Damage and Failure of Composite Materials

Damage and Failure of Composite Materials
Understanding damage and failure of composite materials is critical for reliable

and cost-effective engineering design. Bringing together materials mechanics and
modeling, this book provides a complete guide to damage, fatigue, and failure of
composite materials. Early chapters focus on the underlying principles governing
composite damage, reviewing basic equations and mechanics theory, before
describing mechanisms of damage such as cracking, breakage, and buckling. In
subsequent chapters, the physical mechanisms underlying the formation and
progression of damage under mechanical loads are described with ample experi-
mental data, and micro- and macro-level damage models are combined. Finally,
fatigue of composite materials is discussed using fatigue-life diagrams. While there
is a special emphasis on polymer matrix composites, metal and ceramic matrix
composites are also described. Outlining methods for more reliable design of
composite structures, this is a valuable resource for engineers and materials
scientists in industry and academia.
Ramesh Talreja is a Professor of Aerospace Engineering at Texas A&M University.
He earned his Ph.D. and Doctor of Technical Sciences degrees from the Technical
University of Denmark. He has contributed extensively to the elds of damage,
fatigue, and failure of composite materials by authoring numerous books and
book chapters as well as by editing several encyclopedic works.
Chandra Veer Singh is an Assistant Professor of Materials Science and Engineering at
the University of Toronto. He earned his Ph.D. in aerospace engineering from Texas
A&M University, and worked as a post-doctoral Fellow at Cornell University. His
research expertise is in damage mechanics of composite materials, atomistic
modeling, and computational materials science. His industry experience includes
R&D at GE Aircraft Engines.
Damage and Failure
of Composite Materials
RAMESH TALREJA
Texas A&M University
CHANDRA VEER SI NGH
University of Toronto
cambri dge uni versi ty press
Cambridge, New York, Melbourne, Madrid, Cape Town,
Singapore, Sa o Paulo, Delhi, Mexico City
Cambridge University Press
The Edinburgh Building, Cambridge CB2 8RU, UK
Published in the United States of America by
Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521819428
#R. Talreja and C. V. Singh 2012
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without
the written permission of Cambridge University Press.
First published 2012
Printed in the United Kingdom at the University Press, Cambridge
A catalogue record for this publication is available from the British Library
Library of Congress Cataloging-in-Publication Data
Talreja, R.
Damage and failure of composite materials / Ramesh Talreja, Chandra Veer Singh.
p. cm.
Includes bibliographical references.
ISBN 978-0-521-81942-8 (Hardback)
1. Composite materialsFatigue. 2. Composite materialsFracture.
I. Singh, Chandra Veer. II. Title.
TA418.9.C6T338 2012
620.1
0
126dc23
2011035578
ISBN 978-0-521-81942-8 Hardback
Cambridge University Press has no responsibility for the persistence or
accuracy of URLs for external or third-party internet websites referred to
in this publication, and does not guarantee that any content on such
websites is, or will remain, accurate or appropriate.
Contents
Preface page ix
1 Durability assessment of composite structures 1
1.1 Introduction 1
1.2 Historical development of damage mechanics of composites 3
1.3 Fatigue of composite materials 5
References 7
2 Review of mechanics of composite materials 9
2.1 Equations of elasticity 9
2.1.1 Straindisplacement relations 9
2.1.2 Conservation of linear and angular momenta 10
2.1.3 Constitutive relations 11
2.1.4 Equations of motion 15
2.1.5 Energy principles 15
2.2 Micromechanics 17
2.2.1 Stiffness properties of a unidirectional lamina 18
2.2.2 Thermal properties of a unidirectional lamina 19
2.2.3 Constitutive equations for a lamina 20
2.2.4 Strength of a unidirectional lamina 21
2.3 Analysis of laminates 24
2.3.1 Straindisplacement relations 25
2.3.2 Constitutive relationships for the laminate 26
2.3.3 Stresses and strains in a lamina within a laminate 28
2.3.4 Effect of layup conguration 28
2.4 Linear elastic fracture mechanics 29
2.4.1 Fracture criteria 30
2.4.2 Crack separation modes 31
2.4.3 Crack surface displacements 32
2.4.4 Relevance of fracture mechanics for damage analysis 33
References 34
3 Damage in composite materials 36
3.1 Mechanisms of damage 37
3.1.1 Interfacial debonding 37
3.1.2 Matrix microcracking/intralaminar (ply) cracking 39
3.1.3 Interfacial sliding 39
3.1.4 Delamination/interlaminar cracking 41
3.1.5 Fiber breakage 42
3.1.6 Fiber microbuckling 42
3.1.7 Particle cleavage 44
3.1.8 Void growth 44
3.1.9 Damage modes 45
3.2 Development of damage in composite laminates 46
3.3 Intralaminar ply cracking in laminates 49
3.4 Damage mechanics 50
References 52
4 Micro-damage mechanics 57
4.1 Introduction 57
4.2 Phenomena of single and multiple fracture: ACK theory 58
4.2.1 Multiple matrix cracking 61
4.2.2 Perfectly bonded ber/matrix interface: a modied shear
lag analysis 65
4.2.3 Frictional ber/matrix interface 67
4.3 Stress analysis (boundary value problem) for cracked laminates 68
4.3.1 Complexity and issues 68
4.3.2 Assumptions 71
4.4 One-dimensional models: shear lag analysis 73
4.4.1 Initial shear lag analysis 74
4.4.2 Interlaminar shear lag analysis 77
4.4.3 Extended shear lag analysis 79
4.4.4 2-D shear lag models 80
4.4.5 Summary of shear lag models 80
4.5 Self-consistent scheme 84
4.6 2-D stress analysis: variational methods 87
4.6.1 Hashins variational analysis 87
4.6.2 Effect of residual stresses 96
4.6.3 [0
m
/90
n
]
s
vs. [90
n
/0
m
]
s
laminates 97
4.6.4 Improved variational analysis 97
4.6.5 Related works 101
4.6.6 Comparison between 1-D and 2-D stress-based models 101
4.7 Generalized plain strain analysis McCartneys model 104
vi Contents
4.8 COD-based methods 110
4.8.1 3-D laminate theory: Gudmundsons model 111
4.8.2 LundmarkVarna model 117
4.9 Computational methods 119
4.9.1 Finite element method (FEM) 120
4.9.2 Finite strip method 121
4.9.3 Layerwise theory 123
4.10 Other methods 124
4.11 Changes in thermal expansion coefcients 125
4.12 Summary 126
References 126
5 Macro-damage mechanics 134
5.1 Introduction 134
5.2 Continuum damage mechanics (CDM) of composite materials 138
5.2.1 RVE for damage characterization 139
5.2.2 Characterization of damage 141
5.2.3 A thermodynamics framework for materials response 144
5.2.4 Stiffnessdamage relationships 148
Case 1: Cracking in one off-axis orientation 152
Case 2: Cross-ply laminates 152
Evaluation of material constants 153
5.3 Synergistic damage mechanics (SDM) 155
5.3.1 Two damage modes 156
5.3.2 Three damage modes 165
5.4 Viscoelastic composites with ply cracking 170
5.5 Summary 176
References 177
6 Damage progression 179
6.2 Experimental techniques 180
6.3 Experimental observations 185
6.3.1 Initiation of ply cracking 185
6.3.2 Crack growth and multiplication 187
6.3.3 Crack shapes 189
6.3.4 Effect of cracking 189
6.3.5 Loading and environmental effects 191
6.3.6 Cracking in multidirectional laminates 193
6.4 Modeling approaches 194
6.4.1 Strength-based approaches 194
6.4.2 Energy-based approaches 198
6.4.3 Strength vs. energy criteria for multiple cracking 210
vii Contents
6.5 Randomness in ply cracking 211
6.6 Damage evolution in multidirectional laminates 217
6.7 Damage evolution under cyclic loading 223
6.8 Summary 229
References 230
7 Damage mechanisms and fatigue-life diagrams 237
7.2 Fatigue-life diagrams 237
7.3 On-axis fatigue of unidirectional composites 238
7.4 Effects of constituent properties 241
7.5 Unidirectional composites loaded parallel to the bers 242
7.5.1 Polymer matrix composites (PMCs) 242
7.5.1.1 Experimental studies of mechanisms 247
7.5.2 Metal matrix composites (MMCs) 250
7.5.3 Ceramic matrix composites (CMCs) 252
7.6 Unidirectional composites loaded inclined to the bers 257
7.7 Fatigue of laminates 259
7.7.1 Angle-ply laminates 260
7.7.2 Cross-ply laminates 261
7.7.3 General multidirectional laminates 263
7.8 Fatigue-life prediction 265
7.8.1 Cross-ply laminates 266
7.8.2 General laminates 273
7.9 Summary 273
References 274
8 Future directions 276
8.1 Computational structural analysis 276
8.2 Multiscale modeling of damage 278
8.2.1 Length scales of damage 280
8.2.2 Hierarchical multiscale modeling 282
8.2.3 Implication on multiscale modeling: Synergistic damage
mechanics 286
8.3 Cost-effective manufacturing and defect damage mechanics 287
8.3.1 Cost-effective manufacturing 288
8.3.2 Defect damage mechanics 291
8.4 Final remarks 296
References 298
Author index 301
Subject index 303
viii Contents
Preface
The eld of composite materials has advanced steadily from the early develop-
ments during the 1970s when laminate plate theory and anisotropic failure criteria
were in focus to todays diversication of composite materials to multifunctional
and nanostructured composite morphologies. Throughout the 1970s and 1980s
several books appeared along with courses that were developed and taught at
advanced levels dealing with mechanics of composite materials and structures.
The failure analysis was mostly limited to descriptions of strength that extended
previous continuum descriptions of metal yielding and failure. Beginning around
the mid-1980s, micromechanics and continuum damage mechanics were applied
to multiple cracking observed in composite materials. Under the overall descrip-
tion of damage mechanics a urry of activities took place as evidenced by
conferences and symposia. Other than several conference proceedings that
recorded such activities, a collection of seminal contributions to the eld appeared
in a volume (Damage Mechanics of Composite Materials, R. Talreja, ed., Composite
Materials Series, R.B. Pipes, series ed., Vol. 9, Amsterdam: Elsevier Science
Publishers, 1994). The two main avenues of approach to damage in composite
materials and its effect on materials response, now referred to as micro-damage
mechanics (MIDM) and macro-damage mechanics (MADM), were presented in a
balanced form in that volume. In the years since then, many developments have
taken place that have brought this eld to such level of maturity that a book
coherently presenting the material was felt to be timely. It is hoped that this book
will help provide impetus for teaching advanced courses in composite damage at
universities as well as support short courses for professional development of
engineers in industry. The wealth of material covered can also help new research-
ers in advancing the eld further. To this end, the last chapter provides some
guidance in identifying gaps and needs for further work.
The structure of the book is as follows. Chapter 1 lays down the overall strategy
for durability assessment of composite structures, emphasizing the needs and
motivating the content of the book to follow. Chapter 2 provides an easy reference
to the basic continuum mechanics topics that are felt to be relevant to the
subsequent treatment. Chapter 3 describes the mechanisms of damage that under-
lie the phenomena aimed for modeling. Many of the physical observations
described there are viewed to be vital to developing proper understanding of the
complex eld of damage in composite materials. Chapters 4 and 5 deal with the
two main approaches stated above, i.e., the MIDM and the MADM. Selected
works from the literature, including the authors own, are given as much treatment
as was found justied to generate coherency without overly including details.
While these two chapters focus on descriptions of damage and the constitutive
property changes caused by it, Chapter 6 is devoted to the progression of damage.
The crack multiplication is a distinctive feature of damage in composite materials
that distinguishes it from single crack growth in monolithic materials, and there-
fore justies treatment in a chapter by itself. Chapter 7 is on fatigue of composite
materials. This eld suffers from the historical treatments of metal fatigue and is
unfortunately the least understood part of damage in composite materials. Multi-
axial fatigue illustrates the situation well where the literature displays little under-
standing of the mechanisms underlying failure. While a separate book on fatigue
of composite materials is needed to do full justice to the eld, a single chapter here
is added to draw attention to the mechanisms-based concepts for proper interpret-
ation and modeling. Finally, Chapter 8 presents a summary of the book and
points to the directions in which further advances are seen to be necessary.
Particular emphasis is given to the computational incorporation of damage mod-
eling in durability assessment as well as taking account of the manufacturing-
induced defects in an integrated manner.
Although the authors have written this book, the credit goes to many research-
ers who have worked on various aspects of damage in composite materials. Their
collective contributions have made it possible for us to present what we have seen
as a coherent story at this time. The eld is evolving, and future versions of the
story will hopefully spur further development and, most importantly, transfer of
this knowledge to industry will take place.
x Preface
1 Durability assessment of composite
structures
1.1 Introduction
Composite structures for mechanical and aerospace applications are designed to
retain structural integrity and remain durable for the intended service life. Since
the early 1970s important advances have been made in characterizing and model-
ing the underlying mechanical behavior and developing tools and methodologies
for predicting the fracture and fatigue of composite materials. This book provides
an exposition of the concepts and analyses related to this area and presents recent
results. The next chapters treat damage in composite materials as observed by a
variety of techniques, followed by modeling at the micro and macro levels. Fatigue
is treated separately because of its particular complexities that require systematic
interpretation schemes developed for the purpose. A chapter is added in the
beginning to provide convenient access to the mechanics concepts needed for the
modeling analyses in later chapters.
Here we present an overview of the durability assessment process for composite
structures. Figure 1.1 depicts the connectivity and ow of the elements of this
process. To begin, one usually conducts stress analysis of the component using the
initial constitutive behavior of the composite along with the service loading on
the component as input. In contrast to monolithic materials, such as metals, the
constitutive behavior of a composite can change due to damage incurred in
service. The stress analysis combined with prior experience allows identifying
critical sites (hot spots) in the component that are prone to be the sites of
failure. Further examination of these sites in terms of the local stress/strain/
temperature excursions combined with the composite material composition at
those sites helps to identify the possible mechanisms of damage that can result.
Examples of such mechanisms are microcracking of the matrix, delamination
(separation of layers at interfaces), aging (of the polymer matrix), etc. Chapter 3
describes these mechanisms in some detail. The next step is to analyze the conse-
quences of the mechanisms on the material response and in turn on the structural
performance. Chapters 4 and 5 deal with different models to predict the damage-
induced material response changes. Since the scales at which damage occurs are
small in comparison to the characteristic geometrical size of the hot spots,
models must account for the multiple length (or size) scales. The differentiation
of scales is conventionally described as micro (the scale of damage) and
macro (the scale at which structural response is characterized). Since connectiv-
ity between these scales must be established, an intermediate scale called meso is
dened as needed by the particular model used. In micromechanics the concept of
representative volume element (RVE) has been proposed. The size of this
element is commonly taken to be the meso scale. Chapters 4 and 5 describe
the three scales in the context of different models. Chapter 6 is focused on the
initiation and progression of damage. Together the three chapters provide the
content of the subject known as damage mechanics, which as indicated in
Figure 1.1 is central to durability assessment.
The common output of the damage mechanics models is a description of the
material response, often described as stiffness degradation, caused by damage.
This description necessarily involves averaging over the so-called RVE. Thus the
materials response, or averaged constitutive behavior description, forms the new
input to the stress analysis that was conducted initially using pristine (undamaged)
material properties. The resulting iterative process of stress analysis should be an
inherent feature of composite structural analysis, although the industry practice
currently does not fully implement this procedure. Another output of the damage
mechanics analysis is strength degradation, i.e., reduction in the load-bearing
capability of the structure due to damage. Depending on the functional requirements
of a given structure, degradation of stiffness or strength would be the path to loss
of structural integrity. A typical example of a stiffness-critical structure is an
aircraft wing that must deform appropriately to perform its aerodynamic function,
while a fuselage is strength-critical as its design requirement is to contain the pressure
within it.
While monolithic materials such as metals fail due to unstable growth of a
crack, the heterogeneous internal structure of a composite leads to formation of
multiple cracks. A generic heterogeneous solid is illustrated in Figure 1.2 in three
states: pristine (undamaged) to the left in the gure shows a representative region
Figure 1.1. A durability analysis scheme for composite structural components.
2 Durability assessment of composite structures
of the solid within which heterogeneities (reinforcements) are indicated symbolic-
ally as lled circles, and two states that have multiple cracks resulting from
debonding of reinforcements (middle gure) and from local failure of the matrix
induced by defects and/or stress concentrations. Consider the external loading on
a composite structure resulting in tractions t on the surface bounding the repre-
sentative region of the composite shown. If the response to these tractions in terms
of the bounding surface displacements is given by u in the pristine state, then the
surface displacements of the multiple cracks (commonly expressed as crack
opening displacements, COD, and crack sliding displacement, CSD) within
the volume will change this to u
1
or u
2
depending on the type of damage (see
Figure 1.2). The local environment around the cracks inuences the COD and
CSD of distributed cracks within the volume. This local environment is typically
described as a constraint (i.e., moderation) to the crack surface displacements
and is expressed in terms of the variables of heterogeneities. If the heterogeneous
solid with multiple cracks is homogenized over the representative region, then the
stressstrain response averaged over the RVE is given by the averaged stiffness
properties that change (degrade) with increasing number of cracks and the con-
straint to the crack surface displacements. This stiffness degradation is the subject
of damage mechanics, as discussed above in describing the durability assessment
procedure depicted in Figure 1.1.
1.2 Historical development of damage mechanics of composites
Although the eld of solid mechanics applied to heterogeneous solids was developed
in the late 1950s and early 1960s, and became known as micromechanics, the specic
situations encountered in composite materials such as those with continuous ber
reinforcements were not addressed until much later. The concepts developed in
micromechanics turned out to be useful for multiple cracking in composite materials
and are recommended as essential background (see the text by Nemat-Nasser and
Hori [1]). However, the rst pioneering work that claried the phenomenon of
multiple cracking in the presence of ber/matrix interfaces in reinforced composites
was by Aveston et al. [2] published as a conference proceedings paper in 1971.
u
t
t
t
u
1
u
2
Figure 1.2. A heterogeneous solid in pristine (undamaged) state (left) and in two possible
multiple cracking states (middle and right).
3 1.2 Historical development of damage mechanics of composites
This work, which became known as the ACK theory, treated multiple parallel cracks
normal to bers in a matrix with all bers in one direction loaded in tension along
bers. The model produced an expression for the overall strain at which multiple
cracking occurs based on a simplied stress analysis and energy balance concepts.
The expression provided a basis for assessing the roles of ber and matrix properties,
their volume fractions, and the ber diameter in resisting multiple cracking.
The ACK model was motivated by the observation of multiple cracking in brittle
matrix composites such as cement reinforced with steel wires. For polymer matrix
composites, the application of the ACK theory was at rst not clear since a ply with
unidirectional bers of glass or carbon does not have the right conditions for multiple
cracking when loaded in tension along bers. Garrett and Bailey in 1977 [3] found
that the multiple cracking observed in cross-ply laminates of glass ber-reinforced
polyester under axial tension could in fact be described well by the case of fully
bonded interfaces treated by Aveston and Kelly [4] in a follow on paper to the ACK
model. This required replacing the matrix and bers in that model by the transverse
and longitudinal plies, respectively. Garrett and Bailey then repeated with appropri-
ate modication the one-dimensional stress analysis and energy balance consider-
ations used in [4]. Thus began a long series of works that applied the one-dimensional
stress analysis, known as shear lag analysis, which assumes axial load transfer from
cracked to uncracked plies by the shear stress at the interfaces.
The inadequacy of the shear lag analysis to properly provide stresses in the
cracked cross-ply laminate was a severe limitation until a variational analysis-
based two-dimensional approximation appeared in the English literature [5]. This
spurred further work of more accuracy [6] and extension to partially debonded
frictional interfaces [7], while extension to cracked plies of other than transverse
orientation required other approaches [8]. The analyses that use local ply stress
solutions to evaluate overall stiffness degradation are grouped together in micro-
damage mechanics (MIDM) and are treated in Chapter 4.
In some ways parallel to the MIDM emerged another approach that became
known as continuum damage mechanics (CDM). Its beginnings are not attrib-
uted to composite materials but to metals undergoing creep. Kachanov in 1958
[9] put forth a concept of a eld of internal material discontinuity responsible
for distributed local stress enhancement leading to overall creep strain. Later,
the internal state was called damage and a (hidden) scalar variable D was
associated with it. The continuum now had an internal damage state and
because of its irreversible nature its treatment required thermodynamics, in
particular the Second Law, which places conditions on the entropy changes.
Kachanovs work stayed relatively unknown until Lemaitre and Chaboche [10]
applied it to analysis of various structural materials with distributed cavities
and cracks. Krajcinovic [11] further enhanced the eld by connecting it to
concepts known from fracture mechanics and plasticity and by elaborating
the thermodynamics implications. For composite materials of technological
interests that are constructed with specic symmetries such as orthotropic, the
rst work to apply CDM was by Talreja [12] and its companion paper that
validated the stiffness degradation relationships by experimental data [13]. The
CDM concept for composites is illustrated in Figure 1.3. The reinforcements in
a composite are regarded as a stationary microstructure and are homogenized
as an anisotropic medium in which the damage entities such as cracks are
embedded. Further homogenization is done by smearing out the damage
entities into an internal eld, which is represented by a pair of vectors, whose
dyadic product averaged over all damage entities in a RVE provides the
characterization of damage. Since the early papers [12, 13] the CDM eld for
composite materials has developed steadily, more recently in a version named
as synergistic damage mechanics (SDM) where the micromechanics is judi-
ciously applied to enhance the applicability of CDM. All this is the subject of
Chapter 5 on macro-damage mechanics (MADM).
As depicted in Figure 1.1, the stress analysis of critical structural sites
requires stressstrain relationships that reect the presence of damage. These
relationships are developed by combining stiffness degradation and damage evo-
lution. The subject of damage evolution is complex with its own challenges.
Therefore Chapter 6 is devoted exclusively to its treatment.
1.3 Fatigue of composite materials
It is natural to assume that the complexities of damage in composite materials
observed under quasi-static loading would be enhanced when the loading is
applied in a cyclic manner. The experience with metal fatigue indicates that the
fracture surface of a sample failed in fatigue shows distinctly different features
than if failed in the application of a monotonically increasing load. The fracture
surface of a unidirectional ber-reinforced composite loaded along bers mono-
tonically or cyclically does not give clear indication of mechanisms preceding
Stationary microstructure
Homogenization
of
stationary
microstructure
Homogenized continuum
with damage
Damage entity
Evolving
microstructure
RVE
a
n
P
Figure 1.3. Illustration of the CDM concept for composites.
5 1.3 Fatigue of composite materials
failure in either case. In more general ber architectures, such as laminates and
woven fabric composites, following the events from the rst (initiation) to the last
(separation by breakage) is generally difcult. However, as advances in nondes-
tructive observation techniques are made, increasing clarity in mechanisms is
