66 .
Problem 6.2 Consider the same material and fiber distribution of Problem 6.1, but choose
an RVE with faces rotated 45◦ with respect√to the horizontal
√ and vertical direction in Fig-
ure 6.1. Therefore, the RVE size will be 2 2 a2 by 2 2 a3 and it will include two fibers
(one full and four quarters). Be careful to select a correct RVE that is periodic.
i. compute the first 3 columns of the stiffness matrix in (6.10).
ii. compute C66 .
Problem 6.3 Compute E1 , E2 , ν12 , ν23 , G12 , G23 , using the stiffness matrices calculated in
Problems 6.1 and 6.2. Compare and explain the results.
Problem 6.4 Perform the averaging (6.8) of the stiffness matrices calculated in Problems
6.1 and 6.2. Then, compute E1 , E2 , ν12 , ν23 , G12 , G23 , using the averaged matrices. Compare
and explain the results.
Problem 6.5 Compute G12 as in Example 6.3 but using symmetry boundary conditions to
discretize only one quarter of the RVE.
Problem 6.6 Compute G23 by using Example 6.4 as a guide.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Chapter 7
Viscoelasticity
Our interest in viscoelasticity is motivated by observed creep behavior of polymer

matrix composites (PMC), which is a manifestation of viscoelasticity. The time-
dependent response of materials can be classified as elastic, viscous, and viscoelastic.
On application of a sudden load, which is then held constant, an elastic material
undergoes instantaneous deformation. In a one-dimensional state of stress, the
elastic strain is ε = Dσ, where D = 1/E is the compliance or inverse of the modulus
E. The deformation then remains constant. Upon unloading, the elastic strain
reverses to its original value, thus all elastic deformation is recovered.
The viscous material flows at a constant rate ε̇ = σ/η where η = τ E0 is the
Newton viscosity, E0 is the initial modulus,
R and τ is the time constant of the
material. The accumulated strain ε = ε̇dt cannot be recovered by unloading.
A viscoelastic material combines the behavior of the elastic and viscous material
in one, but the response is more complex than just adding the viscous strain to the
elastic strain. Let H be the Heaviside function defined as
H(t − t0 ) = 0 when t < t0

H(t − t0 ) = 1 when t ≥ t0 (7.1)
Upon step loading σ = H(t − t0 ) σ0 , with a constant load σ0 , the viscoelastic

material experiences a sudden elastic deformation, just like the elastic material.
After that, the deformation grows by a combination of recoverable and unrecoverable
viscous flow.
A simple series addition of viscous flow and elastic strain (Maxwell model, Fig-
ure 7.1(a), with η = τ E0 ) yields totally unrecoverable viscous flow plus recoverable
elastic deformation
σ(t) σ̇(t)
ε̇(t) = + (7.2)
τ E0 E0
A simple parallel combination of elastic and viscous flow (Kelvin model, Fig-
ure 7.1(b), with η = τ E) yields totally recoverable deformation with no unrecover-
able viscous flow
σ(t) = τ E ε̇(t) + Eε(t) (7.3)
179
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 7.1: Viscoelastic models: (a) Maxwell, (b) Kelvin, (c) standard solid, (d)
Maxwell-Kelvin.
but the deformation does not recover instantaneously.

Materials with unrecoverable viscous flow, such as (7.2), are called liquids even
though the flow may occur very slowly. Glass is a liquid material over the time
span of centuries; the thickness of window panes in medieval cathedrals is thicker
at the bottom and thinner at the top, thus revealing the flow that took place over
the centuries under the load imposed by gravity. Materials with fully recoverable
viscous deformations, such as (7.3), are called solids. We shall see that structural
design is much easier with solid materials than with liquid materials.
Please take heed of the common misconception introduced in early mechan-
ics of materials courses that most structural materials are elastic. Only perfectly
crystalline materials are elastic. Most materials are viscoelastic if observed for suf-
ficiently long periods of time, or at a sufficiently high temperature. In other words,
most real materials are viscoelastic.
For elastic materials, the compliance D is the inverse of the modulus E, both of
which are constants, and they are related by
DE = 1 (7.4)
For viscoelastic materials in the time domain, the compliance is called D(t) and
it is related to the time-varying relaxation E(t) in a similar but not so simple way,
as it will be shown in Section 7.3. Note that the relaxation E(t) takes the place
of the modulus E. A brief derivation of the relationship between compliance and
relaxation is presented next, in order to facilitate the presentation of viscoelastic
models in Section 7.1. When both the compliance D and the relaxation E are
functions of time, (7.4) simply becomes
D(t)E(t) = 1 (7.5)
Both D(t) and E(t) are functions of time and thus it is not possible to operate
algebraically on (7.5) to get either function explicitly in terms of the other. To find
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Viscoelasticity 181
one from the other, take the Laplace transform (see Section 7.3) to get
s2 D(s)E(s) = 1 (7.6)
Since both D(s) and E(s) are algebraic functions of s, and the time t is not
involved, it is possible to operate algebraically to get
1
E(s) = (7.7)
s2 D(s)
Finally, the relaxation in the time domain is the inverse Laplace of (7.7) or
E(t) = L−1 [E(s)] (7.8)
Similarly, the compliance D(t) can be obtained from the relaxation E(t) as

−1 1
D(t) = L (7.9)
s2 L[E(t)]
where L[ ] indicates the Laplace transform and L−1 [ ] indicates the inverse Laplace
transform.
7.1 Viscoelastic Models

The viscoelastic material models presented in this section are convenient curve fits
of experimental data. In the time domain, the usual experiments are the creep and
relaxation tests. In the creep test, a constant stress σ0 is applied and the ensuing
strain is measured. The ratio of measured strain to applied stress is the compliance
D(t) = ε(t)/σ0 . In the relaxation test, a constant strain ε0 is applied and the stress
needed to maintain that strain is measured. The ratio of measured stress to applied
strain is the relaxation E(t) = σ(t)/ε0 .
7.1.1 Maxwell Model

To derive the compliance of the Maxwell model [43], a creep test is performed under
constant stress σ0 applied at the ends of the model shown in Figure 7.1(a). The
rate of strain is given by (7.2). Integrating with respect to time we get
Z t
1 σ0
ε(t) = σ0 dt + (7.10)
τ E0 0 E0
where E0 is the elastic constant of the spring, τ is the time constant of the material,
and η0 = τ E0 in Figure 7.1(a). The spring and dashpot are subject to the same
load and to the same constant stress σ0 , so evaluating the integral yields
σ0 t σ0
ε(t) = + (7.11)
τ E0 E0
Then, the compliance is
1 t
D(t) = + (7.12)
E0 τ E0
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Table 7.1: Some common Laplace transforms

Function f(t)=L−1 {f(s)} f(s)=L{f(t)}
constant a a/s
linear at a/s2
derivative df /dt sf(s)-f(0)
exponential exp(at) 1/(s − a)
Rt
convolution integral 0 f (t − τ )g(τ )dτ L{f }L{g}
To derive the relaxation of the Maxwell model, take the Laplace transform of
(7.12), using Table 7.1 or MATLAB R
, to get
1 1 sτ + 1
D(s) = + 2 = 2 (7.13)
sE0 s τ E0 s τ E0
At t = 0, the dashpot does not move, so E0 is also the initial elastic modulus of
the material. Now, the relaxation in the Laplace domain is
1 τ E0
E(s) = = (7.14)
s2 D(s) sτ + 1
and the relaxation in the time domain is obtained by taking the inverse Laplace
transform (using Table 7.1 or MATLAB) to get
E(t) = E0 exp(−t/τ ) (7.15)
Note that at t = τ , the relaxation decays to 36.8% of its initial value, and thus
τ is called the time constant of the material.
7.1.2 Kelvin Model

For the Kelvin model, only the creep test is possible, since a relaxation test would
require an infinitely large stress to stretch the dashpot in Figure 7.1(b) to a constant
value in no time. For a creep test, a constant stress σ = σ0 is applied. Then, (7.3)
is an ordinary differential equation (ODE) in ε(t), which is satisfied by ε(t) =
(σ0 /E) [1 − exp(−t/τ )]. Therefore, the compliance D(t) = ε(t)/σ0 is
D(t) = 1/E0 [1 − exp(−t/τ )] (7.16)
Using (7.8), the relaxation function can be written with the aid of the Heaviside
step function H(t) and the Dirac delta function δ(t) as follows
E(t) = EH(t) + Eτ δ(t) (7.17)
where δ(t − t0 ) = ∞ if t = t0 and zero for any other time. The following MATLAB
code yields (7.17):
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Viscoelasticity 183
syms s complex; syms Dt Et t E tau real;

Dt=expand((1-exp(-t/tau))/E)
Ds=laplace(Dt)
Es=1/Ds/s^2
Et=ilaplace(Es)
7.1.3 Standard Linear Solid

To have an initial compliance 1/E0 , a spring is added to the Kelvin model (Fig-
ure 7.1.c). Then, the compliance is

−t
D(t) = 1/E0 + 1/E2 1 − exp (7.18)
τ2
and

−t (E0 + E2 )
E(t) = E∞ + (E0 − E∞ ) exp (7.19)
τ2 E2
where E∞ = (E0−1 + E2−1 )−1 is the equilibrium modulus as time goes to infinity.
To obtain a better correlation, more spring-dashpot elements are added in series,
as in
n
X
D(t) = D0 + Dj [1 − exp (−t/τj )] (7.20)
j=1
where τj are the retardation times [44]. When n → ∞,
Z ∞
D(t) = ∆(τ ) [1 − exp (−t/τ ) dτ ] (7.21)
0
where ∆(τ ) is the compliance spectrum [44].
7.1.4 Maxwell-Kelvin Model

A crude approximation of a liquid material is the Maxwell-Kelvin model, also called
the four-parameter model, described by Figure 7.1(d). Since the Maxwell and Kelvin
elements are placed in series, the compliance is found by adding the compliances of
the two individual modes, as
1 t 1
D (t) = + + [1 − exp (−t/τ 2)] (7.22)
E0 τ1 E0 E2
where E0 is the elastic modulus, τ1 takes the place of τ in (7.12), and E2 , τ2 , take
the place of E, τ, in (7.16). The relaxation modulus is given by [43, page 28]
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i

−1/2 q2 q2
E (t) = P12 − 4P2 q1 − exp (−t/T 1 ) − q1 − exp (−t/T 2 )
T1 T2
η1 = E0 τ1 ; η2 = E0 τ2
η1 η2
q1 = η1 ; q2 =
E2

1 1
q q
T1 = 2
P1 + P1 − 4P2 ; T2 = 2
P1 − P1 − 4P2
2P2 2P2
η1 η1 η2 η1 η2
P1 = + + ; P2 = (7.23)
E0 E2 E2 E0 E2
Another way to determine if a material is a liquid or a solid is to look at its
long-term deformation. If the deformation is unbounded, then it is a liquid. If the
deformation eventually stops, then it is a solid.
7.1.5 Power Law

Another model, which is popular to represent the short-term deformation of poly-
mers is the power law
E(t) = At−n (7.24)
The parameters A and n are adjusted with experimental data. The power law is
popular because it fits well the short-time behavior of polymers and because fitting
the data is very easy; just take a logarithm on both sides of (7.24) so that the
equation becomes that of a line, then fit the parameters using linear regression.
The compliance is obtained by using (7.9) as
D(t) = D0 + Dc (t)
Dc (t) = [AΓ(1 − n)Γ(1 + n)]−1 tn (7.25)
where Γ is the Gamma function [45], D0 = 1/E0 is the elastic compliance and
the subscript ()c indicates the creep component of the relaxation and compliance
functions.
7.1.6 Prony Series

Although the short-term creep and relaxation of polymers can be described well by
the power law, as the time range becomes longer, a more refined model becomes
necessary. One such model is the Prony series, which consists of a number n of
decaying exponentials
n
X
E(t) = E∞ + Ei exp(−t/τi ) (7.26)
i=1
where τi are the relaxation times, Ei are the relaxation moduli, and E∞ is the
equilibrium modulus, if any exists. For example, a Maxwell material is a “liquid”
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Viscoelasticity 185
and thus E∞P= 0. The larger the τi the slower the decay is. Note that at t = 0,
E0 = E∞ + Ei . Equation (7.26) can be rewritten as
n
X
E(t) = E∞ + mi E0 exp(−t/τi ) (7.27)
1
where mi = Ei /E0 are the dimensionless moduli.

The Prony series can be written in terms of the shear modulus and bulk modulus
(G, K, see Section 1.12.5) as follows
n
X
G(t) = G∞ + gi G0 exp(−t/τi )
1
n
X
K(t) = K∞ + ki K0 exp(−t/τi ) (7.28)
1
where G0 , K0 are the initial values of shear and bulk modulus, respectively; Gi , Ki
are shear and bulk moduli of the i-th term; gi = Gi /G0 and ki = Ki /K0 are
dimensionless shear/bulk moduli.
Noting that at t = 0, G∞ = G0 (1− n1 gi ), and K∞ = K0 (1− n1 ki ), the Prony
P P
series can be rewritten as
n n
!
X X
G(t) = G0 1− gi + gi G0 exp(−t/τi )
1 1
n n
!
X X
k(t) = k0 1− ki + ki K0 exp(−t/τi )
1 1
(7.29)
For most polymers and composites it is usual to assume that the Poisson’s ratio
does not change with time, which according to (1.74) and (1.76) is achieved by
setting ki = gi . Also, if ν is constant over time, mi = gi in (7.27).
7.1.7 Standard Nonlinear Solid

While the Prony series can fit any material behavior if a large number of terms are
used, other models are more efficient for fitting purposes, if harder to manipulate
mathematically. For example the Standard Nonlinear Solid model
D(t) = D0 + D10 [1 − exp(−t/τ )m ] (7.30)

can approximate well the long-term compliance in the α-region of polymer creep [46].
At room temperature, this is the region of interest to structural engineers since it
spans the range of time from seconds to years. In contrast the β-region [46], is
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
of interest to sound and vibration experts, among others, since it spans the sub-
second range of times. In other words, for long-term modeling, all compliance
occurring in the β-region can be lumped in the term D0 , with D10 representing all the
compliance that could ever be accumulated in the α-region. Equation (7.30) has four
parameters. When the data spans short times, it may be impossible to determine
all four parameters because the material behavior cannot be distinguished from a
3-parameter power law (7.31). This can be easily understood if (7.30) is expanded
in a power series, truncated after the first term as follows [46]
D(t) = D0 + D10 (t/τ )m [1 − (t/τ )m + ...] ≈ D0 + D1 tm ; D1 = D10 /τ (7.31)
For short times, all higher order powers of t can be neglected. What remains
is a modified power law with only three parameters. Note that for short times,
the parameter τ is combined with D10 to form D1 . If the data cover a short time,
the fitting algorithm will not be able to adjust both τ and D10 in (7.31); virtually
any combination of τ and D10 will work. That means that short-term data must be
modeled by a smaller number of parameters, in this case three.
7.1.8 Nonlinear Power Law

All models described so far represent linear viscoelastic materials. In the context of
viscoelasticity, linear means that the parameters in the model are not a function of
stress (see Section (7.2.1)). That means that the deformation at any fixed time can
be made proportionally larger by increasing the stress. If any of the parameters are
a function of stress, the material is nonlinear viscoelastic. For example a nonlinear
power law takes the form
ε̇ = AtB σ D (7.32)
Take a logarithm to both sides of (7.32) to get a linear equation in two variables
y = Ā + BX1 + DX2 ; Ā = log(A), X1 = log(t), X2 = log(σ) (7.33)
that can be fitted with a multiple linear regression algorithm in MATLAB.

Although most materials are not linearly viscoelastic, they can be approximated
as linear viscoelastic if the range of stress at which the structure operates is narrow.
Example 7.1 Fit the creep data in Table 7.2 with (a) Maxwell (7.12), (b) Power Law
(7.31), and (c) Standard Nonlinear Solid (7.30).
Table 7.2: Creep data for Example 7.1

time [sec] 1 21 42 62 82 102 123 143 163 184 204
D(t) [GPa−1 ] 1.49 1.99 2.21 2.35 2.56 2.66 2.75 2.85 2.92 2.96 3.01
Solution to Example 7.1 To fit the Maxwell model, fit a line to the secondary creep data;
that is, ignore the curvy portion for short times to get E0 = 0.460 GPa, τ = 495 s.
To fit the Power Law, write (7.19) as
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Viscoelasticity 187
2.5
D [GPa-1 ]
1.5 Experimental Data

Maxwell Model
Power Law
Std Nonlinear Solid
1
50 100 150 200
t [s]
Figure 7.2: Viscoelastic fit: Maxwell Model, Power Law, and Standard Nonlinear
Solid.
D(t) − D0 = D1 tm
where D0 = 1.49 GPa−1 is the first datum in Table 7.2 (see also (7.31)). Take a logarithm
to both sides of the above equation and adjust a line using linear regression to get D0 = 1.49
GPa−1 , D1 = 0.1117 (GPa sec)−1 , and m = 0.5.
To fit the Standard Nonlinear Solid you need to use a nonlinear solver to minimize the
error between the predicted (expected) values ei and the experimental (observed) values oi .
Such an error is defined as the sum over all the available data points: χ2 = (ei − oi )2 /o2i .
P
In this way, the following are obtained: D0 = 1.657 GPa−1 , D10 = 1.617 GPa−1 , τ = 0.273
sec, and m = 0.0026.
The experimental data and the fit functions are shown in Figure 7.2.
7.2 Boltzmann Superposition

7.2.1 Linear Viscoelastic Material
A viscoelastic material is linear if superposition applies. That is, given a stress
history
σ(t) = σ1 (t) + σ2 (t) (7.34)
the strain is given by
ε(t) = ε1 (t) + ε2 (t) (7.35)
where ε1 (t), ε2 (t) are the strain histories corresponding to σ1 (t) and σ2 (t), re-
spectively. For linear materials, the creep compliance and relaxation modulus are
independent of stress
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 7.3: Boltzmann superposition of strains.
ε(t)
D(t) =
σ0
σ(t)
E(t) = (7.36)
ε0
For nonlinear materials, D(t, σ) is a function of stress and E(t, ε) is a function

of strain.
For a linear material subjected to a stress σ0 applied at time t = θ0 (Figure 7.3)
we have
ε(t) = σ0 D(t, θ0 ) ; t > θ0 (7.37)
Adding an infinitesimal load step dσ at time θ0 + dθ results in
ε(t) = σ0 D(t, θ0 ) + dσ D(t, θ0 + dθ) ; t > θ0 + dθ (7.38)
If stress changes continuously by dσ over intervals dθ, the summation (7.38) can
be replaced by an integral to yield the accumulated strain as
Z t Z t
dσ
ε(t) = σ0 D(t, θ0 ) + D(t, θ)dσ = σ0 D(t, θ0 ) + D(t, θ) dθ (7.39)
θ0 θ0 dθ
where the discrete times θ0 , θ0 + dθ, etc., are represented by the continuous function
θ. Although aging effects are negligible over each infinitesimal dθ, they are signifi-
cant over time. Therefore, the compliance D(t, θ) is a function of the current time
t and all the time-history represented by θ in D(t, θ).
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Viscoelasticity 189
7.2.2 Unaging Viscoelastic Material

If the ε1 (t, θ) curve has the same shape as the ε1 (t, θ0 ) curve, only translated hori-
zontally, any curve can be shifted to the origin (Figure 7.3)
D(t, θ) = D(t − θ) (7.40)
Equation (7.40) is the definition of unaging material. For a discussion of aging

materials see [43, 47–49]. Equation (7.40) means that all the curves have the same
shape regardless of age θ, only shifted. Note θ in (7.40) is a continuous function
θ < t that denotes the time of application of each load (σ0 , dσ, etc.).
The response ε(t, θ) at a fixed time t is a function of the response at all times
θ < t. Therefore, it is said that the response is hereditary. If the material is aging, t
and θ are independent variables in D(t, θ). For unaging materials, only one variable,
t−θ, is independent, so it does not matter how old the material is (t), it only matters
for how long (t − θ) it has been loaded with dσ(θ).
The creep compliance is the response of the material to stress and always starts
when the stress is applied. If the change is gradual, from (7.39) we have
Z t
ε(t) = D(t − θ) σ̇(θ) dθ (7.41)
0
The relaxation is Z t
σ(t) = E(t − θ) ε̇(θ) dθ (7.42)
0
The time-dependent behavior of linear viscoelastic materials is hereditary, mean-

ing that the behavior at time t depends on what happened to the material since the
beginning of loading at t = 0.
Example 7.2 Consider an unaging material represented by D(t − θ) = 1/E + (t − θ)/η

and loaded with (a) σ0 H(θ) and (b) σ0 H(θ − 1). Find ε(t) in both cases and comment on
the results.
Solution to Example 7.2
⇒ dσ/dt = σ0 δ(0)
(a) σ = σ0 H(θ)
Rt 1 (t − θ)
ε(t) = 0 + σ0 δ(0)dθ
E η
1 t
ε(t) = + σ0 ; t > 0
E η
− 1) ⇒ dσ/dt
(b) σ = σ0 H(θ = σ0 δ(1)
Rt 1 (t − θ)
ε(t) = 0 + σ0 δ(1)dθ
E η
1 (t − 1)
ε(t) = + σ0 ; t > 1
E η
It can be seen that (b) is identical to (a), only shifted; meaning that there is no aging.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
7.3 Correspondence Principle

The Laplace transform of a function f (t) in the time domain (t-domain) maps to
the Laplace domain (s-domain) as f (s). The Laplace transform is defined as
Z ∞
L[f (t)] = f (s) = exp(−st)f (t)dt (7.43)
0
Most of the time, the Laplace transform can be obtained analytically, just using
a table of transforms, such as Table 7.1. Taking the Laplace transform of (7.41,
7.42), yields
ε(s) = L[D(t)] L[σ̇(t)] = sD(s)σ(s) (7.44)

σ(s) = L[E(t)] L[ε̇(t)] = sE(s)ε(s) (7.45)
Multiplying (7.44) times (7.45) yields
s2 D(s)E(s) = 1 (7.46)
or
s D(s) = [s E(s)]−1 (7.47)

where it can be seen that sD(s) is the inverse of sE(s). This is analogous to (7.4)
for elastic materials.
The correspondence principle states that all the equations of elasticity, available
for elastic materials, are valid for linearly viscoelastic materials in the Laplace do-
main. This principle is the basis, for example, of the determination of creep and
relaxation of polymer matrix composites in terms of fiber and matrix properties
using standard micromechanics methods, as shown in Section 7.6.
The inverse mapping from the Laplace domain to the time domain
f (t) = L−1 (f (s)) (7.48)
is more difficult to compute. Decomposition in partial fractions [50] is a useful

technique to break up f (s) into simpler component functions for which the inverse
Laplace can be found analytically. Another useful technique is the convolution
theorem defined in Table 7.1. Also, the limiting value theorems
f (0) = lim [sF (s)]

s→∞
f (∞) = lim [sF (s)] (7.49)
s→0
can be used to evaluate the initial and final response of a material in the time
domain directly in the Laplace domain. Otherwise, the inverse Laplace can be found
numerically using [51] or by the collocation method described in [41, Appendix D].
The Carson transform is defined as
fˆ(s) = sf (s) (7.50)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Viscoelasticity 191
In the Carson domain, the constitutive equations (7.41–7.42) become
ε(s) = D̂(s)σ(s)
σ(t) = Ê(s)ε(s) (7.51)
which are analogous, in the Carson domain, to the stress-strain equations of elastic
materials in the time domain. Furthermore, the relationship between compliance
and relaxation becomes
D̂(s) = 1/Ê(s) (7.52)
7.4 Frequency Domain

The Fourier transform maps the time domain into the frequency domain. It is
defined as Z ∞
F [f (t)] = f (ω) = exp(−iωt)f (t) dt (7.53)
−∞
and its inverse

Z ∞
1
f (t) = √ exp(iωt)f (w)dω (7.54)
2π −∞
Applying the Fourier transform to (7.41–7.42) yields
ε(ω) = D(ω)σ̇(ω)
σ(ω) = E(ω)ε̇(ω) (7.55)
and
1
D(ω) = − (7.56)
ω 2 E(ω)
where D(ω) = D0 + iD00 and E(ω) = E 0 + iE 00 are complex numbers. Here D0 , D00
are the storage and loss compliances, and E 0 , E 00 are the storage and loss moduli.
Using standard complex analysis we get
E0
D0 =
E 02 + E 002
E 00
D00 = 02 (7.57)
E + E 002
The frequency domain has a clear physical meaning. If a sinusoidal stress

σ(ω, t) = σ0 exp(−iωt) is applied to a viscoelastic material, it responds with an
out-of-phase sinusoidal strain ε(ω, t) = ε0 exp(−iωt + φ). Furthermore, the com-
plex compliance is D(ω) = ε(ω, t)/σ(ω, t) and the complex relaxation is simply the
inverse of the complex compliance, E(ω) = σ(ω, t)/ε(ω, t).
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
7.5 Spectrum Representation

The Prony series (7.26) provides a physical interpretation of polymer behavior as a
series of Maxwell models, each with its own decay time. In the limit, a real polymer
has an infinite number of such models [52], so that
Z ∞ Z ∞
H(θ)
E(t) − E∞ = H(θ) exp(−t/θ)d ln θ = exp(−t/θ)dθ (7.58)
−∞ 0 θ
where H(θ) is the relaxation spectrum [53]. In terms of compliance, we have

Z ∞
t L(θ)
D(t) − D0 = + [1 − exp(−t/θ)]dθ (7.59)
η −∞ θ
where L(θ) is the retardation spectrum [53], D0 is the elastic compliance, η is the
asymptotic viscosity of liquids, with η → ∞ for solids (see also [54]).
7.6 Micromechanics of Viscoelastic Composites

7.6.1 One-Dimensional Case
Recall the constitutive equations (7.51) in the Carson domain. By the correspon-
dence principle, all equations of micromechanics for elastic materials are valid in
the Carson domain for linear viscoelastic materials. For example, the Reuss mi-
cromechanical model assumes uniform identical strain in the matrix and fiber (see
discussion on page 152). Therefore, the stiffness of the composite C is a linear
combination of the stiffness of the constituents (fiber and matrix) weighed by their
respective volume fractions Vm , Vf
C = Vm Cm + Vf Cf (7.60)
with Am = Af = I in (6.1). Taking into account the correspondence principle for a
viscoelastic material (Section 7.3), it is possible to write the stiffness tensor in the
Carson domain by analogy with (7.60) simply as
Ĉ(s) = Vm Ĉm + Vf Ĉf (7.61)
From it, the stiffness tensor in the Laplace domain is (see (7.47))
1
C(s) = Ĉ(s) (7.62)
s
Finally, the stiffness tensor in the time domain is obtained by finding the inverse
Laplace transform (7.48) as
C(t) = L−1 [C(s)] (7.63)
Example 7.3 Derive the transverse compliance D2 (t) in the time domain for a unidirec-
tional composite with elastic fibers and a viscoelastic matrix represented by Dm = 1/Em +
t/ηm . Plot Df , Dm (t), and D2 (t) for 0 < t < 0.1, Ef = 10, Vf = 0.5, Em = 5, ηm = 0.05.
Use the Reuss model and discuss the results.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Viscoelasticity 193
Solution to Example 7.3 The elastic behavior of the fiber and viscoelastic behavior of
the matrix are defined as follows:
1
Fiber (elastic): Ef = constant → Df =
Ef
1 1 t
Matrix (Maxwell model (7.12) with Em = E0 , ηm = τ E0 ): = +
Em Em ηm
Take the Laplace transform,
1 1
Df (s) = because is constant.
sEf Ef
1 1
Dm (s) = + 2
sEm s ηm
Then, the Carson transform is
1
D̂f (s) = s Df (s) =
Ef
1 t
D̂m (s) = s Dm (s) = +
Em sηm
Using the Reuss model (page 152) to compute the composite behavior
D̂2 = Vf D̂f + Vm D̂m
1 1 1
D̂2 = Vf + Vm +
Ef Em sηm
Back to the Laplace domain
Vf Vm Vm
D2 (s) = + + 2
sEf sEm s ηm
Back transform to the time domain (inverse Laplace)
Vf Vm (Em t + ηm )
D2 (t) = L−1 (D2 (s)) = +
Ef Em ηm
To make a plot, take Ef = 10, Vf = 0.5, Em = 5, ηm = 0.05, which results in
Df = 0.1 = 1/10
Dm (t) = 0.2 + 20t
D2 (t) = 0.15 + 10t
Since Vf = 0.5, the initial compliance is halfway between those of the fiber and the
matrix. The elastic fiber has a constant compliance. The creep rate of the composite 1/ηc
is 1/2 of the creep rate of the matrix 1/ηm .
7.6.2 Three-Dimensional Case

The constitutive equation for an elastic, isotropic material (1.78) can be written in
terms of just two material parameters λ and µ = G as
σ = (λI(2) ⊗ I(2) + 2µI(4) ) : ε (7.64)

where I(2) and I(4) are the second- and fourth-order identity tensors1 (see Appendix
A). The constitutive equation of isotropic viscoelastic materials can be written in
terms of the viscoelastic Lamé constants λ(s) and µ(s) as follows [54]
Z t Z t
(2) (2)
σ(t) = λ(t − θ)I ⊗I : ε̇(θ)dθ + 2µ(t − θ)I(4) : ε̇(θ)dθ (7.65)
0 0
1
Tensors are indicated by boldface type, or by their components using index notation.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Using the convolution theorem (Table 7.1), the Laplace transform of (7.65) is
σ(s) = sλ(s)I(2) ⊗ I(2) : ε(s) + s 2µ(s)I(4) : ε(s) (7.66)
or in terms of the Carson transform
σ̂(s) = Ĉ(s) : ε̂(s) (7.67)

Assuming that the fiber is elastic, and the matrix is viscoelastic; the latter
represented with a Maxwell model
Dm (t) = 1/Em + t/ηm (7.68)
the Carson transform is
Em + sηm
D̂m = 1/Em + 1/sηm = (7.69)
sηm Em
Using the correspondence principle yields
sηm Em sEm
Êm = 1/D̂m = = (7.70)
Em + sηm Em /ηm + s
Using (1.75) and assuming the Poisson’s ratio νm of the matrix to be constant,
the Lamé constant of the matrix in the Carson domain is
Êm νm
λ̂m = (7.71)
(1 + νm )(1 − 2νm )
and the shear modulus of the matrix is
Êm
µ̂m = (7.72)
2(1 + νm )
Barbero and Luciano [36] used the the Fourier expansion method to get the
components of the relaxation tensor in the Carson domain for a composite with
cylindrical fibers arranged in a square array with fiber volume fraction Vf . The
elastic, transversely isotropic fibers are represented by the transversely isotropic
stiffness tensor C0 defined by (1.70, 1.92) in terms of fiber properties in the axial
and transverse (radial) directions EA , ET , GA , GT , and νT . Defining the matrix
properties in the Laplace e and Carson domain b as λ b m = sλ
em (s) and µ
bm = seµm (s),
the components of the relaxation tensor of the composite in the Carson domain L b∗
become [36]
b ∗ (s) = λ bm − Vf −a24 + a23

L 11
bm + 2 µ
 2 −1
2 µ + 2 λ − C 0 − C0 (a2 − a2 ) 2 (a − a ) λ − C 0
m m 4 3 m
b b
b 33 23 4 3 12
− +
 
a1 2
a1

i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Viscoelasticity 195
 
bm − C 0 (a4 − a3 )
λ 12
b ∗12 (s) = λ
L b m + Vf  
a1
 2 −1
0 0
bm + 2 λm − C33 − C23 (a23 − a24 )
2µ b bm − C 0
2 (a4 − a3 ) λ 12
+
 
a1 a21
 
 2 
2µ b m − C 0 − C 0 a3
bm + 2 λ bm − C 0
λ
33 23 12
b ∗22 (s) = λ
L bm − Vf 
bm + 2 µ 
−

a1 a21

 2 −1
2µ bm − C 0 − C 0 (a2 − a2 )
bm + 2 λ bm − C 0
2 (a4 − a3 ) λ
33 23 3 4 12
+
 
a1 a21
 
 2 
2µ b m − C 0 − C 0 a4
bm + 2 λ bm − C 0
λ
33 23 12
b ∗ (s) = λ
L b m + Vf  −

23
a1 a21
 
 2 −1
2µ bm − C 0 − C 0 (a2 − a2 )
bm + 2 λ bm − C 0
2 (a4 − a3 ) λ
33 23 3 4 12
+
 
a1 a21
 
−1
b ∗ (s) = µ 2 4 S7
L 44 bm − Vf 0 + C0 − 2 S3 − b−1
µ m
bm − C22
2µ 23 2 − 2 νm
−1
b ∗66 (s) = µ 0
−1 S3
L bm − Vf bm −
µ C66 − (7.73)
µ
bm
where
0 0 0 0 0 bm C 0
b2m − 2 µ
a1 = 4 µ bm C33 + 6λ bm − 2 C11
bm µ bm − 2 µ
µ bm C23 + C23 C11 + 4λ 12
02 0 0 0 0 0
− 2C − λ12
bm C − 2 C λ
33
bm + C C − λ
11
bm C
11 33 23
0 0
b3m − 8 µ
a2 = 8 µ b2m C33 b2m λ
+ 12 µ b2m C11
bm − 4 µ
− 2µbm C 0 2 + 4 µ
23
bm C 0 + 4 µ
bm λ 23bm C 0 C 0 11 33
− 8µ bm C 0 − 4 µ
bm λ 02
bm C12 + 2µ 02
bm C33 − 4µ 0 b
bm C11 λm + 8 µ
bm λbm C 0
33 12
0 0 0 0 0
bm − 4 C C λ0 bm − 2 λ 0 0
+ 2λ
bm C C + 4 C C λ
11 33 12 23
bm C C
12 33 11 23
0 02 02 0 0 02 0 02 bm C 0 2 − λ
bm C 0 2
− 2 C23 C12 + C23 C11 + 2 C33 C12 − C11 C33 +λ 33 23
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
b2m + 4 λ 0 µ
bm − 2 C11 bm − 2 µ 0 − C0 λ 0 02
4µ 11 m − λm C33 − C12
bm µ bm C33 b b
a3 =
a2
S6
S3 −
C0 C0 + 2 λbm C 0 2 − 2νm
+ 11 33 12
−
a2 µ
bm
−2 µ 0 + 2λ bm C 0 − C 0 λ
bm − λ 02 0 0 0
bm C23 11 m − C12 + 2 λm C12 + C11 C23
bm µ b b
23
a4 = −
a2
S7
+ (7.74)
bm (2 − 2 νm )
µ
The coefficients S3 , S6 , S7 account for the geometry of the microstructure, in-
cluding the geometry of the inclusions and their geometrical arrangement [33]. For
cylindrical fibers arranged in a square array [34] we have
S3 = 0.49247 − 0.47603Vf − 0.02748Vf2

S6 = 0.36844 − 0.14944Vf − 0.27152Vf2 (7.75)
S7 = 0.12346 − 0.32035Vf + 0.23517Vf2
Note that (7.73) yield six independent components of the relaxation tensor. This
is because (7.73) represent a composite with the microstructure arranged in a square
array. If the microstructure is random (Figure 1.12), the composite is transversely
isotropic (Section 1.12.4) and only five components of the relaxation tensor are
independent. When the axis x1 is the axis of transverse isotropy for the composite,
the averaging procedure (6.7) yields the relaxation tensor with transverse isotropy
as
Ĉ11 = L̂∗11
Ĉ12 = L̂∗12
3 1 ∗ 1 ∗
Ĉ22 = L̂∗22 + L̂ + L̂
4 4 23 2 44
1 3 ∗ 1 ∗
Ĉ23 = L̂∗22 + L̂ − L̂
4 4 23 2 44
Ĉ66 = L̂∗66 (7.76)
where the remaining coefficients are found using (1.70) due to transverse isotropy
of the material. For example, Ĉ44 = (Ĉ22 − Ĉ23 )/2. This completes the derivation
of the relaxation tensor Ĉij = sCij (s) in the Carson domain. Next, the inverse
Laplace transform of each coefficient yields the coefficients of the stiffness tensor in
the time domain as

−1 1
Cij (t) = L Ĉij (7.77)
s
A MATLAB code based on [51] is available in [5, invlapFEAcomp.m] to perform
the inverse Laplace numerically. Another algorithm is provided in [41, Appendix D].
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Viscoelasticity 197
Example 7.4 Consider a composite made with 60% by volume of transversally isotropic
fibers with axial properties EA = 168.4 GP a, GA = 44.1 GP a, νA = 0.443, and transverse
properties ET = 24.8 GP a and νT = 0.005. The epoxy matrix is represented by a Maxwell
model (7.12) with E0 = 4.08 GP a, τ = 39.17 min and νm = 0.311. Plot the relaxation
E2 (t) of the composite as a function of time for 0 < t < 100 minutes, compared to the
elastic value of the transverse modulus E2 .
Solution to Example 7.4 This example has been solved using MATLAB. The elastic and
viscoelastic values of the transverse modulus E2 are shown in Figure 7.4. The calculation
procedure is explained next:
– Program the equations of Section 7.6.2 and use them to calculate the elastic values of
the composite’s elastic properties such as E2 . These equations have been implemented
in PMMViscoMatrix.m.
– Replace the elastic modulus of the matrix E0 by the Maxwell model for the matrix Eq.
(7.15) in the Carson domain Ê0 (See PMMViscoMatrix.m), as follows:
i. The output from the portion of the code implementing (7.73–7.76) are equations
for the relaxation moduli in terms of s in the Carson domain. Note that it is
necessary to declare the variable s as symbolic.
ii. Divide them by s to go back to the Laplace domain.
iii. Back transform to the time domain using the function invlapFEAcomp, which
is derived from [51].
iv. Finally, fit the numerical values of E2 (t) with a viscoelastic model equation.
Usually it is convenient to use the same model equation for the composite re-
laxation as that used for the matrix relaxation; in this case the Maxwell model.
This step is implemented in fitfunFEAcomp.m
The MATLAB codes PMMViscoMatrix.m, invlapFEAcomp.m, and fitfunFEAcomp.m are

available in [5]. The results are shown in Figure 7.4. The complete set of Maxwell param-
eters for the composite are calculated in Example 7.5.
7.7 Macromechanics of Viscoelastic Composites

7.7.1 Balanced Symmetric Laminates
The in-plane viscoelastic behavior of a balanced symmetric laminate can be obtained
using the procedure in Section 1.15 (Apparent Laminate Properties), but in the
Carson domain. Start with the stiffness of the laminas (7.76) in lamina coordinates,
in the Carson domain. Rotate each matrix to laminate coordinates. Then, average
them using (1.102). Using (1.105), find the laminate engineering properties in the
Carson domain and divide by s to go back to the Laplace domain. Finally, take the
inverse Laplace transform to find the laminate stiffness in the time domain. Then,
fit them with a model equation as is done in Example 7.4.
7.7.2 General Laminates

Thanks to the correspondence principle, the stress-resultant vs. strain-curvature
equations from classical lamination theory (CLT, see Chapter 3) are valid for linearly
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
15
E2(t) [GPa] 10
Viscoelastic response
Elastic
0
0 20 40 60 80 100
t [minutes]
Figure 7.4: Elastic and viscoelastic values of the transverse modulus E2 .
Table 7.3: Lamina viscoelastic properties for Example 7.5

(E1 )0 = 102417 MPa (G12 )0 = (G13 )0 = 5553.8 MPa ν12 = ν13 = 0.4010
τ1 = 16551 min τ12 = τ13 = 44.379 min
(E2 )0 = (E3 )0 = 11975 MPa (G23 )0 = 5037.3 MPa ν23 = 0.1886
τ2 = τ3 = 58.424 min τ23 = 54.445 min
viscoelastic laminated composites in the Carson domain. The A, B, D, H matrices

of a laminate in the Carson domain can be computed by using the equations from
first-order shear deformation theory (FSDT, Section 3.1.1). This methodology was
used in [55].
7.8 FEA of Viscoelastic Composites

Most commercial codes have implemented viscoelasticity (creep) for isotropic ma-
terials. This is a severe limitation for users interested in the analysis of viscoelastic
behavior of polymer matrix composites.
However, it is possible to take advantage of the user programmable features
of commercial software in order to implement the formulations presented in this
chapter. This is relatively easy because the approach used in this chapter is not
stress dependent, but a linear viscoelastic approach, and its implementation is not
complicated. A USERMAT subroutine is used in Example 7.5 to implement the
viscoelastic formulation.
Example 7.5 Compute the relaxation response of a [0/908 ]s laminate. The thickness of
each lamina is tk = 1.25 mm. The laminate width is 2b = 20 mm and its length is
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Viscoelasticity 199
2L = 40 mm. Load the sample with a uniform strain x = 0.1 by applying a uniform
displacement at x = L. Use solid elements on each lamina and symmetry conditions. Plot
the laminate stiffness Ex (t) for 0 > t > 150 minutes. Use the lamina material properties
given in Table 7.3, which were computed with the procedure used in Example 7.4.
Solution to Example 7.5 First, compute the viscoelastic engineering properties using the
procedure described in Example 7.4. The resulting Maxwell parameters of the lamina are
shown in Table 7.3.
In ANSYS R
, using an USERMAT subroutine for solid elements, it is possible to imple-
ment the constitutive equation of an orthotropic material with the following time-dependent
properties:
E1 (t) = (E1 )0 exp(−t/τ1 ) ; E2 (t) = E3 (t) = (E2 )0 exp(−t/τ2 )
G12 (t) = G13 (t) = (G12 )0 exp(−t/τ12 ) ; G23 (t) = (G23 )0 exp(−t/τ23 )
Such subroutine is available in [5, usermat3d 705.f90]. Next, the geometry can be mod-
eled using a command file similar to that used in Example 5.2.
! FEAcomp. Solution of Example 7.5. USERMAT3D

/TITLE,Viscoelastic Analysis [0/90_8]s laminate
*set,ThZ,1.25 ! Thickness of lamina in mm
*set,n90,8 ! 1/2 number 90 layers half laminate
*set,LX,25 ! 1/2 Length of laminate in mm
*set,bY,20.0 ! 1/2 width of laminate in mm
*set,neX,n90*2 ! Number of elements in x/z direction
*set,neY,10 ! Number of elements in y direction
*set,epsX,0.01 ! Uniform strain in x direction
TB,USER,1,1,12, ! Material properties #1, Maxwell Model, 10 variables

TBTEMP,0
! Variable descriptions fot the USERMAT sobroutine
! ------------------------------------------------
! E1 E2 nu12 nu23 G12 G23
! Tau1 Tau2 Tau12 Tau13
TBDATA,,102417,11974.8,0.40096,0.18864,5553.8,5037.3
TBDATA,,16550.7243,58.4242,44.3791,54.4447
TB,STAT,1,,2, ! NUMBER OF STAT VARIABLES

local,11,,0,0,0,90 ! defines 90 degree local cs
local,12,,0,0,0,0 ! defines 0 degree local cs
! Generate Geometry
BLOCK,0,LX,0,bY,0,n90*ThZ ! 90 degrees layer
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
BLOCK,0,LX,0,bY,n90*ThZ,(n90+1)*ThZ ! 0 degress layer


lesize,all,,,neX ! line number divisions = neX
lsel,s,loc,y,0 ! selects lines y=0
lsel,a,loc,y,bY ! add lines y=bY
lsel,r,loc,z,n90*ThZ,(n90+1)*ThZ ! reselect lines volume 2
lsel,r,loc,x,LX ! reselects lines x=LX
lesize,all,,,neX/n90,,1,,,1 ! line number divisions = neX/n90
lsel,s,loc,y,0 ! selects lines y=0
lsel,a,loc,y,bY ! add lines y=bY
lsel,r,loc,z,n90*ThZ,(n90+1)*ThZ ! reselect lines volume 2
lsel,r,loc,x,0 ! reselects lines x=0
lesize,all,,,neX/n90,,1,,,1 ! line number divisions = neX/n90
lsel,s,loc,z,0 ! selects lines z=0
lsel,a,loc,z,n90*ThZ ! add lines z=Thz to selection
lsel,a,loc,z,(n90+1)*ThZ ! add lines z=2Thz to selection
lsel,r,loc,x,0 ! reselects lines x=0
LESIZE,ALL,,,neY,8,1,,,1 ! define element size in selected lines
lsel,s,loc,z,0 ! selects lines z=0
lsel,a,loc,z,n90*ThZ ! add lines z=Thz to selection
lsel,a,loc,z,(n90+1)*ThZ ! add lines z=2Thz to selection
lsel,r,loc,x,LX ! reselects lines x=LX
LESIZE,ALL,,,NEy,(1/8),1,,,1 ! define ele. size in selected lines
lsel,all ! select all lines
VMESH,1 ! Meshes 90 degrees layer

OUTRES,ALL,1 ! STORE RESULTS FOR EVERY SUBSTEP
KBC,1 ! Specifies stepped loading within a load step

NSUBST,1 ! 1 = Number of substeps in this load step
TIME,1e-6 ! Define time near to zero
ASEL,S,LOC,X,0
ASEL,A,LOC,Y,0
ASEL,A,LOC,Z,0
ASEL,S,LOC,X,LX
DA,ALL,UX,(epsX*LX) ! Impose displacement on the end = epsX*LX
!SFA,ALL,,PRESS,-100 ! or could impose a load
NSLA,s,1
CP,1,UX,all
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Viscoelasticity 201

NSUBST,10,25,10 ! 10 = Number of substeps in this load step

TIME,300 ! Time at the end of load step

NUMVAR,200
ESOL,2,2721,12588,s,x,s0layer
ESOL,3,1,2,s,y,s90layer
NSOL,4,54,u,x,displacement
PLVAR,2,3
lines,1000
PRVAR,2,3,4
The results are shown in Figures 7.5 and 7.6. The following APDL generates results to
plot stress components along a user defined path.

PPATH,1,0,0,0,ThZ,0 ! 1st point of the path location
PPATH,2,0,0,bY,ThZ,0 ! 2nd point of the path location
PDEF,zero,EPSW,,AVG ! Compute zero axis (optional)
PLPATH,Sz,Sxz,Syz,zero ! Plot Sz,Sxz,Syz
/page,1000,,1000 ! Define print list without skips between pages
PRPATH,Sz,Sxz,Syz ! Print Sz,Sxz,Syz

PPATH,1,0,LX/2,0,ThZ,0 ! 1st point of the path location
PPATH,2,0,LX/2,bY,ThZ,0 ! 2nd point of the path location
PDEF,Sz ,S,Z,AVG ! Compute Sz
PDEF,Syz,S,YZ,AVG ! Compute Syz
PLPATH,Sz,Syz,zero ! Plot Sz,Syz
/page,1000,,1000 ! Define print list without skips between pages
PRPATH,Sz,Syz ! Print Sz,Syz
Example 7.6 Consider a composite made with 40% by volume of isotropic graphite fibers
with properties Ef = 168.4 GP a, νf = 0.443 and epoxy matrix represented by a Maxwell
model with E0 = 4.082 GP a, τ = 39.15 min and νm = 0.311 (independent of time).
Construct a finite element micromechanical model using a hexagonal microstructure (see
Example 6.3), subject to shear strain γ4 = 0.02 applied suddenly at t = 0. Tabulate the
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
0 50 100 150 200 250 300

1200 60
0 degree lamina
90 degree lamina
1000 50
σx (lamina 0) [MPa]
800 40
600 30
400 20
σx
200 10
0 0
0 50 100 150 200 250 300
t [min]
Figure 7.5: Time stress evolution in 0◦ -lamina, and 90◦ -lamina.
average stress σ4 over the representative volume element (RVE) at times t = 0, 20, 40, 60, 80,
and 100 minutes.
Solution to Example 7.6 The fiber can be represented using standard elastic properties
in ANSYS. The matrix should be modeled using the USERMAT provided in Example 7.5
(usermat3d 705.f90). Therefore, the only part of the ANSYS model definition that changes
with respect to Example 6.3 is the definition of the material, as follows
MP,EX,1,168.4e-3 ! Fiber material properties in Tera Pascals [TPa]

MP,PRXY,1,0.443
TB,USER,2,1,12, ! Material properties #2, Maxwell Model, 10 variables

TBTEMP,0 ! Matrix material properties in Tera Pascals [TPa]
! Variable descriptions for the USERMAT subroutine
! ------------------------------------------------
! E1 E2 nu12 nu23 G12 G23
! Tau1 Tau2 Tau12 Tau13
TBDATA,,4.082e-3,4.082e-3,0.311,0.311,1.556e-3,1.556e-3
TBDATA,,39.15,39.15,39.15,39.15
The APDL macro ceRVE.mac available in [5] is used to define the constraint equations
(CE) for the periodic model. The macro needs the RVE dimensions and the applied strain
as input arguments. In this example, only a strain γ4 = 0.04 is applied. The model is solved,
using different substeps at times t = 0, 20, 40, 60, 80, and 100 minutes.

OUTRES,ALL,1 ! STORE RESULTS FOR EVERY SUBSTEP
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Viscoelasticity 203
0.9
0.8
0.7
0.6
Cx/(Cx)0
0.5
0.4
0.3
0.2
Laminate
0.1 0 degree lamina
90 degree lamina
0
0 50 100 150 200 250 300
t [min]
Figure 7.6: Normalized stiffness C22 (t)/(C22 )o for global laminate, 0◦ -lamina, and
90◦ -lamina.
KBC,1 ! Specifies stepped loading within a load step

NSUBST,1 ! 1 = Number of substeps in this load step
TIME,1e-6 ! Define time near to zero
! ceRVe arguments:
*use,ceRVE,a1,a2,a3,0,0,0,20e-3,0,0
TIME,100 ! Time at the end of load step
To compute the average stress in the RVE, it is possible to use the macro srecover,
described in Example 6.2. The average stress obtained is shown in Table 7.4. The SET,LIST
command lists all the load steps and substeps, each of which can be selected inside POST1
with the command SET,#loadstep,#substep.

/DEVICE,VECTOR,0
/PNUM,MAT,0
SET,1,1
*use,srecover
!S_1 = Szz0
!S_2 = Sxx0
!S_3 = Syy0
S_4 = Sxy0 !Average stress t = 0
!S_5 = Syz0
!S_6 = Sxz0
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
SET,2,1
*use,srecover
SET,2,2
*use,srecover
SET,2,3
*use,srecover
SET,2,4
*use,srecover
SET,2,5
*use,srecover
VSEL,s,,,9
ESLV,S
plesol,s,x,1
plesol,s,xy,1
FINISH !Exit post-processor module
Table 7.4: Average stress σ4 along time

Time [min] 0 20 40 60 80 100
Average σ4 [MPa] 62.6 38.1 23.1 13.9 8.4 5.04
Using an exponential regression it is possible to calculate the values of (G0 )23 = 3.13
GPa and τ = 39.97 min that represent the relaxation of the composite in the 23-shear
direction using a Maxwell model (see Figure 7.7).
Suggested Problems
Problem 7.1 Consider a composite made with 60% by volume of isotropic fibers with prop-
erties Ef = 168.4 GPa and νf = 0.443, and epoxy matrix represented by a power law model
(7.24) with D0 = 0.222 GPa−1 , D1 = 0.0135 (GPa min)−1 , m = 0.17 and νm = 0.311.
Plot the relaxation C22 (t) of the composite as a function of time for 0 < t < 100 minutes.
Compare it to the elastic value of the stiffness C22 of the composite and the elastic stiffness
C22 of the matrix.
Problem 7.2 Consider a composite made with 60% by volume of transversely isotropic
graphite fibers with properties EA = 168.4 GPa, ET = 24.82 GPa, νA = 0.443, νT = 0.005,
GA = 44.13 GPa and epoxy matrix represented by a Maxwell model (7.15) with E0 = 4.082
GPa, τ = 39.15 min and νm = 0.311. Plot the relaxation tensor stiffness components C(t)
of the composite as a function of time for 0 < t < 100 minutes, compared to the elastic
stiffness C of the composite and the elastic stiffness Cm of the matrix.
Problem 7.3 Compute the parameters in the Maxwell model for unidirectional lamina (see
Section 1.14) of carbon/epoxy material used in Problem 7.2. Plot and compare the elastic
and viscoelastic properties: E1 (t), E2 (t) and G12 (t). Show all work in a report.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Viscoelasticity 205
3.5
FEM Data
3 Maxwell Model
2.5
G23 [GPa]
1.5
0.5
0
0 20 40 60 80 100 120
t [min]
Figure 7.7: Average (G0 )23 vs. time.
Problem 7.4 Use the user programmable features to implement the Maxwell model con-
stitutive equations for a transversally orthotropic lamina material under plane stress con-
ditions. Using the viscoelastic materials properties obtained in Problem 7.3, compute the
response of a [±45/902 ]s laminate. The thickness of each lamina is tk = 1.25 mm. Load the
sample with uniform edge loads in the middle laminate surface Nx = Ny = 10 N/mm. Plot
the laminate and the laminas relaxations, as well as the laminas stress σx as a function of
time for 0 > t > 300 minutes.
Problem 7.5 Compute the parameters in the Maxwell model for all the nine engineering
properties of a [0/90]S laminate. Each lamina is 1.25 mm thick. The material is carbon
T300 and Epoxy 934(NR) with Vf = 0.62 and lamina thickness 1.25 mm. Epoxy is repre-
sented by a Maxwell model (7.15) with E0 = 4.082 GPa, τ = 39.15 min and νm = 0.311.
Carbon T300 is transversely isotropic with axial modulus EA = 202.8 GPa, transverse mod-
ulus ET = 25.3 GPa, GA = 44.1 GPa, νA = 0.443, and νT = 0.005, where the subscripts A
and T indicate the axial and radial (transverse) directions of the fiber, respectively.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Chapter 8
Continuum Damage Mechanics
Many modes of damage can be observed in composite materials, including matrix

cracks, fiber breakage, fiber-matrix debonding, and so on. Much work has been
done trying to quantify each of these damage modes, their evolution with respect
to load, strain, time, number of cycles, etc., and their effect on stiffness, remaining
life, etc. Continuum Damage Mechanics (CDM) represents all these failure modes
by the effect they have on the mesoscale behavior (lamina level) of the material.
That is, CDM calculates the degraded moduli of the laminas and laminate in terms
of continuum damage variables. Then, either strength or fracture mechanics failure
criteria are used to detect damage initiation. Finally, damage evolution is predicted
in terms of empirical hardening/softening equations set up in terms of additional
parameters, such as the hardening exponent used for metal plasticity. For example,
a form of CDM is used in Chapter 10 to represent the degradation of the interface
between laminas.
Hardening equations require non-standard experimentation to adjust the ad-

ditional, empirical parameters. Since the parameters are adjusted to the model,
some shortcomings of the model may be masked by the fitting of the additional
parameters. From a thermodynamics point of view, damage variables are the state
variables of the formulation, and they are not measurable. This is in contrast to the
micromechanics of damage models (Chapter 9) and metal plasticity where the state
variables, i.e., crack density and plastic strain, are measurable. From a practical
point of view, CDM major shortcoming is the need for additional experimentation
to determine parameters that are particular to each model. Furthermore, since the
state variables are not measurable, the additional parameters need to be adjusted
to the model through the loss of stiffness, which may not be sufficiently sensitive to
damage [56].
One notable effect of damage is a reduction of stiffness, which can be used to

define damage [57]. One-dimensional models are used in Section 8.1 to introduce
the concepts. The theoretical formulation for the general three-dimensional case is
developed in Sections 8.2–8.4.
207
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
8.1 One-Dimensional Damage Mechanics

The development of a one-dimensional damage mechanics solution involves the def-
inition of three major entities: (1) a suitable damage variable, (2) an appropriate
damage activation function, and (3) a convenient damage evolution, or kinetic equa-
tion.
8.1.1 Damage Variable

Consider a composite rod of nominal area A, e unloaded, and free of any damage
(Figure 8.1.a). Upon application of a sufficiently large load P , damage appears
(Figure 8.1.b). On a macroscopic level, damage can be detected by the loss of
stiffness of the material. In CDM, damage is represented by a state variable, also
called damage variable D, which represents the loss of stiffness [57]
D = 1 − E/E
e (8.1)
where E
e is the initial (virgin) Young’s modulus, and E is the modulus after dam-
1
age. Earlier work [58] conceptualized damage as the reduction of area due to
accumulation of microcracks having the same effect as the actual damage
D = 1 − A/A
e (8.2)
where A,
e A, are the initial and remaining cross-sectional areas, respectively. The
complement to damage is the integrity [59]
Ω = 1 − D = A/A
e (8.3)
which can be interpreted as the remaining cross-sectional area ratio, using the orig-
inal area as basis. It is noted that, in principle, damage is a measurable parameter,
which could be determined by measuring the damaged area, remaining area, or
more practically measuring the initial and remaining moduli. Therefore, in ther-
modynamic terms, damage is a measurable state variable, in the same sense as the
temperature is a measurable state variable that quantifies in macroscopic terms the
random agitation of atoms, molecules, and other elementary particles. While it is
possible, but extremely difficult, to track the agitation of atoms and molecules, it
is very easy to measure the temperature with a thermometer or other device. The
same holds true for damage in composite materials.
The analysis of a structural component is done in terms of the nominal area A, e
which is the only one known to the designer. The remaining area A = (1−D)A is not
e
known a priori. The nominal stress is σ = P/A. e Neglecting stress concentrations at
the tips of the fictitious cracks representing damage in the damaged configuration
(Figure 8.1.b), the value of effective stress2 acting on the remaining area A is σe=
P/A > P/A. e
1
See also (8.10).
2
Even taking into account the stress concentrations, the volume average of the distribution of
effective stress in the representative volume element (RVE, see Chapter 6) is still σ
e = P/A.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Continuum Damage Mechanics 209
~ ~
~ ~
~
Figure 8.1: (a) Unstressed material configuration, (b) stressed material configura-
tion with distributed damage, (c) effective configuration.
Therefore, we can envision a configuration (Figure 8.1.c) free of damage, with

nominal area A,e loaded by nominal stress σ, but internally subjected to effective
stress σ
e and degraded stiffness E. Thus, the effective configuration allows us to
perform a structural analysis using the nominal geometry but effectively taking
into account the increase of effective stress and the decrement of stiffness caused by
damage.
In the undamaged configuration (a), D = 0, σ = σ e, ε = εe, and Hooke’s law is
σ
e = Ee
eε (8.4)
In the effective configuration (c)
σ = E(D)ε (8.5)
The principle of strain equivalence assumes that the strain is the same in the
configurations (b) and (c), or ε = εe. Starting with the nominal stress σ = P/A, e
multiplying by A/A and using (8.3), the relationship between effective stress σ
e and
nominal stress σ (under strain equivalence) is
e (1 − D)
σ=σ (8.6)
Using (8.6), (8.4) and ε = εe in (8.5), the apparent modulus is a function of the
damage D given by
e − D)
E(D) = E(1 (8.7)
The principle of energy equivalence [60] states that the elastic strain energy is
identical in the configurations (b) and (c). That is, σ : ε = σ
e : εe, which is satisfied
by
σ=σ e(1 − D)
(8.8)
εe = ε(1 − D)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Substituting (8.8) in (8.5) yields
e − D)2
E(D) = E(1 (8.9)
which redefines the damage variable as
q
D = 1 − E/E
e (8.10)
Every state variable has a conjugate thermodynamic force driving its growth.
In plasticity, the measurable state variable is the plastic strain tensor εp , which
is driven to grow by its conjugate thermodynamic force, the stress tensor σ. The
thermodynamic damage force Y is defined as conjugate to the state variable D.
A kinetic equation Ḋ(Y ) governs the growth of the state variable D as a function
of its conjugate thermodynamic force Y . In principle, any relevant variable can be
chosen as independent variable Y to define the kinetic equation Ḋ(Y ), as long as
it is independent of its conjugate state variable. When the damage D is a scalar
and it is used to analyze one-dimensional problems, various authors have chosen
independent variables in the form of strain ε [61], effective stress σe [62, 63], excess
energy release rate G − 2γc [64], and so on. However, the choice is better based on
the appropriate form of the thermodynamic principle governing the problem, as it
is shown in Section 8.3.
8.1.2 Damage Threshold and Activation Function

The elastic domain is defined by a threshold value for the thermodynamic force; no
damage occurs below the threshold. When the load state is in the elastic domain,
damage does not grow. When the load state reaches the limit of the elastic domain,
additional damage occurs. Furthermore, the elastic domain modifies its size or
hardens. Typical one-dimensional responses of two materials are shown in Figure
8.2. Initially the elastic domain is defined by the initial threshold values, σ ≤ σ0 and
ε ≤ ε0 . While the load state is inside this domain, no damage occurs. When the load
state is higher than the threshold, damage increases and the threshold changes. The
elastic domain may evolve as hardening or softening. A stress threshold increases
for materials with hardening (see Figure 8.2a), and it decreases for materials with
softening (see Figure 8.2b). On the other hand, a strain or effective stress threshold
always increases for hardening and softening behavior, as shown in Figure 8.2.
The elastic domain can be defined by the damage activation function g as
g = ĝ − γ̂ ≤ 0 (8.11)
where ĝ is a positive function (norm) that depends on the independent variable
(in a one-dimensional case a scalar Y ) and γ̂ is the updated damage threshold for
isotropic hardening. According to the positive dissipation principle (see Section 8.3
and (8.82),(8.97)), the updated damage threshold γ̂ can be written as
γ̂ = γ(δ) + γ0 (8.12)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 8.2: (a) Hardening behavior and (b) softening behavior. No damage occurs
until the strain reaches a threshold value ε0 , and no damage occurs during unloading.
where γ0 denotes the virgin damage threshold, and γ is a positive monotonic func-
tion, called the hardening (or softening) function, that depends on the internal
variable δ, called the damage hardening variable.
8.1.3 Kinetic Equation

The rate of damage accumulation is represented by a kinetic equation. The evolution
of damage and hardening are defined by
∂g ∂g
Ḋ = λ̇ ; δ̇ = λ̇ (8.13)
∂Y ∂γ
where Y is the independent variable and λ̇ ≥ 0 is the damage multiplier that en-
forces consistency among the damage and hardening evolution as defined by (8.13).
Furthermore, the values of λ̇ and g allow us to distinguish among two possible sit-
uations, loading or unloading without damage growth, and loading with damage
growth, according to the Kuhn-Tucker conditions [65]
λ̇ ≥ 0 ; g≤0 ; λ̇g = 0 (8.14)
In other words, the Kuhn-Tucker conditions allow us to differentiate among two

different cases:
i. Undamaging loading or unloading, in the elastic domain. The damage ac-

tivation function is g < 0 and by condition (8.14.c) λ̇ = 0, and by (8.13.a)
Ḋ = 0.
ii. Damage loading. In this case λ̇ > 0 and condition (8.14.c) implies that g = 0.
Then, the value of λ̇ can be determined by the damage consistency condition
g=0 and ġ = 0 (8.15)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Example 8.1 Compute λ̇ for a one-dimensional model under tensile load where the inde-
pendent variable is the effective stress Y = σ
e, the activation function is defined by ĝ = σ
e,
and the hardening function is defined by
γ̂ = (F0 − FR )δ + F0
where F0 and FR are the initial threshold and the strength of the strongest microscopic
element in the material, respectively.
Solution to Example 8.1 The damage activation function g is defined as
g = ĝ − γ̂ = σ
e − [(F0 − FR )δ + F0 ] ≤ 0
Therefore,
∂g ∂g
= +1 ; = −1
∂e
σ ∂γ̂
Using (8.13), the kinetic equations can be written as
∂g ∂g
Ḋ = λ̇ = λ̇ ; δ̇ = λ̇ = −λ̇
∂e
σ ∂γ̂
When new damage appears, the consistency conditions (8.15) yield
g=0 ⇒ γ̂ = σ
e
and
∂g ˙ ∂g ˙
ġ = 0 ⇒ ġ = σ
e+ e˙ − γ̂˙ = 0
γ̂ = σ
∂e
σ ∂γ̂
where
∂γ̂
γ̂˙ = δ̇ = (F0 − FR )(−λ̇) = (FR − F0 )λ̇
∂δ
Substituting into the second consistency condition (8.15) we obtain λ̇ as
1
λ̇ = e˙
σ
FR − F0
8.1.4 Statistical Interpretation of the Kinetic Equation

Let’s assume that individual damaging events are caused by the failure of micro-
scopic elements inside the material (e.g., fiber breaks, matrix cracks, fiber-matrix
debond, etc.). Furthermore, assume each of these material points has a failure
strength σe and that the collection of failure strengths for all these points, i.e., el-
ements failing at a certain stress σ
e over the total number of elements available, is
represented by a probability density f (eσ ) (Figure 8.3.b). The fraction of elements
broken during an effective stress excursion from zero to σ e provides a measure of
damage
Z σ
e
D(e
σ) = f (σ 0 )dσ 0 = F (e
σ) (8.16)
0
where F (eσ ) is the cumulative probability (Figure 8.3.b) corresponding to the prob-
ability density f (eσ ), and σ 0 is a dummy integration variable. Then, the kinetic
equation in terms of effective stress σ e is
dD ˙
Ḋ = σ
e = f (e e˙
σ )σ (8.17)
de
σ
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
~
~
~
~ ~
Figure 8.3: One-dimensional random-strength model.
8.1.5 One-Dimensional Random-Strength Model

As explained in Section 8.1.3, the rate of damage accumulation is represented by
a kinetic equation. Equation (8.17) represents a generic kinetic equation, which
becomes specific once a particular probability density of failure is adopted.
Consider a loose bundle of short fibers embedded in a matrix and subjected to
a uniform stress. The fiber-matrix interfacial strength is assumed to be identical
for all fibers but the embedment length is random. The fiber pull-out strength is
therefore random. Random means that the probability of finding a fiber pulling
out at any value of stress F0 < σ e < FR is constant. In other words, there is no
stress level at which more fibers or less fibers pull out because the probability of
pull out is random. This is represented in Figure 8.3 and given by the equation
σ ) = 1/(FR − F0 ). Substituting εe for σ
f (e e as the independent variable in (8.17), and
assuming strain equivalence ε = εe, we have
E
e
f (e
ε) = ; F0 ≤ σ
e ≤ FR (8.18)
FR − F0
Equation (8.18) yields the model proposed in [61], which represents well the damag-
ing behavior of Haversian bone [66], concrete in tension [67], fiber composites when
damage is controlled by fiber pull out [68], and transverse damage of unidirectional
composites.
Damage Activation Function

For a one-dimensional problem, choosing strain as the independent variable, it is
possible to write ĝ = ε. Therefore, the damage activation function can be written
as
g = ε − γ̂ ≤ 0 (8.19)
where γ̂ is the updated damage threshold. Assuming that no damage occurs un-
til the strain reaches a threshold value ε0 = F0 /E,
e and applying the consistency
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
conditions (8.15) and using (8.19), the updated damage threshold γ̂ is given by the
highest value of strain seen by the material, or
γ̂ = max(ε0 , ε) (8.20)
Kinetic Equation
The kinetic equation (8.17) for the case of random strength (8.18) in terms of strains
εe = ε can be expressed as

dD E/(F
e R − F0 )ε̇ when ; ε > γ̂
Ḋ = ε̇ = (8.21)
dε 0 otherwise
In this case, the independent variable is ε, and using (8.19), the kinetic equation
(8.13) reduces to
Ḋ = λ̇ (8.22)
Using the Kuhn-Tucker conditions and (8.21), the consistency condition (8.15)
reduces to
λ̇ = E/(FR − F0 )ε̇ (8.23)
e
when damage occurs and λ̇ = 0 otherwise. In this particular case, the kinetic
equation is known explicitly (8.22)–(8.23). Therefore, it is not necessary to evaluate
the evolution of hardening (8.12) because hardening is computed explicitly by (8.20).
Note that (8.23) is identical to the solution of Example 8.1 because the hardening
function was chosen deliberately to yield this result.
Secant Constitutive Equation

In this particular case, the damage variable is active when tensile load appears, and
it can be obtained by integrating (8.21) as
e γ̂ − ε0
Dt = E when ε > 0 (8.24)
FR − F0
Note that the damage state does not depend on the actual load state ε, it
only depends on the history of the load state γ̂. In this example, crack closure is
assumed in compression, damage becomes passive, and Dc = 0. Mathematically,
damage under unilateral contact conditions can be defined by the following equation
hεi h−εi
D = Dt + Dc (8.25)
|ε| |ε|
where the McCauley operator hxi is defined as hxi := 21 (x + |x|).
Substituting (8.24) into (8.5), and using strain equivalence, yields the following
constitutive equation

 1−E γ̂ − ε 0
E
e ε when ε > 0
 e
σ = E(D) ε = FR − F0 (8.26)

Eε when ε < 0
 e
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Tangent Constitutive Equation

In a finite element formulation, it is necessary to provide the constitutive equation
in rate form, where the rates of stress σ̇ and strain ε̇ are expressed as functions
of pseudo-time. In this particular case, the tangent constitutive equation can be
obtained by differentiation of the secant constitutive equation as
σ̇ = E(D)ε̇ + Ė(D)ε (8.27)
The term Ė(D) is zero when new damage does not appear, i.e., when there is
elastic loading or unloading. When damaging behavior occurs (8.20) yields γ̂ = ε,
and differentiating E(D) in (8.26) we obtain
Ee2
Ė(D) = − ε̇ (8.28)
FR − F0
Substituting (8.28) into (8.27) if damage occurs, or Ė(D) = 0 if no damage

occurs, the tangent constitutive equation can be written as

2γ̂ − ε0
1−E E
e ε̇ when ε > γ̂
 e
σ̇ = FR − F0 (8.29)
E(D) ε̇ when ε < γ̂

Model Identification
The initial damage threshold ε0 represents the minimum strain to initiate damage
and it is proportional to F0 as follows
F0 = Eε
e 0 (8.30)
Under load control, a tensile specimen breaks at ε = γ̂ = εcr when dσ/dε = 0.