emerging. In the early years of composite fatigue studies in the 1970s, little was
understood of mechanisms and consequently the assumptions made in predictive
models were speculative at best.
One study of a unidirectional glass/epoxy composite made assumptions of
fatigue mechanisms that led to reasonable explanation of the trends in fatigue
life [14]. Following that work, a systematic conceptual framework for interpret-
ation of fatigue damage and failure was proposed by Talreja [15] for more
general cases of loading as well as for more general ber orientations. The
framework took the form of a two-dimensional plot called a fatigue life dia-
gram in which regions of dominant mechanisms were separated. The diagram is
not meant to be a data-tted S-N curve (historically known as a Wo hler dia-
gram) but as a means of interpreting the roles of bers, matrix, and interfaces as
well as of laminate conguration parameters such as ply orientation, sequence,
and thickness.
Since the unidirectional composite (or ply) is a basic unit in laminates, the
fatigue life diagram for this composite under tensiontension loading forms the
baseline diagram from which more general cases evolve. This diagram is illustrated
in Figure 1.4 and discussed in detail in Chapter 7. As shown, the vertical axis of
the diagram is the maximum strain attained at the rst application of maximum
stress in a load controlled fatigue test. This quantity forms a proper reference to
the loading condition and provides upper and lower limits to the fatigue behavior.
Thus the strain to failure (of ber) forms the upper limit while the strain corres-
ponding to the fatigue limit (primarily a matrix property) forms the lower limit.
These strain values can always be converted to applied stress, but plotting these in
the diagram allows a systematic and proper interpretation of the roles of the
constituents. The regions indicated in the fatigue life diagram provide clarity of
the governing mechanisms dictated by the constituent properties. The construc-
tion of the diagram for unidirectional composites was initially based on systematic
arguments and logical deduction. Physical evidence to support the diagram was
later presented by an elaborate and tedious experimental study [16].
The fatigue life diagram can also serve the purpose of facilitating mechanisms-
based life prediction modeling. For cross-ply laminate this was demonstrated
in [17]. Generally the path to predictive modeling with account of the underlying
damage mechanisms is long and hard. Consequently, the literature has a prepon-
derance of studies that resort to failure criteria that are mostly extensions of
those for static failure with assumed procedures without fundamental validation.
The models are therefore not reliable enough to extend beyond the cases that
formed the impetus for the proposed schemes.
Chapter 7 treats the subject of composite damage with emphasis on mechan-
isms. It is not exhaustive in the sense of including the literature on models for life
prediction. A recent paper [18] has a fairly thorough examination of the main
models for multiaxial fatigue. It reveals the frustrating situation of lack of reliabil-
ity of the models. In Chapter 7 the main ndings of this review are discussed and a
mechanisms-based methodology is proposed.
References
1. S. Nemat-Nasser and M. Hori, Micromechanics: Overall Properties of Heterogeneous
Materials. (Amsterdam: Elsevier, 1993).
2. J. Aveston, G. A. Cooper, and A. Kelly, Single and multiple fracture. In The Properties
of Fiber Composites. (Surrey, UK: IPC Science and Technology Press, National Phys-
ical Laboratory, 1971), pp. 1526.
3. K. W. Garrett and J. E. Bailey, Multiple transverse fracture in 90 degrees cross-ply
laminates of a glass ber-reinforced polyester. J Mater Sci, 12:1 (1977), 15768.
4. J. Aveston and A. Kelly, Theory of multiple fracture of brous composites. J Mater Sci,
8:3 (1973), 35262.
5. Z. Hashin, Analysis of cracked laminates: a variational approach. Mech Mater, 4:2
(1985), 12136.
6. J. Varna and L. A. Berglund, Multiple transverse cracking and stiffness reduction in
cross-ply laminates. J Compos Tech Res, 13:2 (1991), 97106.
7. N. V. Akshantala and R. Talreja, A mechanistic model for fatigue damage evolution in
composite laminates. Mech Mater, 29 (1998), 12340.
8. L. N. McCartney, Model to predict effects of triaxial loading on ply cracking in general
symmetric laminates. Compos Sci Technol, 60 (2000), 225579 (see Errata in Compos Sci
Technol, 62:9 (2002), 12734).
9. L. M. Kachanov, On the creep rupture time. Izv Akad Nauk SSR, Otd Tekhn Nauk,
8 (1958), 2631.
10. J. Lemaitre and J. L. Chaboche, Mechanique des Materiaux Solide. (Paris: Dunod, 1985).
11. D. Krajcinovic, Continuous damage mechanics. Appl Mech Rev, 37 (1984), 15.
12. R. Talreja, A continuum-mechanics characterization of damage in composite-
materials. Proc R Soc London A, 399:1817 (1985), 195216.
Figure 1.4. Fatigue life diagram of a unidirectional ber-reinforced composite subjected to
cyclic tension in the ber direction.
7 References
13. R. Talreja, Transverse cracking and stiffness reduction in composite laminates.
J Compos Mater, 19 (1985), 35575.
14. C. K. H. Dharan, Fatigue failure mechanisms in a unidirectionally reinforced compos-
ite material. In Fatigue in Composite Materials, ASTM STP 569. (Philadelphia, PA:
ASTM, 1975), pp. 17188.
15. R. Talreja, Fatigue of composite materials: damage mechanisms and fatigue-life dia-
grams. Proc R Soc London A, 378 (1981), 46175.
16. E. K. Gamstedt and R. Talreja, Fatigue damage mechanisms in unidirectional carbon
bre-reinforced plastics. J Mater Sci, 34 (1999), 253546.
17. N. V. Akshantala and R. Talreja, A micromechanics based model for predicting fatigue
life of composite laminates. Mater Sci Eng A, 285 (2000), 30313.
18. M. Quaresimin, L. Susmel, and R. Talreja, Fatigue behaviour and life assessment of
composite laminates under multiaxial loadings. Int J Fatigue, 32 (2010), 216.
2 Review of mechanics of
composite materials
In this chapter the fundamental aspects of elasticity, strength, and fracture of
composite solids are reviewed. Although this information is available in numerous
texts, more comprehensively and in greater detail than here, a brief exposition is
provided for convenient reference. For further in-depth treatment, the reader may
consult, e.g., [15] for theory of elasticity and continuum mechanics, [612] for
mechanics of composite materials, and [1317] for fracture mechanics.
2.1 Equations of elasticity
2.1.1 Straindisplacement relations
Figure 2.1 illustrates the initial and deformed congurations of a body whose
representative material point P is described with respect to a xed rectangular
Cartesian frame by coordinates X
j
and x
i
respectively, j, i = 1, 2, 3. The compon-
ents of displacement of the point are given by
u
i
= x
i
X
j
d
ij
; (2:1)
where X
j
are the coordinates of the material point in the initial undeformed
conguration, x
i
are the coordinates of the material point in the nal deformed
conguration, and d
ij
is the Kronecker delta. The Lagrangian description of
displacement at time t is expressed in terms of the X
j
coordinates as
u
i
= x
i
X
1
; X
2
; X
3
; t ( ) X
j
d
ij
: (2:2)
The components of the GreenLagrange strain tensor are given by
E
ij
=
1
2
u
i;j
u
j;i
u
i;k
u
j;k
_ _
; (2:3)
where u
i;j
=
@u
i
@X
j
; etc., and repeated indices imply summation.
When u
i;j
1, E
ij
reduces to the innitesimal strain tensor e
ij
given by
e
ij
=
1
2
@u
i
@x
j
@u
j
@x
i
_ _
=
1
2
u
i;j
u
j;i
_ _
: (2:4)
From Eq. (2.4) it is seen that the strain tensor is symmetric. Thus, there are six
independent strain components, which in the innitesimal version are three normal
strains (e
11
, e
22
, and e
33
), and three shear strains (e
12
= e
21
, e
23
= e
32
, and e
13
= e
31
).
To ensure single-valued displacements u
i
, the strain components e
ij
cannot be
assigned arbitrarily but must satisfy certain integrability or compatibility condi-
tions, given by
e
ij;kl
e
kl;ij
e
ik;jl
e
jl;ik
= 0: (2:5)
Of the 81 equations included in Eq. (2.5), only six are independent. The remainder
are either identities or repetitions due to symmetry of e
ij
. For the special case of
plane stress conditions, the only surviving compatibility equation is
e
11;22
e
22;11
2e
12;12
= 0: (2:6)
2.1.2 Conservation of linear and angular momenta
Ingeneral, the forces exertedona continuumbody are body forces andsurface forces.
Body forces, such as gravitational and magnetic forces, act on all particles within the
volume of the body and are described in terms of force intensity per unit mass or per
unit volume, while surface forces are contact forces that act across an internal surface
or an external (bounding) surface. The continuum description of surface forces is
given by the traction vector t acting on a surface element dS with a unit normal n (see
Figure 2.2(a)). Let dP be the total force exerted on dS by the material points on the
side of dS toward which n is pointing. The traction vector t is then dened as
t = lim
dS0
dP
dS
: (2:7)
At an internal point P there are innitely many surface elements, each with a different
unit normal vector. According tothe Cauchy theorema tractionvector onany of these
Initial configuration
Deformed configuration
X
x
u
X
2
X
1
O
P
P
X
3
Figure 2.1. Initial and deformed geometry of a continuum body.
10 Review of mechanics of composite materials
planes can be expressed in terms of the traction vectors on three orthogonal planes
passingthroughthe point P. InaCartesianreference frame, the three planes are chosen
parallel to the coordinate planes and the resultant traction vectors on these planes are
decomposed along the three coordinate axes. These 3 3 = 9 components taken
together form the components of the second rank stress tensor associated with the
considered point P. They are indicated in Figure 2.2(b) where their positive directions
are shown. Inindex notation, they are denoted by s
ij
, where the rst index refers to the
directionof the unit normal onthe surface (the face of the cube inFigure 2.2(b)) andthe
second index stands for the direction of the resolved traction component. The stress
components withtwoequal indices, e.g., s
11
, are callednormal stresses while those with
unequal indices, e.g., s
23
, are termed shear stresses.
The traction vector components are related to the stress tensor components by
the following equation
t
i
= s
ij
n
j
; (2:8)
where n
j
are components of the normal vector associated with the traction vector.
The conservation of linear momentum at a material point inside the continuum
body gives the following relation
s
ji; j
f
i
= r u
i
; (2:9)
where f
i
are components of the body force vector, and r is the mass density. For
quasi-static problems the right-hand side of Eq. (2.9) vanishes, and if the body
forces are neglected, the equations of equilibrium reduce to
s
ji; j
= 0: (2:10)
When there are no body moments, the conservation of angular momentum results
in the symmetry of the stress tensor, i.e.,
s
ij
= s
ji
: (2:11)
2.1.3 Constitutive relations
For an elastic material there exists a positive-denite, single-valued, potential
function of strains e
kl
, dened as
s
22
s
23
s
21
s
13
s
11
s
32
s
31
P
n
t
dS
(a) (b)
s
33
s
12
Figure 2.2. (a) Traction vector; (b) a volume element with components of stress tensor.
11 2.1 Equations of elasticity
U =
_
e
kl
0
s
ij
de
ij
: (2:12)
This function is termed as the strain energy density. U is independent of the
loading path and thus a function of nal strains only. Differentiating Eq. (2.12)
with respect to the strains, the stress tensor can be written as
s
ij
=
@U
@e
ij
: (2:13)
If we consider a linear elastic material, then U can be written as a quadratic
function in e
kl
U e
kl
( ) =
1
2
C
ijkl
e
ij
e
kl
; (2:14)
where C
ijkl
is a fourth-order tensor of material stiffness coefcients known as
the stiffness tensor. Using Eqs. (2.13) and (2.14), one obtains the generalized
Hookes law
s
ij
= C
ijkl
e
kl
: (2:15)
A potential function of stresses known as the complementary energy density is
dened as
U
+
s
ij
_ _
= s
ij
e
ij
U: (2:16)
Differentiation of Eq. (2.16) with respect to stress tensor yields the relation
e
ij
=
@U
+
@s
ij
: (2:17)
Analogous to Eq. (2.14), U
*
can also be represented as a quadratic function as
U
+
s
ij
_ _
=
1
2
S
ijkl
s
ij
s
kl
; (2:18)
where S
ijkl
are components of the compliance tensor. Using Eqs. (2.17) and (2.18),
one obtains the inverse constitutive law
e
ij
= S
ijkl
s
kl
: (2:19)
In all, the stiffness matrix C
ijkl
has 81 coefcients. However, not all of these
coefcients are independent. Note rst that the symmetry of the strain compon-
ents (e
kl
= e
lk
) leads to C
ijkl
= C
ijlk
, which reduces the number of coefcients from
81 to 54. Similarly, the symmetry of the stress tensor further reduces the number of
these coefcients to 36. Finally, differentiating Eq. (2.14) twice with respect to
strains, one obtains
C
ijkl
=
@
2
U
@e
ij
@e
kl
: (2:20)
Since the order of differentiation in the above equation is arbitrary, one infers
that
C
ijkl
= C
klij
; (2:21)
which reduces the number of independent material coefcients to 21.
The coefcient matrix C
ijkl
is expressed in compact form by using the Voigt
notation, in which stress and strain tensor components are denoted using a single
subscript, whereas two subscripts are used to denote the stiffness tensor. With this,
the constitutive relation, Eq. (2.15), can be written as s
p
= C
pq
e
q
; p, q = 1,2, . . ., 6,
or in expanded matrix form as
s
1
s
2
s
3
s
4
s
5
s
6
_
_
_
_
=
C
11
C
12
C
13
C
14
C
15
C
16
C
21
C
22
C
23
C
24
C
25
C
26
C
31
C
32
C
33
C
34
C
35
C
36
C
41
C
42
C
43
C
44
C
45
C
46
C
51
C
52
C
53
C
54
C
55
C
56
C
61
C
62
C
63
C
64
C
65
C
66
_
_
_
_
e
1
e
2
e
3
e
4
e
5
e
6
_
_
_
_
; (2:22)
where C
pq
= C
qp
, and
s
1
= s
11
; s
2
= s
22
; s
3
= s
33
; s
4
= s
23
; s
5
= s
31
; s
6
= s
12
e
1
= e
11
; e
2
= e
22
; e
3
= e
33
; e
4
= 2e
23
; e
5
= 2e
31
; e
6
= 2e
12
C
11
= C
1111
; C
22
= C
2222
; : : : ; etc:
(2:23)
These constitutive relationships are for an anisotropic material. If material sym-
metry exists, then further reduction occurs in the number of independent coef-
cients of the stiffness matrix. It should be noted that the stiffness matrix in the
Voigt notation does not follow the transformation rule for tensors. The fourth-
order stiffness tensor C
ijkl
transforms as
C
/
ijkl
=
ip
jq
kr
ls
C
pqrs
; (2:24)
where
ij
is the matrix of direction cosines associated with coordinate trans-
formation from one coordinate system (x
1
, x
2
, x
3
) to another (x
/
1
; x
/
2
; x
/
3
).
A material with one plane of symmetry is called monoclinic, and if this plane is
parallel to the x
1
x
2
plane then it can be shown that the constitutive relation is
given by
s
1
s
2
s
3
s
4
s
5
s
6
_
_
_
_
=
C
11
C
12
C
13
0 0 C
16
C
21
C
22
C
23
0 0 C
26
C
31
C
32
C
33
0 0 C
36
0 0 0 C
44
C
45
0
0 0 0 C
54
C
55
0
C
61
C
62
C
63
0 0 C
66
_
_
_
_
e
1
e
2
e
3
e
4
e
5
e
6
_
_
_
_
: (2:25)
Here, the stiffness matrix has 13 independent material coefcients. If a mater-
ial has two mutually orthogonal planes of symmetry, then the plane
orthogonal to these planes is also a plane of symmetry. In this case, the
material symmetry is described as orthotropic, and the number of independent
constants in the stiffness matrix reduces to nine. The stressstrain relations
when the symmetry planes are parallel to the three coordinate planes take the
following form
s
1
s
2
s
3
s
4
s
5
s
6
_
_
_
_
=
C
11
C
12
C
13
0 0 0
C
21
C
22
C
23
0 0 0
C
31
C
32
C
33
0 0 0
0 0 0 C
44
0 0
0 0 0 0 C
55
0
0 0 0 0 0 C
66
_
_
_
_
e
1
e
2
e
3
e
4
e
5
e
6
_
_
_
_
: (2:26)
In terms of the engineering elastic constants the inverse strainstress relations for
the orthotropic case become as follows
e
1
e
2
e
3
e
4
e
5
e
6
_
_
_
_
=
1
E
1
n
21
E
2
n
31
E
3
0 0 0
n
12
E
1
1
E
2
n
32
E
3
0 0 0
n
13
E
1
n
23
E
2
1
E
3
0 0 0
0 0 0
1
G
23
0 0
0 0 0 0
1
G
31
0
0 0 0 0 0
1
G
12
_
_
_
_
s
1
s
2
s
3
s
4
s
5
s
6
_
_
_
_
; (2:27)
where E
1
, E
2
, E
3
are Youngs moduli in the three material symmetry directions
(x
1
, x
2
, x
3
) respectively, n
ij
; i ,= j; are the six Poissons ratios dened in the
conventional way, e.g., n
12
= e
2
/ e
1
with s
1
applied, and G
23
, G
31
, and G
12
are
shear moduli in the x
2
x
3
, x
1
x
3
, and x
1
x
2
planes, respectively. The compliance
matrix in Eq. (2.27), being the inverse of a symmetric matrix, is also symmetric.
From this symmetry follows the reciprocal relationship,
n
ij
E
i
=
n
ji
E
j
(no sumon i; j); (2:28)
which can be used to eliminate three of the six Poissons ratios.
If the material is isotropic in a plane, i.e., with same elastic properties in all
directions in the plane, it is called transversely isotropic. Let the x
2
x
3
plane be the
plane of isotropy, i.e., E
3
= E
2
; n
31
= n
12
; G
31
= G
12
; G
23
=
E
2
2 1 n
23
( )
: The com-
pliance tensor is then given by
S [ [ =
S
11
S
12
S
12
0 0 0
S
12
S
22
S
23
0 0 0
S
12
S
23
S
22
0 0 0
0 0 0 2 S
22
S
23
( ) 0 0
0 0 0 0 S
66
0
0 0 0 0 0 S
66
_
_
_
_
=
1
E
1
n
12
E
1
n
12
E
1
0 0 0
n
12
E
1
1
E
2
n
23
E
2
0 0 0
n
12
E
1
n
23
E
2
1
E
2
0 0 0
0 0 0
E
2
2 1 n
23
( )
0 0
0 0 0 0
1
G
12
0
0 0 0 0 0
1
G
12
_
_
_
_
:
(2:29)
As seen above, a transversely isotropic material has ve independent stiffness
coefcients, viz. E
1
, E
2
, n
23
, n
12
, and G
12
. For a completely isotropic material there
are only two independent material coefcients, namely the Youngs modulus (E)
and Poissons ratio (n) or, alternatively, the Lame constants (l and m). The
constitutive relations can now be written as
s
ij
= le
kk
d
ij
2me
ij
; (2:30)
where d
ij
is the Kronecker delta. Alternatively,
e
ij
=
1
E
1 n ( )s
ij
ns
kk
d
ij
_
: (2:31)
2.1.4 Equations of motion
The equations governing the motion of a deformable body can be obtained by
combining kinematic relations, Eq. (2.4), equilibrium equations, Eq. (2.10), and
the constitutive relations, Eq. (2.15). For the particular case of linear elastic
isotropic materials, they can be written as
l m ( )u
j; ji
mu
i; jj
f
i
= r u
i
: (2:32)
These equations are known as Naviers equations. The displacement eld obtained
from these equations is unique and results into the determination of strains and
stresses by use of kinematic and constitutive relations.
2.1.5 Energy principles
Energy principles for a continuum body allow formulating the relationships
between stresses, strains or deformations, displacements, material properties,
and external effects in the form of energy or work done by internal and external
forces. They are also useful for obtaining approximate solutions of complex
boundary value problems, e.g., nite element methods. Detailed treatment of these
concepts can be found in [1820].
Principle of virtual work
In the context of an elastic boundary value problem, consider a solid continuum
body (Figure 2.3), occupying a volume V and bounded by surface S = S
t
S
u
, to
be in static equilibrium under prescribed body forces f
i
over volume V, surface
tractions t
i
on S
t
, and displacements u
i
over remaining portion of the boundary S
u
.
For a statically admissible stress eld ~ s
ij
(such that ~ s
ij; j
= 0 in V, and
~
t
i
= ~ s
ij
n
j
on S
t
) and a kinematically admissible displacement eld ^ u
i
(such that
ê
ij
=
1
2
^ u
i; j
^ u
j;i
_ _
), the principle of virtual work states
_
S
~
t
i
^ u
i
dS
_
V
f
i
^ u
i
dV =
_
V
~ s
ij
ê
ij
dV: (2:33)
It should be noted that the displacement eld ^ u
i
and the stress eld ~ s
ij
are
completely independent of each other.
Principle of minimum potential energy
For a kinematically admissible displacement eld ^ u
i
, the potential energy of a
linear elastic continuum body under the action of conservative forces f
i
and
prescribed surface tractions t
i
on S
t
is dened as
^ u
i
( ) =
1
2
_
V
^ s
ij
ê
ij
dV
_
S
t
i
^ u
i
dS
_
V
f
i
^ u
i
dV: (2:34)
The principle of minimum potential energy states that among all the kinematically
admissible displacement elds the actual displacement eld minimizes the poten-
tial energy. Thus, if u
i
represents the actual displacement eld, then
^ u
i
( ) _ u
i
( ) : (2:35)
V
t
j
n
i
S
t
S
u
u
i
f
i
Figure 2.3. A continuum body loaded with body forces inside its volume, and traction and
displacement on the boundary.
Principle of minimum complementary energy
For a statically admissible stress eld ^ s
ij
, the complementary potential energy of a
linear elastic body is dened as
+
^ s
ij
_ _
=
1
2
_
V
^ s
ij
ê
ij
dV
_
S
^
t
i
u
i
dS; (2:36)
where
^
t
i
= ^ s
ij
n
j
is the reaction on S
u
. The principle of minimum complementary
energy states that among all the statically admissible stress elds the actual stress
eld minimizes the complementary potential energy. Thus, if u
i
represents the
actual displacement eld, then
+
^ s
ij
_ _
_
+
s
ij
_ _
: (2:37)
For actual stress, strain, anddisplacement elds, additionof Eqs. (2.34) and(2.36) yields
u
i
( )
+
s
ij
_ _
=
_
V
s
ij
e
ij
dV
_
S
t
i
u
i
dS
_
V
f
i
u
i
dV: (2:38)
The right-hand side of Eq. (2.38) vanishes by virtue of the principle of virtual
work. Hence,
u
i
( ) =
+
s
ij
_ _
: (2:39)
Using Eqs. (2.35), (2.37), and (2.39), we obtain the lower and upper bounds to the
potential energy of a continuum body
^ u
i
( ) _ u
i
( ) =
+
s
ij
_ _
_
+
^ s
ij
_ _
: (2:40)
For the purpose of illustration, the potential and complementary energies for a
typical loaddisplacement response are shown in Figure 2.4.
2.2 Micromechanics
Micromechanics is a well-developed advanced eld that treats the response of a
heterogeneous solid based on the behavior of its constituents and their geometrical
congurations. For a detailed exposition the reader may refer to, e.g., [8]. Here a
brief summary of simple micromechanics estimates of the linear elastic properties
P
* P, D
D
Figure 2.4. A typical loaddisplacement diagram.
17 2.2 Micromechanics
of a unidirectional ber-reinforced composite is provided. These estimates are
useful in selecting bers and matrix materials and their volume fractions. In many
structural applications a unidirectional composite, fabricated as a thin layer,
called lamina or ply, is used as a basic unit and a laminate is constructed by
stacking these layers as illustrated in Figure 2.5.
2.2.1 Stiffness properties of a unidirectional lamina
Linear elastic properties of a lamina can be referred to a coordinate system (x
1
, x
2
,
x
3
) where the x
1
-axis is aligned with bers, x
2
-axis is transverse to bers in
the plane of the lamina, and the x
3
-axis is normal to the plane of lamina
(see Figure 2.6). Noting that the lamina has orthotropic symmetry, the nine
independent elastic constants, as described in Section 2.1.3 above, in this reference
system are the three Youngs moduli (E
1
, E
2
, E
3
), the three Poissons ratios (n
12
,
n
13
, n
23
), and the three shear moduli (G
12
, G
13
, G
23
). For a subset of these constants
that represents in-plane properties, i.e., E
1
, E
2
, n
12
, n
21
, and G
12
, in the x
1
x
2
plane,
the following expressions hold
E
1
= E
f
V
f
E
m
V
m
; (2:41)
n
12
= n
f
V
f
n
m
V
m
; (2:42)
1
E
2
=
V
f
E
f
V
m
E
m
; (2:43)
1
G
12
=
V
f
G
f
V
m
G
m
; (2:44)
where E, n, G, and V stand for the Youngs modulus, Poissons ratio, shear
modulus, and volume fraction, respectively, with the subscripts f and m indicating
bers and matrix, respectively. The minor Poissons ratio n
21
can be estimated
using the reciprocal relationship n
21
= n
12
(E
2
/E
1
).
Equations (2.41) and (2.42) have the form of the familiar rule of mixtures and
Eqs. (2.43) and (2.44) follow that rule for the inverse of the respective properties.
A unidirectional lamina
Lamination
Laminate
(a) (b) (c)
Figure 2.5. Stacking of a number of laminae makes up a laminate.
The rst two expressions predict the experimental properties usually well while the
third and fourth expressions are found to be less accurate. Halpin and Kardos [21]
and Halpin and Tsai [22] proposed semi-empirical relationships based on numer-
ical computations. These relations can be used in place of Eqs. (2.43) and (2.44),
and are together expressed as
p
p
m
=
1 xV
f
1 V
f
; (2:45)
where
=
p
f
p
m
1
p
f
p
m
x
: (2:46)
Here p represents E
2
or G
12
, and p
f
and p
m
are the corresponding moduli for ber
and matrix, respectively. The tting parameter x needs to be determined by
comparing predictions with experimental data.
More advanced micromechanics approaches, such as HashinShtrikman vari-
ational bounds [2329], MoriTanaka model [30], composite sphere and cylinder
assemblage model [31, 32], self-consistent method [33], method of cells [3436], etc.
have also been developed in the past four decades. Interested readers are referred to
texts on micromechanics, e.g., [8, 37, 38], for detailed treatment of these approaches.
2.2.2 Thermal properties of a unidirectional lamina
Simple micromechanics estimates for the linear coefcient of thermal expansion of
a lamina can be obtained in the same way as the linear elastic properties. The
expressions obtained are as follows
a
1
=
1
E
1
a
f
E
f
V
f
a
m
E
m
V
m
( );
a
2
= 1 n
f
( )a
f
V
f
1 n
m
( )a
m
V
m
a
1
n
12
;
(2:47)
x
2
x
1
x
y
x
3
, z
Figure 2.6. Coordinate systems for a unidirectional ply. The material system is denoted by
x
1
, x
2
, x
3
; while the laminate system is denoted by x, y, z.
where a
1
and a
2
are the thermal expansion coefcients in the ber and transverse
directions, respectively, and E
1
and n
12
are given by Eqs. (2.41) and (2.42).
2.2.3 Constitutive equations for a lamina
A lamina is thin compared to other dimensions of the entire laminate. Therefore,
the lamina can be assumed to be in a state of generalized plane stress. Conse-
quently, all the through-thickness stress components are zero, i.e., s
4
= s
5
=
s
6
= 0. In such a case, the constitutive relation for an individual lamina referred
to the three axes of symmetry can be written in Voigt notation as
s
1
s
2
s
6
_
_
_
_
_
_
=
Q
11
Q
12
0
Q
12
Q
22
0
0 0 Q
66
_
_
_
_
e
1
e
2
e
6
_
_
_
_
_
_
; (2:48)
with
Q
11
=
E
1
1n
12
n
21
; Q
22
=
E
2
1n
12
n
21
; Q
12
=
n
12
E
2
1n
12
n
21
=
n
21
E
1
1n
12
n
21
; Q
66
=G
12
: (2:49)
The inverse constitutive relation for the lamina is given by
e
ij
= S
ijkl
s
kl
=
e
1
e
2
e
6
_
_
_
_
_
_
=
1
E
1
n
21
E
2
0
n
12
E
1
1
E
2
0
0 0
1
G
12
_
_
_
_
s
1
s
2
s
6
_
_
_
_
_
_
: (2:50)
The above constitutive relations are written in the lamina coordinate system (i.e.,
with x
1
along the ber direction, x
2
normal to the ber direction, and x
3
along the
lamina thickness). The constitutive relation for the lamina in another coordinate
system (xyz), which, for instance, could be aligned with the coordinate system
chosen for the laminate, is
s
xx
s
yy
s
xy
_
_
_
_
_
_
=
Q
11