Then, using (8.29.a), the only unknown parameter in the model can be computed
as
FR = 2Eε
e cr (8.31)
The material parameters F0 and FR can be calculated from the experimental
data using (8.30) and (8.31) with E
e being the undamaged modulus of the material.
The measurable values ε0 and εcr can be obtained easily from material testing at
the macroscopic level.
For the particular case ε0 = 0, using (8.24) and (8.31) at ε = εcr , the critical
damage at failure under tensile load is
Dcr = 0.5 (8.32)
Therefore, the critical effective stress is
σ
eT cr = Eε
e cr = 0.5FR (8.33)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 8.4: One-dimensional random-strength model for Example 8.2.3
and using (8.7) the critical applied stress is
σT cr = 0.25FR (8.34)
Therefore, in a material with initial threshold ε0 = 0, a tensile specimen under

load control fails when D = 1/2 and applied stress FR /4.
A conservative estimate of transverse tensile strength of a fiber-reinforced lam-
ina can be obtained assuming that the fiber-matrix bond strength is negligible. In
the limit, only the matrix between fibers carries the transverse load, with the fibers
acting as holes. In this limit case, the matrix links can be assumed to have a ran-
dom distribution of strength (8.18). Therefore, the random-strength model (8.29)
applies, and the critical damage for transverse tensile loading of unidirectional fiber-
reinforced lamina can be estimated by (8.32) as D2t cr = 0.5. At the present time,
there is no model available to estimate the critical transverse-direction compression

damage D2c cr .
Example 8.2 A beam of rectangular cross-section, width b = 100 mm and height 2h =

200 mm is subjected to pure bending. The bending moment at failure is 25.1 M N mm. The
beam is made of carbon/epoxy composite with randomly oriented short fibers with undamaged
Young’s modulus E e = 46 GP a. Find the bending moment at failure in terms of FR in (8.21).
Assume that the material does not damage in compression and it has a random distribution
of strength in tension, with the strongest material element having unknown strength FR > 0
and F0 = 0. Determine FR using the data given.
Solution to Example 8.2 This problem was solved in [68]. With reference to Figure 8.4,
M is the applied bending moment, and yc , yt , are the distances from the neutral axis to the
stress resultants Nc , Nt , on the tensile and compression portions of the beam.
Denoting by εt and εc the tension and compression strain on the outer surfaces of the
beam, y0 the distance from the mid-plane to the neutral surface, and assuming linear strain
3
Reprinted from Mechanics of Materials, vol. 8 (1998), D. Kracjcinovic, Damage Mechanics,
Figure 2.11, p. 134, copyright (1998), with permission from Elsevier.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
distribution through the thickness we have

y − y0 −y + y0
ε(y) = εt or ε(y) = εc
h − y0 h + y0
Since there is no damage in compression, the compression stress distribution is linear,
and the resulting compression stress resultant is
1
Nc = b(h + y0 )Eε
e c
2
and the distance yc is
1
(y0 − 2h)
yc =
3
As the tensile side of the beam damages, the neutral axis moves away from the mid-
surface. The tensile stress resultant is obtained using (8.26) and integrating the stress
between y0 and h as
Z h Z h
1
Nt = dNt = b E(D)ε(y)dy = (h − y0 )b 3 − 2(E/F
e R )εt Eε e t
y0 y0 6
where E
e is the undamaged elastic moduli. The distance yt is
h
4h − 2y0 − (E/F
Z
1 e R )εt (3h + y0 )
yt = ydNt =
Nt y0 6 − 4(E/F
e R )εt
The force and moment equations of equilibrium are
Nc + Nt = 0
Nc yc + Nt yt = M
Using the force equilibrium equation and assuming linear strain distribution through the
thickness, it is possible to obtain the strains εt and εc in terms of y0 as
6hy0 FR 6hy0 (h + y0 ) FR
εt = − ; εc =
(h − y0 )2 E
e (h − y0 )3 E e
Using the above relation, it is possible to reduce the moment equilibrium equation to a
single cubic equation in y0
−y0 (4h2 + 9hy0 + 3y02 ) 2

M= bh FR
(h − y0 )3
The ultimate bending moment can be determined by differentiation with respect to y0
dM
=0
dy0
that yields y0 cr = −0.175 h at beam failure. Therefore, the rupture bending moment is
Mcr = 0.2715 bh2 FR

A simple test (ASTM D790 or D6272) can be used to obtain the bending moment at
failure; in this example Mcr = 25.1 106 N mm. Therefore, FR can be estimated as FR =
92 M P a. As it is customary in structural engineering, the equivalent bending strength is
defined as
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 8.5: One-dimensional longitudinal tensile strength model.
Mcr
σBcr = = 0.407 FR
S
where S is the section modulus (for a rectangular beam S = 23 bh2 ). Note that according to
(8.34), the tensile strength of the same material assuming the same kinetic equation (8.26)
would be σT cr = 0.25 FR . This gives a ratio of equivalent bending to tensile strength equal to
σB cr /σT cr = 1.63, which is in good agreement with experimental data σB cr /σT cr = 1.6 [69]
obtained for unreinforced concrete and also with the value σB cr /σT cr = 1.5 recommended
by the ACI Code [70].
8.1.6 Fiber Direction, Tension Damage

If a lamina is subjected to tensile stress in the fiber direction, it is reasonable
to assume that the matrix carries only a small portion of the applied load and
no damage is expected in the matrix during loading. Then, the ultimate tensile
strength of the composite lamina can be accurately predicted by computing the
strength of a bundle of fibers.
Fiber strength is a function of the gauge length used during fiber strength tests.
The length scale that determines how much of the fiber strength is actually used in
a composite is the ineffective length δ. Starting at a fiber break point, the ineffective
length is that length over which a fiber recovers a large percentage of its load (say,
90%). Rosen [71] recognized this fact and proposed that the longitudinal ultimate
strength of fibers embedded in a ductile-matrix can be accurately predicted by the
strength of a dry bundle of fibers with length δ.
A dry bundle is defined as a number of parallel fibers of some given length and
diameter which, if unbroken, carry the same load (Figure 8.5). After a fiber within
a dry bundle fails, the load is shared equally by the remaining unbroken fibers. A
dry bundle typically refers to fibers which have not yet been combined with the
matrix. As tensile load is slowly applied to a dry bundle of fibers, the weaker fibers
(with large flaw sizes) begin to fail and the stress on the remaining unbroken fibers
increases accordingly. The Weibull expression [72]
m
δ σ
σ ) = 1 − exp −
e
F (e (8.35)
L0 σ e0
is often used to describe the cumulative probability F (e
σ ) that a fiber of length δ will
fail at or below an effective stress σ
e. The values of σ
e0 and m, which represent the
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
characteristic strength of the fiber, and the dispersion of fiber strength, respectively,
can be determined from fiber strength experiments performed with a gauge length
L0 . Equation (8.35) can be simplified as
F (e σm)
σ ) = 1 − exp (−δαe (8.36)
where
Γ(1 + 1/m) m 1

1
α= = (8.37)
e0m
L0 σ σ
eav L0
where Γ(x) is the Gamma function [45], σ eav is the average strength for a gauge
length L0 . Equation (8.36) provides the percentage of fibers in a bundle which
are broken as a function of the actual (or apparent) stress in the unbroken fibers.
The percentage of fibers which are unbroken is 1 − F (e σ ). The apparent stress or
bundle stress σ = σb is equal to the applied load divided by the total fiber cross-
sectional area. It is also equal to the product of the stress in unbroken fibers and
the percentage of fibers which are unbroken
σ = σb = σ σm)
e exp(−δαe (8.38)
The value σ
emax which maximizes (8.38), can be easily determined and is given
by
emax = (δαm)−1/m
σ (8.39)
The maximum (or critical) bundle strength σcr is determined by substituting

(8.39) into (8.38)
σcr = (δαm)−1/m exp(−1/m) = (δαme)−1/m (8.40)
where e is the basis of the natural logarithms. The composite longitudinal tensile
strength is [1, (4.82)]

Em
F1t = Vf + (1 − Vf ) σcr (8.41)
Ef
where Vf is the fiber volume fraction, and Ef and Em are the fiber and matrix
elastic Young’s moduli, respectively.
Combining (8.36) and (8.39), we get
cr
D1t = 1 − exp(−1/m) (8.42)
Therefore, the critical or maximum damage D1t cr for longitudinal tensile load-
ing can be computed as the area fraction of broken fibers in the lamina prior to
catastrophic failure [62, 63], which turns out to be a function of the Weibull shape
modulus m only.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Example 8.3 The data sheet of carbon fiber T300 from TorayTM Carbon Fibers, Inc. gives
average tensile strength of the fiber σav = 3.53 GP a, and tensile modulus Ef = 230 GP a.
Also, the same data sheet provides results of tensile tests of a unidirectional (UD) composite
with epoxy Em = 4.5 GP a and fiber volume fraction Vf = 0.6. The tensile strength reported
is F1t = 1860 M P a. Using this experimental data, and assuming a Weibull shape parameter
m = 8.9, identify the damage model under tensile load. Then, formulate the damage model
and implement it in ANSYS R
for a one-dimensional bar element. Finally, obtain the strain
vs. stress response of the UD composite.
Solution to Example 8.3

MODEL IDENTIFICATION From (8.41) and using the experimental data available,
it is possible to obtain σcr as
F1t
σcr = Em
= 3060 M P a
Vf + Ef (1 − Vf )
Then, the product δα can be obtained using (8.40) as
(σcr )−m
δα = = 3.92 10−33
me
The properties Ef = 230 GP a, m = 8.9, and δα = 3.92 × 10−33 are sufficient for the
identification of the model.
MODEL FORMULATION Following a procedure similar to that shown in Section 8.1.5
to implement a damage model, the following items are needed
Damage Activation Function In this example, the effective stress is chosen as the
independent variable. Therefore, the damage activation function can be written
as
e − γ̂ ≤ 0
g=σ (8.43)
where γ̂ is the updated damage threshold. Assuming an initial threshold value
σ0 = 0, from the consistency conditions (8.15) and (8.19), γ̂ is given by the
highest value of effective stress seen by the material
γ̂ = max(0, σ
e) (8.44)
Secant Constitutive Equation In this example, the kinetic equation (8.1.3) is

available in integral form and given explicitly by (8.36) as
D = 1 − exp (−δαγ̂ m ) when σ

e > 0; ε > 0 (8.45)
where the damage state does not depend on the actual load state σ e; it only
depends on the load history state γ̂.
Substituting (8.45) into (8.5) and (8.7), and using strain equivalence, yields the
constitutive equation
σ = E(D) ε = exp (−δαγ̂ m ) E

eε when ε>0 (8.46)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Tangent Constitutive Equation The tangent constitutive equation can be obtained

by differentiating the secant constitutive equation as
σ̇ = E(D)ε̇ + Ė(D)ε (8.47)
The factor Ė(D) is zero when no new damage appears, i.e., during elastic load-
ing or unloading. When damage occurs, (8.44) yields γ̂ = Eε,
e and differentiat-
ing E(D) in (8.46) we obtain
Ė(D) = −δαmγ̂ m−1 exp (−δαγ̂ m ) E

e 2 ε̇ (8.48)
The tangent constitutive equation is obtained by substituting (8.48) into (8.47)

when damage occurs, or Ė(D) = 0 when no new damage appears. Therefore,
the tangent constitutive equation can be written as
(
(1 − δαmγ̂ m ) exp (−δαγ̂ m ) E
e ε̇ when ε > γ̂/E
e
σ̇ = (8.49)
E(D) ε̇ when ε < γ̂/E
e
NUMERICAL ALGORITHM The one-dimensional damage model is implemented in

ANSYS using the subroutine usermat1d 803.f90, available in [5]. The following
items describe the procedure used to explicitly evaluate the damage constitutive equa-
tion.4
i. Read the strain at time t

εt
ii. Compute the effective stress (assuming strain equivalence)
σ
et = Eε
e t
iii. Update the threshold value
γ̂t = max(γ̂t−1 , σ
et )
iv. Compute the damage variable
Dt = 1 − exp (−δα(γ̂t )m )
v. Compute the nominal stress
σt = (1 − Dt ) E
e εt
vi. Compute the tangent stiffness

(
dam (1 − δαm(γ̂t )m ) exp (−δα(γ̂t )m ) E
e when γ̂t > γ̂t−1
Et =
(1 − Dt ) E
e when γ̂t = γ̂t−1
4
See Section 8.4.1 for those cases for which it is not possible to integrate the constitutive equation
explicitly.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
MODEL RESPONSE The APDL code below and user subroutine usermat1d 803.f90,
available in [5], are used to model a one-dimensional bar representative of a carbon
fiber UD composite. The nominal stress-strain response is shown with a solid line in
Figure 8.6. The UD composite fails at εcr = 1.5%, in good agreement with the strain
to failure reported by Toray. Your results must be similar to those shown in Figure
8.6.
/TITLE, Tensile response bundle Carbon Fiber T300, FEAcomp Ex. 8.3
!=== USER MATERIAL DECLARATION ==================

TB,USER,1,1,3, ! DECLARES USERMAT 1, MAT 1, PROPERTIESS 3
TBTEMP,0
TBDATA,,230000,8.9,3.92e-033 ! PROPERTIES: E, m, delta_alpha
TB,STAT,1,,2, ! NUMBER OF STATE VARIABLES 2
!================================================
ET, 1, 180 ! LINK180, link element for analysis

R,1,1 ! Real constant #1, Area = 1
NGEN,6,1,1,,,2 ! Generate 5 additional nodes
! distance between adjacent nodes 2mm
E,1,2 ! Generate element 1 by node 1 to 2
EGEN,5,1,1 ! Generate element 2,3,4 and 5

ANTYPE,STATIC
OUTRES,ALL,1 ! Store results for each substep
D,1,all ! Define b.c. on node 1, totally fixed
D,6,UX,0.25 ! Define horizontal displacement on node 6.
SOLVE ! Solve load step
/POST26 ! Start time-historic post-process

NSOL,2,6,U,X, UXnode6 ! Load displacements node 6
RFORCE,3,6,F,X, FXnode6 ! Load reaction force node 6
XVAR,2 ! displacement x-graph variable
PLVAR,3 ! plot reaction as y-graph variable
lines,1000 !
PRVAR,2,3 ! list displacements and reactions
FINISH ! Exit post-process module
8.1.7 Fiber Direction, Compression Damage

Many models have been proposed trying to improve the prediction of compres-
sion strength of composites first introduced by Rosen [73]. The literature encom-
passes fiber buckling modes [17, 22, 74, 75], kink-band models [76], and kink-bands
induced by buckling [77]. In fiber buckling models, it is assumed that buckling of
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
3500
← σ =3060 MPa
3000 cr
2500
2000
σ [MPa]
1500 ∠(1-D cr)Ef

where D =10.63%
cr
1000
500
0
0 0.5 1 1.5 2 2.5
ε [%]
Figure 8.6: Fiber tensile damage model response.
Figure 8.7: One-dimensional longitudinal compression strength model.
the fibers initiates a process that leads to the collapse of the material [73]. Rosen’s
model has been refined with the addition of initial fiber misalignment and non-
linear shear stiffness [74]. Experimental evidence suggests that fiber buckling of
perfectly aligned fibers (Rosen’s model) is an imperfection sensitive problem (see
Section 4.1.1). Therefore, small amounts of imperfection (misalignment) cause large
reductions in the buckling load, thus the reduction of the compression strength with
respect to Rosen’s prediction. Each fiber has a different value of fiber misalignment.
The probability of finding a fiber with misalignment angle α is given by a Gaussian
distribution [22, 78].
An optical technique [13] can be used to measure the misalignment angle of
each fiber in the cross-section. The resulting distribution of fiber misalignment was
shown to be Gaussian (Figure 8.7) by using the cumulative distribution function
(CDF) plot and the probability plot [22]. Therefore, the probability density is
exp(−z2 ) α
f (z) = √ ; z= √ (8.50)
Λ 2π Λ 2
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
where Λ is the standard deviation and α is the continuous random variable, in

this case being equal to the misalignment angle. The CDF gives the probability of
obtaining a value smaller than or equal to some value of α, as follows
Z z
2 0
F (z) = erf(z) = √ exp(−z 2 )dz 0 (8.51)
π 0
where erf (z) is the error function.
The relationship between the buckling stress and the imperfection (misalign-
ment) is known in stability theory as the imperfection sensitivity curve. Several
models from the literature can be used to develop this type of curve. The deter-
ministic model, similar to the one presented by Wang [74] is developed in [17] but
using the representation of the shear response given by Equation (8.52).
The shear stress-strain response of polymer-matrix composites can be repre-
sented [22, 77] by

G12
σ6 = F6 tanh γ6 (8.52)
F6
where γ6 is the in-plane shear strain. Furthermore, G12 is the initial shear stiff-
ness and F6 is the shear strength, which are obtained by fitting the stress-strain
experimental data. Complete polynomial expansions [79] fit the experimental data
well but they are not antisymmetric with respect to the origin. This introduces
an artificial asymmetric bifurcation during the stability analysis [75]. Shear exper-
imental data can be obtained by a variety of techniques including the ±45 coupon,
10◦ off-axis, rail shear, Iosipescu, Arcan, and torsion tests [80]. The nonlinear shear
stress-strain curve should be measured for the actual composite being tested in
compression.
Barbero [17] derived the equilibrium stress σeq as a function of the shear strain
and the misalignment angle as
√ √
F6 ( 2 − 1)a + ( 2 + 1)(b − 1)
σ
eeq (α, γ6 ) =
2(γ6 + α) 1−a+b
√
a = exp( 2g)√ − exp(2g) (8.53)
b = exp(2g + 2g)
γ6 G12
g=
F6
with G12 and F6 as parameters. Note that if the shear behavior is assumed to be
linear σ
e6 = G12 γ6 [81], then (8.53) does not have a maximum with respect to γ6 and
thus misaligned fibers embedded in a linear elastic matrix do not buckle. On the
contrary, by using the hyperbolic tangent representation of shear (8.52), a maximum
with respect to γ6 is shown in (8.53). The maxima of the curves σ e(γ6 ) as a function
of the misalignment angle α is the imperfection sensitivity curve, which represents
the compression strength of a fiber (and surrounding matrix) as a function of its
misalignment. For negative values of misalignment, it suffices to assume that the
function is symmetric σ e(−α) = σ e(α).
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
The stress carried by a fiber reduces rapidly after reaching its maximum because
the load-carrying capacity of a buckled fiber is much lower than the applied load.
Several models can be constructed depending on the assumed load that a fiber
carries after buckling. A lower bound can be found assuming that buckled fibers
carry no more load because they have no post-buckling strength. According to
the imperfection sensitivity equation (8.53), fibers with large misalignment buckle
under low applied stress. If the post-buckling strength is assumed to be zero, the
applied stress is redistributed onto the remaining, unbuckled fibers, which then
carry a higher effective stress σe(α). At any time during loading of the specimen,
the applied load σ (applied stress times initial fiber area) is equal to the effective
stress times the area of fibers that remain unbuckled
σ = σ̄(α)[1 − D(α)] (8.54)
where 0 ≤ D(α) ≤ 1 is the area of the buckled fibers per unit of initial fiber area.
For any value of effective stress, all fibers having more than the corresponding value
of misalignment have buckled. The area of buckled fibers D(α) is proportional to
the area under the normal distribution located beyond the misalignment angle ±α.
Equation (8.54) has a maximum that corresponds to the maximum stress that
can be applied to the composite. Therefore, the compression strength of the com-
posite is found as
Z α
0 0
σc = max σ̄(α) f (α )dα (8.55)
−α
where σ̄(α) is given by Equation (8.53) and f (α0 ) is given by (8.50). The maximum
of (8.54), given by Equation (8.55) is a unique value for the compression strength
of the composite that incorporates both the imperfection sensitivity and the distri-
bution of fiber misalignment. Note that the standard deviation Λ is a parameter
that describes the actual, measured, distribution of fiber misalignment, and it is
not to be chosen arbitrarily as a representative value of fiber misalignment for all
the fibers.
Since the distribution given in (8.50) cannot be integrated in closed form, (8.55)
is evaluated numerically. However, it is advantageous to develop an explicit formula
so that the compression strength can be easily predicted. Following the explicit for-
mulation in [17], the compression strength of the unidirectional composite, explicitly
in terms of the standard deviation of fiber misalignment Λ, the in-plane shear stiff-
ness G12 , and the shear strength F6 is
F1c
= (1 + 4.76Ba )−0.69 (8.56)
G12
where 4.76 and −0.69 are two constants chosen to fit the numerical solution to the
exact problem [17], with the dimensionless group Ba given by
G12 Λ
Ba = (8.57)
F6
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
The misalignment angle of the fibers that buckle just prior to compression failure
is given by [17, (23)]
αcr = a/b
a = 1019.011G12 C22 Λ3 − 375.3162C23 Λ4 − 845.7457G212 C2 Λ2
+g 282.1113G12 C2 Λ2 − 148.1863G212 Λ − 132.6943C22 Λ3
b = 457.3229C23 Λ3 − 660.77G12 C22 Λ2 − 22.43143G212 C2 Λ (8.58)
2 2 2
p 161.6881C2 Λ − 138.3753G12 C2 Λ − 61.38939G12
+g
g = C2 Λ (8.0C2 Λ − 9.424778G12 )
C2 = −G212 / (4F6 )
Additionally, the shear strain at failure is

r
2 +
3 πF6 αcr
γcr = −αcr + αcr (8.59)
2 G12
In summary, when a fiber-reinforced lamina is compressed, the predominant
damage mode is fiber buckling. However, the buckling load of the fibers is lower
than that of the perfect system because of fiber misalignment, so much that a small
amount of fiber misalignment could cause a large reduction in the buckling load. For
each misalignment angle α, the composite area-fraction with buckled fibers D(α),
corresponding to fibers with misalignment angle greater than α, can be taken as a
measure of damage. If the fibers are assumed to have no post-buckling strength,
then the applied stress is redistributed onto the remaining unbuckled fibers, which
will be carrying a higher effective stress. The applied stress, which is lower than
the effective stress by the factor (1 − D), has a maximum, which corresponds to
the compressive strength of the composite. Therefore, it is possible to compute the
critical damage D1c for longitudinal compressive loading as

cr αcr
D1c = 1 − Ω1c = 1 − erf √ (8.60)
Λ 2
where erf is the error function, Λ is the standard deviation of the actual Gaussian
distribution of fiber misalignment (obtained experimentally [78]), and αcr is the
critical misalignment angle at failure. The three-dimensional theoretical formulation
is developed in the next three sections.
8.2 Multidimensional Damage and Effective Spaces

The first step in the formulation of a general multidimensional damage model is
to define the damage variable as well as the effective stress and strain spaces, as
shown in this section. The second step is to define the form of either the Helmholtz
free energy or the Gibbs energy and from them derive the thermodynamic forces
conjugate to the state variables representing damage and hardening, as shown in
Section 8.3. The third step is to derive the kinetic laws governing the rate of damage
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
hardening, which are functions of the damage and hardening potentials, as shown
in Section 8.4.
Experimental knowledge of the degradation and subsequent material response is
used to guide the selection on the variable used to represent damage. A second-order
damage tensor D can be used to represent damage of orthotropic fiber-reinforced
composite materials, following Kachanov-Rabotnov’s approach [59, 82]. For com-
posite materials reinforced with stiff and strong fibers, damage can be accurately
represented by a second-order tensor5 with principal directions aligned with the ma-
terial directions (1, 2, 3) [83, 84]. This is due to the fact that the dominant modes
of damage are microcracks, fiber breaks, and fiber-matrix debond, all of which can
be conceptualized as cracks either parallel or perpendicular to the fiber direction.6
Therefore, the damage tensor can be written as
D = Dij = Di δij no sum on i (8.61)

where Di are the eigenvalues of D, which represent the net stiffness reduction along
the principal material directions ni , and δij is the Kronecker delta (δij = 1 if i = j, or
zero otherwise). The integrity tensor is also diagonal, and using energy equivalence
(8.8),we have
p
Ω = Ωij = 1 − Di δij no sum on i (8.62)
The integrity tensor is always symmetric and positive, because the net area
reduction must be positive definite during damage evolution [85]. Both tensors
are diagonal when represented in the principal system. Introducing a symmetric
fourth-order tensor, M, called the damage effect tensor, as
1
M = Mijkl = (Ωik Ωjl + Ωil Ωjk ) (8.63)
2
The transformation of stress and strain between the effective and damaged con-
figurations is accomplished as follows
ε=M:ε
e
e = M−1 : σ
σ εe = M : εe
e (8.64)
p p
ε =M:ε
e
where an over-bar indicates that the quantity is evaluated in the effective configura-
tion and the superscripts e, p, denote quantities in the elastic and plastic domains,
respectively.
By the energy equivalence hypothesis [60], it is possible to define the constitutive
equation in the effective configuration (Figure 8.1.c) as
5
Tensors are denoted by boldface type, or by their components with index notation.
6
Strictly speaking, damage is transversely isotropic since cracks can also be aligned along any
direction in the 2-3 plane.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
σ
e =C εe
e :e ; εe = C
e e −1 : σ
e =S
e :σ
e (8.65)
where the fourth-order tensors C and S denote the secant stiffness tensor and com-
pliance tensor, respectively. The stress-strain equations in the damaged configura-
tion (Figure 8.1.b) are obtained by substituting (8.65) into (8.64),
σ=M:σ e =M:C e :eεe , εe = M−1 : S

εe = M−1 : e e :σ
e,
σ=M:C e : M : εe , e −1 −1
ε = M : S : M : σ,
e (8.66)
σ = C : εe εe = S : σ
with
C=M:C
e :M S = M−1 : S
e : M−1 (8.67)
The explicit form of these tensors are presented in Appendix B. Given that
the tensor M is symmetric, the secant stiffness and compliance tensors are also
symmetric.
8.3 Thermodynamics Formulation

The damage processes considered in this chapter can be described by a series of
equilibrium states reached while the system traverses a nonequilibrium path due to
the irreversibility of damage and plasticity. In general, the current state of a system
(e.g., stress, stiffness, compliance) depends on the current state (e.g., strain) as well
as on the history experienced by the system. This is the case for viscoelastic ma-
terials discussed in Chapter 7. However, for damaging and elastoplastic materials,
the current state can be described in terms of the current strain and the effects
of history on the material, which in this chapter are characterized by the damage
tensor D and the plastic strain tensor εp .
8.3.1 First Law

The first law of thermodynamics states that any increment of internal energy of the
system is equal to the heat added to the system minus the work done by the system
on its surroundings
δU = δQ − δW (8.68)
The system under consideration in this section is a representative volume ele-

ment (RVE), which is the smallest volume element that contains sufficient features
of the microstructure and irreversible processes, such as damage and plasticity, to
be representative of the material as a whole. Further discussions about the RVE
can be found in Chapter 6.
In rate form, (8.68) is
U̇ = Q̇ − Ẇ (8.69)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
where Z
d
U̇ = ρudV (8.70)
dt
Ω
Here ρ is the density, Ω is the volume of the RVE, and u is the internal energy
density, which is an internal variable and a potential function.7
For a deformable solid, the rate of work done by the system is minus the product
of the stress applied on the system times the rate of strain
Z
Ẇ = − σ : ε̇ dV (8.71)
Ω
where ε is the total strain (see (8.125)).

The heat flow into the RVE is given by
Z Z
Q̇ = ρrdV − q · ndA (8.72)
Ω ∂Ω
where r is the radiation heat per unit mass, q is the heat flow vector per unit
area, and n is the outward normal vector to the surface ∂Ω enclosing the volume Ω.
Since the volume Ω of the RVE does not change with time, and using the divergence
theorem,8 the first law at the local level becomes
ρu̇ = σ : ε̇ + ρr − ∇ · q (8.73)
The internal energy accounts for all the energy stored into the system. For
example, a system undergoing elastic deformation δεe , raising temperature δT , and
damage in the form of cracks of area growing by δAc , undergoes a change of internal
energy u given9 by
δu = σ : δεe + Cp δT − (G − Gc )δAc (8.74)
where G is the strain energy release rate, Gc is the surface energy needed to create
the increment of the two surfaces of an advancing crack, and Cp = Cv is the specific
heat capacity of the solid.
In general ε = ε(σ, u, sα ) where sα are internal variables. Let’s assume for the
time being that the system is adiabatic, i.e., ρr − ∇ · q = 0. Further, if there are no
dissipative effects or heat transfer, then u is a function of ε only, u = u(εe ), where
εe is the elastic strain. In such a case, the internal energy density reduces to the
strain energy density, which in rate form is
ϕ̇(ε) = σ : ε̇e (8.75)
and the complementary strain energy density is
ϕ̇∗ (ε) = σ : ε̇e − ϕ̇ = σ̇ : εe (8.76)
7
The values of the potential functions depend on the state and not on the path or process
followed
R by the system
R to reach such a state [86].
8
( ∂Ω q · ndA = Ω ∇ · qdV ) ; div(q) = ∇ · q = ∂qi /∂xi .
9
Thermodynamics custom and [87] are followed here in representing the internal energy with
the letter u, not to be confused with the displacement vector u used elsewhere.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
8.3.2 Second Law

The second law of thermodynamics formalizes the fact that heat flows from hot to
cold. Mathematically, the heat flow q has the opposite direction to the gradient10
of temperature T , which is formally written as
q · ∇T ≤ 0 (8.77)
where the equal sign holds true only for adiabatic processes, i.e., when there is no
heat exchange and thus no thermal irreversibility.
Let’s visualize a process of heat transfer from a hot reservoir to a cold reservoir,
happening in such a way that no heat is lost to, and no work is exchanged with the
environment. Once heat has flowed to the cold reservoir, it is impossible to transfer
it back to the hot reservoir without adding external work. That is, the process of
heat transfer is irreversible even though, on account of the first law energy balance
(8.73), no energy has been lost. For future use (8.77) can be written11 as
q · ∇T −1 ≥ 0 (8.78)
The second law justifies the introduction of a new internal variable, the entropy
density s = s(u, ε), which is also a potential function [88]. According to the second
law, the entropy density rate is ṡ ≥ 0, where the equal sign holds true only for
adiabatic processes.
Assume the specific entropy s = s(u, ε) is a potential function such that for a
reversible process [88]
δq
ds = (8.79)
T rev
with δQ R= Ω ρ δq dΩ, where δq = r − ρ−1 ∇ · q is the heat input per unit mass,
R
and S = Ω s ρ dΩ is the entropy. We use δ, not d, to emphasize that δq is not the

differential (perfect or total) of any (potential) function.
As a preamble to the definition of conjugate variables (see (8.86, 8.92, 8.99)),
note that using (8.79), the first law can be rewritten for a reversible process on an
ideal gas (pv = RT ), as the Gibbs equation for an ideal gas,
du = T ds − p dv (8.80)
where v is the specific volume (volume per unit mass). It can be seen in (8.80) that
v is conjugate to −p for calculating work input for an ideal gas and s is conjugate
to T for calculating thermal energy input.
For a cyclic reversible process returning to its initial state characterized by state
H H δq
variables (e.g., u, T, ε), by virtue of (8.79) we have ds = T rev = 0. Since
H
s is a potential function but q is not, for an irreversible process we have ds = 0
H δq
but T < 0, as corroborated by experiments. The heat δq entering at
irrev
10
The gradient of a scalar yields a vector, ∇T = ∂T /∂xi .
11
∇T −1 = −T −2 ∇T .
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
temperature Ti provides less entropy input δq/Ti than the entropy output δq/To
leaving the same cycle at temperature To < Ti (see also [89, Example 6.2]). Since
entropy
H is a potential function, and therefore a state variable, it always satisfies
ds = 0. Therefore, a negative net entropy supply must be compensated by internal
entropy production. The entropy of a system can be raised or lowered by adding or
extracting heat (in the form of δq/T ) but it is always raised by internal irreversible
processes such as crack formation and so on (positive dissipation principle).
Adiabatic systems do not exchange heat with the surroundings (δq = 0), so the
only change in entropy is due to internal irreversibility ṡ ≥ 0, where the equal sign
holds for reversible processes only. Note that any system and its surroundings can
be made adiabatic by choosing sufficiently large surroundings, e.g., the universe.
For an arbitrary system, the total entropy rate is greater than (or equal to) the net
entropy input due to heat
r 1 q
ṡ ≥ − ∇· (8.81)
T ρ T
The left-hand side of (8.81) represents the total entropy rate of the system. The
right-hand side of (8.81) represents the external entropy supply rate. The difference
is the internal entropy production rate
r 1 q
γ̇s = ṡ − + ∇· ≥0 (8.82)
T ρ T
Equation 8.82 is called the local Clausius-Duhem inequality. Noting that ∇ ·
T −1 q = T −1 ∇ · q + q∇ T −1 results in

1 1
γ̇s = ṡ − (ρr − ∇ · q) + q · ∇ T −1 ≥ 0 (8.83)
ρT ρ
where the first two terms represent the local entropy production due to local dissi-
pative phenomena, and the last term represents the entropy production due to heat
conduction12 [88]. Assuming it is possible to identify all local dissipative phenom-
ena, their contributions can be written as products of conjugate variables pα s˙α ≥ 0,
and (8.83) can be written as
ρT γ̇s = pα s˙α + T q · ∇T −1 ≥ 0 (8.84)

where α = 1 . . . n, spans the total number of dissipative phenomena considered.
Note that the dissipation is a scalar given by the contracted product of a thermo-
dynamic force pα times the increment of a measurable state variable sα . The state
variable, also called thermodynamic flux, describes univocally the effects of history
(e.g., yield, damage) on the material. Note that γs is defined as an entropy, not as
a dissipation heat, so that it is a potential function, while q is not.
For the particular case of damage due to penny-shaped cracks growing self sim-
ilarly [64], the state variable is the crack area Ac and the thermodynamic force is
12
Even absent local dissipative phenomena, q · ∇T −1 ≥ 0 represents the well-known fact that
heat flows opposite to the temperature gradient ∇T .
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
the energy available to grow the cracks pc = G − Gc , which is equal to the differ-
ence between the energy release rate (ERR) G and the critical ERR Gc = 2γc , the
latter being equal to twice the surface energy because two surfaces must be created
every time a crack appears (see Chapter 10). In this case, the dissipation (heat) is
ρT γ̇ = pc Ȧc .
From the first law (8.73), considering an adiabatic process (ρr − ∇ · q) and using
the chain rule u̇ = ∂u/∂ε : ε̇ we have

∂u
σ−ρ : ε̇ = 0 (8.85)
∂ε
Since ε̇ = 0 would be a trivial solution, the stress tensor, conjugate to strain, is
defined as
∂u
σ=ρ (8.86)
∂ε
The Clausius-Duhem inequality (8.83) for an isothermal (∇T = 0) system re-
duces to
1
γ̇s = ṡ − (ρr − ∇ · q) ≥ 0 (8.87)
ρT
and using the first law we get
ρT γ̇s = ρT ṡ − (ρu̇ − σ : ε) ≥ 0 (8.88)

The Helmholtz free energy (HFE) density is defined as
ψ(T, ε, sα ) = u − T s (8.89)
which is also a potential function. The corresponding extensive function is the
13
R
Helmholtz free energy A = Ω ρψdV . The rate of change of HFE density is
ψ̇ = u̇ − Ṫ s − T ṡ (8.90)
and introducing (8.88), with γ̇s = 0 at an equilibrium state, we get
ρψ̇ = −ρsṪ + σ : ε̇ (8.91)
from which an alternative definition of stress, conjugate to strain, is found as

∂ψ
σ=ρ =C:ε (8.92)
∂ε
where the secant elastic stiffness, which is affected by dissipative phenomena, in-
cluding damage, is defined as
∂2ψ
C(sα ) = ρ (8.93)
∂ε2
13
The nomenclature of [87] has been used.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Using the first law (8.73) in the internal entropy production per unit volume, or
local Clausius-Duhem inequality (8.83), and expanding ∇ · (qT −1 ) = T −1 ∇ · q + q ·
∇T −1 , we get
q
ρT γ̇s = · ∇T −1 − ρ ψ̇ + sṪ − ρ−1 σ : ε̇ ≥ 0 (8.94)
T
Realizing that ∇T −1 = −∇T /T 2 , the Clausius-Duhem inequality becomes

q
T ργ̇s = σ : ε̇ − ρ ψ̇ + sṪ − · ∇T ≥ 0 (8.95)
T
Since the HFE density is a function of the internal variables ε, T, sα , we have

∂ψ ∂ψ ∂ψ
ψ̇ = : ε̇ + Ṫ + ṡα ; α = 1 . . . n (8.96)
∂ε T,sα ∂T ε,sα ∂sα ε,T

∂
where represents the partial derivative with respect to y at constant x.
∂y x
Inserting (8.96) into (8.95), using (8.89), (8.92), and ∇T −1 = −∇T /T 2 , the
second law can be written as follows
∂ψ
γ̇ = ρT γ̇s = −ρ sα + T q · ∇T −1 ≥ 0 (8.97)
∂sα
where γ̇ is the heat dissipation rate per unit volume. Comparing (8.97) to (8.84)
it becomes clear that −ρ∂ψ/∂sα = pα are the thermodynamic forces conjugated to
sα , which provides a definition for the thermodynamic forces.
The complementary free-energy density, or Gibbs energy density, is defined as
χ = ρ−1 σ : ε − ψ (8.98)
which is also a potential function. The corresponding extensive function is the Gibbs
14
R
energy G = Ω ρχdV . From (8.98) it follows the definition of strain, conjugate
to stress, and the definition of the thermodynamic forces, conjugate to the state
variables sα , as
∂χ ∂χ ∂ψ
ε=ρ ; pα = ρ = −ρ (8.99)
∂σ ∂sα ∂sα
where sα includes the damage variables and consequently pα includes the thermo-
dynamic damage forces (see Example 8.4).
The secant elastic compliance, which is affected by dissipative phenomena, in-
cluding damage, is defined by
∂2χ
S(sα ) = ρ (8.100)
∂σ 2
14
The nomenclature of [87] has been used.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Example 8.4 The following Gibbs free energy is proposed to represent the onset and accu-
mulation of transverse matrix cracks resulting from transverse tension and in-plane shear
loads:
" #
1 σ12 σ22 σ62

νe21 νe12 σ1 σ2
χ= + 2 e
+ 2 − +
2ρ E e1 (1 − D2 ) E 2 (1 − D6 ) 2G12 Ee2 Ee1 1 − D2
where Ee1 , E
e2 , νe12 , νe21 and G
e 12 are the undamaged in-plane elastic orthotropic properties of
a unidirectional lamina where the subscript ()1 denotes the fiber direction and ()2 denotes the
transverse direction. The damage variables D2 and D6 represent the effect of matrix cracks.
The proposed distinguishes between active (D2+ ) and passive damage (D2− ), corresponding
to the opening or closure of transverse matrix cracks, respectively. The determination of
the active damage variable is based on the following equation:
hσ2 i h−σ2 i
D2 = D2+ + D2−
|σ2 | |σ2 |
where hxi is defined as hxi = 12 (x + |x|).
For a lamina in a state of plane stress, subjected to in-plane stress only, without fiber
damage (D1 = 0), and using the energy equivalence principle (8.8), derive expressions for
(a) the secant stiffness tensor, (b) the effective stress, and (c) the thermodynamic forces
associated to the model. Use tensor components of strain (ε1 , ε2 , ε6 ).
Solution to Example 8.4 The constitutive model is defined as the derivative of the Gibbs
free energy with respect to the stress tensor
∂χ
ε =ρ =S:σ
∂σ
where the compliance tensor S is defined as:
∂2χ
S=ρ
∂σ 2
The compliance tensor for plane stress S in Voigt contracted notation is
1 νe21
 
− 0
 E
e1 Ee2 (1 − D2 ) 
 
 ν 12 1 
−
e
S =
2
0 
 E1 (1 − D2 ) E2 (1 − D2 )
e e 

 1 
0 0 2
2G12 (1 − D6 )
e
The damage variables appear in S12 , S21 , S22 and S66 and 6 = γ6 /2. To perform tensor
products using matrix multiplications, see (A.14) and (A.20). Using the energy equivalence
principle and (8.67), the compliance matrix can be written as
S = M−1 : S
e : M−1
where the undamaged compliance is
 
1 νe21
− 0

 Ee1 Ee2 

 νe12 1 
S=
e − 0 

 Ee1 E
e2 

 1 
0 0
2G12
e
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
and where the effective damage tensor M, written in contracted notation, multiplied by the
3×3 version of the Reuter matrix (A.13) is
 
1 0 0
M = 0 (1 − D2 ) 0 
0 0 (1 − D6 )
The stiffness tensor C is obtained by
C=M:C
e :M
where, for this particular example, the secant stiffness tensor is

 
E
e1 e2 (1 − D2 )
νe12 E
0
 1 − νe21 νe12 1 − νe21 νe12 
e2 (1 − D2 )2
 
C= e1 (1 − D2 )
νe21 E E 
 0 
 1 − νe21 νe12 1 − νe21 νe12 
e 12 (1 − D6 ) 2
0 0 2G
The effective stress σ
e is related to the nominal stress σ by the effective damage tensor
e = M−1 : σ, which yields
M using σ

T σ2 σ6
σ
e = σ1 , ,
1 − D2 1 − D6
The thermodynamic forces are obtained by using Y = ρ∂χ/∂D, which for this particular
example yield
0
 
 
2
  σ σ1 σ2 νe12
  

2
 Y1   −
 

 3 e 2 e

Y= Y2 = (1 − D2 ) E2 (1 − D2 ) E 1
Y6
   2

σ6

 


 

3 e
 
(1 − D6 ) 2G 12
8.4 Kinetic Law in Three-Dimensional Space

The damage variable D introduced in Section 8.2 is a state variable that represents
the history of what happened to the material. Next, a kinetic equation is needed
to predict the evolution of damage in terms of the thermodynamic forces. Kinetic
equations can be written directly in terms of internal variables as in (8.21) or as
derivatives of potential functions. For three-dimensional problems, it is convenient
to derive the kinetic law from a potential function, similar to the flow potential used
in plasticity theory.
Two functions are needed. A damage surface g(Y(D), γ(δ)) = 0 and a convex
damage potential f (Y(D), γ(δ)) = 0 are postulated. The damage surface delimits
a region in the space of thermodynamic forces Y where damage does not occur
because the thermodynamic force Y is inside the g-surface. The function γ(δ)
accomplishes the expansion of g and f needed to model hardening. The damage
potential controls the direction of damage evolution (8.102).
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
If the damage surface and the damage potential are identical (g = f ), the
model is said to be associated and the computational implementation is simplified
significantly. For convenience, the damage surface is assumed to be separable in the
variables Y and γ, and written as the sum (see (8.11)–(8.12))
g(Y(D), γ(δ)) = ĝ(Y(D)) − (γ(δ) + γ0 ) (8.101)
where Y is the thermodynamic force tensor, γ(δ) is the hardening function, γ0 is

the damage threshold, and δ is the hardening variable.
As a result of damage, ĝ may grow but the condition g < 0 must be satisfied.
This is possible by increasing the value of the hardening function γ(δ), effectively
allowing ĝ(Y(D)) to grow. Formally, the hardening function γ(δ) can be derived
from the dissipation potential as per (8.99), (8.124), provided the form of the po-
tential can be inferred from knowledge about the hardening process. Alternatively,
the form of the function (e.g., polynomial, Prony series, etc.) can be chosen so
that the complete model fits adequately the experimental data available. The latter
approach is more often followed in the literature.
When g = 0, damage occurs, and a kinetic equation is needed to determine the
magnitude and components of the damage Ḋ. This is accomplished by
∂D ∂f
Ḋ = = λ̇ (8.102)
∂Y ∂Y
where λ̇ yields the magnitude of the damage increment and ∂f /∂Y is a direction
in Y-space. To find the damage multiplier λ̇, it is postulated that λ̇ is also involved
in the determination of the rate of change of the hardening variable as follows
∂g
δ̇ = λ̇ (8.103)
∂γ
There are two possible situations regarding g and λ̇:
i. If g < 0, damage is not growing and λ̇ = 0, so Ḋ = 0.
ii. If g = 0, damage occurs and λ̇ > 0, so Ḋ > 0.
These are summarized by the Kuhn-Tucker conditions
λ̇ ≥ 0 ; g≤0 ; λ̇g = 0 (8.104)
The value of λ̇ can be determined by the consistency condition, which leads to

∂g ∂g
ġ = : Ẏ + γ̇ = 0 ; g=0 (8.105)
∂Y ∂γ
On the other hand, the rates of thermodynamic forces and hardening function
can be written as
∂Y ∂Y
Ẏ = : ε̇ + : Ḋ
∂ε ∂D
(8.106)
∂γ
γ̇ = δ̇
∂δ
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
or in function of λ̇, introducing (8.103) and (8.104) into (8.106) as follows
∂Y ∂Y ∂f
Ẏ = : ε̇ + λ̇ :
∂ε ∂D ∂Y
(8.107)
∂γ ∂g
γ̇ = λ̇
∂δ ∂γ
Introducing (8.107) into (8.105) we obtain the following equation

∂g ∂Y ∂Y ∂f ∂g ∂γ ∂g
ġ = : : ε̇ + λ̇ : + λ̇ =0 (8.108)
∂Y ∂ε ∂D ∂Y ∂γ ∂δ ∂γ
Next, ∂f /∂γ = ∂g/∂γ = −1, (8.108) can be written as

∂g ∂Y ∂g ∂Y ∂f ∂γ
ġ = : : ε̇ + : : + λ̇ = 0 (8.109)
∂Y ∂ε ∂Y ∂D ∂Y ∂δ
Therefore, the damage multiplier λ̇ can be obtained as
d
L : ε̇ when g = 0
λ̇ = (8.110)
0 when g < 0
where
∂g ∂Y
:
d
L =− ∂Y ∂ε (8.111)
∂g ∂Y ∂f ∂γ
: : +
∂Y ∂D ∂Y ∂δ
Equations (8.103), (8.104), and (8.110) yield the pair D, δ, in rate form as
∂f
Ḋ = Ld : : ε̇ ; δ̇ = −λ̇ (8.112)
∂Y
The tangent constitutive equation can be obtained by differentiation of the con-
stitutive equation σ = C : ε, which yields
σ̇ = C : ε̇ + Ċ : ε (8.113)
where the last term represents the stiffness reduction. Next, the last term in (8.113)
can be written as
∂C
Ċ : ε = : Ḋ : ε (8.114)
∂D
Introduce (8.112) and rearrange
∂C ∂f
Ċ : ε = : ε : Ld : : ε̇ (8.115)
∂D ∂Y
Since (ε, D) are state variables, and thus independent variables,
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
∂ε
=0 (8.116)
∂D
Therefore,
∂σ ∂f
Ċ : ε = : Ld : : ε̇ (8.117)
∂D ∂Y
Finally, reintroduce the above into (8.113) to get
σ̇ = Ced : ε̇ (8.118)
where the damaged tangent constitutive tensor, Ced , can be written as follows

 C if Ḋ ≤ 0
Ced = ∂σ d ∂f (8.119)
 C+ :L : if Ḋ ≥ 0
∂D ∂Y
The internal variables D, δ, and related variables, are found using numerical
integration, usually using a return-mapping algorithm as explained in Section 8.4.1.
As explained in Sections 8.1.3 and 8.4, a number of internal material param-
eters are needed to define the damage surface, damage potential, and hardening
functions. These parameters cannot be obtained directly from simple tests, but
rather the model is identified by adjusting the internal parameters in such a way
that model predictions fit well some observed behavior that can be quantified exper-
imentally. Model identification is very specific to the particular model formulation,
material, availability of experiments, and feasibility of conducting relevant exper-
iments. Therefore, model identification can be explained only on a case-by-case
basis, as is done in Example 8.3.
8.4.1 Return-Mapping Algorithm

A return-mapping algorithm [90–92] is used to solve for the variables λ̇, δ̇, Ḋ, δ, and
D, in numerically approximated form.
The internal variables are updated by a linearized procedure between two con-
secutive iterations (k − 1 and k). The first-order linearization of (8.109) yields

∂g ∂Y ∂f ∂γ
(g)k − (g)k−1 = : : + ∆λk = 0 (8.120)
∂Y ∂D ∂Y ∂δ k−1
Successful iterations yield [g]k = 0 and
−(g)k−1
∆λk = (8.121)
∂g ∂Y ∂f ∂γ
: : +
∂Y ∂D ∂Y ∂δ k−1
The complete algorithm used for a typical integration of constitutive equations
is shown next:
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
i. Retrieve the strain (ε)n−1 from the previous increment and the strain incre-
ment (∆ε)n for the current increment from the finite element method (FEM)
code. The updated strain is calculated as
(ε)n = (ε)n−1 + (∆ε)n
ii. Retrieve the state variables from the previous step and start the return-
mapping algorithm by setting the predictor iteration k = 0
(D)n0 = (D)n−1 ; (δ)n0 = (δ)n−1
iii. Update the secant stiffness and Cauchy stress, which are used to calculate the
thermodynamic forces and damage hardening at this point
(C)nk = (M)nk : C
e : (M)n
k
(σ)k = (C)k : ()n
n n
(Y)nk ; (γ)nk
iv. The damage threshold is evaluated at this point
(g)k = g ((Y)nk , (γ(δ))nk , γ0 )
There are two possible cases:
(a) If (g)k ≤ 0, there is no damage, then ∆λk = 0. Go to (viii).
(b) If (g)k > 0, there is damage evolution, then ∆λk > 0. Go to (v).
v. Damage evolution. Starting at iteration k, the damage multiplier is found

from (g)k = 0 as
−(g)k−1
∆λk =
∂g ∂Y ∂f ∂γ
: : +
∂Y k−1 ∂D k−1 ∂Y k−1 ∂δ k−1
vi. Update the state variables using ∆λk

n n ∂f
Dij = Dij + ∆λk
k k−1 ∂Y k−1

∂f
(δ)nk = (δ)nk−1 + ∆λk = (δ)nk−1 − ∆λk
∂γ k−1
vii. End of linearized damage process. Go to (iii).
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
viii. Compute the tangent stiffness tensor
∂σ n ∂f n

ed
n n d
n
C = (C) + : L :
∂D ∂Y
ix. Store the stress and state variables to be used on the next load increment
(σ)n = (σ)nk ; (D)n = (D)nk ; (δ)n = (δ)nk
x. End of the integration algorithm.
Example 8.5 Implement the damage model developed in Example 8.4 into a user material
subroutine for a 2-D plane stress element with damage in the directions 2 (transverse) and
6 (shear). Use a return-mapping algorithm as shown in Section 8.4.1. Furthermore, use
the following damage activation function
v !2 !2
u
u GIc Y2 E
e2 GIc Y2 E
e2 Y6 G
e 12
g = ĝ − γ̂ = t 1− 2 + + − γ̂ ≤ 0
GIIc F2t GIIc F2t 2 F6 2
where GIc and GIIc are the critical energy release in mode I and in mode II, respectively,
F2t and F6 are the transverse tensile strength and the shear strength, respectively. Also, use
the following damage hardening function

δ
γ̂ = γ + γ0 = c1 exp − 1 + γ0 ; γ0 − c1 ≤ γ b ≤ γ0
c2
where γ0 defines the initial threshold value, c1 and c2 are material parameters. For this
particular damage model, the model parameters for AS4/8852 carbon/epoxy are given Tables
8.1 and 8.2.
Table 8.1: Elastic and strength properties for AS4/8852 unidirectional lamina
E
e1 E
e2 G
e 12 νe12 F2t F6
171.4 GPa 9.08 GPa 5.29 GPa 0.32 62.29 MPa 92.34 MPa
Table 8.2: Critical energy release, and hardening parameters for AS4/8852 unidi-
rectional lamina
GIc GIIc γ0 c1 c2
170 J/m2 230 J/m2 1.0 0.5 -1.8
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
100
1.6
1.4
50
1.2
σ [MPa]
1
0
6
Y
0.8
6
0.6
0.4 -50
0.2
0 -100
0 0.5 0 50
Y σ [MPa]
2 2
Figure 8.8: Initial damage surface in thermodynamic and stress spaces.
Solution to Example 8.5 This model represents damage caused by transverse-tensile and
in-plane-shear stress. Longitudinal tension/compression have no effect. Therefore, the
model is defined in the thermodynamic force space Y2 , Y6 . The shape of the damage surface
for AS4/8852 lamina is shown in Figure 8.8.
To implement the return mapping algorithm shown in Section 8.4.1, expressions for
∂f /∂Y, ∂g/∂Y, ∂f /∂γ, ∂g/∂γ,∂γ/∂δ, and ∂Y/∂D are needed.
Assuming f = g, the derivative of the potential function and the damage surface with
respect to the thermodynamic forces is given by
0
 

 r ! 

 1 GIc 1 2E2 GIc E2

 

∂g ∂f 1− +

= = ĝ G IIc 4F2t Y2 G IIc (F2t )
2
∂Y ∂Y  

 1 

G12

 

ĝ
and the derivative of the damage surface with respect to the damage hardening function is
∂g ∂f
= = −1
∂γ ∂γ
Also, the derivative of the hardening function γ with respect to conjugate variable δ is
needed
∂γ c1 δ
= exp
∂δ c2 c2
Next, the derivative of the thermodynamic forces w.r.t the internal damage variables is
written as

∂Y ∂Y ∂Y ∂σ
= + :
∂D ∂D σ=const ∂σ ∂D

i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Furthermore, the derivative of the thermodynamic forces w.r.t strain is written as

∂Y ∂Y ∂σ ∂Y
= : = :C
∂ε ∂σ ∂ε ∂σ
The following are written in contracted notation as
 
0 0 0
 −σ2 νe12 2σ2 σ1 νe12 
∂Y  2 e 3
− 2
0 
=  (1 − D2 ) E (1 − D2 ) E (1 − D2 ) E
 
1
e2 e1
∂σ

 
 2σ6 
0 0 3 e
(1 − D6 ) G 12
and
 
Ee2 νe12
 0 − ε2 0
 1 − νe12 νe21 

∂σ  
= Ee1 νe21 2 (1 − D2 )E e2 
∂D  0 − ε1 − ε2 0 
 1 − νe12 νe21 1 − νe12 νe21 

0 0 −4(1 − D6 ) G
e 12 ε6
The damage model is implemented in ANSYS using subroutine usermatps 805.f90,
available in [5], which can be used in conjunction with plain stress element (PLANE182 or
PLANE183) and laminate shells (SHELL181 or SHELL281).
First, follow the instructions in Appendix C.1.1 to make a dynamic link library (DLL)
with usermatps 805.f90, which you have to copy as usermatps.f90 in your work directory
before making the new DLL with AnsUserMatEjb.bat.
Next, use the APDL commands shown below and available in [5] for this example:
/TITLE, Matrix cracking Carbon/Epoxy, FEAcomp Ex. 8.5

!=== USERMAT DECLARATION SECTION ==================================

TB,USER,1,1,13, ! DECLARES USAGE OF USERMAT 1, MAT 1, PROPERTIES 13
TBTEMP,0
TBDATA,,171.4e3,9.08e3,0.32,,5.29e3, ! E1,E2,nu12,nu23,G12,G23
TBDATA,,62.29,92.34,170e-3,230e-3,0.5,-1.8 ! F2t,F6,GIc,GIIc,c1,c2
TBDATA,,1 ! gamma0
TB,STAT,1,,3 ! NUMBER OF STATE VARIABLES 3
!==================================================================
ET,1,182,,,3 ! PLANE182, plane elements with plane stress

R,1,1 ! Real constant #1, thickness = 1
N,2,1,0 ! Define node 2,
N,3,1,1
N,4,0,1
E,1,2,3,4 ! Generate element 1 by node 1 to 4
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
ANTYPE,STATIC
OUTRES,STAT,1 ! Store results of damage variables
D,1,all ! Define b.c. on node 1, totally fixed
! case 1
! apply one-dimensional strain in x-direction
!D,2,UX,0.02
!D,3,UX,0.02
!D,4,UX
! case 2
! apply one-dimensional strain in y-direction
!D,2,UY
!D,3,UY,0.02
!D,4,UY,0.02
! case 3
! apply in-plane shear stress
D,2,UY
D,3,UX,0.04
D,4,UX,0.04
D,3,UY,0.0
D,4,UY,0.0

/POST26 ! Start time-history post-process
! Stress X vs strain X
!ANSOL,2,3,S,X, UXnode ! stress-x
!ANSOL,3,3,EPEL,X, FXnode ! strain-x
! Stress Y vs strain Y
!ANSOL,2,3,S,Y, UXnode ! stress-y
!ANSOL,3,3,EPEL,Y, FXnode ! strain-y
! Stress XY vs strain XY
ANSOL,2,3,S,XY, UXnode ! stress-xy
ANSOL,3,3,EPEL,XY, FXnode ! strain-xy
XVAR,3 ! x-graph variable: strain

PLVAR,2 ! plot, y-graph variable: stress
LINES,1000 !
PRVAR,2,3 ! list stress and strain
!FINISH ! Exit post-process module
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
140
Transverse stress
120 In-plane shear stress
100
Stress [MPa]
80
60
40
20
0
0 0.5 1 1.5 2
Strain [%]
Figure 8.9: Response to in-plane shear stress and to transverse tensile stress.
The model response under one-dimensional transverse stress and under only in-plane
shear stress is shown in Figure 8.9.
8.5 Damage and Plasticity

For polymer matrix composites reinforced by strong and stiff fibers, damage and its
conjugate thermodynamic force can be described by second-order tensors D and Y.
Furthermore, the hardening processes that take place during plasticity and damage
imply additional dissipation, so that
ρπ = T ρπs = σ : ε̇p + R ṗ + Y : Ḋ + γ δ̇ (8.122)

where (R, p) is the thermodynamic force-flux pair associated to plastic hardening,
and (γ, δ) is the thermodynamic force-flux pair associated to damage hardening,
and ρπ is the dissipation heat due to irreversible phenomena.
For the particular case of (8.122), from (8.92) and (8.99), the following definitions
for the thermodynamic forces are obtained
∂ψ ∂ψ ∂χ ∂ψ ∂χ
σ=ρ = −ρ p ε=ρ Y = −ρ =ρ (8.123)
∂ε ∂ε ∂σ ∂D ∂D
as well as definitions for the hardening equations
∂χ ∂ψ ∂π ∂χ ∂ψ ∂π
γ=ρ = −ρ =ρ R=ρ = −ρ =ρ (8.124)
∂δ ∂δ ∂δ ∂p ∂p ∂p
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
The additive decomposition [64]
ε = εe + εp (8.125)
can be rewritten taking into account that the elastic component of strain can be
calculated from stress and compliance, so that
ε = S : σ + εp (8.126)
Therefore, the strain-stress law in incremental and rate form are
δε = S : δσ + δS : σ + δεp
ε̇ = S : σ̇ + Ṡ : σ + ε˙p (8.127)
showing that an increment of strain has three contributions: elastic, damage, and
plastic. The elastic strain occurs as a direct result of an increment in stress, the
damage strain is caused by the increment in compliance as the material damages,
and the plastic strain occurs at constant compliance. The elastic unloading stiffness
does not change due to plasticity but it reduces due to damage. Following this
argument, it is customary [93] to assume that the free energy and complementary
free energy can be separated as follows
ψ(ε, εp , p, D, δ) = ψ e (εe , D, δ) + ψ p (εp , p)

χ(σ, εp , p, D, δ) = χe (σ, D, δ) + χp (εp , p) (8.128)
Suggested Problems
Problem 8.1 Using the formulation and properties of Example 8.2, obtain a graphical
representation of the evolution of strain vs. nominal stress (ε vs. σ) and the evolution of
strain vs. effective stress (ε vs. σ
e) for a point on the top surface of the beam and for another
point on the bottom surface of the beam. Comment on the graphs obtained.
Problem 8.2 Implement a USERMAT for a one-dimensional CDM model active in the
x1 -direction only. Use 2D plane stress constitutive equations. Leave the x2 -direction, Pois-
son’s, and shear terms as linear elastic with no damage. Verify the program by recomputing
Example 8.2 and the plots obtained in Problem 8.1. Note that to obtain the same values,
the Poisson’s ratio should be set to zero.
Problem 8.3 The Gibbs free energy is defined in expanded form and using Voigt contracted
notation, as:
"
1 σ12 σ22 σ62
χ= 2 e
+ 2 e
+ −
2ρ (1 − D1 ) E 1 (1 − D2 ) E 2 (1 − D1 ) (1 − D2 ) G
e 12

νe21 νe12 σ1 σ2
− +
Ee2 Ee1 (1 − D1 ) (1 − D2 )
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Table 8.3: Elastic properties for composite material lamina
E
e1 Ee2 G
e 12 νe12
171.4 GPa 9.08 GPa 5.29 GPa 0.32
Table 8.4: Identification model parameters for composite material lamina
H1 H2 γ0 c1 c2
0.024 8.36 1.0 1.5 -2.8
where Ee1 , E
e2 , νe12 , νe21 , and Ge 12 are the undamaged in-plane elastic orthotropic properties of
a unidirectional lamina where the subindex ()1 denotes the fiber direction and ()2 denotes
the transverse direction. (a) Obtain the secant constitutive equations, C and S, using the
given Gibbs free energy. (b) Obtain the thermodynamic forces Y1 and Y2 associated to D1
and D2 . (c)If M is represented using Voigt contracted notation and multiplied by a Reuter
matrix as
 
(1 − D1 ) 0 0
M = 0 (1 − D2 ) √ 0√ 
0 0 1 − D1 1 − D2
check if this definition of M can be used as the damage effect tensor in a damage model
using the principle of energy equivalence. Justify and comment on your conclusion.
Problem 8.4 The damage activation function, for the model shown in Problem 8.3, is
defined as
q
g := ĝ − γ̂ = Y1 2 H1 + Y2 2 H2 − (γ + γ0 )
where H1 and H2 are model parameters that depend on elastic and strength material prop-
erties, and Y1 and Y2 are the thermodynamic forces associated to the damage variables D1
and D2 , respectively. The damage hardening depends on δ according to

δ
γ̂ = γ + γ0 = c1 exp − 1 + γ0
c2
where γ0 defines the initial threshold value, c1 and c2 are material parameters. All necessary
material parameters are shown in Tables 8.3 and 8.4.
a) Using a flowchart diagram, describe the algorithm, with all necessary steps to imple-
ment it as a constitutive subroutine in a finite element package.
b) Compute the analytic expressions necessary to implement the model in a USERMAT.
c) Program the algorithm using the USERMAT capability for a plane stress constitutive
equation.
d) Finally, using ANSYS, plot a single curve of apparent stress σ2 vs. apparent strain
ε2 for a RVE loaded only with ε2 .
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
e) Using APDL code, describe the process used to solve the problem in ANSYS.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Chapter 9
Discrete Damage Mechanics
Prediction of damage initiation and propagation is tackled in Chapter 8 using contin-

uum damage mechanics (CDM). Then, a particular CDM approach, called cohesive
zone model (CZM), is explained in Chapter 10. Application of CZM methodol-
ogy to the problem of intralaminar damage is made in Abaqus’ progressive damage
approach (PDA). A methodology to determine the critical energy release rates for
transverse tension and shear using Abaqus’ PDA is available in [56]. Also, a pro-
gressive CDM model for intralaminar damage is announced for ANSYS R
15.
Alternatives to CDM include: micromechanics of damage, crack opening dis-
placement methods, computational micromechanics, and synergistic methods. While
CDM (Chapter 8) homogenizes the damage and treats it phenomenologically, the
alternative methods attempt to represent the actual geometry and characteristics
of damage. Accurate physical representation of the fracture phenomena is the most
salient feature of alternative models, thus motivating their inclusion in this chapter.
Prediction of transverse matrix cracking in laminated composites has been ex-
tensively studied for the particular case of symmetric [0m /90n ]S laminates under
membrane loads, for which matrix cracking is found in the 90◦ laminas (trans-
verse laminas). Extensions to other laminate configurations such as [0/ ± θ/0]S and
[0/θ1 /θ2 ]S , models featuring cracks in the off-axis θ laminas have been developed,
but they are still limited to symmetric laminates subjected to in-plane loading.
Micromechanics of Damage Models (MMD) find an approximate elasticity so-
lution for a laminate with a discrete crack or cracks [94–115]. The solutions are
approximate because kinematic assumptions are made, such as a linear [116] or bi-
linear [117] distribution of interlaminar shear stress through the thickness of each
lamina, as well as particular spatial distributions of in-plane displacement func-
tions [114], stresses, and so on. The state variable is the crack density in the
cracking lamina, defined as the number of cracks per unit distance perpendicular
to the crack surface. Therefore, the state variable is measurable. One advantage
of MMD is that the reduction of laminate moduli as a function of crack density
is calculated without resorting to adjustable parameters as in the case of CDM. In
CDM, those parameters have to be found from often numerous and difficult physical
experiments, thus invalidating the main premise of virtual testing.
The main disadvantage of MMD is that most of the solutions available are
249
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
limited to symmetric laminates under membrane loads with only one or two laminas
cracking. A generalization to the case of multiple cracking laminas is presented in
this chapter by resorting to the concept of synergistic methods, explained below.
Crack Opening Displacement (COD) methods [118–125] are based on the theory
of elastic bodies with voids [126]. The distinct advantage of COD models is that
the laminate stiffness can be calculated for any laminate configuration, even non-
symmetric laminate stacking sequence (LSS), subject to any deformation, including
bending, featuring matrix cracking in any of its laminas [127]. The main disadvan-
tage of COD methods is that they often rely on parametric finite element analysis
(FEA), and thus their applicability is limited to the range of materials, LSS, loads,
and boundary conditions used in the underlying parametric/verification studies.
Numerical solutions, such as FEA, provide 3D solutions without the kinematic
simplifications of MMD and COD models [105, 121, 124, 128–131]. However, FEA
solutions require a new mesh and boundary conditions for each LSS, crack orienta-
tion, and so on, making them too cumbersome for practical application. Another
numerical approach is Monte Carlo simulation, where the probabilistic distribution
of flaws in material is considered [132–134]. Unfortunately, Monte Carlo simulations
require additional parameters that have to be adjusted by fitting the results of the
model to experimental damage evolution data. Such data is scarce.
Synergistic Damage Mechanics (SDM) methods combine elements of different
modeling strategies such as CDM and MMD [132, 133, 135–139], bringing the best
features of each of the models involved. For example, in this chapter the laminate
stiffness reduction is computed via MMD methods and the generalization to multiple
cracking laminas is made via CDM concepts, but unlike CDM models, no adjustable
parameters are needed.
9.1 Overview
In the following we describe how to use two material properties, the fracture tough-
ness in modes I and II, GIc , GIIc , to predict the damaging behavior and transverse
tensile and in-plane shear failure of a unidirectional fiber reinforced lamina embed-
ded in a laminate. The constraining effect of adjacent laminas is taken into account,
leading to apparent transverse strength F2t being a function of ply thickness. The
crack initiation strain, crack density evolution as a function of stress (strain) up
to crack saturation, and stress redistribution to adjacent laminas is predicted accu-
rately.
The physics of matrix cracking under transverse tension and in-plane shear is
as follows. No matter how much care is taken during the production process, there
are always defects in the material. These defects may be voids, microcracks, fiber–
matrix debonding, and so on, but all of them can be represented by a typical matrix
crack of representative size 2a0 , as shown in Figure 9.1.
When subject to load, matrix cracks grow parallel to the fiber orientation, as
shown in Figure 9.2, where it can be seen that cracks are aligned with the fiber
direction in the ±55◦ laminas. These sets of parallel cracks reduce the stiffness of the
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Discrete Damage Mechanics 251
1
2a0L 2a0
Figure 9.1: Representative crack geometry.
Figure 9.2: Matrix cracks in the ±55 laminas of a [0/ ± 554 /01/2 ]S laminate loaded
by an increasing tensile strain (top to bottom) along the 0◦ direction [140].
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
cracked lamina, which then sheds its share of the load onto the remaining laminas.
In each lamina, the damage caused by this set of parallel cracks is represented by
the crack density, defined as the inverse of the distance between two adjacent cracks
λ = 1/(2l), as shown in Figure 9.3. Therefore, the crack density is the only state
variable needed to represent the state of damage in the cracked lamina. Note that
the actual, discrete cracks are modeled by the theory, which is thus named discrete
damage mechanics (DDM).
The basic ingredients of the DDM model for transverse tension and in-plane
shear damage are listed below:
i. In each lamina i, the state variable is the crack density λi . Two damage
variables D2 (λi ) and D6 (λi ) are defined for convenience but they are not in-
dependent variables; instead, they are computed in terms of the crack density.
The set of crack densities for the laminate is denoted by λ = λi with i = 1...N ,
where N is the number of laminas in the laminate.
ii. The independent variable is the midsurface1 strain = {1 , 2 , γ12 }T .
iii. The damage activation function, which separates the damaging states from
the undamaging states is written as follows