Q
12

Q
16
Q
12

Q
22

Q
26
Q
16

Q
26

Q
66
_
_
_
_
e
xx
e
yy
2e
xy
_
_
_
_
_
_
; (2:51)
where

Q
ij
are known as reduced stiffness coefcients. These are related to Q
ij
,
dened by Eq. (2.49), by the transformation rules for stresses and strains. Thus,
s
1
s
2
s
6
_
_
_
_
_
_
=
m
2
n
2
2mn
n
2
m
2
2mn
mn mn m
2
n
2
_
_
_
_
s
xx
s
yy
s
xy
_
_
_
_
_
_
= T [ [
s
xx
s
yy
s
xy
_
_
_
_
_
_
(2:52)
e
1
e
2
e
6
_
_
_
_
_
_
=
m
2
n
2
mn
n
2
m
2
mn
2mn 2mn m
2
n
2
_
_
_
_
e
xx
e
yy
e
xy
_
_
_
_
_
_
= T [ [
1
_ _
T
e
xx
e
yy
e
xy
_
_
_
_
_
_
(2:53)
where m = cos y, n = sin y, where y is the angle between the x- and x
1
-axes (Figure
2.6). Then by inverting Eqs. (2.52) and (2.53), substituting these in Eq. (2.51), and
on using Eq. (2.48) one obtains
Q
_
= T [ [
1
Q [ [ T [ [
T
_ _
1
: (2:54)
[T]
1
is simply given by changing y to y, i.e., [T(y)]
1
= [T(y)]. Expanding the
above relation, we have
Q
11
= Q
11
m
4
2 Q
12
2Q
66
( )m
2
n
2
Q
22
n
4
;
Q
22
= Q
11
n
4
2 Q
12
2Q
66
( )m
2
n
2
Q
22
m
4
;
Q
12
= Q
11
Q
22
4Q
66
( )m
2
n
2
Q
12
m
4
n
4
_ _
;
Q
16
= Q
11
Q
12
2Q
66
( )m
3
n Q
12
Q
22
2Q
66
( )mn
3
;
Q
26
= Q
11
Q
12
2Q
66
( )mn
3
Q
12
Q
22
2Q
66
( )m
3
n;
Q
66
= Q
11
Q
22
2Q
12
2Q
66
( )m
2
n
2
Q
66
m
4
n
4
_ _
:
(2:55)
The transformation rules described above enable us to express engineering moduli
for the lamina referred to arbitrary in-plane axes (xy) in terms of moduli in the
principal (x
1
x
2
) directions as
1
E
x
=
m
4
E
1
n
4
E
2
1
G
12
2n
12
E
1
_ _
m
2
n
2
;
1
E
y
=
n
4
E
1
m
4
E
2
1
G
12
2n
12
E
1
_ _
m
2
n
2
;
n
xy
E
x
=
n
12
E
1
1 2n
12
E
1
1
E
2
1
G
12
_ _
m
2
n
2
;
1
G
xy
=
1
G
12
4m
2
n
2
1 n
12
E
1
1 n
21
E
2
1
G
12
_ _
:
(2:56)
To account for thermal stresses, we need to modify strains in Eq. (2.51) to include
thermal strains, as
e
xx
e
yy
2e
xy
_
_
_
_
_
_
=
e
0
xx
e
0
yy
2e
0
xy
_
_
_
e
th
xx
e
th
yy
2e
th
xy
_
_
_
_
; (2:57)
where the superscripts 0 and th denote mechanical and thermal strains, respectively, with
e
th
xx
e
th
yy
2e
th
xy
_
_
_
_
= T [ [
e
e
th
1
e
th
2
2e
th
12
_
_
_
_
_
_
= T [ [
e
a
1
a
2
0
_
_
_
_
_
_
T : (2:58)
2.2.4 Strength of a unidirectional lamina
Phenomenological failure (strength) criteria that use experimental data to deter-
mine material constants have been proposed for composite materials along the
lines of those used for metals such as the von Mises yield criterion. Failure
mechanisms in composite materials are, however, signicantly more complex,
resulting in a large number of criteria. Here, some common criteria will be stated
for reference; the interested reader is encouraged to consult [39] for more in-depth
treatment.
For a unidirectional ber-reinforced lamina, the ve basic strength parameters
under in-plane loading are as follows:
X = ultimate tensile strength in the ber direction
X
/
= ultimate compressive strength in the ber direction
Y = ultimate tensile strength transverse to bers
Y
/
= ultimate compressive strength transverse to bers
S = ultimate shear strength in the lamina plane.
These parameters are obtained by experimental testing. See, e.g., [39, 40] for
further details.
Maximum stress theory
According to this theory, a lamina fails if
s
1
=
X s
1
> 0;
X
/
s
1
< 0;
_
s
2
=
Y s
2
> 0;
Y
/
s
2
< 0;
_
s
6
[ [ = S:
(2:59)
For combined loading, theoretical predictions of the theory are inaccurate because
the maximum stress criterion does not account for stress interactions. For an off-
axis normal loading, i.e., loading axis inclined to bers, this theory can be applied
by transforming the stresses to the principal material directions and then using the
criteria in Eq. (2.59).
Maximum strain theory
This theory states that failure occurs when
e
1
=
X
e
e
1
> 0;
X
/
e
e
1
< 0;
_
e
2
=
Y
e
e
2
> 0;
Y
/
e
e
2
< 0;
_
e
6
[ [ = S
e
;
(2:60)
where X
e
= X=E
1
; X
/
e
= X
/
=E
1
; Y
e
= Y=E
2
; Y
/
e
= Y
/
=E
2
; and S
e
= S=G
12
are the
ultimate failure strains analogous to the stress-based parameters mentioned above.
Distortional energy (TsaiHill) criterion
This criterion is based on the distortional energy failure (yield) theory of
von Mises. Hill [41] further developed this yield criterion for anisotropic
materials and Azzi and Tsai [42] modied it to describe failure of a composite
lamina as follows
s
2
1
X
2

s
1
s
2
X
2

s
2
2
Y
2

s
2
6
S
2
= 1; (2:61)
where s
1
and s
2
are the tensile normal stresses along bers and normal to
bers, respectively, and s
6
is the in-plane shear stress. When the normal
stresses are compressive, the compressive strength values in Eq. (2.61) are to
be used.
TsaiWu criterion
A polynomial function of stress components can be formulated with the multiply-
ing terms of the polynomial expressing strength properties. Restricted to quadratic
terms of in-plane stress components, such a function is known as the TsaiWu
criterion [43] and can be expressed as
F
1
s
1
F
2
s
2
F
11
s
2
1
F
22
s
2
2
F
66
s
2
6
2F
12
s
1
s
2
= 1: (2:62)
The product terms s
1
s
6
and s
2
s
6
are not present in Eq. (2.62) because the
multiplying coefcients to these terms can be shown to vanish. Also, the linear
term in s
6
is absent because of the shear strength being independent of the sign
of shear stress, which renders the coefcient of this term to be zero.
The six material constants in the TsaiWu criterion require two tests (tension
and compression) in the ber direction, two similar tests normal to bers, an in-
plane shear test, and a biaxial normal load test.
Hashins criterion
Hashin [44] formulated three-dimensional failure criteria for unidirectional ber
composites in terms of quadratic stress polynomials. The terms used in the
polynomials were functions of the stress invariants for transversely isotropic
symmetry. Thus the cross-sectional plane of a unidirectional ber composite was
assumed as an isotropic plane. For relatively thick layers this may be a good
assumption.
A unidirectional ber composite was assumed to fail in one of four possible
separate modes: tensile ber mode (s
1
> 0), compressive ber mode (s
1
< 0),
tensile matrix mode (s
2
+ s
3
> 0), and compressive matrix mode (s
2
+ s
3
< 0)
For a thin unidirectional ber composite layer (lamina), the four failure criteria
are given by
s
2
X
_ _
2
s
6
S
_ _
2
= 1; s
1
> 0;
s
1
= X
/
; s
1
< 0;
s
2
Y
_ _
2
s
6
S
_ _
2
= 1; s
2
> 0;
s
2
2S
/
_ _
2
Y
/
2S
/
_ _
2
1
_ _
s
2
Y
/

s
6
S
_ _
2
= 1; s
2
< 0;
(2:63)
where S
/
is the strength in transverse shear, while S here is the same in axial shear.
The difference between the two shear strengths is not fully unambiguous.
Over the years, a wide variety of failure criteria have been proposed. There is no
single failure theory that seems to capture all the complexities of composite failure.
A world-wide failure exercise was conducted to evaluate applicability of most
theories by comparing their predictions with test data [45].
2.3 Analysis of laminates
Laminates used in most engineering applications are fabricated by stacking plies in
different orientations. An example of a laminate with layup [0/90/45]
s
is shown in
Figure 2.7. A commonly used method of determining stresses and strains for such
laminates is based on the classical laminate plate theory (CLPT). More advanced
theories are treated in [9, 46]. The geometrical conditions needed for the application
of the CLPT are: (a) the individual plies are of uniform thickness, (b) they are
perfectly bonded to their neighboring plies, and (c) the total thickness of the
laminate follows the so-called thin plate assumption, which states that the thickness
dimension is much smaller than other structural dimensions (width and length).
The kinematic assumptions of the CLPT derive from the Kirchhoff assumptions,
which state that (a) a line element normal to the mid-plane in the undeformed state
0
90
45
45
90
0
y
z
x
z
0
z
1
z
2
z
n
q
Figure 2.7. Stacking of unidirectional plies in different orientations to make a
multidirectional [0/90/45]
s
laminate. The subscript s denotes that the laminate is symmetric
about the mid-plane.
of the plate remains straight and perpendicular to the mid-plane after deformation,
and (b) such a line element does not change its length when the plate deforms.
2.3.1 Straindisplacement relations
The Kirchoff assumptions stated above lead to the x-, y-, and z-displacements u, v,
and w, respectively, in the coordinate system shown in Figure 2.7 as follows
u(x; y; z) = u
0
(x; y) z
@w
0
(x; y)
@x
;
v(x; y; z) = v
0
(x; y) z
@w
0
(x; y)
@y
;
w(x; y; z) = w
0
(x; y);
(2:64)
where (u
0
, v
0
, w
0
) are the displacements of the laminate mid-plane. The corres-
ponding straindisplacement relations are given by
e
xx
=
@u
@x
=
@u
0
@x
z
@
2
w
0
@x
2
;
e
yy
=
@v
@y
=
@v
0
@y
z
@
2
w
0
@y
2
;
e
zz
=
@w
@z
= 0;
e
xy
=
1
2
@u
@y
@v
@x
_ _
=
1
2
@u
0
@y