GI (λ, , ∆T ) GII (λ, , ∆T )
g = max , −1≤0 (9.1)
GIc GIIc
where g ≤ 0 represents the undamaging domain. The critical energy release

rates (ERR) are not easily found in the literature but they can be fit to
available experimental data in the form of crack density vs. applied strain or
laminate modulus vs. applied strain [56].
iv. The damage threshold is embedded into g, and represented by the (invariant)
material properties GIc , GIIc . Before damage starts, λ = 0 and (9.1) is a
damage initiation criterion, similar to [141] but without mode interaction.
With λ = 0, the strain for which g = 0 is the strain for crack initiation. Once
damage starts, (9.1) becomes a damage activation function by virtue of the
automatic hardening described below.
v. The hardening function is embedded into the damage activation function

g. For a given value of strain, the calculated values of energy release rate
GI (λ), GII (λ) are monotonically decreasing functions of λ. Therefore, as soon
as λ grows, GI (λ), GII (λ) decrease, making g < 0 and thus stopping further
damage until the driving thermodynamic force, i.e., the strain, is increased by
the application of additional load [83].
vi. No damage evolution function need to be postulated, with the advantage

that no new empirical parameters are needed. Simply the crack density λ
1
The analysis presented in this section is for symmetric laminates under membrane forces. A
formulation for unsymmetric laminates and/or laminates under bending is being reported in [127].
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Top View
y
2
1
x
RVE
1
2l
Side View
z 2l RVE
Homogenized Laminae
h/2
k Lamina
y
symmetry plane
Figure 9.3: Representative unit cell used in discrete damage mechanics.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
adjusts itself to a value that will set the laminate in equilibrium with the
external loads for the current strain while satisfying g = 0. A return mapping
algorithm (Section 8.4.1) achieves this by iterating until g = 0 and updating
∂g
the crack density with iterative increments calculated as ∆λ = −g/ ∂λ .
vii. The crack density grows until the lamina is saturated with cracks (λ → ∞).
At that point the lamina loses all of its transverse and shear stiffness (D2 ≈
1, D6 ≈ 1), at which point all of the load is already transferred to the remaining
laminas in the laminate. The analysis of the cracked lamina is stopped when
the crack density reaches λlim = 1/hk , where hk is the thickness of lamina k;
i.e., when cracks are closely spaced at a distance equal to the lamina thickness.
Having described the ingredients of the model, it now remains to show how to
calculate the various quantities. The solution begins by calculating the degraded2
stiffness of the laminate Q = A/h for a given crack density λk in a cracked lamina
k, where A is the in-plane laminate stiffness matrix, and h is the thickness of the
laminate.
The following conventions are used in this section:
– (i) denotes any lamina in the laminate.
– (k) denotes the cracking lamina.
– (m) denotes any lamina other than the cracking one (m 6= k).
– A superscript in parentheses (i) denotes the lamina number; not a power or
order of differentiation.
– xj with j = 1, 2, 3, denote the coordinates x1 , x2 , x3 , or j = 1, 2, 6 for quantities
expressed in Voigt contracted notation.
– u(xj ), v(xj ), w(xj ), with j = 1, 2, 3 are the three components of the displace-
ment.
– hat p̂ denotes the thickness average of quantity p, where the thickness is
mentioned or it is obvious from context.
– tilde pe denotes the virgin value of quantity p.
– overline p denotes the volume average of quantity p.
9.2 Approximations
Most practical laminates are symmetric and the most efficient use of them is by
designing the structure to be loaded predominantly with membrane loads [1, Chap-
ter 12]. Therefore, the solution presented here is for a symmetric laminate under
membrane loads. In this case,
∂w(i) ∂w(i)
= =0 (9.2)
∂x1 ∂x2
where u(xj ), v(xj ), w(xj ), with j = 1, 2, 3 are the displacements of a point in lamina
i as a function of the coordinates xj with j = 1, 2, 3. Furthermore, the thickness
2
Also called “damaged,” “reduced,” or “homogenized.”
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
h(i) of the laminas are assumed to be small, so that the plane stress assumption
holds
(i)
σ3 = 0 (9.3)
Since all cracks are parallel to the fiber direction and practical designs avoid
thick laminas, it can be expected that the cracks occupy the entire thickness of the
lamina. Any crack smaller than the lamina thickness is unstable both through the
lamina thickness and along the fiber direction [1, Section 7.2.1].
Since the objective is to calculate the laminate stiffness reduction due to cracks,
it suffices to work with thickness averages of the variables. A thickness average is
denoted by
Z Z
1 0
φ̂ = 0 φ dx3 ; h = dx3 (9.4)
h h0
where h0 can be the lamina or laminate thickness, denoted by h(i) , h, respectively.
Specifically,
– û(i) (xj ), v̂ (i) (xj ), ŵ(i) (xj ), are the thickness-average displacements in lamina i
as a function of the in-plane coordinates xj with j = 1, 2.
(i) (i) (i)
– ˆ1 (xj ), ˆ2 (xj ), γ̂12 (xj ), are the thickness-average strains in lamina i.
(i) (i) (i)
– σ̂1 (xj ), σ̂2 (xj ), τ̂12 (xj ), are the thickness-average stress in lamina i.
Out-of-plane (intralaminar) shear stress components appear due to the the per-
turbation of the displacement field caused by the crack. These are approximated
by linear functions through the thickness of the lamina i, as follows
i−1,i
i−1,i x3 − x3

(i) i−1,i i,i+1
τ13 (x3 ) = τ13 + τ13 − τ13
h(i)
x − xi−1,i
(i) i−1,i i,i+1 i−1,i 3 3
τ23 (x3 ) = τ23 + τ23 − τ23 (9.5)
h(i)
where x3i−1 is the thickness coordinate at the bottom of lamina i, i.e., at the in-
i−1,i
terface between lamina i − 1 and lamina i, and τ13 is the shear stress at the
interface between the i − 1 and the i lamina. This assumption, which is common
to several other analytical models, is called the shear lag assumption. The linear
approximation has been shown to yield accurate results [131].
The shear lag equations are obtained from the constitutive equations for out-of-
plane shear strains and stresses using weighted averages [137, Appendix A],
(i) (i−1) ( i−2,i−1 )
û − û(i−1)

(i−1) S 45 S 55 τ23
= h 6 i−2,i−1
v̂ (i) − v̂ (i−1) S44 S45 τ13
" (i−1) (i) # ( i−1,i )
h(i−1) S45 S55 h(i) S45 S55 τ23
+ 3 + 3 i−1,i
S44 S45 S44 S45 τ13
(i) ( i,i+1 )
(i) S 45 S55 τ23
+ h6 i,i+1 (9.6)
S44 S45 τ13
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Inverting (9.6), the intralaminar stresses are written in terms of displacements

at the interfaces, as follows
n−1
Xh in o
i,i+1
τ23 i−1,i
− τ23 = [H]−1 2i−1,2j−1 − [H]−1 2i−3,2j−1 û(j+1) − û(j)
j=1
h in o
+ [H]−1 2i−1,2j − [H]−1 2i−3,2j v̂ (j+1) − v̂ (j)
n−1
Xh in o
i,i+1
τ13 i−1,i
− τ13 = [H]−1 2i,2j−1 − [H]−1 2i−2,2j−1 û(j+1) − û(j)
j=1
h in o
+ [H]−1 2i,2j − [H]−1 2i−2,2j v̂ (j+1) − v̂ (j) (9.7)
in terms of the 2(N − 1) by 2(N − 1) coefficient matrix H.
9.3 Lamina Constitutive Equation

The stress-strain law for the cracking lamina k is that of an intact material, i.e.,

(k) e (k) ˆ(k) − α(k) ∆T
σ̂i = Q ij j j (9.8)
(k)
where α(k) is the coefficient of thermal expansion (CTE) of lamina k, σi =
(k) (k) T
n o
(k)
σ1 , σ2 , τ12 , and tilde denotes a virgin property. The strain-displacement
equations are
 (k) (k)


 1 = u,1 

(k) (k) (k)
= 2 = v,2 (9.9)
 (k)
 (k) (k) 

γ12 = u,2 + v,1
For the remaining laminas (m 6= k), the constitutive equations can be obtained
(m)
using (9.8) and the stiffness matrix Qij , written in terms of their previously calcu-
(m) (m)
lated damage values D2 , D6 , defined in (9.32), and rotated to the k coordinate
system using the usual transformation equations [1, Section 5.4]
e (m) (m) e (m)

 
Q 11 1 − D2 Q12 0
(m) e (m) (m) e (m)
Q(m) = [T (−θ)]  1 − D2 Q 1 − D2  [T (−θ)]T
 
12 Q22 0
 
(m) e (m)
0 0 1 − D6 Q66
(9.10)
9.4 Displacement Field

The objective now is to solve for the average displacements û(i) (xj ), v̂ (i) (xj ); j = 1, 2,
in all laminas i for a given crack density set λ and applied strain . Taking into
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
account that the intralaminar shear stresses are assumed to vary linearly through
the thickness of each lamina, the equilibrium equations (1.15) for each lamina can
be written as follows

(i) (i) i,i+1 i−1,i
σ̂1,1 + τ̂12,2 + τ̂13 − τ̂13 /hi = 0 (9.11)

(i) (i) i,i+1 i−1,i
τ̂12,1 + σ̂2,2 + τ̂23 − τ̂23 /hi = 0 (9.12)
Using the the strain-displacement equations (9.9), the constitutive equations

(9.8), and the shear lag equations (9.7) into the equilibrium equations (9.11)–(9.12)
leads to a system of 2N partial differential equations (PDE) in û(i) (xj ), v̂ (i) (xj ).
The PDE has particular solutions of the form
û(i) = ai sinh λe x2 + a x1 + b x2
v̂ (i) = bi sinh λe x2 + b x1 + a∗ x2 (9.13)
where e is the eigenvalue number. The general solution can be written as
 (1)       

 û 
 
 a1 
 
 a 
 
 b 

û(2)
       








 a2 




 a 




 b 


       


 . 




 . 




 . 




 . 


       



 . 






 . 






 . 






 . 





 . 




 . 




 . 




 . 


 (n)
  2N      
û an a b
 X      
= Ae sinh (ηe x2 ) + x1 + x2 (9.14)
 v̂ (1)   b1   b   a∗ 
  e=1      
v̂ (2) b2 b a∗

 
 
 
 
 
 
 


 
 
 
 
 
 
 

  
.
 
.
 
.




 . 





























 . 




 . 




 . 




 . 


       


 .





 . 




 . 




 . 


a∗

 (n)       
v̂
  bn 
e
 b   
which substituted into the PDE leads to the eigenvalue problem

α1 β1 aj 2 ζ26 ζ22 aj 0
+η = (9.15)
α2 β2 bj ζ66 ζ26 bj 0
where j = 1...2N ; η are the 2N eigenvalues and {aj , bj }T are the 2N eigenvectors
of (9.15).
It turns out that two of the eigenvalues are always zero (corresponding to the
linear terms in (9.14)), which can be taken to be the last two in the set, thus
remaining only 2N − 2 independent solutions. Then, the general solution of the
PDE system is built as the linear combination of the 2N − 2 independent solutions
as follows
(i) 2N −2
û X ai a b
= Ae sinh (ηe x2 ) + x1 + x2 (9.16)
v̂ (i) bi e b a∗
e=1
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
where Ae are unknown coefficients in the linear combination. It can be seen that
the general solution contains 2N + 1 unknown coefficients, including the scalars
a, b, a∗ , and the sets Ae with e = 1...2N − 2. To determine these coefficients, one
needs 2N + 1 boundary conditions on the boundary of the representative volume
element (RVE) in Figure 9.3. Note the that RVE spans a unit length along the
fiber direction x1 , a distance 2l between successive cracks (along x2 ) and the whole
thickness h of the symmetric laminate.
Two very important parameters are introduced through the boundary condi-
tions, namely the crack density λ and the stress σ̂ = N/h applied to the laminate,
where N is the in-plane force per unit length.3 The crack density enters through the
dimension of the RVE, which has a width of 2l = 1/λ. The applied stress (or strain)
enters through the force equilibrium on the RVE. In summary, there are 2N + 1
boundary conditions that lead to a system of 2N + 1 algebraic equations that can
be solved for the 2N + 1 coefficients in (9.16). Therefore, the average displacements
in all laminas are now known from (9.16) for given values of crack density λ and
applied load σ̂ = N/h.
9.4.1 Boundary Conditions for ∆T = 0
First consider the case of mechanical loads and no thermal loads. To find the values
of Ae , a, a∗, b, the following boundary conditions are enforced: (a) stress-free at
the crack surfaces, (b) external loads, and (c) homogeneous displacements. The
boundary conditions are then assembled into an algebraic system as follows
n oT
[B] Ae , a, a* , b = {F } (9.17)
T
where [B] is the coefficient matrix of dimensions 2N + 1 by 2N + 1; Ae , a, a* , b

represents the 2N + 1 unknown coefficients, and {F } is the right hand side (RHS)
or force vector, also of dimension 2N + 1.
(a) Stress-Free at the Crack Surfaces
The surfaces of the cracks are stress-free
1/2
(k)
∫ σ̂2 (x1 , l) dx1 = 0 (9.18)
−1/2
1/2
(k)
∫ τ̂12 (x1 , l) dx1 = 0 (9.19)
−1/2
3
Not to be confused with the number of laminas N.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
(b) External Loads
In the direction parallel to the surface of the cracks (fiber direction x1 ) the load is
supported by all the laminas
N Zl
1 X (i)
hi σ̂1 (1/2, x2 )dx2 = hσ̂1 (9.20)
2l
i=1 −l
In the direction normal to the crack surface (x2 direction) only the uncracking
(homogenized) laminas carry load
Z1/2
(m)
X
hm σ̂2 (x1 , l) dx1 = hσ̂2 (9.21)
m6=k 1/2
Z1/2
(m)
X
hm τ̂12 (x1 , l)dx1 = hτ̂12 (9.22)
m6=k 1/2
(c) Homogeneous Displacements
For a homogenized symmetric laminate, membrane loads produce a uniform dis-

placement field through the thickness, i.e., all the uncracking laminas are subjected
to the same displacement
û(m) (x1 , l) = û(r) (x1 , l) ; ∀m 6= k (9.23)

(m) (r)
v̂ (x1 , l) = v̂ (x1 , l) ; ∀m 6= k (9.24)
where r is an uncracked lamina taken as the reference. In the computer implemen-

tation, lamina 1 is taken as the reference unless lamina 1 is cracking, in which case
lamina 2 is taken as the reference lamina.
9.4.2 Boundary Conditions for ∆T 6= 0
Next, consider the case of thermal loads, which add a constant term to the boundary
conditions. Constant terms do not affer the matrix [B], but rather subtract from
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
the forcing vector {F }, as follows

 
P (k) (k)

 ∆T Q̄1j ᾱj 

j=1,2,6

 

 
 P (k) (k) 
∆T Q̄1j ᾱj

 


 


 j=1,2,6 

(i) (i)

 P P 



 ∆T Q̄1j ᾱj 


i6=(k) j=1,2,6

 

 
 P P (i) (i) 
∆T Q̄2j ᾱj

 


 

 i6=k j=1,2,6 
{∆F }∆T 6=0 = ∆T
P P (i) (i)
Q̄6j ᾱj (9.25)
 
i6=k j=1,2,6

 


 



 0 


 



 0 






 ... 






 ... 



0

 


 

0
 
In this way, the strain calculated for a unit thermal load (∆T = 1) is the
degraded CTE of the laminate for the current crack density set λ.
9.5 Degraded Laminate Stiffness and CTE

In this section, we calculate the degraded stiffness of the laminate Q = A/h for
a given crack density λk in a cracked lamina k, where A is the in-plane laminate
stiffness matrix, and h is the thickness of the laminate. First, the thickness-averaged
strain field in all laminas can be obtained by using the kinematic equations (9.9),
namely by differentiating (9.16). Then, the compliance of the laminate S in the
coordinate system of lamina k can be calculated one column at a time by solving
for the strains (9.9) for three load cases, a, b, and c, all with ∆T = 0, as follows
     
 1   0   0 
a b c
σ̂ = 0 ; σ̂ = 1 ; σ̂ = 0 ; ∆T =0 (9.26)
0 0 1
     
Since the three applied stress states are unit values, for each case, a, b, c, the
volume average of the strain (9.9) represents one column in the laminate compliance
matrix
a b c

x x x
S= a y b c  (9.27)
 y y
aγ bγ cγ
xy xy xy
where x, y, are the coordinates of lamina k (Figure 9.3). Next, the laminate stiffness
in the coordinate system of lamina k is
Q = S −1 (9.28)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
To get the degraded CTE of the laminate, one sets σ̂ = {0, 0, 0}T and ∆T = 1.
The resulting strain is equal to the CTE of the laminate, i.e., {αx , αy , αxy }T =
{x , y , γxy }T .
9.6 Degraded Lamina Stiffness

The stiffness of lamina m, with m 6= k, in the coordinate system of lamina k (see
(m) (m)
Figure 9.3) is given by (9.10) in terms of the previously calculated values D2 , D6 ,
(k)
given by (9.32). The stiffness of the cracking lamina Q is yet unknown. Note
that all quantities are expressed in the coordinate system of lamina k.
The laminate stiffness is defined by the contribution of the cracking lamina k
plus the contribution of the remaining N − 1 laminas, as follows
n
(k) hk hm
X
Q=Q + (1 − δmk )Q(m) (9.29)
h h
m=1
where the delta Dirac is defined as δmk = 1 if m = k, otherwise 0. The left-hand side
(LHS) of (9.29) is known from (9.28) and all values of Q(m) can be easily calculated
since the m laminas are not cracking at the moment. Therefore, one can calculate
the degraded stiffness Q(k) of lamina k as follows
n
" #
h X hm
Q(k) = Q− (1 − δmk )Q(m) (9.30)
hk h
m=1
where Q without a superscript is the stiffness of the laminate.

To facilitate later calculations, the stiffness Q(k) can be written using concepts
of continuum damage mechanics (Section 8.2) in terms of the stiffness of the un-
(k) (k)
damaged lamina and damage variables D2 , D6 , as follows
e (k) e (k)
 
Q 11 (1 − D2 ) Q 12 0
Q(k) = (1 − D2 ) Q
 e (k) (1 − D2 ) (k)
Q22
e 0

(9.31)
12 
(k)
0 0 (1 − D6 ) Q66
e
(k)
with Dj calculated for a given crack density λk and applied strain 0 , as follows
(k) (k) (k)

Dj (λk , 0 ) = 1 − Qjj /Q
e
jj ; j = 2, 6; no sum on j (9.32)
where Qe (k) is the original value of the undamaged property and Q(k) is the degraded4
value computed in (9.30), both expressed in the coordinate system of lamina k.
4
Homogenized.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
The coefficient of thermal expansion of the cracking lamina k is calculated in a

similar fashion, as follows
 
1 X
α(k) = (k) S (k) hQα − h(m) Q(m) α(m)  (9.33)
t m6=k
−1
with S = Q(k) . The corresponding thermal damage is calculated as

α(k) (k)
Dj = 1 − αj (k) /e
αj ; j = 2, 6 (9.34)
9.7 Fracture Energy

Under displacement control, the energy release rate (ERR) is defined as the partial
derivative of the strain energy U with respect to the crack area A (see (10.1)).
According to experimental observations on laminated, brittle matrix composites
(e.g., using most toughened epoxy matrices), cracks develop suddenly over a finite
length, and thus are not infinitesimal. Then, Griffith’s energy principle is applied on
its discrete (finite) form in order to describe the observed, discrete (finite) behavior
of crack growth, as follows
∆UI
GI = −
∆A
∆UII
GII =− (9.35)
∆A
where ∆UI , ∆UII , are the change in laminate strain energy during mode I and mode
II finite crack growth, respectively; and ∆A is the is the newly created (finite) crack
area, which is one half of the new crack surface. Counting crack area as one-half
of crack surface is consistent with the classical fracture mechanics convention for
which fracture toughness Gc is twice of Griffith’s surface energy γc .
To calculate the ERR, it is convenient to use the laminate stiffness Qij in the
coordinate system (c.s.) of the cracked lamina, because in this way, the ERR can be
decomposed into opening and shear modes. Since the laminate stiffness is available
from the analysis as a function of crack densityλ, the ERR can be calculated, for
a fixed strain level (load), and using [111] and [142, Section 3.2.10], into (9.35), we
arrive at
V
GI = − (2 − α2 ∆T ) ∆Q2j (j − αj ∆T ) ; opening mode (9.36)
2∆A
V
GII =− (6 − α6 ∆T ) ∆Q6j (j − αj ∆T ) ; shear mode (9.37)
2∆A
where V, ∆A, are the volume of the RVE and the increment of crack area, respec-
tively; ∆Qij is the change in laminate stiffness corresponding to the change in crack
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
area experienced; and all quantities are laminate average quantities expressed in
the c.s of the cracked lamina in order to allow for ERR mode decomposition [111].
In the current implementation of the model, which is used in Example 9.1,
∆A = hk is the area of one new crack appearing halfway between two existing
cracks. In this case the crack density doubles and ∆Q = Q(2λ) − Q(λ) < 0.
Alternative crack propagation strategies are considered in [127].
It can be seen that the proposed methodology provides the key ingredients for
the computation of the ERR; namely the degraded stiffness and degraded CTE of
the laminate, both as a function of crack density.
The damage activation function (9.1) can now be calculated for any value of λ
and applied strain x , y , γxy . Note that the computation of the ERR components
derives directly from the displacement solution (9.16) for a discrete crack (Figure
9.3). When this formulation is used along with the finite element method (FEM),
it does not display mesh dependency on the solution and does not require the
arbitrary specification of a characteristic length [138], in contrast to formulations
based on smeared crack approximations [56]. The effect of residual thermal stresses
is incorporated into the formulation. The code is available as a user material for
ANSYS [138], which is used in Example 9.1. Other implementations include a shell
user element for ANSYS [139] and a user general section (UGENS) for Abaqus TM .
9.8 Solution Algorithm

The solution algorithm consists of (a) strain steps, (b) laminate-iterations, and
(c) lamina-iterations. The state variables for the laminate are the array of crack
densities for all laminas i and the membrane strain . At each load (strain) step,
the strain on the laminate is increased and the laminas are checked for damage.
9.8.1 Lamina Iterations

When matrix cracking is detected in lamina k, a return-mapping algorithm (RMA)
(Section 8.4.1) is invoked to iterate and adjust the crack density λk in lamina k
in such a way that gk returns to zero while maintaining equilibrium between the
external forces and the internal forces in the laminas. The iterative procedure works
as follows. At a given strain level for the laminate and given λk for lamina k,
calculate the value of the damage activation function gk and the damage variables,
which are both functions of λk . The RMA calculates the increment (decrement) of
crack density as
∂gk
∆λk = −gk / (9.38)
∂λ
until gk = 0 is satisfied within a given tolerance, for all k = 1...n, where n is the
number of laminas in the laminate. The analysis starts with a negligible value of
crack density present in all laminas (λ = 0.02 cracks/mm were used in the examples).
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Table 9.1: Properties for Example 9.1

Property Value
E1 [GPa] 44.7
E2 [GPa] 12.7
G12 [GPa] 5.8
ν12 0.297
ν23 0.41
Ply thickness [mm] 0.144
GIc [kJ/m2 ] 0.254
GIIc [kJ/m2 ] 1.400
CTE1 [1E-6/c ircC] 3.7
CTE2 [1E-6/◦ C] 30
∆T [◦ C] 0
9.8.2 Laminate Iterations

To calculate the stiffness reduction of a cracked lamina (k -lamina), all of the other
laminas (m-laminas) in the laminate are considered not damaging during the course
of lamina-iterations in lamina k, but with damaged properties calculated according
(m)
to the current values of their damage variables Di . Given a trial value of λk ,
(k)
the analytical solution provides gk , Di , for lamina k assuming all other laminas do
not damage while performing lamina iterations in lamina k. Since the solution for
lamina k depends on the stiffness of the remaining laminas, a converged iteration
for lamina k does not guarantee convergence for the same lamina once the damage
in the remaining laminas is updated. In other words, within a given strain step, the
stiffness and damage of all the laminas are interrelated and they must all converge.
This can be accomplished by laminate-iterations; that is, looping over all laminas
repeatedly until all laminas converge to g = 0 for all k.
Example 9.1 Consider a [0/908 /0/908 /0] laminate made of Glass/Epoxy with prop-
erties given in Table 9.1 subjected to a membrane strain x 6= 0, y = γxy = 0. Vi-
sualize the crack density in lamina k = 2 for an uniform applied strain x = 0.48%.
Also, plot the average laminate stress σx = Nx /h, where Nx , h are the stress resul-
tant and the total laminate thickness, respectively.
Solution to Example 9.1 Since ANSYS does not have a built-in capability for
calculating crack density, we have to use a plugin. In this case, we use a user
material subroutine for a state of plane stress that implements Discrete Damage
Mechanics. The DDM plugin is available in [5, USERMATLib.DLL] and the theory
behind it is explained in this chapter.
The DDM plugin employs 3+9N parameters, and 3N state variables, as follows:
– The whole laminate thickness, which will be entered later as a REAL CON-
STANT associated to the MESH. This is not really a requirement for the
plugin, but ANSYS needs this value for plane stress elements, both PLANE
and SHELL, so it is mentioned here as part of the requirements.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
– The plugin requires an input of 3 + 9 ∗ N material properties, where N is the

number of laminas in the symmetric part of the laminate, i.e., in one half of
the LSS. The properties are ordered as follows, starting with the first lamina,
k = 1 (bottom surface), and continuing until the lamina N (middle surface):
GIc Critical value of ERR in mode I.

GIIc Critical value of ERR in mode II. Use GIIc > 4 GIc if data is not
available.
∆T Change in temperature from the temperature at which GIc , GIIc , were
measured to the operating temperature.
E1 Longitudinal modulus.
E2 Transverse modulus.
G12 In-plane shear modulus.
ν12 In-plane Poisson’s ratio.
ν23 Intralaminar Poisson’s ratio. Note; G23 = E2 /2/(1 + ν23 ).
α1 Longitudinal coefficient of thermal expansion.
α2 Transverse coefficient of thermal expansion.
θk Lamina orientation with respect to the laminate c.s.
tk Lamina thickness.
– Next, the plugin calculates 3 ∗ N state variables, starting with the first lamina,
k = 1 (bottom surface), and continuing until the lamina N (middle surface):
λk Crack density in lamina k.

D2 Transverse damage, lamina k.
D6 Shear damage, lamina k.
This example illustrates how to create a model using ANSYS/APDL, run the
Job with the plugin, and visualize the results. Instructions on how to use a .dll
with ANSYS are given in Appendix C.1.1. The APDL input file is available in [5,
FEAcomp Ex901.inp] and it is fully explained next.
i. Parametric modeling in APDL

First, the model is set parametrically, as follows. True parameters are:
– The applied strain at the end of the time step,

– The initial crack density. Note that DDM requires a small amount of
damage (material defects, etc.) to start the analysis.
– The shell dimensions, which in this case it is a square.
– The ply thickness. A lamina may use one or more plies to achieve the
desired thickness.
– The number of laminas (call layers in the ANSYS documentation).
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
– The number of properties, which for DDM is calculated in terms of the

number of laminas.
The APDL code to set up the model parametrically is shown next.
/TITLE, FEAcomp Ex. 9.01, USERMATLib.DLL

!=== PARAMETERS ==================================================
appliedStrain = 2. ! percent
L0 = 0.02 ! initial crack density
ShellDimensionX = 1. ! model dimensions
ShellDimensionY = 1. ! mm
tk =.144 ! ply thickness
NL = 3 ! number layers half laminate
Nprops = 3+9*NL ! # material properties
!=== NEXT VALUES GO IN TBDATA ====================================
GIc = .254
GIIc = 1.E16
deltaT = 0.
E1 = 44700 ! MPa
E2 = 12700
G12= 5800
nu12 =.297
nu23 =.410
CTE1 =3.7
CTE2 =30.
!Angle with TBDATA for each layer
!Thickness with TBDATA for each layer
ii. TBDATA
The 3 + 9N required parameters are supplied via TBDATA line in the APDL
file, as follows:
!=== USERMAT DECLARATION SECTION ==================================

TB,USER,1,1,Nprops, ! DECLARES USAGE OF USERMAT 1, MAT 1,
TBTEMP,0 ! ref. temperature
TBDATA,,GIc,GIIc,detaT,E1,E2,G12 ! 6 values per TBDATA line
TBDATA,,nu12,nu23,CTE1,CTE2,0,tk
TBDATA,,E1,E2,G12,nu12,nu23,CTE1
TBDATA,,CTE2,90,8*tk,E1,E2,G12
TBDATA,,nu12,nu23,CTE1,CTE2,0,tk/2
TB,STAT,1,,3*NL ! NUMBER OF STATE VARIABLES
To complete the USERMAT declaration, the plugin requires the 3N state vari-
ables to be initialized to a small value, i.e.,
! INITIALIZE THE STATE VARIABLES

TBDATA,,L0,L0,L0,L0,L0,L0
TBDATA,,L0,L0,L0
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
iii. Creating the mesh

Since the strain field inside the shell is uniform (in x and y), we use only
one PLANE182 element to model a unit cell with dimensions a = 100 × b =
100 mm, where a, b, are the dimensions of the model along the x, y, directions
respectively.
!=== MESH =========================================================

ET,1,182,,,3 ! PLANE182, plane elements with plane stress
R,1,2*NL*tk ! Real const. #1, thickness of whole laminate
N,2,ShellDimensionX,0 ! Define node 2,
N,3,ShellDimensionX,ShellDimensionY
N,4,0,ShellDimensionY
E,1,2,3,4 ! Generate element 1 by node 1 to 4
iv. Solution
Use the OUTRES,SVAR,1 command to store the values of state variables for
every substep.

ANTYPE,STATIC
OUTRES,SVAR,1 ! Store results of state variables
Then, set up a uniform deformation x using displacement boundary condi-

tions.
! Define one-dimensional stress in 1-axes direction

D,1,all ! Define b.c. on node 1, fixed
D,2,UY,0.00 ! Symmetry
D,4,UX,0.00 ! Symmetry
D,2,UX,appliedStrain*ShellDimensionX/100 ! applied displacement
D,3,UX,appliedStrain*ShellDimensionX/100 ! applied displacement
Since damage is a nonlinear problem, one has to choose some time of incre-
mentation within the STEP. If AUTOS,ON, the solution is obtained for optimum
but uneven substep size. If AUTOS,OFF, the solution is obtained for equal sub-
steps, which makes it easier to plot the solution.
AUTOS, OFF ! Automatic substeps OFF (min supstep=1/desired)

NSUBST,100,200,100 ! substeps: desired, max.#, min.#
v. Field Visualization
The filed postprocessor (/POST1) can be used to produce a contour plot of
state variables, as shown in Figure 9.4 which in this case is not very interesting
because the values are uniform over the x-y domain. The T IM E = 0.24 was
selected purposely to coincide with the initiation of damage at x = 0.48%.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i

/GRA,FULL ! NEEDED FOR PLOTTING SVARS
RSYS,SOLU ! RSYS: ACTIVATE RESULTS IN SOLUTION COORD. SYSTEM
SET,,,,,0.24 ! SET,,,,,TIME : SELECT TIME
PLESOL,SVAR,4 ! PLESOL: CONTOUR PLOT STATE VAR NUMBER
vi. Time Visualization

The time postprocessor (/POST26) can be used to produce a time plot of state
variables, as shown in Figure 9.5.
/POST26 ! Start time-history post-process

ANSOL,2,3,EPEL,X,EpsXNod3 ! Var #2, Node 3, Strain, X-dir, label
ANSOL,4,3,S,X,SxNod3 ! Var #3, Node 3, Stress, X-dir, label
/AXLAB,X,STRAIN
/AXLAB,Y,STRESS
XVAR,2 ! plot #2 as abscissa
PLVAR,3 ! plot #3 as ordinate
and also a printout, that can be saved to a file for further use.
! list time(default), strain=2, reactions=3

PRVAR,2,3
FINISH ! Exit post-process module
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 9.4: Uniform crack density λ = 0.033923 crack/mm, in layer 2 (SV AR = 4),
shown for applied strain x = 0.48205%, for Example 9.1.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 9.5: Average laminate stress σx = Nx /h vs. applied strain x , for Exam-
ple 9.1.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Table 9.2: Laminates for Exercises 9.1–9.4

Laminate number LSS
1 [0/908 /0/908 /0]
2 [0/704 / − 704 /0/ − 704 /704 /0]
3 [0/554 / − 554 /0/ − 554 /554 /0]
4 [02 /908 /02 ]
5 [15/ − 15/908 / − 15/15]
6 [30/ − 30/908 / − 30/30]
7 [40/ − 40/908 / − 40/40]
Suggested Problems
Problem 9.1 Calculate the critical laminate strain cx for which the first crack appears in
laminates 1 to 3 in Table 9.2, all made of Glass/Epoxy with properties given in Table 9.1
subjected to a membrane strain x 6= 0, y = γxy = 0. For each laminate, in what lamina
the first crack appears?
Problem 9.2 Using the results from Exercise 9.1, calculate the the in-situ transverse strength
is
F2t of the laminas [1, Section 7.2.1].
Problem 9.3 Calculate the critical laminate strain cx for which the first crack appears in
laminates 4 to 7 in Table 9.2, all made of Glass/Epoxy with properties given in Table 9.1
subjected to a membrane strain x 6= 0, y = γxy = 0. For each laminate, in what lamina
the first crack appears?
Problem 9.4 Using the results from Exercise 9.3, calculate the the in-situ transverse strength
is
F2t of the laminas [1, Section 7.2.1].
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Chapter 10
Delaminations
Delamination is a frequent mode of failure affecting the structural performance of

composite laminates. The interface between laminas offers a low-resistance path
for crack growth because the bonding between two adjacent laminas depends only
on matrix properties. Delamination may originate from manufacturing imperfec-
tions, cracks produced by fatigue or low velocity impact, stress concentration near
geometrical/material discontinuity such as joints and free edges, or due to high
interlaminar stresses.
In laminates loaded in compression, the delaminated laminas may buckle, and
cracks propagate due to interaction between delamination growth and buckling.
The presence of delaminations may reduce drastically the buckling load and the
compressive strength of the composite laminates [143] (Figure 10.1). Delamina-
tions may also be driven by buckling in laminates under transverse loading [144].
The analysis of delamination buckling requires the combination of geometrically
nonlinear structural analysis with fracture mechanics.
According to its shape, delaminations are classified into through-the-width or
strip [144–151], circular [151–157], elliptic [158], rectangular [159], or arbitrary [160,
161]. Depending on its location through the laminate thickness, delaminations are
classified into thin film, symmetric split [143, 146, 147], and general [148, 151, 154,
155, 157]. In addition, analysis of combined buckling and growth for composite
laminates containing multiple delaminations under in-plane compressive loading
has been carried out [162, 163]. Experimental results on delamination buckling are
presented in [164, 165].
Other delamination configurations that have been investigated in the literature
are the beam-type delamination specimens subjected to bending, axial, and shear
loading [164–170] which form the basis for experimental methods used to measure
interlaminar fracture strength under pure mode I, mode II, and mixed mode con-
ditions in composites, adhesive joints, and other laminated materials (Figure 10.2).
In plates with piezoelectric sensors or actuators, an imperfect bonding between
the piezoelectric lamina and the base plate may grow under mechanical and/or
electrical loading. As a consequence, the adaptive properties of the smart system
can be significantly reduced since debonding results in significant changes to the
static or dynamic response [171, 172]. Finally, delamination growth may be caused
273
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
g pat
axial load kl i n
st buc
N* po
delamination
h
pat
Nc ng growth path
N N
i
ckl
bu
pre
N N
axial shortening
Figure 10.1: Delamination buckling in a compressed laminate.
F F
Double Cantilever Beam (DCB) F End Loaded Split (ELS)

Mode I Mode II
F2
Asymmetric End Loaded Split (AELS) F Asymmetric End Loaded Split (AELS) F1
Mixed Mode Mixed Mode
Figure 10.2: Beam-type delamination specimens.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Delaminations 275
by dynamic effects, such as vibration and impact. For instance, the dynamics effects
resulting from the inertia of the laminate on the growth process resulting from the
buckling of the delamination has been investigated for a circular delamination and
time-dependent loadings [173].
Delaminations can be analyzed by using cohesive damage models (Section 10.1)
and fracture mechanics (Section 10.2). A cohesive damage model implements in-
terfacial constitutive laws defined in terms of damage variables and a damage
evolution law. Cohesive damage elements are usually inserted between solid ele-
ments [174–177] or beam/shell elements [176].
In the fracture mechanics approach, the propagation of an existing delamination
is analyzed by comparing the amount of energy release rate (ERR) with the fracture
toughness of the interface. When mixed mode conditions are involved, the decom-
position of the total ERR into mode I, mode II, and mode III components becomes
necessary due to the mixed-mode dependency of interface toughness [168, 178]. A
number of fracture mechanics-based models have been proposed in the literature
to study delamination, including three-dimensional models [179–181] and simplified
beam-like models [143, 145, 170, 182, 183].
Fracture mechanics allows us to predict the growth of a pre-existing crack or de-
fect. In a homogeneous and isotropic body subjected to a generic loading condition,
a crack tends to grow by kinking in a direction such that a pure mode I condition
at its tip is maintained. On the contrary, delaminations in laminated composites
are constrained to propagate in its own plane because the toughness of the interface
is relatively low in comparison to that of the adjoining material. Since a delamina-
tion crack propagates with its advancing tip in mixed mode condition, the analysis
requires a fracture criterion including all three mode components (Section 10.1.2).
The elastic strain energy per unit volume (density, in J/m3 ) is definedR as U0 =
1/2σij ij . The strain energy (in J) is defined as the volume integral U = V U0 dV .
The energy required to form, or to propagate, a crack is equal to the elastic energy
released by the solid during crack formation. The energy released is the difference
between the elastic strain energy available before and after the crack is formed, i.e.,
−∆U = Uaf ter − Ubef ore . The rate of energy released per unit of crack area A is
given, in J/m2 by
∆U
G=− (10.1)
∆A
where A is one half the surface area created. The theory of crack growth may be
developed by using one of two approaches due to Griffith and Irwin, respectively.
The Griffith energy approach uses the concept of Energy Release Rate G as the
(computable) energy available for fracture on one hand, and the material property
Gc , which is the energy necessary for fracture, on the other hand. A crack grows
when
G ≥ Gc (10.2)
where for completeness note that Gc = 2γc , where γc is the critical fracture energy
per unit surface crack area, and A is one half of the crack area formed, i.e., the area
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
d
ds
Elastic zone Stress transfer Stress-free

(D = 0) zone (0 < D < 1) zone (D = 1)
Figure 10.3: Cohesive zone model to simulate crack propagation.
of only one of the faces of the crack.