@v
0
@x
_ _
z
@
2
w
0
@x@y
;
e
xz
=
1
2
@u
@z

@w
@x
_ _
= 0;
e
yz
=
1
2
@v
@z

@w
@y
_ _
= 0:
(2:65)
The nonzero equations can be written in the following form
e
xx
e
yy
g
xy
_
_
_
_
_
_
=
e
0
xx
e
0
yy
g
0
xy
_
_
_
_
z
k
xx
k
yy
k
xy
_
_
_
_
_
_
; (2:66)
where e
0
xx
; e
0
yy
; g
0
xy
_ _
are the mid-plane strains and k
xx
; k
yy
; k
xy
_ _
are the laminate
curvatures, given by
e
0
xx
e
0
yy
g
0
xy
_
_
_
_
=
@u
0
@x
@v
0
@y
@u
0
@y

@v
0
@x
_
_
_
_
and
k
xx
k
yy
k
xy
_
_
_
_
_
_
=
@
2
w
0
@x
2
@
2
w
0
@y
2
2
@
2
w
0
@x@y
_
_
_
_
: (2:67)
25 2.3 Analysis of laminates
2.3.2 Constitutive relationships for the laminate
Using the lamina constitutive relations described earlier, the constitutive equation
for the kth (k = 1, 2, . . .) layer of the laminate can be written as
s
(k)
=

Q
_
(k)
e
(k)
: (2:68)
In the above equation, the square bracket represents a 33 matrix and the curly
bracket is for a 31 vector. The strains in the kth ply are given by
e
(k)
= e
0
_ _
z k : (2:69)
The thermal strains can be added to these strains, such that
e
(k)
= e
0
_ _
z k a
k
T: (2:70)
The kth ply stresses on using Eq. (2.68) can now be written as
s
(k)
=

Q
_
(k)
e
0
_ _
a
k
T
_ _
z

Q
_
(k)
k : (2:71)
At the laminate level the force and moment resultants are dened as
N [ [ =
N
xx
N
yy
N
xy
_
_
_
_
=
_
h=2
h=2
s
xx
s
yy
s
xy
_
_
_
_
dz;
M [ [ =
M
xx
M
yy
M
xy
_
_
_
_
=
_
h=2
h=2
s
xx
s
yy
s
xy
_
_
_
_
z dz:
(2:72)
In terms of ply stresses that generally vary from ply to ply, we have
N
xx
N
yy
N
xy
_
_
_
_
_
_
=
N
k=1
_
z
k1
z
k
s
xx
s
yy
s
xy
_
_
_
_
_
_
dz; (2:73)
which gives us
N
xx
N
yy
N
xy
_
_
_
N
th
xx
N
th
yy
N
th
xy
_
_
_
_
=
N
k=1
_
z
k1
z
k
Q
11

Q
12

Q
16
Q
12

Q
22

Q
26
Q
16

Q
26

Q
66
_
_
_
_
e
0
xx
e
0
yy
g
0
xy
_
_
_
_
dz
N
k=1
_
z
k1
z
k
Q
11

Q
12

Q
16
Q
12

Q
22

Q
26
Q
16

Q
26

Q
66
_
_
_
_
k
xx
k
yy
k
xy
_
_
_
_
z dz;
(2:74)
where the force resultants due to thermal stresses are given by
N
th
xx
N
th
yy
N
th
xy
_
_
_
_
=
N
k=1
_
z
k1
z
k
Q
11

Q
12

Q
16
Q
12

Q
22

Q
26
Q
16

Q
26

Q
66
_
_
_
_
a
x
T
a
y
T
0
_
_
_
_
_
_
dz: (2:75)
The relations in Eq. (2.74) can be rewritten in more compact form by using
matrices [A] and [B] as follows
N
xx
N
yy
N
xy
_
_
_
_
_
_
N
th
xx
N
th
yy
N
th
xy
_
_
_
_
=
A
11
A
12
A
16
A
12
A
22
A
26
A
16
A
26
A
66
_
_
_
_
e
0
xx
e
0
yy
g
0
xy
_
_
_
B
11
B
12
B
16
B
12
B
22
B
26
B
16
B
26
B
66
_
_
_
_
k
xx
k
yy
k
xy
_
_
_
_
_
_
:
(2:76)
Similarly,
M
xx
M
yy
M
xy
_
_
_
_
_
_
=
N
k=1
_
z
k1
z
k
s
xx
s
yy
s
xy
_
_
_
_
_
_
zdz; (2:77)
i.e.,
M
xx
M
yy
M
xy
_
_
_
M
th
xx
M
th
yy
M
th
xy
_
_
_
_
=
N
k=1
_
z
k1
z
k
Q
11

Q
12

Q
16
Q
12

Q
22

Q
26
Q
16

Q
26

Q
66
_
_
_
_
e
0
xx
e
0
yy
g
0
xy
_
_
_
_
zdz
N
k=1
_
z
k1
z
k
Q
11

Q
12

Q
16
Q
12

Q
22

Q
26
Q
16

Q
26

Q
66
_
_
_
_
k
xx
k
yy
k
xy
_
_
_
_
z
2
dz;
(2:78)
where
M
th
xx
M
th
yy
M
th
xy
_
_
_
_
=
N
k=1
_
z
k1
z
k
Q
11

Q
12

Q
16
Q
12
Q
22
Q
26
Q
16

Q
26

Q
66
_
_
_
_
a
x
T
a
y
T
0
_
_
_
_
_
_
zdz: (2:79)
Introducing a new matrix [D], Eq. (2.78) can be rewritten as
M
xx
M
yy
M
xy
_
_
_
_
_
_
M
th
xx
M
th
yy
M
th
xy
_
_
_
_
=
B
11
B
12
B
16
B
12
B
22
B
26
B
16
B
26
B
66
_
_
_
_
e
0
xx
e
0
yy
g
0
xy
_
_
_
D
11
D
12
D
16
D
12
D
22
D
26
D
16
D
26
D
66
_
_
_
_
k
xx
k
yy
k
xy
_
_
_
_
_
_
:
(2:80)
The material coefcients (A
ij
, B
ij
, D
ij
) are known as the extensional stiffness, the
extensionbending coupling stiffness, and the bending stiffness coefcients, respect-
ively. These are given by
A
ij
; B
ij
; D
ij
_ _
=
_ h
2
h
2
Q
ij
1; z; z
2
_ _
dz; (2:81)
27 2.3 Analysis of laminates
or
A
ij
=
N
k=1
Q
ij
z
k1
z
k
( ) ;
B
ij
=
1
2
N
k=1
Q
ij
z
2
k1
z
2
k
_ _
;
D
ij
=
1
3
N
k=1
Q
ij
z
3
k1
z
3
k
_ _
:
(2:82)
The laminate constitutive relations can now be written in compact form as
N
M
_ _
=
A [ [ B [ [
B [ [ D [ [
_ _
e
0
_ _
k
_ _
(2:83)
where {N}, {M} include thermal resultants.
2.3.3 Stresses and strains in a lamina within a laminate
The laminate constitutive relations in Eq. (2.83) can be reverted to yield the mid-
plane strains and curvatures in terms of the stress and moment resultants.
A partial inversion is rst done by inverting Eq. (2.76) and substituting into
Eq. (2.80) to obtain
e
0
M
_ _
=
A
+
B
+
C
+
D
+
_ _
N
k
_ _
; (2:84)
where
A
+
= A
1
; B
+
= A
1
B; C
+
= BA
1
= B
+
( )
T
; D
+
= D BA
1
B; (2:85)
and the brackets for matrix/vector representation have been dropped for conveni-
ence. Solving Eq. (2.84) for k and its substitution back gives
e
0
k
_ _
=
A
/
B
/
B
/
D
/
_ _
N
M
_ _
; (2:86)
where
A
/
= A
+
B
+
D
+
( )
1
B
+
( )
T
; B
/
= B
+
D
+
( )
1
; D
/
= D
+
( )
1
: (2:87)
Once mid-plane strains and curvatures are known, the strains and stresses in each
lamina can be determined using Eqs. (2.70) and (2.68), respectively.
2.3.4 Effect of layup conguration
The sequence of ply layup has a signicant impact on the stiffness properties of the
designed laminate. Some interesting ply congurations are described below.
v Balanced laminate: If for each +y-ply, we have another identical ply of
the same thickness, but y orientation, we have A
16
= A
26
= 0. Such
laminates are known as balanced laminates. If additionally these plies are
at the same distance about the mid-plane (one above and another below
the mid-plane), then D
16
= D
26
= 0. An example of a balanced
laminate is [0/+45/45/90
2
/0]
T
, where the subscript T denotes total laminate
sequence.
v Symmetric laminate: If a laminate has plies stacked in such a way that through
its thickness the plies are symmetrical about the mid-plane, then B
ij
= 0. Thus,
such laminates will not exhibit any extensionbending coupling, e.g., [0/30/
45
2
/90
2
/45
2
/30/0]
T
= [0/30/45
2
/90]
s
, where the subscript s represents sym-
metry about mid-plane.
v Cross-ply laminate: If the plies are stacked in two orthogonal directions, e.g., in
longitudinal (0
) and transverse (90
) directions, the laminate thus built is

known as a cross-ply laminate, e.g., [0
2
/90
4
/0]
T
.
v Quasi-isotropic laminate: If plies of identical properties and thickness are
oriented in such a way that the angle between any two adjacent plies is equal
to p/n, where n is the number of plies equal to or greater than three, then [A]
becomes directionally independent, thereby showing isotropy in the in-plane
material properties. This construction does not imply that the matrices [B] and
[D] are also isotropic, e.g., [0/45/90]
s
.
For the special case of a symmetric balanced laminate, the in-plane engineering
moduli can be obtained from the [A], [B], and [D] matrices using
E
x
=
1
h
A
11

A
2
12
A
22
_ _
; E
y
=
1
h
A
22

A
2
12
A
11
_ _
;
n
xy
=
A
12
A
22
; G
xy
=
A
66
h
;
(2:88)
where h is the total laminate thickness.
The laminate analysis summarized here is inadequate at and near free edges in
laminates. The stress state in a laminate near a free boundary is three-dimensional
and cannot be assumed to be well described by plane stress or plane strain
assumptions. The through-thickness normal and shear stresses can, in some cases,
be signicant and could cause laminate failure.
2.4 Linear elastic fracture mechanics
The basic concepts of fracture mechanics are useful in analyzing damage and
failure in composite materials. Here we will briey review those concepts and
list some of the commonly used results from the linear elastic fracture mechanics.
For more detailed treatment of fracture mechanics comprehensive texts, e.g.,
[13, 15, 17, 47] are suggested.
29 2.4 Linear elastic fracture mechanics
2.4.1 Fracture criteria
The traditional strength of the materials approach to structural design and mater-
ial selection is based on the notion of yield or failure stress (strength) of a given
material. The fracture mechanics approach instead recognizes the presence of
material aws whose unstable growth could cause catastrophic failure. To deter-
mine conditions for this type of failure the local stress eld in the vicinity of aws
(modeled as cracks) is analyzed. The singularity of this stress eld is characterized
by the so-called stress intensity factor and its critical value is associated with
unstable crack growth. Alternatively, energy balance considerations are used to
nd the so-called energy release rate, per unit increment in the crack surface, and
its critical value is associated with the condition of unstable crack growth. For
linear elastic materials undergoing brittle failure the two approaches produce the
same failure criteria.
The stress intensity criterion
Consider an innite plate with a through thickness edge crack of size a subjected
to a remote tensile stress as shown in Figure 2.8. For a linear elastic plate material
the stress eld in close vicinity of the crack tip is given by
s
xx
=
K
I
2pr
_ cos
y
2
_ _
1 sin
y
2
_ _
sin
3y
2
_ _ _ _
;
s
yy
=
K
I
2pr
_ cos
y
2
_ _
1 sin
y
2
_ _
sin
3y
2
_ _ _ _
;
t
xy
=
K
I
2pr
_ cos
y
2
_ _
sin
y
2
_ _
cos
3y
2
_ _
;
(2:89)
where r and y are as shown in the gure and K
I,
known as the stress intensity
factor, is given by
K
I
= s
pa
_
: (2:90)
where the subscript I denotes the opening mode (mode I). It can be noted that the
stress eld is singular at the crack tip with a r
1/2
singularity.
The condition of failure, i.e., unstable crack growth, is assumed when
K
I
_ K
IC
: (2:91)
K
IC
, known as the critical stress intensity factor, or fracture toughness, is a
parameter representing the material resistance to fracture, and can be obtained
experimentally.
The energy criterion
In the energy-based approach, one considers a cracked body and examines the
changes brought about by an incremental crack growth in the potential energy of
applied forces the stored elastic strain energy and the crack surface energy. The
condition for unstable crack growth is then expressed as
G _ G
C
; (2:92)
where G is the energy available for crack growth per unit of crack surface area,
called the energy release rate, and G
C
is its critical value, which depends on the
material in which the crack is advancing. G
C
is viewed as the resistance to crack
growth induced by the material. For a linear elastic material undergoing small-
scale yielding at the crack front, the energy release rate is found to be related to the
stress intensity factor, described above, as
G =
K
2
I
E
/
; (2:93)
where E
/
= E for plane stress condition, and E
/
=
E
1 n
2
for plane strain condition.
2.4.2 Crack separation modes
A crack is activated, i.e., it produces stresses at its front, when the two crack
surfaces separate. The separation can take place in combination of three
a
x
y
r
s
yy
t
xy
q
s
xx
s
s
Figure 2.8. Edge crack in a plate in tension.
independent modes, denoted as modes I, II, and III, illustrated in Figure 2.9. In
mode I, also called the crack opening mode, the two crack surfaces separate
symmetrically about the crack plane. Mode II is a sliding mode, in which the two
crack surfaces remain in contact and slide past each other in the crack plane.
Finally, mode III, described as the tearing mode, is driven by out-of-plane shear,
resulting in displacement of the two crack surfaces in the x
3
-direction.
Any displacement of the crack surfaces for a general loading can be viewed as
a superposition of these three modes. Denoting the stress intensity factors in individ-
ual modes as K
I
, K
II
, and K
III
, the energy release rate for mixed-mode is given by
G =
K
2
I
E
/

K
2
II
E
/

1 n
E
K
2
III
; (2:94)
where
K
I
= s
11
pa
_
; K
II
= s
12
pa
_
; K
III
= s
13
pa
_
; (2:95)
where the stresses s
11
, etc. refer to the axes shown in Figure 2.9.
2.4.3 Crack surface displacements
The displacement jump across the two crack surfaces, expressed as
u
i
= u
i
u
i
; (2:96)
where u
i
and u
i
represent the displacements of the upper and lower crack
surfaces, respectively, is a quantity of interest in fracture analysis. For the opening
mode of crack separation, mode I, illustrated in Figure 2.10, i = 2, this quantity is
described as crack opening displacement (COD). For an innite isotropic homo-
geneous medium the COD value is given by
u
2
= k
1
x
1
a
_ _
2
_
; (2:97)
which describes an elliptical crack opening prole.
(a) (b) (c)
x
3
x
2
x
1
Figure 2.9. Crack separation modes: (a) opening; (b) sliding; and (c) tearing.
2.4.4 Relevance of fracture mechanics for damage analysis
Fracture mechanics developed and matured well before damage mechanics
emerged. For both elds, the impetus came from the need to analyze failure of
metals. Fracture mechanics initially addressed brittle failure from sharp defects
based on idealized stress analysis of cracks. In contrast, damage mechanics was
concerned with the effect of distributed voids and cracks on the average response of
a solid. For composite materials, the complexity of failure processes involving
a multitude of cracks gave rise to further development of damage mechanics. Today,
damage mechanics of composite materials stands on its own as a mature eld solidly
founded in thermodynamics and having a variety of analytical and computational
methodologies associated with it. Fracture mechanics has aided the development of
damage mechanics of composite materials in providing energy-based concepts for
addressing evolution of failure states. However, the stress analysis of cracks, char-
acterized by stress intensity factors, is less relevant to composite damage analysis.
Other than a few cases where single crack growth is a dominant failure mechanism,
such as delamination emanating from free edges in laminates, crack front singular-
ities are of little interest. Indeed, individual cracks constituting damage modes are
usually arrested at interfaces. Therefore, their growth is of little interest. Instead,
energy dissipation occurs due to crack multiplication. Therefore, an appropriate
energy-based analysis, needed to treat this type of situation, does not resort to stress
intensity factors, as is the case in brittle fracture of metals.
x
1
x
2
u
2
+
u
2
s
s
+
Figure 2.10. Crack opening displacement for a crack of size 2a.
References
1. Y. C. Fung, A First Course in Continuum Mechanics. (Englewood Cliffs, NJ: Prentice-
Hall, Inc., 1977).
2. L. E. Malvern, Introduction to the Mechanics of a Continuous Medium. (Englewood
Cliffs, NJ: Prentice-Hall, Inc., 1969).
3. I. S. Sokolnikoff, Mathematical Theory of Elasticity, 2ndedn. (NewYork: McGraw-Hill, 1956).
4. A. E. H. Love, A Treatise on the Mathematical Theory of Elasticity. (New York: Dover
Publications, 1944).
5. S. P. Timoshenko and J. N. Goodier, Theory of Elasticity. (NewYork: McGraw-Hill, 1951).
6. R. M. Jones, Mechanics of Composite Materials, 2nd edn. (Philadelphia, PA: Taylor &
Francis, 1999).
7. R. M. Christensen, Mechanics of Composite Materials. (Dover Publications, 2005).
8. J. Qu and M. Cherkaoui, Fundamentals of Micromechanics of Solids. (Hoboken, NJ:
Wiley, 2006).
9. J. N. Reddy, Mechanics of Laminated Composite Plates and Shells: Theory and Analysis,
2nd edn. (Boca Raton, FL: CRC Press, 2004).
10. I. M. Daniel and O. Ishai, Engineering Mechanics of Composite Materials. (Oxford:
Oxford University Press, 1994).
11. A. K. Kaw, Mechanics of Composite Materials. (Boca Raton, FL: CRC Press, 1997).
12. L. P. Kollar and G. S. Springer, Mechanics of Composite Structures. (Cambridge:
Cambridge University Press, 2003).
13. D. Broek, Elementary Engineering Fracture Mechanics, 4th revised edn. (Dordrecht,
Netherlands: Kluwer Academic Publishers, 1991).
14. T. L. Anderson, Fracture Mechanics: Fundamentals and Applications, 2nd edn. (Boca
Raton, FL: CRC Press, 1994).
15. E. E. Gdoutos, Fracture Mechanics: An Introduction, 2nd edn. Solid Mechanics and its
Applications, Vol. 123. (Dordrecht: Norwell, MA: Springer, 2005).
16. K. Friedrich, Application of Fracture Mechanics to Composite Materials. (Amsterdam:
Elsevier, 1989).
17. R. W. Hertzberg, Deformation and Fracture Mechanics of Engineering Materials, 3rd
edn. (New York: Wiley, 1989).
18. K. Washizu, Variational Methods in Elasticity and Plasticity, 2nd edn. (Oxford: Perga-
mon Press, 1974).
19. C. L. Dym and I. H. Shames, Solid Mechanics: A Variational Approach. (New York:
McGraw-Hill, 1973).
20. J. N. Reddy, Energy Principles and Variational Methods in Applied Mechanics, 2nd edn.
(Hoboken, NJ: Wiley, 2002).
21. J. C. Halpin and J. L. Kardos, HalpinTsai equations review. Polym Eng Sci, 16:5
(1976), 34452.
22. J. C. Halpin and S. W. Tsai, Environmental Factors Estimation in Composite Materials
Design. US Airforce Technical Report, AFML TR (1967). pp. 67423.
23. Z. Hashin, On elastic behaviour of bre reinforced materials of arbitrary transverse
phase geometry. J Mech Phys Solids, 13:3 (1965), 11934.
24. Z. Hashin and S. Shtrikman, A variational approach to the theory of the
effective magnetic permeability of multiphase materials. J Appl Phys, 33:10 (1962),
312531.
25. Z. Hashin and S. Shtrikman, A variational approach to the theory of the elastic
behaviour of multiphase materials. J Mech Phys Solids, 11:2 (1963), 12740.
26. Z. Hashin and S. Shtrikman, On some variational principles in anisotropic and non-
homogeneous elasticity. J Mech Phys Solids, 10:4 (1962), 33542.
27. Z. Hashin and S. Shtrikman, A variational approach to the theory of the elastic
behaviour of polycrystals. J Mech Phys Solids, 10:4 (1962), 34352.
28. Z. Hashin, Variational principles of elasticity in terms of the polarization tensor. Int
J Eng Sci, 5:2 (1967), 21323.
29. Z. Hashin, Analysis of composite materials a survey. J Appl Mech, Trans ASME, 50:3
(1983), 481505.
30. T. Mori and K. Tanaka, Average stress in matrix and average elastic energy of
materials with mistting inclusions. Acta Metall, 21:5 (1973), 5714.
31. Z. Hashin, The elastic moduli of heterogeneous materials. J Appl Mech, T-ASME,
29 (1962), 14350.
32. R. M. Christensen and K. H. Lo, Solutions for effective shear properties in 3-phase
sphere and cylinder models. J Mech Phys Solids, 27:4 (1979), 31530.
33. R. Hill, A self-consistent mechanics of composite materials. J Mech Phys Solids,
13:4 (1965), 21322.
34. J. Aboudi, Micromechanical analysis of composites by the method of cells. Appl Mech
Rev, 42:7 (1989), 193221.
35. J. Aboudi, Micromechanical analysis of composites by the method of cells update.
Appl Mech Rev, 49:10 Part 2 (1996), S8391.
36. J. Aboudi, Micromechanical prediction of the effective coefcients of thermo-
piezoelectric multiphase composites. J Intell Mater Syst Struct, 9:9 (1999), 71322.
Materials, 2nd edn. (Amsterdam: North Holland, 1999).
38. T. Mura, Micromechanics of Defects in Solids, 2nd edn. (New York: Springer, 1987).
39. C. T. Sun, Strength analysis of unidirectional composites and laminates. In Fiber
Reinforcements and General Theory of Composites, ed. T. W. Chou. (Amsterdam:
Elsevier, 2000), pp. 64166.
40. L. A. Carlsson, D. F. Adams, and R. B. Pipes, Experimental Characterization of
Advanced Composite Materials. (Boca Raton, FL: CRC Press, 2003).
41. R. Hill, The Mathematical Theory of Plasticity. (New York: Oxford University Press,
1998).
42. V. D. Azzi and S. W. Tsai, Anisotropic strength of composites, Exp Mech, 5 (1965),
28388.
43. S. W. Tsai and E. M. Wu, General theory of strength for anisotropic materials.
J Compos Mater, 5:January (1971), 5880.
44. Z. Hashin, Failure criteria for unidirectional ber composites. J Applied Mech, Trans
ASME, 47 (1980), 32934.
45. P. D. Soden, A. S. Kaddour, and M. J. Hinton, Recommendations for designers and
researchers resulting from the world-wide failure exercise. Compos Sci Technol, 64:34
(2004), 589604.
46. J. N. Reddy, Mechanics of Laminated Composite Plates: Theory and Analysis. (Boca
Raton, FL: CRC Press, 1997).
47. T. L. Anderson, Fracture Mechanics: Fundamentals and Applications, 3rd edn. (Boca
Raton, FL: Taylor & Francis, 2005).
35 References
3 Damage in composite materials
All structures are designed for a purpose. If the purpose is to carry loads, then a
designer must assure that the structure has sufcient load-bearing capacity. If the
structure is to function over a period of time, then it must be designed to meet its
functionality over that period without losing its integrity.
These are generic structural design issues irrespective of the material used. There
are, however, signicant differences in design procedures depending on whether the
material used is a so-called monolithic material, e.g., a metal or a ceramic, or
whether it is a composite material with distinctly different constituents. The hetero-
geneity of microstructure as well as the anisotropy of properties provide signi-
cantly different characteristics to composite materials in how they deform and fail
when compared to metals or ceramics. This chapter will review those characteris-
tics. However, before proceeding we need to introduce certain denitions.
Fracture: Conventionally, fracture is understood to be breakage of material,
or at a more fundamental level, breakage of atomic bonds, manifesting itself in
formation of internal surfaces. Examples of fracture in composites are ber
breakage, cracks in matrix, ber/matrix debonds, and separation of bonded
plies (delamination). The eld known as fracture mechanics deals with condi-
tions for formation and enlargement of the surfaces of material separation.
Damage: Damage, on the other hand, refers to a collection of all the irreversible
changes brought about in a material by a set of energy dissipating physical or
chemical processes, resulting from the application of thermomechanical load-
ings. Damage may inherently be manifested by atomic bond breakage. Unless
specied differently, damage is understood to refer to distributed changes.
Examples of damage in composites are multiple ber-bridged matrix cracking
in a unidirectional composite, multiple intralaminar cracking in a laminate,
local delamination distributed in an interlaminar plane, and ber/matrix inter-
facial slip associated with multiple matrix cracking. These damage mechanisms
will be explained in some detail later in this chapter. The eld of damage
mechanics deals with conditions for initiation and progression of distributed
changes as well as with consequences of those changes on the response of a
material (and by implication, a structure) to external loading.
Failure: The inability of a given material system (and consequently, a structure
made from it) to perform its design function. Fracture is one example of a
possible failure; but, generally, a material could fracture (locally) and still
perform its design function. Upon suffering damage, e.g., in the form of
multiple cracking, a composite material may still continue to carry loads and,
thereby, meet its load-bearing requirement but fail to deform in a manner
needed for its other design requirements, such as vibration characteristics and
deection limits. It is a common practice for engineers to predict composite
failure based on any of the multitude of lamina failure criteria described in the
previous chapter. These criteria only predict the nal event of failure, and
generally cannot characterize the damage mechanisms leading to the nal
failure. In reality, the failure event in a composite structure is preceded and
inuenced by the progressive occurrence and interaction of various damage
mechanisms.
Structural integrity: The ability of a load-bearing structure to remain intact and
functional upon the application of loads. In contrast to metals, remaining intact
(not breaking up in pieces) for composites is not necessarily the same as
remaining functional. For instance, composites can lose their functionality by
suffering degradation in their stiffness properties while still carrying signicant
loads.
Durability: A term very close in meaning to structural integrity. Specically,
durability is dened as the ability of a structure to retain adequate properties
(strength, stiffness, and environmental resistance) throughout its life to the
extent that any deterioration can be controlled and repaired [1]. The long-term
durability of a composite structure is an important design requirement in civil,
infrastructure, and aircraft industries.
3.1 Mechanisms of damage
The heterogeneous microstructure of composites, the large differences between
constituent properties, the presence of interfaces as well as directionality of
reinforcement that induces anisotropy in overall properties, are reasons for the
complexities observed in geometrical features of micro-level failure (microcracks)
in composites. Additionally, when interfaces are present, such as between bers
and matrix and between plies in a laminate, the stress transfer via interfaces
provides conditions for multiple cracking (to be discussed later). The wealth of
observations reported in the literature on various cracking processes, collectively
referred to as damage mechanisms, are summarized below for the purpose of
treatments in later chapters related to deformation and failure of composite
materials at a macro level.
3.1.1 Interfacial debonding
The performance of a ber-reinforced composite is markedly inuenced by the
properties of the interface between the ber and the matrix resin. The
37 3.1 Mechanisms of damage
adhesion bond at the interfacial surface affects the macroscopic mechanical
properties of the composite. The interface plays a signicant role in stress
transfer between ber and matrix. For instance, if the bers are weakly held
by the matrix, the composite starts to form a matrix crack at a relatively low
stress. On the other hand, if the bers are strongly bonded to the matrix, the
matrix cracking is delayed and the composite fails catastrophically because of
ber fracture as the matrix cracks. The constraint between the ber and the
matrix also inuences other damage mechanisms such as interfacial slipping,
and ber pull-out. Controlling interfacial properties can thus provide a way to
control the performance of a composite structure. In unidirectional compos-
ites, debonding occurs at the interface between ber and matrix when the
interface is weak. Figure 3.1 shows debond surfaces observed in a ber-
reinforced composite [2].
The longitudinal interfacial debonding behavior of single-ber composites has
been studied in detail by the use of the pull-out [37] and fragmentation [811]
tests. The mechanics of ber/matrix interfacial debonding in a unidirectional
ber-reinforced composite is depicted in Figure 3.2. When fracture strain of the
ber is greater than that of the matrix, a crack originating at a point of stress
concentration, e.g., voids, air bubbles, or inclusions, in the matrix is either halted
by the ber, if the stress is not high enough (Figure 3.2(b)), or it may pass
around the ber without destroying the interfacial bond. As the applied load
increases, the ber and matrix deform differentially, resulting in a buildup of
large local stresses in the ber. This causes local Poisson contraction and even-
tually when the shear stress developed at the interface exceeds the interfacial
shear strength, debonding extending over a distance along the ber results
(Figure 3.2(c)). Shear lag and cohesive zone models are commonly used
approaches to predict initiation of debonding and stress transfer at the interface
[3, 5, 1217].
20m
Figure 3.1. Debonds in a ber-reinforced composite. Reprinted, with kind permission, from
Compos Sci Technol, Vol. 59, E. K. Gamstedt and B. A. Sjo gren, Micromechanisms in
tension-compression fatigue of composite laminates containing transverse plies, pp. 16778,
copyright Elsevier (1999).
3.1.2 Matrix microcracking/intralaminar (ply) cracking
Fiber-reinforced composites offer high strength and stiffness properties in the
longitudinal direction. Their properties, however, in the transverse directions are
generally low. As a result, they readily develop cracks along bers. These cracks
are usually the rst observed form of damage in ber-reinforced composites [19].
In laminates with plies in different ber orientations, these cracks can form from
defects in a given ply and grow traversing the thickness of the ply and running
parallel to the bers in that ply. The terms matrix microcracks, transverse cracks,
intralaminar cracks, and ply cracks are invariably used to refer to these very same
cracks. Such cracks are found to be caused by tensile loading, fatigue loading, as
well as by changes in temperature or by thermal cycling. They can originate from
ber/matrix debonds or manufacturing-induced defects such as voids and inclu-
sions [20] (see Figure 3.3). Matrix cracks can also form in ceramic matrix compos-
ites (CMC), and in short ber composites (SFC). The eld of damage mechanics
deals with prediction of formation, growth, and effects of matrix cracks on overall
material behavior. Analysis, design, and behavior of composites subjected to
intralaminar cracking will be dealt with in detail in the subsequent chapters.
Figure 3.4 illustrates matrix cracks observed on the free edges of continuous
ber and woven fabric polymer composite laminates induced due to fatigue
loading [21, 22]. Although matrix cracking does not cause structural failure by
itself, it can result in signicant degradation in material stiffness and can also
induce more severe forms of damage, such as delamination and ber breakage,
and give pathways for entry of uids.
3.1.3 Interfacial sliding
Interfacial sliding between constituents in a composite can take place by differential
displacement of the constituents. One example of this is when bers and matrix in a
composite are not bonded together adhesively but by a shrink-t mechanism
due to difference in thermal expansion properties of the constituents. On
(a) (b) (c)
Figure 3.2. Mechanics of interfacial debonding in a simple composite: (a) perfect laminate;
(b) differential deformation of ber and matrix crack causes high stresses at ber/matrix
interface; (c) shear stress exceeds the interfacial shear strength nucleating a debond.
Reprinted from [18], with kind permission from Maney Publishing.
thermomechanical loading, the shrink-t (residual) stresses can be removed,
leading to a relative displacement (sliding) at the interface. The relief of interfacial
normal stress can also occur when a matrix crack tip approaches or hits the
interface.
Debond
Matrix
Void
Fiber
5 m 10 m
(a)
(b)
Figure 3.3. Matrix crack initiation from: (a) ber debonds; (b) void results. Reprinted, with
kind permission, from Compos Sci Technol, Vol. 57, C. A. Wood and W. L. Bradley,
Determination of the effect of seawater on the interfacial strength of an interlayer E-glass/
graphite/epoxy composite by in situ observation of transverse cracking in an environmental
SEM, pp. 103343, copyright Elsevier (1997).
(a)
(b)
Figure 3.4. Examples of matrix cracks observed in (a) continuous ber and (b) woven fabric
polymer composite laminates [22]. Part (a) reprinted, with kind permission, from Compos Sci
Technol, Vol. 68, D. T. G. Katerelos, J. Varna, and C. Galiotis, Energy criterion for modeling
damage evolution in cross-ply composite laminates, pp. 231824, copyright Elsevier (2008).
When the two constituents are bonded together adhesively, interfacial sliding
can occur subsequent to debonding if a compressive normal stress on the interface
is present. The debonding can be induced by a matrix crack, or it can result from
growth of interfacial defects. Thus, interfacial sliding that follows debonding can
be a separate damage mode or it can be a damage mode coupled with matrix
damage.
Interfacial sliding in ceramic matrix composites (CMCs) can be signicant if the
temperature change imposed is high and the thermal expansion mismatch between
the bers and matrix is also large. When the matrix in a CMC cracks, the resulting
interfacial debonding affects interfacial sliding, causing interactive effect on the
composite deformation [23].
3.1.4 Delamination/interlaminar cracking
Interlaminar cracking, i.e., cracking in the interfacial plane between two adjoining
plies in a laminate, causes separation of the plies (laminae) and is referred to as
delamination. In composite laminates, delamination can occur at cut (free) edges,
such as at holes, or at an exposed surface through the thickness. When loaded in
the plane, the laminate develops through-thickness normal and shear stresses at
the traction-free surface extending a short distance into the laminate plane. These
stresses can result in local cracking in the interlaminar planes. Delaminations can
also form as a result of low-velocity impact [2426]. In contrast to metals, in
polymer composite laminates delamination can occur below the surface of a
structure under a relatively light impact, such as that from a dropped tool, while
the surface appears undamaged to visual inspection [25, 27, 28]. The growth of
delamination cracks under the subsequent application of external loads leads to a
rapid deterioration of the mechanical properties and may cause catastrophic
failure of the composite structure [29, 30]. Another source of delamination is the
local interlaminar cracking induced by ply cracks. This delamination can grow
and separate the region between two adjacent ply matrix cracks as illustrated in
Figure 3.5.
Delamination can be a substantial problem in designing composite structures
as it can diminish the role of strong bers and make the weaker matrix
properties govern the structural strength [31]. In initiating delamination the
critical material property is the interlaminar strength, which is determined by
the matrix [26, 31]. Once the interlaminar cracks are formed, their growth is
determined by the interlaminar fracture toughness, which is also governed by
the matrix. If delamination is viewed as decohesion of the cohesive zone
between the separating plies, then both the matrix strength and the fracture
toughness act as material parameters [32]. As a design approach, delamination
can be reduced either by improving the interlaminar strength and fracture
toughness or by modifying the ber architecture to reduce the driving forces
for delamination [33, 34].
3.1.5 Fiber breakage
The failure (separation) of a ber-reinforced composite ultimately comes from
breakage of bers. In a unidirectional composite loaded in tension along bers the
individual bers fail at their weak points and stress redistribution between bers
and matrix occurs, affecting other bers in the local vicinity of the broken bers
and possibly breaking some. The ber/matrix interface transfers the stress from
the broken ber back to the ber at a certain distance, making another ber break
possible if the strength is exceeded by the stress. The ber breakage process is of
a statistical nature because of the nonuniformity of ber strength along the ber
length and the stress redistribution. When plies of unidirectional bers are stacked
in a laminate, the stress on bers is enhanced in the vicinity of ply cracks in the
adjacent plies, causing a narrow distribution of ber failure sites [35]. A greater
number of ber breaks per unit volume is found closer to the interface where the
ply crack terminates than away from the interface where the local stress concen-
tration falls off [35].
The ultimate tensile strength of a ply within a general laminate is difcult to
predict from the tensile strength of bers due to the statistical nature of ber
failure and the progression of ber failures [36, 37]. Fracture (crack growth)
properties such as the fracture toughness of a composite depend not only on the
failure properties of the constituents but signicantly also on the efciency of
bonding across the interface [38].
3.1.6 Fiber microbuckling
When a unidirectional composite is loaded in compression, the failure is governed
by a mechanism known as microbuckling of bers. There are two idealized
basic modes of microbuckling deformation, denoted extensional and shear
modes [39], illustrated in Figure 3.6, depending upon whether the bers deform
out of phase or in phase with one another. The corresponding compressive
strength for the onset of instability is given as
Figure 3.5. Interlaminar delamination crack formed due to joining of two adjacent matrix
cracks in a ber-reinforced composite laminate.
s
c
2V
f
V
f
E
f
E
m
31 V
f
s
; 3:1
for the extension mode, and
s
c