The Irwin (local) approach is based on the concept of stress intensity factor,
which represents the stress field in the neighborhood of the crack tip. These two
approaches are equivalent and, therefore, the energy criterion may be rewritten in
terms of stress intensity factors. Further, a number of path independent integrals
have been proposed to calculate the ERR, such as the J -integral [184].
The elastic strain energy released ∆U during crack propagation, and therefore
used to create the new surface area, can be calculated as the work required to close
the crack, i.e.,
∆U = Wclosure (10.3)
The crack closure method of computation provides the basis for the Virtual
Crack Closure Technique (VCCT) described in Section 10.2.
10.1 Cohesive Zone Method

The cohesive zone method (CZM) is based on the assumption that the stress transfer
capacity between the two separating faces of a delamination is not lost completely at
damage initiation, but rather is a progressive event governed by progressive stiffness
reduction of the interface between the two separating faces (Figure 10.3).
The interface between the two possible separating faces of the laminated material
is modeled with cohesive material behavior, i.e., cohesive zone method (CZM). There
are two types of elements capable of using a CZM material model in ANSYS R
:
Interface elements are based on a traction-separation (σ − δ) constitutive behav-

ior [13, Section 4.12.1]. They can occupy a finite thickness between the two
surfaces that they join or they can have zero thickness. In any case, the con-
tact between the two surfaces is known before the analysis and the surfaces
are thus joined by the interface elements.
Contact elements have zero thickness and can detect contact, separation, pen-
etration, and slip between a contact surface and a target surface [13, Sec-
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Delaminations 277
Figure 10.4: Stress transfer model for cohesive zone model.
tion 4.12.2], [13, Section 14.170.1]. The CZM model can be used only with
bonded contact [13, Section 14.174.7].
The cohesive behavior is described in terms of a traction-separation equation

(Figure 10.4). As the name implies, this approach replaces the engineering stress-
strain (σ − ) equation with a traction-separation (σ − δ) equation. The thickness
of the element is set to zero by defining coincident opposite nodes of the cohesive
element. However, even if the opposite nodes are initially coincident, they are
still separate entities, and they separate during the deformation of the laminated
composite. The separate faces of the laminated plies can be thought of as being
connected to each other through the stiffness of the cohesive element. During the
deformation, the resulting separation between the connected faces of the plies are
proportional to the stiffness of the cohesive element.
Both, the interface element with finite thickness (Example 10.1) and the contact
element with zero thickness (Example 10.2), can use the CZM traction-separation
constitutive equation. The element stiffness matrix requires the stiffness K e of the
interface material, also called penalty stiffness, but the element stiffness matrix is
not formulated as usual by integration over the volume of the element because the
initial volume of the element is zero. The CZM can be visualized as a spring between
the initially coincident nodes of the element. However, the stiffness of the element is
part of the structural stiffness, and the element will undergo deformation during the
loading of the laminate. The initially coincident nodes will open (mode I: opening)
or slide (model II: shear and III: tearing) relative to each other (Figure 10.5). The
nodal separation between the elements are always known by solving the discretized
structure.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
y y y
x x x
(a) (b) (c)

Figure 10.5: Crack propagation modes: (a) I-opening, (b) II-shear, (c) III: tearing.
If the initial thickness of the element is zero, the deformation state of the element
can not be described by the classical definition of strain. Instead, the measure of
the deformation becomes the separation δ between the faces connected through the
element, and this makes possible the use of the (σ − δ) traction-separation equation
instead of the classical engineering (σ − ) equation.
10.1.1 Single Mode Cohesive Model

The CZM is formulated assuming the three crack propagation modes are uncoupled,
even if multiple modes are active simultaneously, as described in Section 10.1.2. In
this section we consider the case of a single mode deformation at an interface of
the laminated material, either mode I, II, or III (Figure 10.5). The formulation is
similar for any of the three modes. The surface tractions at the interface are σi with
i = I, II, III denoting the three modes of crack propagation. The corresponding
separation between the opposite faces of the CZM element are denoted as δi , each
related to the surface traction σi through the interface stiffness Ki , also called
penalty stiffness. Therefore, material stiffness values are necessary for each of the
loading modes, i.e., KI , KII , KIII . A discussion on how to choose numerical values
for Ki is provided in [185].
The behavior of the material at the interface is assumed to be linear-elastic up
to the onset of damage (OA in Figure 10.4) and damaging-elastic after that (OB in
Figure 10.4). Consequently, the stress-separation is described by
σi = Ki δi (10.4)
and
Ki = (1 − Di ) K
fi (10.5)
where Di are the damage variables, and K fi in [N/mm3 ] are the stiffness values
of the undamaged material at the interface, relating the stresses σi to the relative
separation δi between the opposite faces of the element. The stiffness values K fi
are additional material properties needed for CZM, which are different from the
Young’s modulus E for normal deformation, or G for shear deformation.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Delaminations 279
For each mode, there is a critical level of stress σi0 and separation δi0 when the
damage at the interface starts (point A in Figure 10.4), called damage onset. At
damage onset, the two laminas do not separate completely into a physical crack
but rather the interface material starts losing its stiffness. Here, σi0 represents the
strength of the interface with one value of strength for each crack propagation mode:
σI0 , σII
0 , σ 0 . In CZM, these modes are called damage modes because the CZM does
III
not use fracture mechanics. Instead, CZM replaces the fracture mechanics problem
by a continuum damage mechanics problem (Chapter 8). The interface strengths
are therefore additional material parameters required by CZM.
Therefore, the damage initiation criteria are
σi = σi0 (10.6)
and the separations at damage onset are calculated as
δi0 = σi0 /K
fi (10.7)
After damage onset, the interface material starts losing its stiffness (OB in
Figure 10.4), according to (10.5). Note that there are as many damage variables
as damage modes: DI , DII , DIII , which are state variables (8.61) to be determined
during the analysis. Their physical interpretation is given by (10.5) as measures
of stiffness degradation (see also Section 8.2). The damage variables satisfy the
following conditions:
– D = 0 up to damage onset (OA in Figure 10.4) while the interface material is

undamaged, thus retaining its initial stiffness.
– 0 < D < 1 during degradation of the interface material (AF in Figure 10.4),
when the material is gradually loosing its stiffness.
– D = 1 at fracture (point C in Figure 10.4), when there is no remaining stiffness

for the interface material, which means no stress transfer capacity is provided
by the interface. This corresponds to the fracture of the cohesive connection
between the two faces of the CZM element.
With reference to Figure 10.4 we have:
0, δi ≤ δi0

Di = (10.8)
1, δi = δic
In stress-strain space, the cohesive behavior depicted in Figure 10.4 displays

stress softening. That is, after damage onset the stresses σi in the damaged interface
have lower values than the peak values σi0 .
The CZM uses the classical assumption of elastic damage typical of continuum
damage mechanics (CDM, Chapter 8). Therefore, unloading from any point such
as B on the line AF in Figure 10.4 will return to the origin without permanent
deformation upon unloading.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Eventually, there will be total fracture of the cohesive bond (point C in Fig-
ure 10.4) when the stiffness of the interface reduces to zero. Due to the fact that
total loss of stiffness, and thus total loss of cohesive stress transfer, does not take
place until point C, the CZM is correlated with the Griffith crack propagation cri-
terion (10.2) at point C. The correlation is made by considering that the area under
the (σ −δ) curve in Figure 10.4 is equal to the critical ERR Gic in Griffith’s principle
(10.2). In this way, the separation at fracture δic can be calculated as
2Gic
δic = (10.9)
σi0
Since there is a critical ERR value for each damage mode, Gic , with i =
I, II, III, there will be three separations at fracture–one per mode. The three
ERR values are material properties required by the CZM, in addition to the three
values of strength σi0 and the three values of interface stiffness K
fi . From the point
of view of the amount of experimental data needed to perform an analysis, this is
a disadvantage of CZM because it requires nine experimental values. On the other
hand, discrete damage mechanics requires only the three values of ERR to predict
both the onset and the evolution of damage (Chapter 9).
Substituting (10.4) into (10.5) and rearranging results in
σi δi0
Di = 1 − (10.10)
σi0 δi
By similarity of triangles BB0 C and AA0 C, we have
σi δic − δi
= (10.11)
σi0 δic − δi0
which substituted into (10.10) yields
δic (δi − δi0 )

Di = (10.12)
δi (δic − δi0 )
In this way, the damage variables Di are calculated as a function of the relative
separation between the faces of the laminate δi , which are provided by the finite
element solution and the values δi0 , δic calculated previously.
In summary, there are four distinct stages in the material behavior described by
Figure 10.4:
– linear elastic undamaged material behavior (line OA), with the associated
constitutive equation (10.4).
– damage initiation (point A), with the associated criterion (10.6).
– damage evolution (line AF), with the associated damage evolution equation
(10.5), (10.12).
– fracture (crack formation), with the associated crack formation criterion (10.2).
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Delaminations 281
As previously stated, the formulation presented in this section works only for
pure mode I, or II, or III. The general case of the mixed mode loading is presented
in the following section.
10.1.2 Mixed Mode Cohesive Model

When the interface of the laminated material is under mixed modes, all three trac-
tion components σI , σII , σIII and all three separation components δI , δII , δIII are
active. In other words, mixed mode implies that two or more pairs (σi , δi ) are non-
vanishing, with i = I, II, III. However, to reduce the burden of experimentation,
it is assumed1 that the penalty stiffness (10.7) is the same for all modes Ke =K ei.
Mixed mode ratios can be defined between pairs of mode components. For
example, in terms of separations
δII δIII
βδII = ; βδIII = (10.13)
δI δI
or in terms of ERRs
GII GIII
βGII = P3 ; βGIII = P3 (10.14)
1 Gi 1 Gi
Regardless of the definition used, mixed mode ratios are just parameters char-
acterizing the mixed mode state, which allow for a simplification of the analysis by
assuming that decohesion progresses at constant mixed mode ratios. It is further as-
sumed that the modes are uncoupled even though they occur simultaneously. That
is, the stress-separation relationship for each uncoupled mode is again expressed by
(10.4), separately for each mode.
Next, a mixed mode separation is defined by the L2 norm of the mode separa-
tions, i.e., v
uM
uX
δ =t m δ2 i (10.15)
i=1
where M is the number of modes involved (i.e., 2 or 3 modes). Next, the single
mode damage initiation criterion in (10.6) is replaced, for example, by a quadratic
stress criterion
M
σi 2
X
=1 (10.16)
i=1
σi0
For the case of mode I and II only, M = 2 and the equivalent mixed mode
separation at damage onset δm 0 is found as follows. First, rewrite the damage initi-
ation criteria (10.16) in terms of separations using (10.7) and (10.4)–(10.5), taking
into account that Di = 0 up to damage initiation. Therefore, mixed mode damage
initiation is predicted by
1
As far as we know, there is no experimental evidence to support this simplification.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
2 2
δI δII
+ =1 (10.17)
δI0 0
δII
Next, rewrite (10.15) using the first of (10.13) to get
δm
δI = p (10.18)
1 + β2
and using (10.13) again
δm
δII = β p (10.19)
1 + β2
Now, substitute (10.18) and (10.19) into (10.17), taking into account that since
(10.17) represents damage initiation, one should write δm 0 for δ . Therefore,
m
s
0 1 + β2
δm 0 )2
= (δI0 )2 (δII 0 )2 + β 2 (δ 0 )2 (10.20)
(δII I
The quantities δi0 in (10.20) represent the separations at damage onset during
single mode loading, calculated with (10.7) and β is the mixed mode ratio, which
is assumed to be constant during the damage process.
A mixed mode crack propagation criterion is now needed to replace the single
mode criterion (10.2). A possible choice is to use an ERR power criterion as follows
3
G i αi
X
=1 (10.21)
Gic
i=1
which attempts to predict fracture under mixed mode conditions, similarly to point
C in Figure 10.4 for the single mode situation. To reduce the burden of experimen-
tation, it is customary to assume that the exponents are the same for all modes,
i.e., αi = α.
Each of the single mode component Gi can be calculated by one of two method-
ologies. Some authors [174, 176] calculate each single mode component ERR by
considering the area OABB’ in Figure 10.4, thus including the recoverable energy
OBB’ in the definition of Gi . This approach is indirectly related to linear elastic
fracture mechanics (LEFM). Other authors [185] use a damage mechanics approach
where each single mode component ERR is calculated by considering only the un-
recoverable energy represented by the area OAB. Both approaches lead to the same
results for a single mode delamination since loss of adhesion occurs at δic where
both approaches predict the same values for Gi (point B reaches point C). How-
ever, different results are obtained for mixed mode delamination because the crack
propagates when an interaction criterion is satisfied, the later involving the ratios
Gi /Gic .
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Delaminations 283
In principle, the approach based on LEFM should lead to conservative predic-

tions of the load-carrying capacity for mixed-mode delamination. On the other
hand, since the total energy dissipated during the delamination at each point is
not released instantaneously as assumed in LEFM, the damage mechanics based
definition appears appropriate especially when the size of nonlinear fracture process
zone ahead the delamination front is not negligible as it may occur in the case of
laminated composite materials where the damage zone may be comparable to or
larger than the single-ply thickness, which generally scales with the near-tip stress
field.
In the sequel, the damage mechanics approach is used. That is, for each single
mode component separation δi , the single mode component ERR Gi is calculated
as the area OAB in Figure 10.4, which represents dissipated energy, i.e.,
AOAB = AOAF − AOBF (10.22)
where AOAF is the single mode critical ERR Gic , and AOBF can be calculated based
on the geometry in Figure 10.4 as
1 1 δ c − δi c
AOBF = BB 0 × OF = Ki δi0 ci δ (10.23)
2 2 δi − δi0 i
where δic = OF , ki δi0 = σi0 , and σi /σi0 is given by (10.11). Based on (10.22), (10.23),
the single mode component ERR at the moment of mixed mode fracture is calculated
as
1 δ c − δi
Gi = Gic − Ki δi0 δic ci (10.24)
2 δi − δi0
The assumed mode decomposition (10.24) is necessary so that each single mode
component ERR Gi corresponding to mixed mode fracture can be expressed as a
function of the single mode separation δi , i.e., Gi = Gi (δi ). All other quantities in
(10.24) are known, as follows:
– Gic is the single mode critical ERR (material parameter).
– δi0 is the separation at damage onset under single mode loading (point A in
Figure 10.4), given by (10.7)
– δic is the separation at fracture under single mode loading (point C in Fig-
ure 10.4), given by (10.9).
The single mode ERR components of the mixed mode condition in (10.24) has to
satisfy the energy criterion (10.21) at the moment of fracture (crack propagation).
For the case of two modes, and assuming αi = α = 2 in (10.21), the mixed mode
separation at fracture (point C) is calculated in [185, (15)] as
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
√
c 1+β
δm = ×
β 2 (δI0F )2 + (δII
0F )2
0 0F 2 0
δI (δII ) + βδII (δI0F )2
(10.25)
q
+ δI0F δII
0F 0F 2 0 2 0 0 2 0 2 2 0F
(δII ) − (δII ) + 2βδI δII − β (δI ) + β (δI )2
where δI0F = δIc − δI0 , δII

0F = δ c − δ 0 .
II II
Once the mixed mode separation at fracture (10.25) is calculated based on the
selected criterion, and the mixed mode separation at damage onset is known based
on (10.20), the damage variable for mixed mode conditions can be expressed in a
similar manner as for the case of the single mode condition (10.12) by satisfying the
requirement in (10.8), i.e.,
0 )
δ c (δm − δm
Dm = m c 0 )
(10.26)
δm (δm − δm
0 and δ c are calculated based on (10.20)
where the onset and fracture separations δm m
and (10.25) respectively; and δm is the current level of separation under mixed
mode conditions obtained using (10.15) in terms of the single modes separations δi
provided by the FE model. The stiffness degradation for the cohesive material is
then calculated according with (10.5). The stress softening evolution of the cohesive
material under mixed mode conditions is similar to the one depicted in Figure 10.4,
0 , δ c for δ 0 , δ c .
by substituting δm m
Example 10.1 A laminated double cantilever beam (DCB) 100 mm long and 20 mm wide
is made up of two laminas bonded by adhesive lamina of negligible thickness. Each lamina
is 1.5 mm thick. Apply a loading system to induce delamination and mode I crack growth
through the adhesive lamina. Assuming linear elastic behavior, create a 2D model of the
DCB using interface elements INTER202 to represent the adhesive lamina. Use the PLANE182
for the two laminas. The lamina material properties are: E1 = 135.3 GP a, E2 = E3 =
9 GP a, ν12 = ν13 = 0.24, ν23 = 0.46, G12 = G23 = 4.5 GP a; G13 = 3.3 GP a. The adhesive
properties are: σI0 = σII
0
= 25.0 M P a and GIc = GIIc = 280 J/m2 .
Solution to Example 10.1 Using (10.9) to calculate the separation at fracture (Figure
10.4), with a bilinear (BILI) model [186, Section 3.22.2], [13, Section 4.12.1.2], we have
2 × 280N m/m2
δIc = δII
c
= = 0.0224 mm
25 × 106 N/M 2
The APDL file, which is available in [5, Ex101bili.log], is as follows
/TITLE,Example 10.1: DCB specimen under mode I loading

/UNITS,MPa ! Units are in mm, N and MPa
/PREP7
ET,1,PLANE182 !DEFINE PLANE STRESS/STRAIN ELEMENT - 2D
KEYOPT,1,1,2 !ENHANCE STRAIN FORMULATION
KEYOPT,1,3,2 !PLANE STRAIN
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Delaminations 285
ET,2,PLANE182
KEYOPT,2,1,2
KEYOPT,2,3,2
ET,3,INTER202 !DEFINE INTERFACE ELEMENT - 2D
MP,EX,1,135.3E3 !MATERIAL PROPERTIES LAMINA

MP,EY,1,9.0E3
MP,EZ,1,9.0E3
MP,GXY,1,4.5E3
MP,GYZ,1,4.5E3
MP,GXZ,1,3.3E3
MP,PRXY,1,0.24
MP,PRXZ,1,0.24
MP,PRYZ,1,0.46
TB,CZM,2,,,BILI !MATERIAL PROPERTIES ADHESIVE

TBDATA,1,25.0,0.0224,-25.0,0.0224,1.0,1.0
RECTNG,0,100,0,1.5 !DEFINE AREA 1.5x100 mm

RECTNG,0,100,0,-1.5 !DEFINE AREA 1.5x100 mm
LSEL,S,LINE,,2,8,2
LESIZE,ALL, , ,2 !NUMBER ELEMENTS VERTICAL
LSEL,INVE
LESIZE,ALL, , ,400 !NUMBER ELEMENTS HORIZONTAL
ALLSEL,ALL
TYPE,1
MAT,1
LOCAL,11,0,0,0,0
ESYS,11
AMESH,2 !MESH
CSYS,0
TYPE,2
ESYS,11
AMESH,1 !MESH
CSYS,0
NSEL,S,LOC,X,30,100
NUMMRG,NODES
ESLN
TYPE,3
MAT,2
CZMESH,,,1,Y,0, !MESH INTERFACE ELEMENTS
ALLSEL,ALL
NSEL,S,LOC,X,100 !CONSTRAINTS
D,ALL,ALL
NSEL,ALL
FINISH
/SOLU
ESEL,S,TYPE,,2
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
NSLE,S
NSEL,R,LOC,X
NSEL,R,LOC,Y,1.5 !DISPLACEMENT ON TOP
D,ALL,UY,6
NSEL,ALL
ESEL,ALL
ESEL,S,TYPE,,1
NSLE,S
NSEL,R,LOC,X
NSEL,R,LOC,Y,-1.5 !DISPLACEMENT ON BOTTOM
D,ALL,UY,-6
NSEL,ALL
ESEL,ALL
NLGEOM,ON
AUTOTS,ON
TIME,1
NSUBST,500,500,500 !LOAD STEPS
OUTRES,ALL,ALL
SOLVE
FINISH
/POST26
NSEL,S,LOC,Y,1.5
NSEL,R,LOC,X,0
*GET,NTOP,NODE,0,NUM,MAX
NSEL,ALL
NSOL,2,NTOP,U,Y,UY
RFORCE,3,NTOP,F,Y,FY
PROD,4,3, , ,RF, , ,20 !width=20
/AXLAB,X,DISPLACEMENT [mm]
/AXLAB,Y,REACTION FORCE [N]
/XRANGE,0,6
/YRANGE,0,65
XVAR,2
PLVAR,4 !PLOT FORCE vs DISPLACEMENT
PRVAR,UY,RF !LIST FORCE & DISPLACEMENT vs TIME
FINISH
/POST1
PLDISP !PLOT DEFORMED PLATE
FINISH
The deformed shape is illustrated in Figure 10.6. Instead of bilinear material model
BILI, an exponential material model EXPO could be used [5, Ex101expo.log]. The required
data [186, Section 3.22.1] includes the maximum normal and shear separations (point C in
Figure 10.4). Taking into account that φn = J in [177], and J = G for a linear elastic
material [187], and using [13, (4-368),(4-369)], we have
GIc GIc
δnc = = 0.004 mm ; δtc = √ = 0.005 mm
e σI0 2e σI0
where e = 2.7182818 is the base of the natural logarithms.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Delaminations 287
Figure 10.6: Deformed DCB using interface elements (Example 10.1).
The APDL code to define the exponential material model is
TB,CZM,2,,,EXPO !MATERIAL PROPERTIES ADHESIVE

TBDATA,1,25.0,0.004,0.005
The resulting force-separation plot should look like the dash line in Figure 10.7. The
drop of the reaction force occurs when the cohesive elements start to degrade (damage). As
the cohesive elements degrade, the reaction force reduces.
Example 10.2 Solve Example 10.1 using contact elements TARGE169 and CONTA171.
Solution to Example 10.2 Using a cohesive zone material model in terms of critical frac-
ture energies [186, CBDE, Section 3.22.3], σI0 = 25.0 M P a and GIc = 0.28 N mm. The
artificial damping was chosen as η = 10−4 to obtain a smooth force-separation plot. The
APDL file, which is available in [5, EX102.inp], is as follows
/TITLE,Example 10.2: DCB specimen under mode I loading

/UNITS,MPa ! Units are in mm, N, and MPa
/PREP7
ET,1,PLANE182 !DEFINE PLANE182 ELEMENT - 2D
ET,3,TARGE169 !DEFINE TARGE169 ELEMENT - 2D
ET,4,CONTA171 !DEFINE CONTA171 ELEMENT - 2D
KEYOPT,4,12,5 !BONDED (ALWAYS)

MP,EY,1,9.0E3
MP,EZ,1,9.0E3
MP,GXY,1,4.5E3
MP,GYZ,1,4.5E3
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
70
60
Reaction Force [N] 50
40
Interface
30 Contact
VCCT
20
10
0
0 2 4 6 8
Displacement [mm]
Figure 10.7: Force-separation plot using interface elements (Example 10.1), contact
elements (Example 10.2), and VCCT (Example 10.3).
MP,GXZ,1,3.3E3
MP,PRXY,1,0.24
MP,PRXZ,1,0.24
MP,PRYZ,1,0.46
TB,CZM,2,,,CBDE !MATERIAL PROPERTIES ADHESIVE

TBDATA,1,25.0,0.28,,,1.0E-4

LSEL,S,LINE,,2,8,2
LSEL,INVE
ALLSEL,ALL
TYPE,1 !MESH
MAT,1
LOCAL,11,0,0,0,0
ESYS,11
AMESH,2
CSYS,0
TYPE,2 !MESH
ESYS,11
AMESH,1
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Delaminations 289
CSYS,0
NSEL,S,LOC,X,30,100
TYPE,3
MAT,2
ESEL,S,TYPE,,2
NSLE,S
NSEL,R,LOC,Y
ESURF !TARGET ELEMENTS
TYPE,4
ESEL,S,TYPE,,1
NSLE,S
NSEL,R,LOC,Y
NSEL,R,LOC,X,30,100
ESURF !CONTACT ELEMENTS
ALLSEL,ALL
D,ALL,ALL
NSEL,ALL
FINISH
/SOLU
ESEL,S,TYPE,,2
NSLE,S
NSEL,R,LOC,X
D,ALL,UY,6
NSEL,ALL
ESEL,ALL
ESEL,S,TYPE,,1
NSLE,S
NSEL,R,LOC,X
D,ALL,UY,-6
NSEL,ALL
ESEL,ALL
NLGEOM,ON
TIME,1
NSUBST,500,500,500 !LOAD STEPS

OUTRES,ALL,ALL
SOLVE
FINISH
/POST26
NSEL,S,LOC,Y,1.5
NSEL,R,LOC,X,0
NSEL,ALL
NSOL,2,NTOP,U,Y,UY
PROD,4,3, , ,RF, , ,20 !width=20
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure 10.8: Deformed DCB using contact elements (Example 10.2).
/XRANGE,0,6
/YRANGE,0,65
XVAR,2
FINISH
/POST1
PLDISP !PLOTS DEFORMED PLATE
FINISH
The deformed shape is illustrated in Figure 10.8. The resulting force-separation plot
should look like the solid line in Figure 10.7.
10.2 Virtual Crack Closure Technique

The virtual crack closure technique (VCCT) can be used analyze delaminations in
laminated materials using a fracture mechanics approach. The method implements
Linear Elastic Fracture Mechanics (LEFM). Only brittle crack propagation is mod-
eled. The energy dissipated by the formation of plastic zones at the crack tip is not
considered.
The condition for crack propagation is based on the Griffith principle (10.2).
For the case of single mode deformation under mode I conditions, the crack grows
when
GI
≥1 (10.27)
GIc
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Delaminations 291
1
Load
F25
3 2
4 Crack 5
w16
propagation
F25
Load
d 6
Figure 10.9: Virtual crack closure technique (VCCT).
where GI is the ERR for mode I crack formation and GIc a material property
representing the critical ERR for mode I crack formation.
The definition of ERR is given by (10.1). In VCCT, the Irwin principle (10.3)
is used to calculate the change in strain energy ∆U , which is considered to be equal
to the work required for crack closure Wclosure .
By substituting (10.3) and (10.1) into (10.27), the condition for crack propaga-
tion under mode I loading becomes
Wclosure /∆A
≥1 (10.28)
GIc
The crack closure work Wclosure is calculated from the FE nodal separations and
forces as illustrated in Figure 10.9. Initially the crack surfaces are rigidly bonded.
The nodal forces at the coincident nodes 2 − 5 are calculated from the FE solution.
The hypothesis of self-similar crack propagation is used, which says that during
crack propagation, the crack configuration between nodes 2 − 3 − 4 − 5 will be
similar to the crack configuration between nodes 1 − 2 − 5 − 6. This implies that
the separation between nodes 2 − 5 after crack propagation will be equal to the
separation between nodes 1 − 6 before crack propagation: v2,5 = v1,6 . If the nodes
2 − 5 open (crack propagation), the elastic work required to close the crack is
1 1
Wclosure = F2,5 v2,5 = F2,5 v1,6 (10.29)
2 2
By substituting (10.29) in (10.28), the condition for crack propagation becomes
F2,5 v1,6 1
≥1 (10.30)
2∆A GIc
The area ∆A of the newly formed crack is ∆A = d × b, where d is the length of
the finite element undergoing crack propagation, and b is the width of the crack.
The VCCT method works similarly for the modes II or III, by considering the
corresponding components of the separations and nodal forces. A refined formula-
tion called the Jacobian Derivative Method (JDM) is available in [188].
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
The crack propagation criterion (10.30) applies for single mode loading only,
as it is derived from the single mode criterion in (10.2). For mixed mode loading,
the single mode crack propagation criterion (10.2) has to be replaced with a mixed
mode criterion. For example, one could use the power equation (10.21), where the
critical ERRs Gic , with i = I, II, III, are material properties and the ERRs Gi
are calculated similarly to (10.30) by using VCCT. Alternatively, one could use the
Benzeggagh-Kenane (BK) equation [189] or the Reeder equation [178, 190].
Example 10.3 Solve Example 10.1 using the virtual crack closure technique (VCCT).
Solution to Example 10.3 In this example, a VCCT model [8, Section 12.1.1] is imple-
mented using INTER202 elements. The fracture energy of the adhesive is GIc = 0.28 N mm.
The APDL file, which is available in [5, EX103.inp], is as follows
/TITLE,Example 10.3: DCB specimen under mode I loading using VCCT

/UNITS,MPa ! Units: [mm, N, MPa]
/PREP7
KEYOPT,1,1,2 !ENHANCE STRAIN FORMULATION
ET,2,PLANE182
KEYOPT,2,1,2
KEYOPT,2,3,2
ET,3,INTER202 !DEFINE INTER202 ELEMENT - 2D

MP,EY,1,9.0E3
MP,EZ,1,9.0E3
MP,GXY,1,4.5E3
MP,GYZ,1,4.5E3
MP,GXZ,1,3.3E3
MP,PRXY,1,0.24
MP,PRXZ,1,0.24
MP,PRYZ,1,0.46
TB,CGCR,1,,3,LINEAR !MATERIAL PROPERTIES ADHESIVE

TBDATA,1,0.28,0.28,0.28

LSEL,S,LINE,,2,8,2
LSEL,INVE
ALLSEL,ALL
TYPE,1
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Delaminations 293
MAT,1
LOCAL,11,0,0,0,0
ESYS,11
AMESH,2 !MESH
CSYS,0
TYPE,2
ESYS,11
AMESH,1 !MESH
CSYS,0
NSEL,S,LOC,X,30,100
NUMMRG,NODES
ESLN
TYPE,3
MAT,1
CZMESH,,,1,Y,0, !MESH INTERFACE ELEMENTS
ALLSEL,ALL
D,ALL,ALL
NSEL,ALL
ESEL,S,ENAME,,INTER202 !SELECT INTERFACE ELEMENTS

CM,CPATH,ELEM !DEFINE CRACK GROWTH PATH
NSEL,ALL
NSEL,S,LOC,X,30
NSEL,R,LOC,Y,0
CM,CRACK1,NODE
ALLS
FINISH
/SOLU
RESC,,NONE
ESEL,S,TYPE,,2
NSLE,S
NSEL,R,LOC,X
D,ALL,UY,0.9
NSEL,ALL
ESEL,ALL
ESEL,S,TYPE,,1
NSLE,S
NSEL,R,LOC,X
D,ALL,UY,-0.9
NSEL,ALL
ESEL,ALL
AUTOTS,ON
TIME,1
CINT,NEW,1 !DEFINE CRACK ID
CINT,TYPE,VCCT
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
CINT,CTNC,CRACK1 !DEFINE CRACK TIP NODE COMPONENT

CINT,SYMM,OFF !SYMMETRY OFF
CINT,NORM,0,2 !DEFINE CRACK PLANE NORMAL
CGROW,NEW,1 !DEFINE CRACK-GROWTH INFORMATION

CGROW,CID,1
CGROW,FCOPTION,MTAB,1 !DEFINE FRACTURE CRITERION FOR CRACK-GROWTH
CGROW,CPATH,CPATH
CGROW,DTIME,1.0e-4
CGROW,DTMIN,1.0e-4
CGROW,DTMAX,1.0e-4
NSUB,4,4,4
ALLSEL,ALL
OUTRES,ALL,ALL
SOLVE
TIME,2
ESEL,S,TYPE,,2
NSLE,S
NSEL,R,LOC,X
D,ALL,UY,6
NSEL,ALL
ESEL,ALL
ESEL,S,TYPE,,1
NSLE,S
NSEL,R,LOC,X
D,ALL,UY,-6
NSEL,ALL
ESEL,ALL
NSUBST,500,500,500
OUTRES,ALL,ALL
SOLVE
FINISH
/POST26
NSEL,S,LOC,Y,1.5
NSEL,R,LOC,X,0
NSEL,ALL
NSOL,2,NTOP,U,Y,UY
PROD,4,3, , ,RF, , ,20 !width=20
/XRANGE,0,6
/YRANGE,0,65
XVAR,2
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Delaminations 295
Figure 10.10: Deformed DCB using VCCT (Example 10.3).
FINISH
/POST1
PLDISP !PLOT DEFORMED PLATE
FINISH
The deformed shape is illustrated in Figure 10.10. The resulting force-separation plot
should look like the dotted line in Figure 10.7.
Suggested Problems
Problem 10.1 Retrieve the values of reaction force vs. separation for Examples 10.1 and
10.2. Plot them together and discuss the difference. Your plot should be similar to the one
Problem 10.2 Retrieve the values of reaction force vs. separation for Example 10.3 and
plot them along those of Examples 10.1 and 10.2. Discuss the differences. The comparative
plot should be similar to the one shown in Figure 10.7.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Appendix A
Tensor Algebra
Tensor operations are needed for the derivation of some of the equations in this
textbook. Since most of these operations are not easily found in textbooks, they
are presented here for reference [191].
A.1 Principal Directions of Stress and Strain

Since stress and strain tensors are symmetric and of second order, they have three
real principal values and three orthogonal principal directions. The principal values
λq and directions nqi of the stress tensor σij satisfy the following
[σij − λq δij ]nqi = 0 (A.1)

nqi nqj =1 (A.2)
where δij is the Kronecker delta (δij = 1 if i = j, zero otherwise). Each of the
principal directions is described by its direction cosines with respect to the original
coordinate system.
The principal directions are arranged by rows into a matrix [A]. Then, the
diagonal matrix [A∗ ] of the principal values is
[A∗ ] = [a][A][a]T (A.3)
It can be shown that [a]−1 = [a]T , where [a] is the transformation matrix given
by (1.21)
A.2 Tensor Symmetry