G
m
1V
f
; 3:2
for the shear mode, where E and G denote Youngs modulus and shear modulus,
respectively, and the subscripts f and m designate ber and matrix, respectively.
These expressions for idealized deformation modes do not generally agree with
experimental data for compression strength. It has been argued that in practical
composites the manufacturing process tends to cause misalignment of bers,
which can induce localized kinking of bers. The kinking process is driven by
local shear, which depends on the initial misalignment angle f
0
[40]. The critical
compressive stress corresponding to instability then is given by
s
c

t
y
f
0
; 3:3
where t
y
represents the in-plane shear strength (yielding). Budiansky [41] con-
sidered the kink band geometry (see Figure 3.7) and derived the following estimate
for the kink band angle b in terms of the transverse modulus E
T
and shear
modulus G of a composite layer:
2
p
1
2
G s
c
E
T
<tan
2
b<
G s
c
E
T
: 3:4
To account for shear deformation effects, Niu and Talreja [42] modeled the ber
as a generalized Timoshenko beam with the matrix as an elastic foundation. It was
observed that not only an initial ber misalignment but also any misalignment in
the loading system can affect the critical stress for kinking.
Extensional Mode Shear Mode
Figure 3.6. Extensional and shear modes of ber microbuckling.
3.1.7 Particle cleavage
If brittle particles (e.g., ceramics) are placed in a ductile but strong and tough
matrix, particle cleavage is the main mode of damage in initial stages of deform-
ation. This mode of damage is found in particulate metal matrix composites.
Cleavage refers to the breakage of the reinforcing particle. The cleavage crack
typically forms perpendicular to the global maximum principal stress. The damage
analysis has been performed assuming viscoplastic material behavior [43]. Failure
of many practically relevant particulate two-phase composites can typically be
attributed to cleavage fracture of the brittle particles followed by ductile crack
growth in the matrix [44]. To account for particle geometry and distribution,
statistical methods are employed to predict inclusion fracture. To fully character-
ize brittle fracture of a particle embedded in a ductile metallic matrix, careful
computational modeling (FEA) sometimes becomes necessary (see, e.g., [4547]).
3.1.8 Void growth
A composite structure may contain an appreciable amount of manufacturing-
induced defects. For polymer matrix composites, the defects induced during
manufacturing can be in the ber architecture, e.g., ber misalignment, irregular
ber distribution in the cross section, and broken bers; in the matrix, e.g., voids;
or at the ber/matrix interface, e.g., disbonds and delaminations. Voids are one of
the primary defects found virtually in all types of composite materials. The
formation of voids is controlled by manufacturing parameters, such as vacuum
pressure, cure temperature, cure pressure, and resin viscosity.
The presence of voids, even at low volume fractions, is found to have a
signicant detrimental effect on the overall material behavior. The exural, trans-
verse, and shear properties are affected the most. Their shape, size, and distribu-
tion also play role in material degradation. Micromechanics homogenization
methods, such as MoriTanaka [48], are commonly used to estimate the average
composite property assuming voids as inclusions with zero properties. More
sophisticated methods have also been developed to analyze the effect of voids on
overall composite elastic and failure properties [49].
Voids can also lead to appreciable inelastic deformations in the material locally,
which can act as precursors to initiation of damage processes, such as crazing,
b
W
j
0
s
s
Figure 3.7. Kink band geometry assumed in Budiansky [41]. Reprinted, with kind
permission, from Computers & Structures, Vol. 16, B. Budiansky, Micromechanics,
pp. 312, copyright Elsevier (1983).
shear yielding, brillation, and local fracture. These damage processes in the nal
stage may have signicant inuence on the deformation response and failure
properties of the composite material.
In composites with metallic and polymer matrices, the matrix phase undergoes
ductile fracture due to nucleation, growth and coalescence of voids and cavities.
These voids grow and expand due to high local inelastic strains and high stress
triaxiality in the matrix. Ductile fracture models, such as RiceTracy [50] can be
used to model the initiation and growth of voids in ductile matrices [51]. These
voids can sometimes coalesce to form matrix cracks, and may also cause ber
matrix debonds.
3.1.9 Damage modes
The damage mechanisms described above have different characteristics depending
on a variety of geometric and material parameters. Each mechanism has different
governing length scales and evolves differently when the applied load is increased.
Interactions between individual mechanisms further complicate the damage pic-
ture. As the loading increases, stress transfer takes place from a region of high
damage to that of low damage, and the composite failure results from the critical-
ity of the last load-bearing element or region. For clarity of treatment, the full
range of damage can be separated into damage modes, treating them individually
followed by examining their interactions.
Which damage mechanisms become active in a given life period of a composite
structure depends mainly on the properties of the base material (e.g., matrix),
architecture, orientation, distribution, and volume fraction of the reinforcing agent
(ber), the properties of the interface, and loading and environmental conditions.
Intralaminar and interlaminar cracking, ber fracture, and microbuckling are the
dominant damage mechanisms in long ber composites. Short ber composites
show three basic mechanisms of interfacial failure [52], as depicted in Figure 3.8:
Mode a: Localized matrix yielding at the interface due to the stress concen-
tration at the ber end (see Figure 3.8(a)). Typically, this occurs in combination
with debonding of the ber end and the formation of a penny-shaped crack.
Mode b: If the interface is relatively weak, an interface crack propagates from
the debonded ber end (Figure 3.8(b)). This is different than the ber end
penny-shaped crack and remains closed upon increase in tensile loading on
the composite, and the load transfer occurs by frictional stress transfer.
Mode g: If the interface is relatively strong, a conical matrix crack propagates
from the debonded ber end at an angle y
c
to the ber axis (Figure 3.8(c)). This
matrix crack opens with increasing applied load and suppresses load transfer
across the crack faces.
For particulate composites, the major damage mechanisms are dewetting
(debonding) of the particle and cavity nucleation [53] (see Figure 3.9). At a critical
tensile load, the particles separate from the matrix causing dewetting. Dewetting
of the particle eventually leads to cavity formation which grows on subsequent
loading. Dewetting introduces volume dilatation and results in nonlinearity in the
stressstrain behavior. For well-bonded particles, cavities and cracks may form
entirely within the matrix [54].
Damage modes in continuous ber laminates are thus rich in complexity. These
will be described below in the context of their evolution with loading.
3.2 Development of damage in composite laminates
A schematic description of damage development in composite laminates in tension
is depicted in Figure 3.10, where the ve identiable damage mechanisms are
indicated in the order of their occurrence. Although the gure is developed on
the basis of fatigue experiments [5560], it provides the basic details for quasi-
static loading as well.
In the early stage of damage accumulation, multiple matrix cracking dominates
in the layers which have bers aligned transverse to the applied load direction.
(a)
(b)
(c)
q
c
q
c
Figure 3.8. Failure mechanisms of interfacial failure in short ber/epoxy composites:
(a) mode a; (b) mode b; (c) mode g. Reprinted, with kind permission, from Compos Sci
Technol, Vol. 60, S. Sirivedin, D. N. Fenner, R. B. Nath, and C. Galiotis, Matrix crack
propagation criteria for model short-carbon bre/epoxy composites, pp. 283547,
Static tensile tests on cross-ply laminates have shown that the transverse matrix
cracks can initiate as early as at 0.4% applied strain depending upon the laminate
conguration. They initiate from the locations of defects such as voids, or areas of
high ber volume fraction or resin rich areas. Ply cracks grow unstably through
the width direction and quickly span the specimen width. As the applied load is
increased (or the specimen is loaded cyclically), more and more cracks appear. The
accumulation of ply cracks in a cracked ply is depicted in Figure 3.11. Initially
these cracks are irregularly spaced and isolated from each other, i.e., have no
interaction among themselves. However, as cracks become closer they start inter-
acting, i.e., the in-between tensile stresses diminish and can no longer build up to
earlier levels. Thus further increase in load is required to produce new cracks. This
is well illustrated in Figure 3.12 by plots of diminishing crack spacing versus load
or number of cycles. The conguration to which crack density saturates, often
reached only under fatigue loading, has been termed the characteristic damage
state (CDS) [5759]. This state seems to mark the termination of the intralaminar
cracking. The uniqueness of the CDS for a given laminate irrespective of the
loading path has, however, not been found to hold in all cases [61].
Subsequent loading causes initiation of cracks transverse to the primary (intra-
laminar) cracks lying in plies adjacent to the ones with those primary cracks (see
Figure 3.10). These cracks, known as secondary cracks, are small in size and they
can cause interfacial debonding, thereby initiating interlaminar cracks. The inter-
laminar cracks are also initially small, isolated and distributed in the interlaminar
planes. Subsequently, some interlaminar cracks merge into strip-like zones leading
to large scale delaminations. This results into loss of the integrity of the laminate
s s s
s s s
Matrix
Particle
Debond
Cavity
Figure 3.9. Damage mechanisms in particulate composites. Reprinted, with kind permission,
from Int J Solids Struct, Vol. 32, G. Ravichandran and C. T. Liu, Modeling constitutive
behavior of particulate composites undergoing damage, pp. 97990, copyright Elsevier (1995).
47 3.2 Development of damage in composite laminates
in those regions. Further development of damage is highly localized, increasing
unstably, and involving extensive ber breakage. The nal failure event is mani-
fested by the formation of a failure path through the locally failed regions and is
therefore highly stochastic.
The damage prior to localization is sometimes referred to as sub-critical
damage. The intralaminar (ply) cracking in this stage causes loss of stiffness
properties in the laminate and can by itself lead to loss of functionality (failure)
of the composite structure. The eld of damage mechanics addresses the initi-
ation and progression of the sub-critical damage. Later chapters will be devoted to
this subject. The next section discusses the phenomenon of multiple cracking and
its effects on overall (average) laminate response.
100
PERCENT OF LIFE
5. Fracture
D
A
M
A
G
E
CDS
0
0
1. Matrix Cracking
0
0
0
0
0
0
3. Delamination
2. Crack coupling-
Interfacial debonding
4. Fiber Breakage
Figure 3.10. Development of damage in composite laminates [62].
Figure 3.11. Accumulation of intralaminar cracks in an off-axis ply of a composite laminate.
Based on X-ray radiographs reported in [64].
3.3 Intralaminar ply cracking in laminates
One of the earliest observations of ply cracking in laminates was reported by
Broutman and Sahu [65]. However, the rst major explanation of multiple matrix
cracking was proposed by Aveston, Cooper, and Kelly [66, 67]. They argued that
multiple fracture occurs in a brous composite when one of the constituents (ber or
matrix) fractures at a much lower elongation than the other and when the unbroken
constituent is able to take the additional load; otherwise single fracture results. Later,
a group of experimentalists, Garret, Bailey, Parvizi, and colleagues [6874] carried
out signicant tests to analyze the ply cracking behavior in cross-ply laminates. These
experiments showed the initiation of microcracking in glass-ber reinforced polyster
and epoxy cross-ply laminates. It was observed that for thick 90
-plies, transverse
cracks initiated at the edge of the specimen and propagated instantly through the
width of entire cross section. As the 90
-plies were made thinner, the strain to initiate

transverse cracks increased (see Figure 3.13). For very thin 90
-plies (< 0.1 mm),

cracks were suppressed and the laminates failed before crack initiation. One of these
experiments [74] involved a microscopy study into the origins of matrix cracks and
revealed that they nucleate from the processing aws, voids and the regions of high
ber volume fraction, and progress through ber-matrix debonding.
The thickness effect on crack initiation can be explained in terms of the
constraint posed by uncracked plies over displacement of crack surfaces
in cracked plies (Figure 3.13). On one hand, as the thickness of (cracked) 90
-plies
0
0
0
0.2
4.0
6.0
8.0
10.0
C
r
a
c
k

s
p
a
c
i
n
g

(
m
m
)
2.0
0.4 0.6 0.8 1.0
100 200 300 400 500 600 700
Applied stress (MPa)
Cycles ( 10
6
)
Fatigue data
Quasistatic data
Figure 3.12. Spacing of cracks in 45
-plies of [0/90/ 45]

s
graphite/epoxy laminates as a
function of quasi-static and fatigue loading [57]. Reprinted, with kind permission, from
Damage in Composite Materials, ASTM STP 775, copyright ASTM International, 100 Barr
Harbor Drive, West Conshohocken, PA 19428.
49 3.3 Intralaminar ply cracking in laminates
increases, the relative constraint from (uncracked) 0
-plies decreases leading to ply

cracking at lower applied strains. On the other hand, thicker 0
-plies exert a larger

constraint on opening of cracks in 90
-plies, thereby delaying the crack initiation

in those plies. For qualitative understanding, Talreja [75] classied this constraint
in four categories: A no constraint; B low constraint; C high constraint; and
D full constraint. The stressstrain behavior for each of these four cases varies
greatly and is illustrated in Figure 3.14. On one extreme, it resembles an elastic-
rigid plastic like deformation behavior for constraint type A, and on the other end
a linear elastic behavior for constraint type D.
Over the past four decades, numerous approaches to analyze ply cracking in
composite laminates have been developed. They can be categorized into two broad
categories: micromechanics-based models (Chapter 4), and continuum damage
models (Chapter 5). Based on the dimensionality of the boundary value
problem, micromechanics-based models can be sub-divided into one-dimensional
[59, 7681], two-dimensional [8285], and three-dimensional [8688]. The next
chapter is devoted to these approaches.
3.4 Damage mechanics
Damage mechanics can be broadly dened as the subject dealing with mechanics-
based analyses of microstructural events in solids responsible for changes in their
0 2.0 4.0 6.0 8.0
2n Number of
90 plies
0
0.25
0.50
0.75
1.00
D
C
B
Gr./Ep. [0
4
/ 90
n
]
s
A
0 Ply failure
90 Ply failure
e
FPF
%
Figure 3.13. The strain at rst ply failure (e
FPF
) as a function of the number of transverse
plies in [0
4
/90
n
]
s
laminates. Source: [75].
response to external loading. The general objectives of damage mechanics analy-
sis are as follows:
1. Understand the conditions for initiation of the rst damage event.
2. Predict the evolution of progressive damage.
3. Characterize and quantify damage in the structure.
4. Analyze the effect of damage on thermomechanical response, e.g., by express-
ing stiffness properties as a function of damage.
5. Assess failure (criticality of damage) and durability of the structure.
6. Provide input into overall structural analysis and design.
This chapter provided an overview of damage development in composites.
The next three chapters will describe analysis methods for quasi-static loading.
Chapter 4 will describe micro-damage mechanics (MIDM), whereas Chapter 5
will describe the macro-damage mechanics (MADM). Evolution of damage will
be covered in Chapter 6. Damage in fatigue and models for lifetime prediction will
be treated in Chapter 7.
e
FPF
e
FPF
e
FPF
e
c
e
c
e
c
e
c e
e
e
s s
s s
(a)
(b)
(c)
(d)
Figure 3.14. Schematic stressstrain response of cross-ply laminates at different constraints to
transverse cracking: (a) single fracture, no constraint; (b) multiple fracture, low constraint;
(c) multiple fracture, high constraint; (d) multiple fracture, full constraint (with crack
suppression). Source: [75].
51 3.4 Damage mechanics
References
1. Military handbook, MIL-HDBK-173f: Composite Materials Handbook, Vol. 3, Depart-
ment of Defense, USA, 2002.
2. E. K. Gamstedt and B. A. Sjo gren, Micromechanisms in tension-compression fatigue of
composite laminates containing transverse plies. Compos Sci Technol, 59:2 (1999), 16778.
3. C. H. Hsueh, Interfacial debonding and ber pull-out stresses of ber-reinforced
composites. Mater Sci Eng A, 154 (1992), 12532.
4. C. H. Hsueh, Elastic load transfer from partially embedded axially loaded bre to
matrix. J Mater Sci Lett, 7:5 (1988), 497500.
5. C. H. Hsueh, Modeling of elastic stress transfer in ber-reinforced composites. Trends
Polym Sci, 3:10 (1995), 33641.
6. B. Lauke, W. Beckert, and J. Singletary, Energy release rate and stress eld calculation
for debonding crack extension at the bre-matrix interface during single-bre pull-out.
Compos Interfaces, 3 (1996), 26373.
7. L. M. Zhou, J. K. Kim, and Y. W. Mai, Interfacial debonding and ber pull-out
stresses. J Mater Sci, 27 (1992), 315566.
8. L. T. Drzal, The effect of polymer matrix mechanical properties on the bre-matrix
interfacial strength. Mater Sci Eng A, 126 (1990), 28903.
9. J. M. Whitney and L. T. Drzal, Axisymmetric stress distribution around an isolated
bre fragment. In Toughened Composites, ed. N. J. Johnston. (Philadelphia, PA: ASTM,
1987), pp. 179196.
10. J. P. Fabre, P. Sigety, and D. Jacques, Stress transfer by shear in carbon ber model
composites. J Mater Sci, 26 (1991), 18995.
11. R. B. Henstenburg and S. L. Phoenix, Interfacial shear strength studies using single
lament composite test. Polym Compos, 10 (1989), 389408.
12. C. H. Hsueh, Interfacial debonding and ber pull-out stresses of ber-reinforced
composites. Mater Sci Eng A, 123:1 (1990), 111.
13. S. Ghosh, Y. Ling, B. Majumdar, and R. Kim, Interfacial debonding analysis in
multiple ber reinforced composites. Mech Mater, 32:10 (2000), 561591.
14. P. Raghavan and S. Ghosh, A continuum damage mechanics model for unidirectional
composites undergoing interfacial debonding. Mech Mater, 37:9 (2005), 955979.
15. N. Chandra and H. Ghonem, Interfacial mechanics of push-out tests: Theory and
experiments. Compos A, 32:34 (2001), 575584.
16. G. Lin, P. H. Geubelle, and N. R. Sottos, Simulation of ber debonding with friction in
a model composite pushout test. Int J Solids Struct, 38:4647 (2001), 854762.
17. Y. F. Liu and Y. Kagawa, Analysis of debonding and frictional sliding in ber-reinforced
brittle matrix composites: Basic problems. Mater Sci Eng A, 212:1 (1996), 7586.
18. B. Harris, Micromechanisms of crack extension in composites Metal Sci, 14 (1980),
351362.
19. J. A. Nairn, Matrix microcracking in composites. In Polymer Matrix Composites, ed.
R. Talreja and J. A. E. Manson. (Elsevier Science, 2000), pp. 40332.
20. C. A. Wood and W. L. Bradley, Determination of the effect of seawater on the interfacial
strength of an interlayer E-glass/graphite/epoxy composite by in-situ observation of
transverse cracking in an environmental sem. Compos Sci Technol, 57:8 (1997), 103343.
21. D. T. G. Katerelos, J. Varna, and C. Galiotis, Energy criterion for modelling damage
evolution in cross-ply composite laminates. Compos Sci Technol, 68:12 (2008), 231824.
22. R. Talreja, Damage in woven fabric polymer laminates. (unpublished).
23. D. B. Marshall and A. G. Evans, Failure mechanisms in ceramic-ber/ceramic-matrix
composites. J Amer Ceram Soc, 68:5 (1985), 22531.
24. W. J. Cantwell and J. Morton, The impact resistance of composite materials a review.
Composites, 22 (1991), 34762.
25. W. C. Chung, et al., Fracture behaviour in stitched multidirectional composites. Mater
Sci Eng A, 112 (1989), 15773.
26. D. Liu, Delamination resistance in stitched and unstitched composite planes subjected
to composite loading. J Reinf Plast Compos, 9 (1990), 5969.
27. J. C. Prichard and P. J. Hogg, The role of impact damage in post-impacted compres-
sion testing. Composites, 21 (1990), 50311.
28. B. Z. Jang, M. Cholakara, B. P. Jang, and W. K. Shih, Mechanical properties in
multidimensional composites. Polym Eng Sci, 31 (1991), 406.
29. K. B. Su, Delamination resistance of stitched thermoplastic matrix composite lamin-
ates. In Advances in Thermoplastic Matrix Composite Materials, ASTM STP 1044.
(Philadelphia, PA: ASTM, 1989), pp. 279300.
30. N. S. Choi, A. J. Kinloch, and J. G. Williams, Delamination fracture of multidirec-
tional carbon-ber/epoxy composites under mode I, mode II and mixed-mode I/II
loading. J Compos Mater, 33:1 (1999), 73100.
31. I. Verpoest, M. Wevers, P. DeMeester, and P. Declereq, 2.5D and 3D fabrics for
delamination resistant composite laminates and sandwich structure. SAMPE J, 25
(1989), 516.
32. D. J. Elder, R. S. Thomson, M. Q. Nguyen, and M. L. Scott, Review of delamination
predictive methods for low speed impact of composite laminates. Compos Struct, 66
(2004), 67783.
33. W. S. Chan, Design approaches for edge delamination resistance in laminated compos-
ites. J Compos Tech Res, 14 (1991), 916.
34. A. Mayadas, C. Pastore, and F. K. Ko, Tensile and shear properties of composites by
various reinforcement concepts. In Advancing Technology in Materials and Processes,
SAMPE 30th National Meeting (1985), pp. 128493.
35. R. Jamison, On the interrelationship between ber fracture and ply cracking in
graphite/epoxy laminates. In Composite Materials: Fatigue and Fracture, ASTM STP
907, ed. H. T. Hahn. (Philadelphia, PA: ASTM, 1986), pp. 25273.
36. N. J. Pagano, On the micromechanical failure modes in a class of ideal brittle matrix
composites. Part 1. Coated-ber composites. Compos B, 29:2 (1998), 93119.
composites. Part 2. Uncoated-ber composites. Compos B, 29:2 (1998), 12130.
38. J. K. Kim and Y. W. Mai, High strength, high fracture toughness bre composites with
interface control a review. Compos Sci Technol, 41:4 (1991), 33378.
39. B. W. Rosen, Tensile failure of brous composites. AIAA J, 2 (1964), 198591.
40. A. Argon, Fracture of composites. In Treatise on Materials Science and Technology,
ed. H. Herman. (New York, London: Academic Press, 1972), pp. 79114.
41. B. Budiansky, Micromechanics. Computers & Structures, 16 (1983), 312.
42. K. Niu and R. Talreja, Modeling of compressive failure in ber reinforced composites.
Int J Solids Struct, 37:17 (2000), 240528.
43. C. Broeckmann and R. Pandorf, Inuence of particle cleavage on the creep behaviour
of metal matrix composites. Comput Mater Sci, 9 (1997), 4855.
53 References
44. T. Antretter and F. D. Fischer, Particle cleavage and ductile crack growth in a two-
phase composite on a microscale. Comput Mater Sci, 13 (1998), 17.
45. H. J. Bo hm, A. Eckschlager, and W. Han, Multi-inclusion unit cell models for metal
matrix composites with randomly oriented discontinuous reinforcements. Comput
Mater Sci, 25 (2002), 4253.
46. A. Eckschlager, W. Han, and H. J. Bo hm, A unit cell model for brittle fracture of
particles embedded in a ductile matrix. Comput Mater Sci, 25 (2002), 8591.
47. J. Segurado and J. Llorca, A new three-dimensional interface nite element to simulate
fracture in composites. Int J Solids Struct, 41 (2004), 297793.
48. T. Mori and K. Tanaka, Average stress in matrix and average elastic energy of
materials with mistting inclusions. Acta Metall, 21:5 (1973), 5714.
49. H. Huang and R. Talreja, Effects of void geometry on elastic properties of unidirec-
tional ber reinforced composites. Compos Sci Technol, 65:13 (2005), 196481.
50. J. R. Rice and D. M. Tracey, On the ductile enlargement of voids in triaxial stress elds.
J Mech Phys Solids, 17 (1969), 20117.
51. H. S. Huang and R. Talreja, Numerical simulation of matrix micro-cracking in short
ber reinforced polymer composites: initiation and propagation. Compos Sci Technol,
66:15 (2006), 274357.
52. S. Sirivedin, D. N. Fenner, R. B. Nath, and C. Galiotis, Matrix crack propagation
criteria for model short-carbon bre/epoxy composites. Compos Sci Technol, 60:15
(2000), 283547.
53. G. Ravichandran and C. T. Liu, Modeling constitutive behavior of particulate com-
posites undergoing damage. Int J Solids Struct, 32 (1995), 97990.
54. L. R. Cornwell and R. A. Schapery, SEM study of microcracking in strained solid
propellant. Metallography, 8 (1975), 44552.
55. K. L. Reifsnider and A. Talug, Analysis of fatigue damage in composite laminates. Int
J Fatigue, 2:1 (1980), 311.
56. W. W. Stinchcomb, K. L. Reifsnider, P. Yeung, and J. M. Masters, Effect of ply constraint
on fatigue damage development in composite material laminates. In Fatigue of Fibrous
Composite Materials, ASTM STP 723. (Philadelphia, PA: ASTM, 1981), pp. 6484.
57. J. M. Masters and K. L. Reifsnider, An investigation of cumulative damage develop-
ment in quasi-isotropic graphite/epoxy laminates. In Damage in Composite Materials,
ASTM STP 775, ed. K. L. Reifsnider. (Philadelphia, PA: ASTM, 1982), pp. 4062.
58. K. L. Reifsnider and R. Jamison, Fracture of fatigue-loaded composite laminates. Int
J Fatigue, 4:4 (1982), 187197.
59. A. L. Highsmith and K. L. Reifsnider, Stiffness-reduction mechanisms in composite
laminates. In Damage in Composite Materials, ASTM STP 775, ed. K. L. Reifsnider.
60. R. D. Jamison, K. Schulte, K. L. Reifsnider, and W. W. Stinchcomb, Characterization
and analysis of damage mechanisms in tensiontension fatigue of graphite/epoxy
laminates. In Effects of Defects in Composite Materials, ASTM STP 836. (Philadelphia,
PA: ASTM, 1984), pp. 2155.
life of composite laminates. Mater Sci Eng A, 285:12 (2000), 30313.
62. R. Talreja, Internal variable damage mechanics of composite materials. In Yielding,
Damage, and Failure of Anisotropic Solids, ed. J. P. Boehler. (London: Mechanical
Engineering Publications, 1990), pp. 50933.
63. Z. Hashin, Analysis of damage in composite materials. In Yielding, Damage, and Failure
of Anisotropic Solids, ed. J. P. Boehler. (London: Mechanical Engineering Publications,
1990), pp. 331.
64. A. S. D. Wang, Fracture mechanics of sublaminate cracks in composite materials. In
Composites Technology Review. (Philadelphia, PA: ASTM, 1984), pp. 4562.
65. L. J. Broutman and S. Sahu, Progressive damage of a glass reinforced plastic during
fatigue. In SPI, 24th Annual Technical Conference, Section 11-D. (Washington, DC: SPI,
1969).
ical Laboratory, 1971).
8:3 (1973), 35262.
68. K. W. Garrett and J. E. Bailey, Effect of resin failure strain on tensile properties of glass
ber-reinforced polyester cross-ply laminates. J Mater Sci, 12:11 (1977), 218994.
69. K. W. Garrett and J. E. Bailey, Multiple transverse fracture in 90
cross-ply laminates
of a glass bre-reinforced polyester. J Mater Sci, 12:1 (1977), 15768.
70. A. Parvizi, K. W. Garrett, and J. E. Bailey, Constrained cracking in glass bre-
reinforced epoxy cross-ply laminates. J Mater Sci, 13:1 (1978), 195201.
71. M. G. Bader, J. E. Bailey, P. T. Curtis, and A. Parvizi, eds. The mechanisms of
initiation and development of damage in multi-axial bre-reinforced plastic laminates.
Proc Third Int Conf Mech Behav Mater (ICM3), Vol. 3. (Cambridge, 1979), pp. 22739.
72. J. E. Bailey, P. T. Curtis, and A. Parvizi, On the transverse cracking and longitudinal
splitting behaviour of glass and carbon bre reinforced epoxy cross-ply laminates and
the effect of Poisson and thermally generated strain. Proc R Soc London A, 366:1727
(1979), 599623.
73. J. E. Bailey and A. Parvizi, On ber debonding effects and the mechanism of
transverse-ply failure in cross-ply laminates of glass ber-thermoset composites.
J Mater Sci, 16:3 (1981), 64959.
74. A. Parvizi and J. E. Bailey, Multiple transverse cracking in glass-ber epoxy cross-ply
laminates. J Mater Sci, 13:10 (1978), 21316.
J Compos Mater, 19:4 (1985), 35575.
76. P. W. Manders, T. W. Chou, F. R. Jones, and J. W. Rock, Statistical analysis of multiple
fracture in [0/90/0] glass ber/epoxy resin laminates. J Mater Sci, 19 (1983), 287689.
77. H. Fukunaga, T. W. Chou, P. W. M. Peters, and K. Schulte, Probabilistic failure strength
analysis of graphite epoxy cross-ply laminates. J Compos Mater, 18:4 (1984), 33956.
78. H. Fukunaga, T. W. Chou, K. Schulte, and P. W. M. Peters, Probabilistic initial failure
strength of hybrid and non-hybrid laminates. J Mater Sci, 19:11 (1984), 354653.
79. P. S. Steif, Parabolic shear lag analysis of a [0/90]
s
laminate. Transverse ply crack growth
and associated stiffness reduction during the fatigue of a simple cross-ply laminate. In
S. L. Ogin, P. A. Smith, and P. W. R. Beaumont (eds.), Report CUED/C/MATS/TR
105, Cambridge University, Engineering Department, UK (September 1984).
80. R. J. Nuismer and S. C. Tan, Constitutive relations of a cracked composite lamina.
J Compos Mater, 22:4 (1988), 30621.
81. S. C. Tan and R. J. Nuismer, A theory for progressive matrix cracking in composite
laminates. J Compos Mater, 23:10 (1989), 102947.
55 References
82. Z. Hashin, Analysis of cracked laminates: a variational approach. Mech Mater, 4:2
(1985), 12136.
83. J. A. Nairn, The strain energy release rate of composite microcracking: a variational
approach. J Compos Mater, 23:11 (1989), 110629.
85. L. N. McCartney, Theory of stress transfer in a 0-degrees-90-degrees-0-degrees cross-
ply laminate containing a parallel array of transverse cracks. J Mech Phys Solids, 40:1
(1992), 2768.
86. P. Gudmundson and W. L. Zang, An analytic model for thermoelastic properties of
composite laminates containing transverse matrix cracks. Int J Solids Struct, 30:23
(1993), 321131.
87. E. Adolfsson and P. Gudmundson, Thermoelastic properties in combined bending and
extension of thin composite laminates with transverse matrix cracks. Int J Solids Struct,
34:16 (1997), 203560.
88. P. Lundmark and J. Varna, Constitutive relationships for laminates with ply cracks in
in-plane loading. Int J Damage Mech, 14:3 (2005), 23559.
4 Micro-damage mechanics
4.1 Introduction
As explained in the previous chapter, damage affects the overall stressstrain
response of the solid continuum body. Damage mechanics pertains to the study
of this effect. Two widely different subelds have emerged over the years in this
eld. One concerns study of damage directly at the scale of formation of cracks,
i.e., the microstructural scale, and hence can be called micro-damage mechanics
(MIDM). The other approach, on the contrary, looks at the overall response at
the macro or structural scale by using some internal variables to characterize
damage, and thus can be termed as macro-damage mechanics (MADM). These
terms were originally coined by Hashin [1]. MADM is the same as continuum
damage mechanics (CDM), which is still the commonly used terminology.
MIDM for composite materials is derived from an older and more mature eld
called micromechanics that deals with overall properties of heterogeneous mater-
ials (see, e.g., [2]). In micromechanics one views heterogeneities such as inclusions
and voids as microstructure and estimates overall properties by various
methods, e.g., averaging schemes such as self-consistent and differential schemes,
or variational methods to obtain bounds to average properties. Microcracks are
treated as limiting geometry of microvoids, such as ellipsoidal voids with one
dimension much smaller than the other two. As illustrated in the previous chapter,
damage in composite materials has signicant complexities concerning the
geometry as well as evolution characteristics such as multiplication of cracks
within a xed volume. For these reasons a simple extension of micromechanics
to damage in composites is generally not possible. A separate eld identied as
MIDM has therefore emerged. This chapter will treat the features of MIDM that
have been developed to specically treat certain cases of damage in composite
materials. Since determining local (micro-level) stress or displacement elds is a
necessary feature of micromechanics, it is expected that not all cases within the
wide range of damage in composites can be handled by MIDM. However, this
limitation can be alleviated by incorporating computational solutions of the local
stress or displacement elds, thereby broadening classical micromechanics to
include so-called computational micromechanics. In the most recent versions of
MIDM this strategy has been used. More on this will be discussed toward the end
of this chapter.
In the following we will rst treat the aspect of damage in composite mater-
ials that is due to the presence of continuous interfaces between dissimilar
materials, such as bers and matrix or plies oriented differently in a laminate.
In fact this aspect is fundamental to understanding damage in composite
materials. Historically, it was rst analyzed in a classical work by Aveston,
Cooper, and Kelly [3] who explained conditions that lead to failure of a
composite from a single crack versus when multiple cracks precede the nal
failure. Their work has become known as the ACK theory. Although the case
treated by them is of simple geometry and loading, namely a unidirectional
ber-reinforced brittle matrix composite loaded in tension along bers, it
explains the basic mechanism underlying multiple cracking in a wide range of
cases. The simplied stress analysis and the associated energy balance consider-
ations in the ACK paper have later been extended to include more accurate
solutions, but little further insight into the multiple cracking mechanism has
resulted by these efforts.
4.2 Phenomena of single and multiple fracture: ACK theory
The mode of failure (separation in two or more pieces) in homogeneous mater-
ials such as metals and ceramics may be described as single fracture in the
sense that the failure is attributable to a single source a crack. Heterogeneous
materials, on the other hand, can fail in the mode of single fracture or sustain
multiple fractures of one of the phases before ultimately separating in two or
more pieces. The latter phenomenon is known as multiple fracture and can
commonly occur in brous composites with brittle matrices, such as cement
plaster, glass, etc.
Damage in composite materials usually initiates with matrix cracking.
Figure 4.1 depicts the isotropic stressstrain response of an unreinforced glass
and of a ber-reinforced glass in tension loading along bers. Other than the
enhancement of the stressstrain response in the ber direction, a striking aspect is
the nonlinearity shown by the reinforced specimen. This nonlinearity is a type of
ductility which occurs due to multiple cracking of the matrix (glass) [4].
Although the phenomenon of multiple fracture was observed earlier, e.g., by
Cooper and Sillwood [5] it was systematically investigated in a landmark paper by
Aveston, Cooper, and Kelly [3]. Their work, the ACK theory, is the basis of the
treatment presented below.
Consider a unidirectional brous composite loaded in tension along bers as
shown in Figure 4.2 and assume the following:
1. Fibers are of the same diameter and uniformly distributed in the matrix.
2. All the bers are aligned parallel to one another.
3. There are no pre-existing aws in the matrix such as voids and cracks.
4. Both the matrix and the bers are linearly elastic.
Assuming that bers and matrix have different failure strain in tension, when one
of the constituents fails, the other will either fail simultaneously or continue
deforming by carrying the additional load. In the latter case, the constituent that
failed rst will fail again at a different site. Thus, there are two necessary condi-
tions for multiple fracture to occur in a composite:
1. One of the constituents has a lower failure strain than the other.
2. When the weaker constituent fails, i.e., when it no longer carries any load,
the stronger constituent must be able to carry the additional load thrown
upon it.
If P
c
is the total tensile load on the composite and P
f
and P
m
represent the load
taken up by the bers and matrix, respectively, then by force balance we have
P
c
= P
f
P
m
: (4:1)
s
m
s
f
s
c
Figure 4.2. A unidirectional brous composite loaded in tension.
0
200
S
t
r
e
s
s
/
M
P
a
400
600
800
1000
0.4 0.8 1.2
Strain (%)
Unreinforced glass
Fibre-
reinforced
glass
Figure 4.1. The stressstrain curves for borosilicate glass alone (dotted line) and
reinforced with aligned carbon bers. Reprinted from [4] with kind permission
from Royal Society Publishing, London.
59 4.2 Phenomena of single and multiple fracture: ACK theory
Dividing by the composites cross-sectional area A, we get
P
c
A
=
P
f
A

P
m
A
; (4:2)
or
P
c
A
=
P
f
A
f
A
f
A

P
m
A
m
A
m
A
: (4:3)
Assuming unit composite length, we obtain
s
c
= s
f
V
f
s
m
V
m
; (4:4)
where V
f
and V
m
are the volume fractions of bers and matrix, respectively.
Depending upon which phase fails rst, two cases as shown in Figure 4.3 arise. Case
1 is whenbers have a lower breaking strainthan the matrix (e
fu
<e
mu
), while Case 2 is
for the opposite (e
mu
< e
fu
). In the rst case, bers will undergo multiple fracture if the
matrix is able to carry the additional load thrown upon it due to ber failure, i.e., if
P
mu
> P
c
[
e
fu
;
i:e:; if s
mu
V
m
> s
fu
V
f
s
/
m
V
m
(4:5)
where P
mu
is the maximum load that can be carried by the matrix, s
fu
and s
mu
are the tensile strength values for bers and matrix, respectively, and
s
/
m
= E
m
e
fu
= (s
mu
=e
mu
)e
fu
is the stress inthe matrix requiredtoproduce a strainequal
to the breaking strain of the bers. In this case, the bers will be successively fractured
intoshorter lengths until the matrix attains its failure strainandat that instant the whole
compositefails. Ontheother hand, for Case2, thematrixwill undergomultiplefractureif
P
fu
> P
c
[
e
mu
;
i:e:; if s
fu
V
f
> s
mu
V
m
s
/
f
V
f
;
(4:6)
A
Multiple
fracture
of fibers
s s
s
f
Single
fracture
V
f
s
m
f
s
m
(a)
A
Multiple
fracture
of matrix
Single
fracture
V
f
(b)
s
m
Figure 4.3. Single and multiple fractures in a unidirectional composite. Fracture stress is
plotted against the ber volume fraction: (a) case 1: e
mu
> e
fu
; (b) case 2: e
mu
< e
fu
.
where P
mu
is the maximum load that can be carried by the matrix, s
fu
and s
mu
are the tensile strength values for bers and matrix, respectively, and
s
/
f
= E
f
e
mu
= (s
fu
=e
fu
)e
mu
is the stress in the bers required to produce a strain
equal to the breaking strain of the matrix.
4.2.1 Multiple matrix cracking
From this point on we will focus on the case of multiple matrix cracking, assuming
e
mu
< e
fu
. In addition to the assumptions described in the previous section, the
following analysis will assume:
1. Fibers remain intact throughout entire loading history.
2. Matrix cracks extend in the entire cross section.
3. Fibers debond completely between adjacent matrix cracks.
If we concentrate on the matrix region between two bers, the force P
m
shed by the
matrix, subsequent to its failure, is carried by bers in the cracked cross section,
and is transferred back to the matrix over a distance x
/
. This load transfer takes
place through shear at the ber/matrix interface with constant shear stress t. The
mechanism of interfacial load transfer is illustrated in Figure 4.4. The load balance
between the total matrix load P
m
and the total shear force at the ber-matrix
interfaces yields
P
m
= s
mu
A
m
= t 2pr x
/
N; (4:7)
where r is the ber radius and N is the number of bers in the composite cross
section of area A.
P
m
x
Figure 4.4. Mechanism of load transfer at the ber/matrix interface.