Minor symmetry provides justification for using contracted notation (Section 1.5).
Minor symmetry refers to identical values of tensor components when adjacent
subscripts are swapped. For example, minor symmetry of the stiffness tensor C
means
Aijkl = Ajilk = Aαβ (A.4)
297
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Major symmetry refers to identical values when adjacent pairs of subscripts are
swapped, or when contracted subscripts are swapped. For example,
Aijkl = Aklij
Aαβ = Aβα (A.5)
A.3 Matrix Representation of a Tensor

A tensor Aijkl with a minor symmetry has only 36 independent constants. There-
fore, it can be represented in contracted notation by a 6 × 6 matrix. Let [a] be the
contracted form of the tensor A. Each element of [a] corresponds to an element in
the tensor A according to the following transformation
aαβ = Aijkl (A.6)
with
α = i when i = j
α = 9 − (i + j) when i 6= j (A.7)
The same transformations apply between β and k and l, or in matrix represen-

tation, as  
A1111 A1122 A1133 A1123 A1113 A1112
 A2211 A2222 A2233 A2223 A2213 A2212 
 
 A3311 A3322 A3333 A3323 A3313 A3312 
[a] = 
 A2311 A2322
 (A.8)
 A2333 A2323 A2313 A2312 

 A1311 A1322 A1333 A1323 A1313 A1312 
A1211 A1222 A1233 A1223 A1213 A1212
It is convenient to perform tensor operations using the contracted form, es-
pecially if the result can be represented also in the contracted form. This saves
memory and time since it is faster to operate on 36 elements than on 81 elements.
Examples of these operations are the inner product of two fourth-order tensors and
the inverse of a fourth-order tensor. However, tensor operations in index notation
do not translate directly into matrix operations in a contracted form. For example,
the double contraction of two fourth-order tensors is
C = A:B
Cijkl = Aijmn Bmnkl (A.9)
Let [a], [b], and [c] the 6×6 matrix representations of the above tensors. Then,
it can be shown that
[a] [b] 6= [c] or

aαβ bβγ 6= cαγ (matrix multiplication) (A.10)
The rest of this appendix presents formulas for adequate representation of tensor
operations in their contracted form.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Appendix 299
A.4 Double Contraction

In (A.9), an element like C1211 can be expanded as
C1211 = A1211 B1111 + A1222 B2211 + A1233 B3311

+2A1212 B1211 + 2A1213 B1311 + 2A1223 B2311 (A.11)
In order to achieve the same result by matrix multiplication, multiply the last
three columns of the matrix [a] by 2, and then perform the multiplication
 
A1111 A1122 A1133 2A1123 2A1113 2A1112

 A2211 A2222 A2233 2A2223 2A2213 2A2212 

 A3311 A3322 A3333 2A3323 2A3313 2A3312 
[c] =  

 A2311 A2322 A2333 2A2323 2A2313 2A2312 

 A1311 A1322 A1333 2A1323 2A1313 2A1312 
A1211 A1222 A1233 2A1223 2A1213 2A1212
 
B1111 B1122 B1133 B1123 B1113 B1112

 B2211 B2222 B2233 B2223 B2213 B2212 

 B3311 B3322 B3333 B3323 B3313 B3312 
  (A.12)

 B2311 B2322 B2333 B2323 B2313 B2312 

 B1311 B1322 B1333 B1323 B1313 B1312 
B1211 B1222 B1233 B1223 B1213 B1212
This transformation can be produced by using the Reuter matrix [R]

 
1 0 0 0 0 0

 0 1 0 0 0 0 

 0 0 1 0 0 0 
[R] =   (A.13)

 0 0 0 2 0 0 

 0 0 0 0 2 0 
0 0 0 0 0 2
Substituting in (A.12) we have
[c] = [a] [R] [b] (A.14)
A.5 Tensor Inversion

First, it is convenient to define the fourth-order identity tensor Iijkl as a tensor that
multiplied innerly by another fourth-order tensor yields this same tensor, or
Iijmn Amnkl = Aijkl (A.15)
If Aijkl has a minor symmetry, the following tensor achieves (A.15)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
1
Iijkl =
(δik δjl + δil δjk ) (A.16)
2
where δij is the Kronecker delta, defined as
δij = 1 if i = j
(A.17)
δij = 0 if i 6= j
In Voigt contracted notation, the fourth-order identity tensor is denoted as [i],

which is equal to the inverse of the Reuter matrix
 
1 0 0 0 0 0

 0 1 0 0 0 0 

 0 0 1 0 0 0  = [R]−1

[i] =  (A.18)

 0 0 0 1/2 0 0 

 0 0 0 0 1/2 0 
0 0 0 0 0 1/2
Now, the inverse of a tensor is a tensor that multiplied by the original tensor
yields the identity tensor, as follows:
Aijmn A−1
mnkl = Iijkl (A.19)
Let us introduce the following notation:
[a]−1 = inverse of the contracted form of Aijkl

−1
a = contracted form of the inverse of Aijkl
If Aijkl has a minor symmetry, the components of a−1

αβ are:
i. Multiply the last three columns of [a] by 2 by using the matrix [R]
ii. Invert the obtained matrix.
iii. Multiply the matrix by [i]
In order words, the matrix a−1 is computed as

= [[a] [R]]−1 [i] = [i] [a]−1 [i]

−1
a (A.20)
A.6 Tensor Differentiation

A.6.1 Derivative of a Tensor with Respect to Itself
Any symmetric second-order tensor Φij satisfies the following:
dΦij = dΦji (A.21)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Appendix 301
Therefore, differentiating a second-order symmetric tensor with respect to itself

is accomplished as follows
∂Φij
= Jijkl (A.22)
∂Φkl
where Jijkl is a fourth-order tensor defined as
Jijkl = 1 if i = k, and j = l
Jijkl = 1 if i = l, and j = k (A.23)
Jijkl = 0 otherwise
In contracted notation, the tensor Jijkl is represented by

 
1 0 0 0 0 0

 0 1 0 0 0 0 

 0 0 1 0 0 0 
[j] =   (A.24)

 0 0 0 1 0 0 

 0 0 0 0 1 0 
0 0 0 0 0 1
A.6.2 Derivative of the Inverse of a Tensor with Respect to the

Tensor
A second-order tensor contracted with its inverse yields the second-order identity
tensor, or Kronecker delta
Aij A−1
jk = δik (A.25)
Differentiating (A.25) with respect to Amn and rearranging terms yields
∂A−1
jk ∂Aij −1
Aij =− A (A.26)
∂Amn ∂Amn jk
Pre-multiplying both sides by A−1

li and rearranging yields
∂A−1
ij ∂Akl −1
= −A−1
ik A (A.27)
∂Amn ∂Amn lj
Finally, using (A.22) yields
∂A−1
ij
= −A−1 −1
ik Jklmn Alj (A.28)
∂Amn
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Appendix B
Second-Order Diagonal Damage

Models
Explicit expressions associated to second-order diagonal damage models are pre-

sented here for completeness.
B.1 Effective and Damaged Spaces

A second-order damage tensor can be represented as a diagonal tensor (see (8.61))
Dij = di δij ; no sum on i (B.1)
in a coordinate system coinciding with the principal directions of D, which may

coincide with the fiber, transverse, and thickness directions, and di are the eigenval-
ues of the damage tensor, which represent the damage ratio along these √ directions.
The dual variable of the damage tensor is the integrity tensor, Ω = I − D, which
represents the undamaged ratio.
The second-order damage tensor D and the integrity tensor Ω are diagonal and
have the following explicit forms
 
d1 0 0
Dij =  0 d2 0  (B.2)
0 0 d3
 √   
1 − d1 √ 0 0 Ω1 0 0
Ωij =  0 1 − d2 √ 0  =  0 Ω2 0  (B.3)
0 0 1 − d3 0 0 Ω3
A symmetric fourth-order tensor, M, called the damage effect tensor, is defined

(see (8.63)) as
1
Mijkl = (Ωik Ωjl + Ωil Ωjk ) (B.4)
2
303
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
The damage effect tensor in contracted form multiplied by the Reuter matrix,
takes the form of a 6×6 array as follows
 2 
Ω1 0 0 0 0 0
 0 Ω22 0 0 0 0 
2
 
 0 0 Ω3 0 0 0 
M = Mαβ =    (B.5)
 0 0 0 Ω2 Ω 3 0 0 

 0 0 0 0 Ω 1 Ω3 0 
0 0 0 0 0 Ω 1 Ω2
The damaged stiffness tensor C multiplied by the Reuter matrix can be written
in explicit contracted notation for an orthotropic material by a 6×6 array as a
function of the undamaged stiffness tensor C as follows
C 11 Ω41 C 12 Ω21 Ω22 C 13 Ω21 Ω23

 
0 0 0
 C 12 Ω21 Ω22 C 22 Ω42 C 23 Ω22 Ω23 0 0 0 
 C 13 Ω21 Ω23 C 23 Ω22 Ω23 4
 
C 33 Ω3 0 0 0 
Cαβ = 
2 2

 0 0 0 2C 44 Ω2 Ω3 0 0 

 0 0 0 0 2 2
2C 55 Ω1 Ω3 0 
0 0 0 0 0 2 2
2C 66 Ω1 Ω2
(B.6)
where C 44 = G23 , C 55 = G13 and C 66 = G12 . The Voigt contracted notation for
fourth-order elasticity tensors is used here: Cαβ replaces Cijkl where α, β take the
values 1, 2, 3, 4, 5, 6, corresponding to the index pairs 11, 22, 33, 23, 13 and 12,
respectively.
The relations between the effective and actual stress components assume the
following expressions
σ 1 = σ1 Ω−2
1 ; σ 4 = σ4 Ω−1 −1
2 Ω3 ;
σ 2 = σ2 Ω−2
2 ;
−1 −1
σ 5 = σ5 Ω1 Ω3 ; (B.7)
σ 3 = σ3 Ω−2
3 ; σ 6 = σ6 Ω−1 −1
1 Ω2 ;
and the strain components
ε1 = ε1 Ω21 ; ε4 = ε4 Ω 2 Ω 3 ;
ε2 = ε2 Ω22 ; ε5 = ε5 Ω 1 Ω 3 ; (B.8)
ε3 = ε3 Ω23 ; ε6 = ε6 Ω 1 Ω 2 ;
where the over-line indicates an effective property.
B.2 Thermodynamic Force Y

By satisfying the Clausius-Duhem inequality, thus assuring non-negative dissipation,
the following thermodynamic forces (see (8.128)) are defined
∂ψ 1 ∂Cklpq 1 ∂Cklpq e
= − εkl −εpkl εpq −εppq = − εekl

Yij = − ε (B.9)
∂Dij 2 ∂Dij 2 ∂Dij pq
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Appendix 305
The second-order tensor of the conjugate thermodynamic forces associated to

the damage variables takes the following form
 
Y11 0 0
Y = Yij =  0 Y22 0  (B.10)
0 0 Y33
or in Voigt contracted notation as
Y = Yα = {Y11 , Y22 , Y33 , 0, 0, 0}T (B.11)
Using (B.9), the explicit expressions for the thermodynamic forces written in
terms of effective strain are found as
1
C 11 εe1 2 + C 12 εe2 εe1 + C 13 εe3 εe1 + 2C 55 εe5 2 + 2C 66 εe6 2

Y11 = 2
Ω1
1
= 2 C 22 εe2 2 + C 12 εe2 εe1 + C 23 εe3 εe2 + 2C 44 εe4 2 + 2C 66 εe6 2

Y22 (B.12)
Ω2
1
= 2 C 33 εe3 2 + C 13 εe3 εe1 + C 23 εe3 εe2 + 2C 44 εe4 2 + 2C 55 εe5 2

Y33
Ω3
The thermodynamic forces written in terms of actual stress are

1 S 11 2 S 12 S 13 2S 55 2 2S 66 2
Y11 = 2 σ1 + σ2 σ1 + σ3 σ1 + σ5 + σ6
Ω1 Ω41 Ω21 Ω22 Ω21 Ω23 Ω21 Ω23 Ω21 Ω22
1 S 22 2 S 12 S 23 2S 44 2 2S 66 2
Y22 = 2 σ2 + σ2 σ1 + σ3 σ2 + σ4 + σ6 (B.13)
Ω2 Ω42 Ω22 Ω21 Ω22 Ω23 Ω22 Ω23 Ω22 Ω21
1 S 33 2 S 13 S 23 2S 44 2 2S 55 2
Y33 = 2 σ3 + σ3 σ1 + σ3 σ2 + σ4 + σ5
Ω3 Ω43 Ω23 Ω21 Ω23 Ω22 Ω23 Ω22 Ω23 Ω21
The derivative of the thermodynamic forces with respect to the damage (∂Y/∂D)
is given by

Y11

 Ω4 0 0 0 0 0 
 1
Y22

 
 0 4 0 0 0 0 
 Ω2 
∂Y  Y

= 0

0
33
0 0 0
 (B.14)
∂D  Ω 4

3

 
 0 0 0 0 0 0 
 
 0 0 0 0 0 0 
0 0 0 0 0 0
The derivative of the thermodynamic forces with respect to the actual strain is
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
given by
 
C εe1 C εe1 C 55 εe5 C 66 εe6
− PΩ112 − 12
Ω21
− 13
Ω21
0 −2 Ω21
−2 Ω21
 1 
 
 C 12 εe2 C 23 εe2 C 44 εe4 C 66 εe6
 − Ω2 − PΩ222 − −2 0 −2

Ω22 Ω22 Ω22

 2 2 
 
∂Y  − C εe3 C 23 εe3 C 44 εe4 C 55 εe5
13
− − PΩ332 −2 −2 0
 
= Ω32 Ω23 Ω23 Ω23
 (B.15)
∂εe 
3 


 0 0 0 0 0 0 

 

 0 0 0 0 0 0 

0 0 0 0 0 0
where
P11 = 2 C 11 εe1 + C 12 εe2 + C 13 εe3
P22 = C 12 εe1 + 2 C 22 εe2 + C 23 εe3 (B.16)
P33 = C 13 εe1 + C 23 εe2 + 2 C 33 εe3
The derivative of the thermodynamic forces with respect to the actual unrecov-
erable strain is given by
∂Y ∂Y
p
=− e (B.17)
∂ε ∂ε
The derivative of the actual stress with respect to damage is given by
0
 
P11 0 0 0 0 0
0
 

 0 P 22 0 0 0 0 

0
 
 0 0 P 33 0 0 0 
∂σ  
= (B.18)
 e e

1 Ω3 C 44 ε4 1 Ω2 C 44 ε4
∂D  0 − − 0 0 0

 2 Ω2 2 Ω3 

 − 1 Ω3 C 55 εe5 Ω C εe5
 
 2 Ω1 0 − 12 1 Ω553
0 0 0 

 
Ω C ε6 e e
1 Ω1 C 66 ε6
− 21 2 Ω66
1
− 2 Ω2 0 0 0 0
where
0 = −C εe − C εe − C εe
P11 11 1 12 2 13 3
0 = −C εe − C εe − C εe
P22 (B.19)
12 1 22 2 23 3
0 = −C εe − C εe − C εe
P33 13 1 23 2 33 3
B.3 Damage Surface

An anisotropic damage criterion expressed in tensorial form, introducing two fourth-
order tensors, B and J defines a multiaxial limit surface in the thermodynamic force
space, Y, that bounds the damage domain. The damage evolution is defined by a
damage potential associated to the damage surface and by an isotropic hardening
function. The proposed damage surface g d is given by
1/2
S 1/2
g d = ŶijN Jijhk Ŷhk
N
+ YijS Bijhk Yhk

− (γ(δ) + γ0 ) (B.20)
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Appendix 307
where γ0 is the initial damage threshold value and γ(δ) defines the hardening.
The derivative of the damage surface with respect to thermodynamic forces is
given by  N
J11 Y11 S 
B11 Y11
 ΦN + ΦS 

J Y N B Y S 
 22 22 22 22 
N
+
ΦS S 
 
∂g d  Φ
 J YN
=  33 33 B33 Y33  (B.21)
∂Y + 
 ΦN ΦS
 


 0 

 0 
0
where q
ΦN = N )2 + J (Y N )2 + J (Y N )2
J11 (Y11 22 22 33 33
q (B.22)
S
Φ = S 2 S 2 S )2
B11 (Y11 ) + B22 (Y22 ) + B33 (Y33
The derivative of the damage surface with respect to damage hardening is
∂g d
= −1 (B.23)
∂γ
B.4 Unrecoverable-Strain Surface

The unrecoverable-strain (yield) surface g p is a function of the thermodynamic forces
in the effective configuration (σ, R). Therefore, the unrecoverable-strain surface is
g p = fij σ i σ j + fi σ i − (R (p) + R0 )
p
(B.24)
where (i = 1, 2...6), R0 is the initial unrecoverable-strain threshold and R is the

hardening function.
The derivative of the unrecoverable-strain surface with respect to effective stress
is given by
1 f1 + 2 f11 σ 1 + 2 f12 σ 2 + 2 f13 σ 3
 
 2
 Φp 

 1 f2 + 2 f22 σ 2 + 2 f12 σ 1 + 2 f23 σ 3 
 
 2
 Φp 

 1 f + 2f σ + 2f σ + 2f σ 
 3 33 3 13 1 23 2 
∂gp   2 Φp

= (B.25)

∂σ

 f4 σ 4 
 

 Φp 


 f5 σ 5 

 Φ p 
 
 f σ
6 6

Φp
where
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Φp = (f1 σ 1 + f2 σ 2 + f3 σ 3 +
+f11 σ 1 2 + f22 σ 2 2 + f33 σ 3 2 +
+2 f12 σ 1 σ 2 + 2 f13 σ 1 σ 3 + 2 f23 σ 2 σ 3 + (B.26)
1/2
+f6 σ 6 2 + f5 σ 5 2 + f4 σ 4 2
The derivative of the yield surface with respect to unrecoverable-strain hardening
is
∂g p
= −1 (B.27)
∂R
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Appendix C
Software Used
Only four software applications are used throughout this textbook. By far the most
used is ANSYS R
Mechanical APDL (referred simply as ANSYS in this textbook).
BMI3 is used only in Chapter 4. MATLAB

c R
is used for symbolic as well as
numerical computations. Finally, Intel Fortran must be available to compile and link
ANSYS with user programmed material subroutines, but its usage is transparent to
the user because it is called by a batch file requiring no user intervention. Of course,
some knowledge of Fortran is required to program new user material subroutines,
but programming is made easier by several example subroutines, which are provided
and used in the examples.
The aim of this section is to present an introduction to the software used in
this textbook, namely ANSYS and BMI3, as well as how to use Intel Fortran to
compile and link user subroutines with ANSYS. It is assumed that the reader can use
MATLAB without help besides that provided by the self-explanatory code included
with the examples, either printed in this textbook or downloadable from the Web
site [5].
Operation of the software is illustrated for a Windows 7 platform but opera-
tion in a Linux environment is very similar. For the sake of space, this section is
very brief. The vendors of these applications have a wealth of information, training
sessions, user groups, and so on, that the reader can use to get familiar with the
software interface. One such source of information is the Web site for this text-
book at http://barbero.cadec-online.com/feacm-ansys/. Another source of
information is the book’s user group at http://tech.groups.yahoo.com/group/
feacomposites/.
C.1 ANSYS Mechanical APDL

ANSYS Mechanical APDL is a commercial finite element analysis (FEA) applica-
tion. It has a friendly graphical user interface (GUI) and an extensive help system.
Once started, the user should have no difficulty navigating menus and so on. Since
all the mouse clicks in the GUI generate ANSYS command lines, which are saved
in a .log file, it is easy to use the GUI to learn what the various commands do.
309
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
The ANSYS help can then be used to enhance the user’s knowledge of the ANSYS
command structure.
In this textbook, selected commands from a .log file created with the GUI are
copied into a .inp file, which upon being read into ANSYS automatically performs
all the tasks of model creation, execution, and post-processing. This methodology
is recommended because the .inp file can be debugged, refined, adapted to similar
situations, recalled later on, and it provides excellent documentation for project
reporting to the client and even auditing by third parties. The .inp files provided
with the examples in this textbook demonstrate their usefulness, even if the GUI
was used to help generate most of them.
The examples in this textbook were produced using ANSYS version 14. They
work identically in Windows and Linux platforms. On Windows, ANSYS is accessi-
ble from the START menu, through two icons: “Mechanical APDL (ANSYS)” and
“Mechanical APDL Product Launcher.” It is best to use at least once the later, as
it allows us to set the default location for the model files.
In this Appendix, it is assumed that the user has created a folder c:\ansys\
user\, where all the model files reside. Therefore, in “Product Launcher,” under
“File Management,” the “Working Directory” should be set to c:\ansys\user.
Clicking on Run invokes the GUI.
The ANSYS GUI has a command bar at the top and a menu list on the left.
Below the command bar, there is single-line command window. ANSYS commands
typed in this window are executed immediately and have the same effect as equiv-
alent GUI operations.
Although the GUI is user friendly, it is very challenging to describe (in a text-
book) all the mouse clicks one has to do in order to set up and solve a problem. It
is also challenging to remember what one did during a previous session using the
GUI. And there is no use trying to write down the myriad mouse clicks needed to
accomplish a task. Fortunately, all GUI operations (mouse clicks, menu selections,
data entry, and so on) are saved by ANSYS into a .log file in the current directory
(c:\ansys\user\). The .log file is a text file that can be edited and cleaned up
of the many commands that represent dead ends that one has reached during a
session. Cleaned up .log files become .inp files, which can be recalled into ANSYS
and executed to reproduce a prior session.
The .inp files can be recalled in three ways. First, each line in the .inp file can
be typed in the command window and executed one at a time (by pressing enter,
of course). This is very useful in order to learn the effect that each command line
has on the model generation, execution, and so on. Second, a portion or the whole
.inp file can be pasted into the command window and executed. Finally, once
a .inp file is polished, the most computationally efficient way to enter a model
is to type the following command /input,file,inp, in the command window.
This will retrieve file.inp and execute it. The equivalent GUI mouse clicks are:
File, Read Input From, OK.
As was mentioned before, this section is very brief. It has been my experience
that students successfully teach themselves ANSYS by figuring out the commands
used in the examples in this textbook, which are available on the Web site [5],
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Appendix 311
along with the help system and the documentation included with ANSYS. Video
recordings illustrating the execution of the examples by using the ANSYS GUI are
available on the Web site [5].
C.1.1 ANSYS USERMAT, Compilation and Execution

Compilation and execution of ANSYS user programmable features (UPF) can be
accomplished following the procedure described in this section or as explained in
[15]. In this section, it is assumed that ANSYS 14.0, Microsoft Visual Studio 2008,
and Fortran Intel 11.1, are all available on a Windows system. Note that the path to
software components will change with time as new versions and different platforms
are released. Therefore, the paths given in this section may have to be adjusted.
For example v140 refers to version 14.0 and intel refers to the IntelTM processor,
which may have to be adjusted to reflect the software and hardware configuration
available to you.
There are 3 methods in ANSYS 14.0 to access user programmable features:
– Custom dynamic link librariy (DLL, described and used in this book).
– Custom .exe (described in the first edition of this book). We no longer favor
this method, but it is clearly explained in the ANSYS documentation in case
the reader wants to use it.
– User programmable features /UPF command. We do not use this method
because it cannot be used from the GUI, only in batch mode, but it is clearly
explained in the ANSYS documentation in case the reader wants to use it.
ANSYS will look for a USERMAT in the form of a USERMATLIB.DLL provided two
things happen:
– You have set up the environment variable ANS USER PATH in Windows, to
point to the place where you want to store your DLLs. We recommend using
the work directory, c:\ANSYS\User.
– A TB,USER, line appears in the .inp file.
One such DLL is available in C:\Program Files\ANSYS Inc\v140\ansys\bin-
\winx64 containing some sample code. To bypass this DLL and use your own, you
must do the following:
– Decide where you want to store your DLL. Our preference is to store them in
our work directory, c:\ANSYS\User.
– Add the environment variable ANS USER PATH to Windows, to point to the
place where you want to store your DLLs.
To add the environment variable, do this:
– Click Start, Control Panel, Advanced System Settings.
– Then, follow the steps illustrated in Figures C.1, where you click Environment
Variables, and Figure C.2, where you can add/edit ANS USER PATH.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Figure C.1: Add environment variable. Step 1. Click Environment Variables.
Figure C.2: Add environment variable. Step 2. Add/edit ANS USER PATH.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Appendix 313
Making your own DLLs
In ANSYS, USERMAT.F calls four specific subroutines:
– USERMAT1D can be used for 1D elements such as bars and links, not beams.
– USERMAT3D is used for 3D solid elements.
– USERMATBM is used for BEAM elements.
– USERMATPS is used for PLANE STRESS elements such as PLANE and SHELL.
You should not modify USERMAT.F, but just the code in one or more of the
specific routines listed above.
To make your own DLLs, copy the following from [5] to your work directory
(folder):
– Copy AnsUserMatEjb.bat to your work directory. This is a modified version

of ANSUSERSHARED.BAT provided by ANSYS. It has been modified to find
the various files needed for compilation, as well as to allow you to write free
formatted code (F90, F95).
– Copy impcomEjb.inc to your work directory. This is an interface to call

ANSYS’s impcom.inc from your own free formatted code. Note: impcom.inc
is an include file called by ANSYS using a compiler directive.
– Copy the folder \EJB to your work directory. This folder contains templates
for all five USERMAT routines, and a backup copy of AnsUserMatEjb.bat
just in case. Do not modify anything in this folder. Just use it as a repository
for templates that you may want to modify.
Then, copy all four specific routines from \EJB to your work directory. Modify
one or more of them to suit your needs. The ones you do not modify will be there
just to fill up a place when you build your DLL. The templates provided implement
routines used in some of the examples. You can look at the source code to learn
about them.
Once you have your own specific USERMATxx ready, you use AnsUserMatEjb.
bat to make your DLL. AnsUserMatEjb.bat will look for a template USERMAT.F in
the \EJB folder and needs the four USERMATxx.F and impcom.inc to be in your work
directory. Double click on AnsUserMatWvu.bat to execute it. If the DLL is built
correctly, it will be compiled and ANSYS will find it. If not, look at compile.log
to see the errors.
Note that .\usermat.F residing in your work directory will be overwritten each
time you use the batch file AnsUserMatWvu.bat, but it is not a problem because
AnsUserMatWvu.bat looks for a fresh copy in \EJB and the four editable files will
be automatically appended. You just edit any of the specific files listed above.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
C.2 BMI3
Most users will run BMI3 within ANSYS as explained in Section C.2.2 but for
troubleshooting it is useful to know how to operate it outside ANSYS, as explained
in Section C.2.1 next.
C.2.1 Stand-Alone BMI3

Native BMI3 code accepts an input file in Abaqus TM format, as long as the input
file is filtered by the program I2B [5]. Not all of the Abaqus commands are accepted
by I2B. For example it only accepts models with concentrated forces on nodes.
Most commercial computer aided design (CAD) packages such as I-DEASTM
and FEMAPTM can output an Abaqus file. Then, it is easy to modify the file
to make it comply with the restrictions of I2B. Run I2B to generate DEMO.inp,
ABAQUS.inp, and DEMO.dat. If ABAQUS.inp were to be executed within Abaqus, it
would give the bifurcation loads Λ(cr) .
The material properties and perturbation parameters are in DEMO.dat. The
last line contains modenum, nodenum, component. This is the mode, node, and
component used as a perturbation parameter. If all are zeros, BMI3 picks the lowest
mode and the node-component combination that yields the largest mode amplitude.
The results are printed in DEMO.out and the mode shapes saved in MODES file.
C.2.2 BMI3 within ANSYS

It is possible to use the program BMI3 directly from ANSYS, with some restrictions:
– Use only element type SHELL99.
– Introduce the laminate properties using ABDH matrices, with

KEYOPT(2)=2.
– Only apply loads on nodes or keypoints using concentrated forces (do not use
moments). If the model has distributed loads, calculate the equivalent nodal
forces and apply them at the nodes.
– Use only one real constant set for all the models.
The procedure to compute the post-critical path parameters using BMI3 within
ANSYS is described next.
– In the working directory (c:\ansys), copy the APDL macro ans2i.mac, and
the programs bmi3.exe and i2b ans.exe from [5].
– Define the model in ANSYS and solve it using the “Eigenvalue buckling anal-
ysis” procedure for obtaining the bifurcation loads Λ(cr) (e.g., Example 4.2).
– Run the APDL macro ans2i simply by entering ans2i in the ANSYS com-
mand line [5] to calculate parameters of the post-critical path.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
Appendix 315
– Look for the c:\ANSYS 14.0 Output Window, which is minimized in the Win-
dows taskbar, and bring it to the foreground.
– In the c:\ANSYS 14.0 Output Window, respond to the two prompts: (i) ac-
tivate or not sorting of the nodes in order to minimize the bandwidth of the
system of equations (sorting along the longest dimension is recommended), (ii)
introduce the mode, node, and component used as the perturbation parameter
s or let BMI3 choose the default. By default, the lowest mode and the node,
component, with the largest mode amplitude is used. If an error message of
“INSUFFICIENT STORAGE” appears, try sorting along another direction.
If that fails, BMI3 needs to be recompiled with larger arrays.
– In addition to the critical load Λ(cr) , BMI3 computes the slope L(1) = Λ(1)
and the curvature L(2) = Λ(2) of the bifurcation mode selected. These results
are shown in the c:\ANSYS 14.0 Output Window and they are printed in
DEMO.out
– Do not close the c:\ANSYS 14.0 Output Window, just minimize it. Other-
wise, it will abort ANSYS. Instead, ANSYS should be closed from the GUI.
Note that the results (bifurcation loads, slopes, and curvatures) appear with a
negative sign. This is usual in stability analysis. If a model is constructed with ten-
sile loads (instead of the usual compression), one can type REVERS=-1 in the ANSYS
command line before executing the APDL macro ANS2I. Another peculiarity of the
BMI3 software is that transverse deflections w (perpendicular to the plate) have an
opposite sign to that used by ANSYS. Since transverse deflections w are often used
as perturbation parameters, the change in sign must be taken into account during
the interpretation of results.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
References
[1] E. J. Barbero. Introduction to Composite Materials Design–Second Edition. http:

//barbero.cadec-online.com/icmd. CRC Press, Boca Raton, FL, 2010.
[2] D. Frederick and T.-S. Chang. Continuum Mechanics. Scientific Publishers, Cam-
bridge, MA, 1972.
[3] F. P. Beer, E. R. Johnston Jr., and J. T. DeWolf. Mechanics of Materials, 3rd Edition.
McGraw-Hill, Boston, MA, 2001.
[4] J. N. Reddy. Energy and Variational Methods in Applied Mechanics. Wiley, New
York, 1984.
[5] E. J. Barbero. Web resource: http://barbero.cadec-online.com/feacm-ansys.
[6] S. S. Sonti, E. J. Barbero, and T. Winegardner. Mechanical properties of pultruded
E-glass–vinyl ester composites. In 50th Annual Conference, Composites Institute,
Society of the Plastics Industry (February) pp. 10-C/1-7, 1995.
[7] E. J. Barbero. Finite Element Analysis of Composite Materials Using Abaqus. CRC
Press, 2013.
[8] ANSYS Inc. ANSYS mechanical APDL structural analysis guide, release 14.0, 2011.
[9] E. J. Barbero. Computer aided design environment for composites. http://www.
cadec-online.com, 2011.
[10] J. N. Reddy. Mechanics of Laminated Composite Plates and Shells, 2nd Edition. CRC
Press, Boca Raton, FL, 2003.
[11] R. J. Roark and W. C. Young. Roark’s Formulas for Stress and Strain, 6th Edition.
McGraw-Hill, New York, NY, 1989.
[12] E. Hinton and D. R. J. Owen. An Introduction to Finite Element Computations.
Pineridge Press, Swansea, UK, 1979.
[13] ANSYS Inc. ANSYS mechanical APDL theory reference, release 14.0, 2011.
[14] Simulia. Abaqus documentation. The default location for the HTML documentation
is http://HOME:2080/v6.10/ where HOME is the url of the server where the documen-
tation is stored during installation. The default location for the PDF documentation
is C:/SIMULIA/Documentation/docs/v6.10/pdf books/index.pdf.
[15] ANSYS Inc. ANSYS mechanical APDL programmer’s manual, release 14.0, 2011.
[16] E. J. Barbero and J. Trovillion. Prediction and measurement of post-critical behavior
of fiber-reinforced composite columns. Composites Science and Technology, 58:1335–
1341, 1998.
317
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
[17] E. J. Barbero. Prediction of compression strength of unidirectional polymer matrix

composites. Journal of Composite Materials, 32(5)(5):483–502, 1998.
[18] A. Puck and H. Schurmann. Failure analysis of FRP laminates by means of physically
based phenomenological models. Composites Science and Technology, 62:1633–1662,
2002.
[19] MIL17.org. The composite materials handbook, web resource, http://www.mil17.
org.
[20] ANSYS Inc. ANSYS parametric design language guide, release 14.0, 2011.
[21] L. A. Godoy. Theory of Stability-Analysis and Sensitivity. Taylor and Francis,
Philadelphia, PA, 2000.
[22] E. J. Barbero, L. A. Godoy, and I. G. Raftoyiannis. Finite elements for three-mode
interaction in buckling analysis. International Journal for Numerical Methods in
Engineering, 39(3):469–488, 1996.
[23] L. A. Godoy, E. J. Barbero, and I. G. Raftoyiannis. Finite elements for post-buckling
analysis. I-The W-formulation. Computers and Structures, 56(6):1009–1017, 1995.
[24] E. J. Barbero, I. G. Raftoyiannis, and L. A. Godoy. Finite elements for post-buckling
analysis. II-Application to composite plate assemblies. Computers and Structures,
56(6):1019–1028, 1995.
[25] I. G. Raftoyiannis, L. A. Godoy, and E. J. Barbero. Buckling mode interaction in
composite plate assemblies. Applied Mechanics Reviews, 48(11/2):52–60, 1995.
[26] E. J. Barbero. Prediction of buckling-mode interaction in composite columns. Me-
chanics of Composite Materials and Structures, 7(3):269–284, 2000.
[27] S. Yamada and J. G. A. Croll. Buckling behavior pressure loaded cylindrical panels.
ASCE Journal of Engineering Mechanics, 115(2):327–344, 1989.
[28] C. T. Herakovich. Mechanics of Fibrous Composites. Wiley, New York, 1998.
[29] J. D. Eshelby. The determination of the elastic field of an ellipsoidal inclusion and
related problems. Proceedings of the Royal Society, A241:376–396, 1957.
[30] J. D. Eshelby. The elastic field outside an ellipsoidal inclusion. Acta Metallurgica,
A252:561–569, 1959.
[31] T. Mori and K. Tanaka. The elastic field outside an ellipsoidal inclusion. Acta Met-
allurgica, 21:571–574, 1973.
[32] R. Hill. A self-consistent mechanics of composite materials. Journal of Mechanics
and Physics of Solids, 13:213–222, 1965.
[33] S. Nemat-Nasser and M. Hori. Micromechanics: Overall Properties of Heterogeneous
Materials. North-Holland, Amsterdam, 1993.
[34] R. Luciano and E. J. Barbero. Formulas for the stiffness of composites with periodic
microstructure. International Journal of Solids and Structures, 31(21):2933–2944,
1995.
[35] R. Luciano and E. J. Barbero. Analytical expressions for the relaxation moduli of
linear viscoelastic composites with periodic microstructure. ASME J. Applied Me-
chanics, 62(3):786–793, 1995.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
References 319
[36] E. J. Barbero and R. Luciano. Micromechanical formulas for the relaxation tensor of
linear viscoelastic composites with transversely isotropic fibers. International Journal
of Solids and Structures, 32(13):1859–1872, 1995.
[37] J. Aboudi. Mechanics of Composite Materials : A Unified Micromechanical Approach.
Vol. 29 of Studies in Applied Mechanics. Elsevier, New York, NY, 1991.
[38] Z. Hashin and S. Shtrikman. A variational approach to the elastic behavior of multi-
phase materials. Journal of Mechanics and Physics of Solids, 11:127–140, 1963.
[39] V. Tvergaard. Model studies of fibre breakage and debonding in a metal reinforced
by short fibres. Journal of Mechanics and Physics of Solids, 41(8):1309–1326, 1993.
[40] J. L. Teply and G. J. Dvorak. Bound on overall instantaneous properties of elastic-
plastic composites. Journal of Mechanics and Physics of Solids, 36(1):29–58, 1988.
[41] E. J. Barbero. Finite Element Analysis of Composite Materials–First Edition. CRC
Press, Boca Raton, FL, 2007.
[42] R. Luciano and E. Sacco. Variational methods for the homogenization of periodic
media. European Journal of Mechanics A/Solids, 17:599–617, 1998.
[43] G. J. Creus. Viscoelasticity: Basic Theory and Applications to Concrete Structures.
Springer-Verlag, Berlin, 1986.
[44] E. J. Barbero. 3-D finite element for laminated composites with 2-d kinematic con-
straints. Computers and Structures, 45(2):263–271, 1992.
[45] Eric W. Weisstein. Gamma function. MathWorld–A Wolfram Web Resource. http:
//mathworld.wolfram.com/GammaFunction.html.
[46] G. D. Dean, B. E. Read, and P. E. Tomlins. A model for long-term creep and the effects
of physical ageing in poly (butylene terephthalate). Plastics and Rubber Processing
and Applications, 13(1):37–46, 1990.
[47] B. F. Oliveira and G. J. Creus. An analytical-numerical framework for the study of
ageing in fibre reinforced polymer composites. Composites B, 65(3-4):443–457, 2004.
[48] E. J. Barbero. Chapter 2: Time-Temperature-Age Superposition Principle for Pre-
dicting Long-Term Response of Linear Viscoelastic Materials, in Creep and Fatigue
in Polymer Matrix Composites. Woodhead, Cambridge, 2010.
[49] L. C. E. Struik. Physical aging in amorphous polymers and other materials. Elsevier
Scientific Pub. Co. ; New York, 1978.
[50] K. Ogatha. Discrete-Time Control System. Prentice Hall, Englewood Cliffs, NJ, 1987.
[51] K. J. Hollenbeck. Invlap.m: A MATLAB function for numerical inversion of Laplace
transforms by the de Hoog algorithm (1998), http://www.mathworks.com/.
[52] R. S. Lakes. Viscoelastic Solids. CRC Press, Boca Raton, FL, 1998.
[53] J. D. Ferry. Viscoelastic Properties of Polymers. 3rd. Edition. Wiley, New York, 1980.
[54] R. M. Christensen. Theory of Viscoelasticity: Second Edition. Dover, Mineola, NY,
2010.
[55] P. Qiao, E. J. Barbero, and J. F. Davalos. On the linear viscoelasticity of thin-walled
laminated composite beams. Journal of Composite Materials, 34(1):39–68, 2000.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
[56] E. J. Barbero, F. A. Cosso, R. Roman, and T. L. Weadon. Determination of

material parameters for Abaqus progressive damage analysis of E-Glass Epoxy
laminates, http://dx.doi.org/10.1016/j.compositesb.2012.09.069. Composites
Part B:Engineering, 46(3):211–220, 2012.
[57] J. Lemaitre and A. Plumtree. Application of damage concepts to predict creep-fatigue
failures. In American Society of Mechanical Engineers, 78-PVP-26, pages 10–26, 1978.
[58] I. N. Rabotnov. Rabotnov: Selected Works - Problems of the Mechanics of a De-
formable Solid Body. Moscow Izdatel Nauka, 1991.
[59] L. M. Kachanov. On the creep fracture time. Izv. Akad. Nauk USSR, 8:26–31, 1958.
[60] N. R. Hansen and H. L. Schreyer. A thermodynamically consistent framework for
theories of elastoplasticity coupled with damage. International Journal of Solids and
Structures, 31(3):359–389, 1994.
[61] J. Janson and J. Hult. Fracture mechanics and damage mechanics: A combined
approach. Journal of Theoretical and Applied Mechanics, 1:S18–28, 1977.
[62] E. J. Barbero and K. W. Kelly. Predicting high temperature ultimate strength of con-
tinuous fiber metal matrix composites. Journal of Composite Materials, 27(12):1214–
1235, 1993.
[63] K. W. Kelly and E. Barbero. Effect of fiber damage on the longitudinal creep of a
CFMMC. International Journal of Solids and Structures, 30(24):3417–3429, 1993.
[64] J. R. Rice. Continuum Mechanics and Thermodynamics of Plasticity in Relation to
Microscale Deformation Mechanisms, chapter 2, pages 23–79. Constitutive Equations
in Plasticity. MIT Press, Cambridge, MA, 1975.
[65] J. C. Simo and T. J. R. Hughes. Computational Inelasticity. Springer, Berlin, 1998.
[66] D. Krajcinovic, J. Trafimow, and D. Sumarac. Simple constitutive model for a cortical
bone. Journal of Biomechanics, 20(8):779–784, 1987.
[67] D. Krajcinovic and D. Fanella. Micromechanical damage model for concrete. Engi-
neering Fracture Mechanics, 25(5-6):585–596, 1985.
[68] D. Krajcinovic. Damage mechanics. Mechanics of Materials, 8(2-3):117–197, 1989.
[69] A. M. Neville. Properties of Concrete, 2nd Edition. Wiley, New York, NY, 1973.
[70] ACI, American Concrete Institute.
[71] B. W. Rosen. The tensile failure of fibrous composites. AIAA Journal, 2(11):1985–
1991, 1964.
[72] W. Weibull. A statistical distribution function of wide applicability. Journal of Applied
Mechanics, 18:293–296, 1951.
[73] B. W. Rosen. Fiber Composite Materials, chapter 3. American Society for Metals,
Metals Park, OH, 1965.
[74] A. S. D. Wang. A non-linear microbuckling model predicting the compressive strength
of unidirectional composites. In ASME Winter Annual Meeting, volume WA/Aero-1,
1978.
[75] J. S. Tomblin, E. J. Barbero, and L. A. Godoy. Imperfection sensitivity of fiber
micro-buckling in elastic-nonlinear polymer-matrix composites. International Journal
of Solids and Structures, 34(13):1667–1679, 1997.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
References 321
[76] D. C. Lagoudas and A. M. Saleh. Compressive failure due to kinking of fibrous

composites. Journal of Composite Materials, 27(1):83–106, 1993.
[77] P. Steif. A model for kinking in fiber composites. I. Fiber breakage via micro-buckling.
International Journal of Solids and Structures, 26 (5-6):549–561, 1990.
[78] S. W. Yurgartis and S. S. Sternstein. Experiments to Reveal the Role of Matrix Prop-
erties and Composite Microstructure in Longitudinal Compression Strength, volume
1185 of ASTM Special Technical Publication, pages 193–204. ASTM, Philadelphia,
PA, 1994.
[79] C. Sun and A. W. Jun. Effect of matrix nonlinear behavior on the compressive strength
of fiber composites. In AMD, volume 162, pages 91–101, New York, NY, 1993. ASME,
American Society of Mechanical Engineers, Applied Mechanics Division.
[80] D. Adams and E. Lewis. Current status of composite material shear test methods.
SAMPE Journal, 31(1):32–41, 1995.
[81] A. Maewal. Postbuckling behavior of a periodically laminated medium in compression.
International Journal of Solids and Structures, 17(3):335–344, 1981.
[82] L. M. Kachanov. Rupture time under creep conditions problems of continuum me-
chanics. SIAM, pages 202–218, 1958.
[83] E. J. Barbero and L. DeVivo. A constitutive model for elastic damage in fiber-
reinforced PMC laminae. Journal of Damage Mechanics, 10(1):73–93, 2001.
[84] E. J. Barbero, F. Greco, and P. Lonetti. Continuum damage-healing mechanics with
application to self-healing composites. International Journal of Damage Mechanics,
14(1):51–81, 2005.
[85] S. Murakami. Mechanical modeling of material damage. Journal of Applied Mechan-
ics, 55:281–286, 1988.
[86] J. M. Smith and H. C. Van Ness. Introduction to Chemical Engineering Thermody-
namics, 3rd Edition. McGraw-Hill, New York, NY, 1975.
[87] E. R. Cohen, T. Cvita, J. G. Frey, B. Holmstrm, K. Kuchitsu, R. Marquardt, I. Mills,
F. Pavese, M. Quack, J. Stohner, H. L. Strauss, M. Takami, and A. J. Thor. Quanti-
ties, Units and Symbols in Physical Chemistry : The IUPAC Green Book, 3rd Edition,
http: // old. iupac. org/ publications/ books/ author/ cohen. html . RSC Pub-
lishing, 2007.
[88] L. E. Malvern. Introduction to the Mechanics of a Continuous Medium. Prentice Hall,
Upper Saddle River, NJ, 1969.
[89] Y. A. Cengel and M. A. Boles. Thermodynamics: An Engineering Approach, 3rd
Edition. McGraw-Hill, New York, NY, 1998.
[90] E. J. Barbero and P. Lonetti. Damage model for composites defined in terms of
available data. Mechanics of Composite Materials and Structures, 8(4):299–315, 2001.
[91] P. Lonetti, R. Zinno, F. Greco, and E. J. Barbero. Interlaminar damage model for
polymer matrix composites. Journal of Composite Materials, 37(16):1485–1504, 2003.
[92] P. Lonetti, E. J. Barbero, R. Zinno, and F. Greco. Erratum: Interlaminar damage
model for polymer matrix composites. Journal of Composite Materials, 38(9):799–800,
2004.
[93] J. Lubliner. Plasticity Theory. Collier Macmillan, New York, NY, 1990.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
[94] J. A. Nairn. The strain energy release rate of composite microcracking: A variational
approach. Journal of Composite Materials, 23(11):1106–1129, 11 1989.
[95] J. A. Nairn, S. Hu, S. Liu, and J. S. Bark. The initiation, propagation, and effect
of matrix microcracks in cross-ply and related laminates. In Proceedings of the 1st
NASA Advanced Composite Technical Conference, pages 497–512, Oct. 29–Nov. 1,
1990.
[96] S. Liu and J. A. Nairn. The formation and propagation of matrix microcracks in cross-
ply laminates during static loading. Journal of Reinforced Plastics and Composites,
11(2):158–178, Feb. 1992.
[97] J. A. Nairn. Microcracking, microcrack-induced delamination, and longitudinal split-
ting of advanced composite structures. Technical Report NASA Contractor Report
4472, 1992.
[98] J. A. Nairn and S. Hu. Damage Mechanics of Composite Materials, volume 9 of
Composite Materials Series, chapter Matrix Microcracking, pages 187–244. Elsevier,
1994.
[99] J. A. Nairn. Applications of finite fracture mechanics for predicting fracture events in
composites. In 5th International Conference on Deformation and Fracture of Com-
posites, 18–19 March 1999.
[100] J. A. Nairn. Polymer Matrix Composites, volume 2 of Comprehensive Composite
Materials, chapter Matrix Microcracking in Composites, pages 403–432. Elsevier
Science, 2000.
[101] J. A. Nairn. Fracture mechanics of composites with residual stresses, imperfect inter-
faces, and traction-loaded cracks. In Recent Advances in Continuum Damage Mechan-
ics for Composites Workshop, volume 61, pages 2159–2167, UK, 20–22 Sept. 2000.
Elsevier.
[102] J. A. Nairn and D. A. Mendels. On the use of planar shear-lag methods for stress-
transfer analysis of multilayered composites. Mechanics of Materials, 33(6):335–362,
2001.
[103] J. A. Nairn. Finite Fracture Mechanics of Matrix Microcracking in Composites, pages
207–212. Application of Fracture Mechanics to Polymers, Adhesives and Composites.
Elsevier, 2004.
[104] Y. M. Han, H. T. Hahn, and R. B. Croman. A simplified analysis of transverse ply
cracking in cross-ply laminates. Composites Science and Technology, 31(3):165–177,
1988.
[105] C. T. Herakovich, J. Aboudi, S. W. Lee, and E. A. Strauss. Damage in composite
laminates: Effects of transverse cracks. Mechanics of Materials, 7(2):91–107, 11 1988.
[106] J. Aboudi, S. W. Lee, and C. T. Herakovich. Three-dimensional analysis of laminates
with cross cracks. Journal of Applied Mechanics, Transactions ASME, 55(2):389–397,
1988.
[107] F. W. Crossman, W. J. Warren, A. S. D. Wang, and G. E. Law. Initiation and growth
of transverse cracks and edge delamination in composite laminates. II-Experimental
correlation. Journal of Composite Materials Supplement, 14(1):88–108, 1980.
[108] D. L. Flaggs and M. H. Kural. Experimental determination of the in situ trans-
verse lamina strength in graphite/epoxy laminates. Journal of Composite Materials,
16:103–116, 03 1982.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
References 323
[109] S. H. Lim and S. Li. Energy release rates for transverse cracking and delamina-
tions induced by transverse cracks in laminated composites. Composites Part A,
36(11):1467–1476, 2005.
[110] S. Li and F. Hafeez. Variation-based cracked laminate analysis revisited and funda-
mentally extended. International Journal of Solids and Structures, 46(20):3505–3515,
2009.
[111] J. L Rebiere and D. Gamby. A decomposition of the strain energy release rate associ-
ated with the initiation of transverse cracking, longitudinal cracking and delamination
in cross-ply laminates. Composite Structures, 84(2):186–197, 2008.
[112] S. C. Tan and R. J. Nuismer. A theory for progressive matrix cracking in composite
laminates. Journal of Composite Materials, 23:1029–1047, 1989.
[113] R. J. Nuismer and S. C. Tan. Constitutive relations of a cracked composite lamina.
Journal of Composite Materials, 22:306–321, 1988.
[114] T. Yokozeki and T. Aoki. Stress analysis of symmetric laminates with obliquely-
crossed matrix cracks. Advanced Composite Materials, 13(2):121–140, 2004.
[115] T. Yokozeki and T. Aoki. Overall thermoelastic properties of symmetric laminates
containing obliquely crossed matrix cracks. Composites Science and Technology,
65(11-12):1647–1654, 2005.
[116] Z. Hashin. Analysis of cracked laminates: A variational approach. Mechanics of
Materials, 4:121:136, 1985.
[117] J. Zhang, J. Fan, and C. Soutis. Analysis of multiple matrix cracking in [±θ/90n ]s
composite laminates, part I, inplane stiffness properties. Composites, 23(5):291–304,
1992.
[118] P. Gudmundson and S. Ostlund. First order analysis of stiffness reduction due to
matrix cracking. Journal of Composite Materials, 26(7):1009–1030, 1992.
[119] P. Gudmundson and S. Ostlund. Prediction of thermoelastic properties of composite
laminates with matrix cracks. Composites Science and Technology, 44(2):95–105,
1992.
[120] P. Gudmundson and W. Zang. An analytic model for thermoelastic properties of
composite laminates containing transverse matrix cracks. International Journal of
Solids and Structures, 30(23):3211–3231, 1993.
[121] E. Adolfsson and P. Gudmundson. Thermoelastic properties in combined bending and
extension of thin composite laminates with transverse matrix cracks. International
Journal of Solids and Structures, 34(16):2035–2060, 06 1997.
[122] E. Adolfsson and P. Gudmundson. Matrix crack initiation and progression in com-
posite laminates subjected to bending and extension. International Journal of Solids
and Structures, 36(21):3131–3169, 1999.
[123] W. Zang and P. Gudmundson. Damage evolution and thermoelastic properties of
composite laminates. International Journal of Damage Mechanics, 2(3):290–308, 07
1993.
[124] E. Adolfsson and P. Gudmundson. Matrix crack induced stiffness reductions in
[(0m/90n/+p/-q)s]m composite laminates. Composites Engineering, 5(1):107–123,
1995.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
[125] P. Lundmark and J. Varna. Modeling thermo-mechanical properties of damaged

laminates. In 3rd International Conference on Fracture and Damage Mechanics, FDM
2003, volume 251–252 of Advances in Fracture and Damage Mechanics, pages 381–
387, Switzerland, 2–4 Sept. 2003. Trans Tech Publications.
[126] M. Kachanov. Elastic Solids with Many Cracks and Related Problems, volume 30 of
Advances in Applied Mechanics, chapter X, pages 260–445. Academic Press, Inc.,
1993.
[127] A. Adumitroaie and E. J. Barbero. Intralaminar damage model for laminates sub-
jected to membrane and flexural deformations. Mechanics of Advanced Materials and
Structures, 2013.
[128] T. Yokozeki, T. Aoki, and T. Ishikawa. Transverse crack propagation in the specimen
width direction of cfrp laminates under static tensile loadings. Journal of Composite
Materials, 36(17):2085–2099, 2002.
[129] T. Yokozeki, T. Aoki, and T. Ishikawa. Consecutive matrix cracking in contiguous

plies of composite laminates. International Journal of Solids and Structures, 42(9-
10):2785–2802, 05 2005.
[130] T. Yokozeki, T. Aoki, T. Ogasawara, and T. Ishikawa. Effects of layup angle and
ply thickness on matrix crack interaction in contiguous plies of composite laminates.
Composites Part A (Applied Science and Manufacturing), 36(9):1229–1235, 2005.
[131] E. J. Barbero, F. A. Cosso, and F. A. Campo. Benchmark solution for degrada-

tion of elastic properties due to transverse matrix cracking in laminated composites,
http://dx.doi.org/10.1016/j.compstruct.2012.11.009. Composite Structures,
98(4):242–252, 2013.
[132] S. Li, S. R. Reid, and P. D. Soden. A continuum damage model for transverse matrix
cracking in laminated fibre-reinforced composites. Philosophical Transactions of the
Royal Society London, Series A (Mathematical, Physical and Engineering Sciences),
356(1746):2379–2412, 10/15 1998.
[133] Janis Varna, Roberts Joffe, and Ramesh Talreja. A synergistic damage-mechanics
analysis of transverse cracking [±θ/904 ]s laminates. Composites Science and Tech-
nology, 61(5):657–665, 2001.
[134] A. S. D. Wang, P. C. Chou, and S. C. Lei. A stochastic model for the growth of
matrix cracks in composite laminates. Journal of Composite Materials, 18(3):239–
254, 05 1984.
[135] S. Li, S. R. Reid, and P. D. Soden. Modelling the damage due to transverse matrix
cracking in fiber-reinforced laminates. In Proceedings 2nd International Conference
on Nonlinear Mechanics (ICNP-2), pages 320–323. Peking University Press, 1993.
[136] D.H. Cortes and E.J. Barbero. Stiffness reduction and fracture evolution
of oblique matrix cracks in composite laminates. http://dx.doi.org/10.1007/
s12356-009-0001-5. Annals of Solid and Structural Mechanics, 1(1)(1):29–40, 2010.
[137] E. J. Barbero and D. H. Cortes. A mechanistic model for transverse damage initiation,
evolution, and stiffness reduction in laminated composites. http://dx.doi.org/10.
1016/j.compositesb.2009.10.001. Composites Part B: Engineering, 41(2):124–132,
2010.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
References 325
[138] E. J. Barbero, G. Sgambitterra, A. Adumitroaie, and X. Martinez. A discrete con-

stitutive model for transverse and shear damage of symmetric laminates with ar-
bitrary stacking sequence. http://dx.doi.org/10.1016/j.compstruct.2010.06.
011. Composite Structures, 93:1021–1030, 2011.
[139] G. Sgambitterra, A. Adumitroaie, E.J. Barbero, and A. Tessler. A robust three-node
shell element for laminated composites with matrix damage. http://dx.doi.org/10.
1016/j.compositesb.2010.09.016. Composites Part B: Engineering, 42(1):41–50,
2011.
[140] A. M. Abad Blazquez, M. Herraez Matesanz, Carlos Navarro Ugena, and E. J. Bar-
bero. Acoustic emission characterization of intralaminar damage in composite lami-
nates. In Asociación Española de Materiales Compuestos MATCOMP 2013, Algeci-
ras, Spain, July 2–5, 2013.
[141] H. T. Hahn. A mixed-mode fracture criterion for composite materials. Composites
Technology Review, 5:26–29, 1983.
[142] Allan F. Bower. Mechanics of Solids. CRC, 2010. Free at http://solidmechanics.org/.
[143] H. Chai, C. D. Babcock, and W. G. Knauss. One dimensional modeling of failure
in laminated plates by delamination buckling. International Journal of Solids and
Structures, 17(11):1069–1083, 1981.
[144] M.-K. Yeh and L.-B. Fang. Contact analysis and experiment of delaminated cantilever
composite beam. Composites: Part B, 30(4):407–414, 1999.
[145] W. L. Yin, S. N. Sallam, and G. J. Simitses. Ultimate axial load capacity of a
delaminated beam-plate. AIAA Journal, 24(1):123–128, 1986.
[146] D. Bruno. Delamination buckling in composite laminates with interlaminar defects.
Theoretical and Applied Fracture Mechanics, 9(2):145–159, 1988.
[147] D. Bruno and A. Grimaldi. Delamination failure of layered composite plates loaded
in compression. International Journal of Solids and Structures, 26(3):313–330, 1990.
[148] G. A. Kardomateas. The initial post-buckling and growth behaviour of internal de-
laminations in composite plates. Journal of Applied Mechanics, 60(4):903–910, 1993.
[149] G. A. Kardomateas and A. A. Pelegri. The stability of delamination growth in com-
pressively loaded composite plates. International Journal of Fracture, 65(3):261–276,
1994.
[150] I. Sheinman, G. A. Kardomateas, and A. A. Pelegri. Delamination growth during pre-
and post-buckling phases of delaminated composite laminates. International Journal
of Solids and Structures, 35(1-2):19–31, 1998.
[151] D. Bruno and F. Greco. An asymptotic analysis of delamination buckling and growth
in layered plates. International Journal of Solids and Structures, 37(43):6239–6276,
2000.
[152] H.-J. Kim. Postbuckling analysis of composite laminates with a delamination. Com-
puter and Structures, 62(6):975–983, 1997.
[153] W. G. Bottega and A. Maewal. Delamination buckling and growth in laminates–
closure. Journal of Applied Mechanics, 50(14):184–189, 1983.
[154] B. Cochelin and M. Potier-Ferry. A numerical model for buckling and growth of
delaminations in composite laminates. Computer Methods in Applied Mechanics,
89(1-3):361–380, 1991.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
[155] P.-L. Larsson. On delamination buckling and growth in circular and annular or-
thotropic plates. International Journal of Solids and Structures, 27(1):15–28, 1991.
[156] W.-L. Yin. Axisymmetric buckling and growth of a circular delamination in a com-
pressed laminate. International Journal of Solids and Structures, 21(5):503–514, 1985.
[157] K.-F. Nilsson, L. E. Asp, J. E. Alpman, and L. Nystedt. Delamination buckling and
growth for delaminations at different depths in a slender composite panel. Interna-
tional Journal of Solids and Structures, 38(17):3039–3071, 2001.
[158] H. Chai and C. D. Babcock. Two-dimensional modeling of compressive failure in
delaminated laminates. Journal of Composite Materials, 19(1):67–98, 1985.
[159] J. D. Withcomb and K. N. Shivakumar. Strain-energy release rate analysis of plates
with postbuckled delaminations. Journal of Composite Materials, 23(7):714–734,
1989.
[160] W. J. Bottega. A growth law for propagation of arbitrary shaped delaminations
in layered plates. International Journal of Solids and Structures, 19(11):1009–1017,
1983.
[161] B. Storåkers and B. Andersson. Nonlinear plate theory applied to delamination in
composites. Journal of Mechanics and Physics of Solids, 36(6):689–718, 1988.
[162] P.-L. Larsson. On multiple delamination buckling and growth in composite plates.
International Journal of Solids and Structures, 27(13):1623–1637, 1991.
[163] M. A. Kouchakzadeh and H. Sekine. Compressive buckling analysis of rectangu-
lar composite laminates containing multiple delaminations. Composite Structures,
50(3):249–255, 2000.
[164] J. M. Comiez, A. M. Waas, and K. W. Shahwan. Delamination buckling: Experiment
and analysis. International Journal of Solids and Structures, 32(6/7):767–782, 1995.
[165] G. A. Kardomateas. Postbuckling characteristics in delaminated kevlar/epoxy lam-
inates: An experimental study. Journal of Composites Technology and Research,
12(2):85–90, 1990.
[166] O. Allix and A. Corigliano. Modeling and simulation of crack propagation in mixed-
modes interlaminar fracture specimens. International Journal of Fracture, 77(2):111–
140, 1996.
[167] D. Bruno and F. Greco. Delamination in composite plates: Influence of shear de-
formability on interfacial debonding. Cement and Concrete Composites, 23(1):33–45,
2001.
[168] J. W. Hutchinson and Z. Suo. Mixed mode cracking in layered materials. Advances
in Applied Mechanics, 29:63–191, 1992.
[169] N. Point and E. Sacco. Delamination of beams: An application to the dcb specimen.
International Journal of Fracture, 79(3):225–247, 1996.
[170] J. G. Williams. On the calculation of energy release rate for cracked laminates.
International Journal of Fracture, 36(2):101–119, 1988.
[171] A. Tylikowsky. Effects of piezoactuator delamination on the transfer functions of
vibration control systems. International Journal of Solids and Structures, 38(10-
13):2189–2202, 2001.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
References 327
[172] C. E. Seeley and A. Chattopadhyay. Modeling of adaptive composites including

debonding. International Journal Solids and Structures, 36(12):1823–1843, 1999.
[173] W. J. Bottega and A. Maewal. Dynamics of delamination buckling. International
Journal of Non-Linear Mechanics, 18(6):449–463, 1983.
[174] G. Alfano and M. A. Crisfield. Finite element interface models for the delamination
analysis of laminated composites: Mechanical and computational issues. International
Journal for Numerical Methods in Engineering, 50(7):1701–1736, 2001.
[175] P. P. Camanho, C. G. Dávila, and M. F. Moura. Numerical simulation of mixed-mode
progressive delamination in composite materials. Journal of Composite Materials,
37(16):1415–1438, 2003.
[176] Z. Zou, S. R. Reid, and S. Li. A continuum damage model for delaminations in
laminated composites. Journal of Mechanics and Physics of Solids, 51(2):333–356,
2003.
[177] X-P Xu and A. Needleman. Numerical simulations of fast crack growth in brittle
solids. Journal of Mechanics and Physics of Solids, 42:1397–1434, 1994.
[178] J. R. Reeder. A bilinear failure criterion for mixed-mode delamination in composite
materials. ASTM STP 1206, pages 303–322, 1993.
[179] H. Chai. Three-dimensional fracture analysis of thin-film debonding. International
Journal of Fracture, 46(4):237–256, 1990.
[180] J. D. Withcomb. Three dimensional analysis of a post-buckled embedded delamina-
tion. Journal of Composite Materials, 23(9):862–889, 1989.
[181] D. Bruno, F. Greco, and P. Lonetti. A 3d delamination modelling technique based on
plate and interface theories for laminated structures. European Journal of Mechanics
A/Solids, 24:127–149, 2005.
[182] J. G. Williams. The fracture mechanics of delamination tests. Journal of Strain
Analysis, 24(4):207–214, 1989.
[183] W.-L. Yin and J. T. S. Wang. The energy-release rate in the growth of a one-
dimensional delamination. Journal of Applied Mechanics, 51(4):939–941, 1984.
[184] J. R. Rice. A path independent integral and the approximate analysis of strain
concentrations by notches and cracks. Journal of Applied Mechanics, 35:379–386,
1968.
[185] C. G. Dávila, P. P. Camanho, and M. F. De Moura. Mixed-mode decohesion elements
for analyses of progressive delamination. In 42nd AIAA/ASME/ASCE/AHS/ASC
Structures, Structural Dynamics, and Materials Conference, April 16 - 19, 2001,
volume 3, pages 2277–2288, Seattle, WA, 2001. AIAA.
[186] ANSYS Inc. ANSYS mechanical APDL materials reference, release 14.0, 2011.
[187] T. L. Anderson. Fracture Mechanics. CRC, Boca Raton, FL, 1995.
[188] E.J. Barbero and J.N. Reddy. Jacobian derivative method for three-dimensional frac-
ture mechanics. Communications in Applied Numerical Methods, 6(7):507–518, 1990.
[189] M. Benzeggagh and M. Kenane. Measurement of mixed-mode delamination fracture
toughness of unidirectional glass/epoxy composites with mixed-mode bending appa-
ratus. Composite Science and Technology, 56:439, 1996.
i i
i i
i i
“K15077” — 2013/11/3 — 21:45

i i
[190] J. Reeder, S. Kyongchan, P. B. Chunchu, and D. R.. Ambur. Postbuckling and

growth of delaminations in composite plates subjected to axial compression. In
43rd AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Mate-
rials Conference, volume 1746, page 10, Denver, CO, 2002.
[191] A. Caceres. Local Damage Analysis of Fiber Reinforced Polymer Matrix Composites,
Ph.D. dissertation. PhD thesis, West Virginia University, Morgantown, WV, 1998.
i i
i i
Materials Science/Mechanical Engineering/Civil Engineering
FINITE ELEMENT ANALYSIS Barbero SECOND EDITION
FINITE ELEMENT ANALYSIS OF COMPOSITE MATERIALS USING ANSYS®

USING ANSYS®
SECOND EDITION
Ever J. Barbero
Designing structures using composite materials poses unique challenges,
especially due to the need for concurrent design of both material
and structure. Professors are faced with two options: textbooks that

teach the theory of advanced mechanics of composites, but lack
computational examples of advanced analysis, and books on finite
element analysis that may or may not demonstrate very limited
applications to composites. But there is a third option that makes the
other two obsolete: Ever J. Barbero’s Finite Element Analysis
of Composite Materials Using ANSYS®, Second Edition.
The Only Finite Element Analysis Book on the Market Using
USING ANSYS
ANSYS® to Analyze Composite Materials
By layering detailed theoretical and conceptual discussions with fully

®
developed examples, this text supplies the missing link between theory
and implementation. In-depth discussions cover all of the major aspects of
advanced analysis, including three-dimensional effects, viscoelasticity, edge
effects, elastic instability, damage, and delamination. This second edition of
the bestseller has been completely revised to incorporate advances in the
state of the art in such areas as modeling of damage in composites. In
addition, all 50+ worked examples have been updated to reflect the newest
version of ANSYS®. Including some use of MATLAB®, these examples
demonstrate how to use the concepts to formulate and execute finite element
analyses and how to interpret the results in engineering terms. In addition,
the source code for each example is available to students for download
online via a companion website featuring a special area reserved for
instructors. A solutions manual is available upon qualifying course adoption.
Ever J. Barbero
Cementing applied computational and analytical experience to a firm
foundation of basic concepts and theory, Finite Element Analysis of
Composite Materials Using ANSYS®, Second Edition offers a
modern, practical, and versatile classroom tool for today’s engineering
classroom.
SECOND
EDITION
K15077
ISBN: 978-1-4665-1689-2
90000
9 781466 516892
K15077_COVER_final.indd 1 11/8/13 9:13 AM

(Barbero, Ever J.) Finite Element Analysis of Comp (B-Ok - Xyz)

Uploaded by

Copyright:

Available Formats

(Barbero, Ever J.) Finite Element Analysis of Comp (B-Ok - Xyz)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Barbero, Ever J.) Finite Element Analysis of Comp (B-Ok - Xyz)

Uploaded by

Copyright:

Available Formats

What challenges are professors faced with when teaching composite materials?

What challenges are professors faced with when teaching composite materials?

What makes the textbook a modern and versatile classroom tool?

What makes the textbook a modern and versatile classroom tool?

Materials Science/Mechanical Engineering/Civil Engineering

FINITE ELEMENT ANALYSIS Barbero SECOND EDITION

FINITE ELEMENT ANALYSIS OF COMPOSITE MATERIALS USING ANSYS®

FINITE ELEMENT ANALYSIS

The Only Finite Element Analysis Book on the Market Using

By layering detailed theoretical and conceptual discussions with fully

K15077_COVER_final.indd 1 11/8/13 9:13 AM

“K15077” — 2013/11/3 — 21:45

K15077_FM.indd 1 11/8/13 1:10 PM

K15077_FM.indd 2 11/8/13 1:10 PM

K15077_FM.indd 3 11/8/13 1:10 PM

No claim to original U.S. Government works

International Standard Book Number-13: 978-1-4665-1690-8 (eBook - PDF)

“K15077” — 2013/11/3 — 21:45

Dedicado a la memoria de mi padre, Ever Francisco Barbero

“K15077” — 2013/11/3 — 21:45

“K15077” — 2013/11/3 — 21:45

Series Preface xiii

List of Symbols xxi

List of Examples xxix

1 Mechanics of Orthotropic Materials 1

1.1 Lamina Coordinate System . . . . . . . . . . . . . . . . . . . . . . . 1

“K15077” — 2013/11/3 — 21:45

viii Finite Element Analysis of Composite Materials

1.12.3 Orthotropic Material . . . . . . . . . . . . . . . . . . . . . . . 20

2 Introduction to Finite Element Analysis 37

3 Elasticity and Strength of Laminates 63

“K15077” — 2013/11/3 — 21:45

3.2.7 LSS for Multidirectional Laminate FEA . . . . . . . . . . . . 82

5 Free Edge Stresses 133

6 Computational Micromechanics 151

“K15077” — 2013/11/3 — 21:45

x Finite Element Analysis of Composite Materials

7.1.7 Standard Nonlinear Solid . . . . . . . . . . . . . . . . . . . . 185

8 Continuum Damage Mechanics 207

9 Discrete Damage Mechanics 249

“K15077” — 2013/11/3 — 21:45

9.8 Solution Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263

A Tensor Algebra 297

B Second-Order Diagonal Damage Models 303

C Software Used 309

“K15077” — 2013/11/3 — 21:45

“K15077” — 2013/11/3 — 21:45

Half a century after their commercial introduction, composite materials are of

Ever Barbero, book series editor

“K15077” — 2013/11/3 — 21:45

“K15077” — 2013/11/3 — 21:45

Finite Element Analysis of Composite Materials Using ANSYS R

“K15077” — 2013/11/3 — 21:45

xvi Finite Element Analysis of Composite Materials

“K15077” — 2013/11/3 — 21:45

on. Their design is especially challenging because, unlike conventional materials

Ever J. Barbero, 2013

“K15077” — 2013/11/3 — 21:45

“K15077” — 2013/11/3 — 21:45

α Average engineering strain components