Thus,
x
/
=
s
mu
t
_ _
A
m
2prN
: (4:8)
Now,
A
m
2prN
=
A
m
=A
2prN=A
=
V
m
:r
2(N:pr
2
=A)
=
V
m
:r
2V
f
: Hence,
x
/
=
s
mu
t
_ _
V
m
V
f
r
2
: (4:9)
The assumptionof constant shear stress at the interface simplies the analysis. However,
its inaccuracycanbe notedbyrealizingthat the shear stress at the point where the matrix
crack meets the interface must vanish if the crack surface is to remain traction free.
Stress distribution in bers and matrix
The balance of forces for a piece of ber of length Dx
/
, over which the change in
the ber stress is Ds
f
, yields
s
f
:pr
2
= t 2pr x
/
: (4:10)
Thus the rate of load transfer is a constant given by
s
f
x
/
=
t
r
; (4:11)
and consequently the ber stress s
f
varies linearly, and correspondingly s
m
also
varies linearly, along the ber axis. The maximum stress in ber occurs at the
matrix crack and can be determined as
s
f;max
= s
f

P
mu
A
f
= s
f
s
mu
A
m
A
f
= s
f
s
mu
V
m
V
f
: (4:12)
Additional strain due to cracking The strain in bers increases from e
mu
at
cracking to the maximum value given by
e
f; max
=
s
f;max
E
f
=
s
f
E
f
s
mu
E
f
V
m
V
f
= e
f

s
mu
E
m
E
m
E
f
V
m
V
f
= e
mu
1 a ( ); (4:13)
with
a =
E
m
E
f
V
m
V
f
: (4:14)
The mean strain over crack spacing 2x
/
is equal to
e
mean;2x
/ =
1
2
e
mu
e
mu
1 a ( ) [ [ = e
mu
1
a
2
_ _
: (4:15)
When the crack spacing reduces to x
/
, the mean strain increases to
e
mean;x
/ = e
mu
1
3a
4
_ _
: (4:16)
Energy considerations in multiple cracking
Consider the unidirectional composite of Figure 4.2 at xed applied load P
c
=
s
c
A. Let its initial conguration be denoted as state 1 and let state 2 refer to its
conguration with multiple matrix cracking. The energy changes in going from
state 1 to state 2 are described below.
Supply of energy
1. DW: the work done per unit cross-sectional area A by an external (xed) load
through specimen extension caused by cracking is given by
W =
1
A
P
c
2x
/
= s
c
2 e
mu
a
2
x
/
_ _
= E
c
e
mu
e
mu
ax
/
= E
c
e
2
mu
ax
/
:
(4:17)
2. DU
m
: the reduction in elastic strain energy of matrix over distance 2x
/
is given by
U
m
= 2
_
x
/
0
1
2
E
m
V
m
e
2
mu

1
2
E
m
V
m
e
mu
x
x
/
_ _
2
_ _
dx
=
2
3
E
m
V
m
e
2
mu
x
/
=
E
m
V
m
3t
e
3
mu
ar:
(4:18)
Consumption of energy
1. Energy spent in formation of matrix cracks. If g
m
is the surface energy per unit
area of crack surface, then the energy spent in formation of a matrix crack per
unit cross-sectional area A is
2g
m
A
m
A
= 2g
m
V
m
: (4:19)
2. Energy spent in ber/matrix interfacial debonding. If we take G
II
to be the
energy released per unit area of the debond surface, then the debond energy g
db
per unit cross-sectional area A can be expressed as
g
db
A = G
II
2pr 2x
/
N; (4:20)
i.e.,
g
db
= G
II

2prN
A
2x
/
= 2G
II

A
f
Ar

V
m
V
f
s
mu
t
r
= 2G
II
V
m
s
mu
t
:
(4:21)
3. U
s
: the energy spent in sliding of the matrix onto the ber surface over a
distance 2x
/
, per unit cross-sectional area A is given by
U
s
=
1
A
N 2
_
x
/
0
v t 2pr dx (4:22)
where Dn is the sliding displacement at x. This sliding displacement is equal to the
difference in displacements of the ber and the matrix. It can be found by
integrating the strain in matrix and ber:
_
x
/
0
tv dx = t
_
x
/
0
e
mu
1 a ( )x
a
2
x
2
x
/
x
/
1
a
2
_ _
x
2
x
/

x
/
2
_ _
dx
=
te
mu
x
/
2
6
1 a ( ):
(4:23)
Thus,
U
s
=
E
f
E
m
V
m
6t
e
3
mu
ra 1 a ( ): (4:24)
4. DU
f
: the increase in the elastic energy of bers due to additional extension
caused by additional ber stress, per cross-sectional area A, is given by
U
f
= U
(2)
f
U
(1)
f
= 2
_
x
/
0
1
2
E
f
V
f
e
mu
a 1
x
x
/
_ _
e
mu
_ _
2
1
2
E
f
V
f
e
2
mu
_ _
dx
= E
f
V
f
e
2
mu
x
/
a 1
a
3
_ _
=
E
f
E
m
V
m
2t
e
3
mu
ra 1
a
3
_ _
:
(4:25)
Conditions for multiple matrix cracking
1. Stress in the matrix is greater than or equal to the matrix failure stress, i.e.,
s
m
_ s
mu
or e
m
_ e
mu
: (4:26)
2. The supply of energy in going from state 1 to state 2 is greater than or equal
to the consumption of energy, i.e.,
2g
m
V
m
g
db
U
s
U
f
_ W U
m
: (4:27)
Substituting Eqs. (4.17)(4.25) derived above into Eq. (4.27), one obtains
2V
m
g
m
G
II
s
mu
t
_ _
_
E
c
E
f
e
3
mu
a
2
r
6t
: (4:28)
It can be argued that the energy term G
II
is much smaller than the other energy
contributions. Assuming G
II
= 0 then gives
2V
m
g
m
_
E
c
E
f
e
3
mu
a
2
r
6t
: (4:29)
Thus, the strain required to cause multiple matrix cracking is given by the
following expression
e
muc
=
12tg
m
E
f
V
2
f
E
c
E
m
V
m
r
_ _
1=3
: (4:30)
Stressstrain response
When the composite is loaded to an applied strain level equal to the failure strain of
the matrix, multiple cracking in the matrix starts occurring. If the matrix has a well-
dened single-valued breaking strain, the cracking will continue at a constant applied
stress E
c
e
mu
until the matrix is broken down into a set of blocks of length between x
/
and 2x
/
. The composite stressstrain behavior subsequent to multiple fracture of the
matrix is shown in Figure 4.5. During multiple matrix cracking (Point A to Point B),
the mean strain varies between e
mu
1
a
2
_ _
and e
mu
1
3a
4
_ _
while going from a crack
spacing of 2x
/
to x
/
. The total strain at the limit of multiple cracking e
mc
is therefore
e
mu
1
a
2
_ _
<e
mc
<e
mu
1
3a
4
_ _
: (4:31)
When the applied load is increased, bers are stretched further and start slipping
through the blocks of matrix (point Bto Point C). Since the matrix can no longer take
any load the Youngs modulus of the specimen is reduced to E
f
V
f
. The composite will
eventually fail at a stress s
fu
V
f
. The failure strain of the composite e
cu
is given by
e
fu

ae
mu
2
_ _
<e
cu
< e
fu

ae
mu
4
_ _
: (4:32)
4.2.2 Perfectly bonded ber/matrix interface: a modied shear lag analysis
The previous analysis was based on the assumption that the bers debond com-
pletely from the matrix during the fracture process. In reality the displacements of
matrix and bers are interrelated. The complete debonding scenario can be viewed
as one extreme, the other extreme being the fully bonded case. For the latter, a
modied shear lag model [6] is applicable.
1+
A
C
fu f
V s
s
D
O
B
1+
4
3a
4 2
2
E
c
E
f
V
f
ae
mu
ae
mu
mu
e
mu
e
mu
e
fu
e
fu
e
e
fu
e
a
Figure 4.5. The stressstrain response subsequent to multiple fracture for a unidirectional
composite according to the ACK theory.
For both cases the fundamental equation governing load transfer between bers
and matrix is obtained from a simple force balance and is, for discrete bers of
radius r in a continuous matrix, given by
dF
dy
=
2V
f
t
i
r
; (4:33)
where dF is the load per cross-sectional area A transferred over the distance dy and
t
i
is the shear stress acting at the interface.
For the unbonded case, the load transfer from ber to matrix is given by
Eq. (4.11), whereas for the bonded case the bers in the plane of the rst crack
will be subjected to an additional stress, Ds
0
, given by
s
0
=
s
a
V
f
E
f
e
mu
; (4:34)
where s
a
is the applied stress. This additional stress has its maximum value at the
plane of the matrix crack and decays with distance from the crack surface. Using a
modied shear lag analysis, Aveston and Kelly [6] showed that the variation of
this additional stress along the ber is given by
s y ( ) = s
0
e

w
_
y
; (4:35)
with
w =
2G
m
E
c
E
f
E
m
V
m
_ _
1
r
2
ln
R
r
_ _ ; (4:36)
where G
m
is the shear modulus of the matrix.
Carrying out a force balance over an element of ber of length dy,
ds pr
2
t
i
2pr dy = 0; (4:37)
from which the rate of change of stress in the ber is given by
ds
dy
=
2
r
t
i
: (4:38)
Differentiating Eq. (4.35) w.r.t. y and substituting into Eq. (4.38), the shear stress
at the interface between the bers and matrix is given by
t
i
=
r
2
s
0
we

w
_
y
: (4:39)
Substituting Eq. (4.39) into Eq. (4.33), and integrating, the load F per cross-
sectional area A transferred to the matrix over any distance l from the crack
surface can be found as
F = V
f
s
0
1 e

w
_
l
_ _
: (4:40)
If Ds
0
_ s
mu
(V
m
/V
f
), the matrix will undergo further cracking into blocks of
length between l and 2l, where l is obtained by setting F = s
mu
V
m
into Eq. (4.40),
l =
1
w
_ ln 1
s
mu
s
0
V
m
V
f
_ _
: (4:41)
The energetic considerations in this case result into the crack initiation strain for
the matrix, as given by
e
mu
=
2g
m
V
m
w
_
aE
c
_ _
1=2
: (4:42)
The effective Youngs modulus for the matrix can be determined by averaging the
stress distribution in the matrix, i.e.,
E
m
=
1
e
mu
1
s
_
s
0
s
m
y ( )dy; (4:43)
where 2s is the mutual crack spacing, and the stress distribution in the matrix is
given by
s
m
y ( ) =

E
m
e
mu
s
V
f
V
m
: (4:44)
Using Eqs. (4.34), (4.43), and (4.44),

E
m
can be derived as [7]
E
m
= E
m

E
m
s

w
_ e
s

w
_
1
_
: (4:45)
The effective modulus for the unidirectional composite can then be determined
using the rule of mixture
E
c
= E
f
V
f

E
m
V
m
: (4:46)
4.2.3 Frictional ber/matrix interface
As suggested by Wang and Parvizi-Majidi [7], another important case is to assume
that the ber/matrix interface is frictional. For this case, the load transfer between
ber and matrix is assumed constant. After the development of matrix cracking,
the additional stress transferred to the ber at the crack surface should be
transferred back to the matrix by a constant interfacial shear stress over the
distance of the limiting matrix crack spacing, x. Accordingly, the stress in the
matrix away from the crack surface will vary linearly, from a zero value at the crack
to a maximum of s
m
at a distance x from the crack. This stress will stay at its
maximum level for distances longer than x. The average stress and the stiffness
values in the matrix can therefore be represented by the following equations
s
m
= s
m
2s x
x
;
E
m
= E
m
2s x
x
:
(4:47)
The overall stiffness for the composite can then be calculated from Eq. (4.46).
The one-dimensional analyses of stress transfer used in the ACK theory and its
later version are inadequate to give accurate prediction, e.g., of the strain to onset
of multiple cracking. The main value of these works was in explaining the phe-
nomenon of multiple cracking. In subsequent sections we will address the problem
of multiple cracking more generally and describe a wide range of approaches
taken since the appearance of the ACK theory.
4.3 Stress analysis (boundary value problem) for cracked laminates
4.3.1 Complexity and issues
Damage analysis of cracked laminates of a general layup is a highly complex task. The
major issues in analyzing damage in a multidirectional laminate are discussed below:
1. Anisotropy and heterogeneity: The commonly used laminate analysis is based
on the assumption that the plies are homogeneous and anisotropic, with the
effective properties of a unidirectional composite. This is a valid assumption
for undamaged laminates for the usual cases of membrane force and moment
loading when the variation of the primary, in-plane, laminate stresses through
the ply thickness is constant or linear. But the presence of intralaminar cracks
may cause local conditions such as high stress gradients to invalidate assump-
tions needed for homogenization of plies [1]. Furthermore, an in-plane stress
analysis will be inadequate to account for interactions between cracks in
different plies.
2. Stress singularity: An assumption of ideal cracks within plies will produce
stress singularity. In reality, however, the crack tips will be blunted due to the
presence of nitely sized bers near the ply interfaces and by local ow of the
matrix at crack tips. Hence, the actual stress eld in the vicinity of the crack
tip may not be accurately given by the usual stress analysis of cracks in
homogeneous elastic medium. In such instances, numerical approaches such
as the nite element method may become necessary.
3. Interaction between cracks: At sufcient crack densities, the stress elds
around two adjacent cracks in a ply start interacting, thereby relaxing the
region between those two cracks. This crack interaction affects the crack
surface displacements as well as the overall stress elds. Accurate modeling
of crack interaction is a complex task.
4. Three-dimensionality of the boundary value problem: Following all the
previous points, and realizing that in a real scenario the ply cracks in a
general laminate may be curved, irregularly spaced, or not fully grown
through the laminate width, a complex 3-D boundary value problem
(BVP) arises. In the simpler case of cracked cross-ply laminate, assuming
that cracks are periodic, straight, and fully grown through the ply thickness
and width, the resulting BVP can be reduced to a generalized plane strain
problem. For the off-axis ply cracking, however, the stress analysis problem
is still a truly 3-D BVP, and any reduction to a 2-D problem will not produce
accurate predictions.
5. Dening RVE size: In homogenizing a heterogeneous solid such as a composite
ply the RVE size must be large enough to contain sufciently many bers to
provide average properties. With bers of typically 0.01 mm in diameter, and a
ply thickness of typically 0.125 mm, the RVE extending across a ply thickness
may or may not sufce, depending on the ber volume fraction and ber
distribution irregularity, but it is implicitly assumed to sufce in the classical
laminate theory. However, when cracks appear within a ply, the local stress
gradients increase sharply, leading to a breakdown of the homogenized ply
properties. Away from the cracks, nevertheless, the properties hold. In
obtaining the overall (average) composite properties with multiple cracks, the
RVE size must be large enough to contain a representation of the cracks. To
satisfy this requirement the RVE must extend in the laminate length direction
while it is limited in the thickness direction by the laminate thickness.
6. Multiscale considerations: Connected with the RVE issue is the consideration
of the multiscale nature of the stress and failure analysis. The RVE scale is the
so-called meso (intermediate) scale, while the scale of discrete entities within
the RVE is the micro scale. The scale at which structural analysis is conducted
is the macro scale. Most of the MIDM is concerned with determining changes
of mesoscale properties using stress elds at the microscale. In fact the
analyses described in the following mostly assume uniform distribution of
cracks, giving repeating unit cells for BVP solutions.
7. Constraint effects: In a cracked laminate, stress perturbations are caused by
the surface displacements of the ply cracks in response to the applied loading.
These surface displacements do not occur freely, as they would if the cracks
were to lie in a homogeneous ply of innite thickness, but are affected by
constraint from the neighboring plies. Understanding these constraint effects
is the key in determining the effective properties of cracked laminate. They
will be discussed in further detail in a subsequent chapter.
8. Complexity of off-axis ply cracking: Unlike in cross-ply laminates, intralami-
nar cracking in off-axis plies of orientations other than 90
can be complex.
Microscopic observations suggest that depending on the off-axis angle, ply
thickness, and orientation of neighboring plies these cracks may be partially
grown and erratic in shape and size, and of nonuniform distribution [8, 9].
Raman Spectroscopy experiments on [0/45]
s
laminates show that a crack
developing in the 45
-ply behaves differently from a similar crack in the 90
-
ply of a cross-ply laminate [10], which seemed to suggest that the initiation and
propagation strains for a 45
crack were different. For the laminates containing

a 90
-ply, the cracks usually initiate in that ply under axial tension, while cracks
in other off-axis plies initiate at higher loading. The observations on multidir-
ectional laminates indicate that the angle between the two adjacent plies may
have signicant effects on damage initiation and progression. When this angle
is small, partially formed cracks are observed before propagation in the ber
69 4.3 Stress analysis (boundary value problem) for cracked laminates
direction. However, fully developed cracks mainly form in the cases of large
intersecting angles [11, 12]. The damage development in the 60
-ply of a [0/60
2
/
90]
s
laminate is shown in Figure 4.6. Moreover, shearextension coupling may
introduce some additional complexity in analysis of off-axis laminates [13, 14].
9. Randomness in cracking process: In general, damage models assume a uni-
form distribution of transverse cracks, i.e., they are assumed to be periodic
and self-similar. In reality the variation in crack spacing can be considerable,
particularly at large crack spacing. Recently, there have been some develop-
ments to account for the inuence of the spatial nonuniformity of matrix
cracking on the stress transfer and the effective mechanical properties of
cracked cross-ply laminates [1517].
10mm
90
60
(a)
x
= 315[MPa]
x
= 0.71%
x
= 347[MPa]
x
= 0.80%
x
= 388[MPa]
x
= 0.89%
x
= 428[MPa]
x
= 1.00%
(b)
(c)
(d)
Figure 4.6. Consecutive matrix cracking behavior in contiguous plies in a [0/60
2
/90]
s
laminate. Parts (a)(d) show damage state at different levels of applied strain. Reprinted,
with kind permission, from Int J Solids Struct, Vol. 42, T. Yokozeki, T. Aoki, and
T. Ishikawa, Consecutive matrix cracking in contiguous plies of composite laminates,
10. Multiple damage mechanisms: In general, composite laminates can display
a multitude of damage modes. Here, our focus is on ply cracking, which
usually occurs much before other damage mechanisms such as delamina-
tions and ber fracture. The inuence of manufacturing-induced defects
such as voids and ber clusters may further complicate the analysis. These
interactions may become important for failure analysis and have been
considered in some studies [1824].
The above list of complex issues concerning cracked laminates, even for simple
congurations, makes the task of stress and failure analysis daunting. The
accuracy of stress analysis by itself may not always be the answer to the
engineering problem at hand. Often approximations and simplications need to
be made in a judicious manner guided by the application. In the following the
efforts made at analyzing damage and its effects are treated in increasing order of
complexity.
4.3.2 Assumptions
Consider a cross-ply laminate loaded in tension (Figure 4.7(a)). At a certain value
of the load, it develops transverse cracks in the 90
-plies. As described in Chapter

3 these cracks instantaneously traverse the thickness of the 90
-plies and span the

composite specimen width, i.e., the extent of the uniformly distributed load in the
width direction. Thus, regarding the width direction to be innite in extent,
the boundary value problem can be reduced to a generalized plane strain problem,
as indicated in Figure 4.7(b). The crack spacing 2l shown in the gure is assumed
to be periodic. Experimental observations indicate that although the crack spacing
is nonuniform in the early stage of the cracking process, it attains uniformity soon
and maintains it until crack saturation.
In addition to the assumptions concerning the crack geometry and spacing, the
ply material is assumed to be homogeneous, linearly elastic, with symmetry prop-
erties dictated by the ber direction. Assuming all bers to be parallel, the cross-
sectional plane becomes a plane of symmetry. The two mutually orthogonal planes
that are orthogonal to the cross-sectional plane, namely the mid-plane and the
through-thickness plane, can both be assumed as planes of symmetry, rendering the
ply an orthotropic symmetry. Alternatively, one can assume the cross-sectional
plane to be the plane of isotropy, which would make the ply transversely isotropic.
Boundary value problem for cracked cross-ply laminates
As stated above, the cracked cross-ply laminate can be treated as a two-
dimensional BVP shown in Figure 4.7(b). The laminate coordinate system xyz
is shown in the gure. For the two-dimensional geometry shown in Figure 4.7(b) the
xz axes are placed with the origin located midway between the cracks, as indi-
cated. The local material coordinate system in a given lamina is x
1
x
2
x
3
, denoted
simply as 123.
71 4.3 Stress analysis (boundary value problem) for cracked laminates
The following are specied for the BVP:
v Ply material transversely isotropic:
E
1
= Longitudinal Youngs modulus
E
2
= Transverse Youngs modulus
n
12
= Major Poissons ratio
G
12
= In-plane shear modulus
n
21
= n
12
E
2
E
1
= Minor Poissons ratio
E
x
= Longitudinal Youngs modulus for the laminate
E
x0
or E
c
= Longitudinal Youngs modulus for the virgin (undamaged) laminate
E
0
x0
= Longitudinal Youngs modulus for the 0
-ply (undamaged laminate)

E
90
x0
= Longitudinal Youngs modulus for the 90
-ply (undamaged laminate)

(4:48)
v Geometry:
A = Cross-sectional area
2t
0
= Thickness of 0
-ply
2t
90
= Thickness of 90
o
-ply
2h = 2 t
0
t
90
( ) = Total laminate thickness
2l = Average spacing between two adjacent cracks
(4:49)
t
0
t
90
h
x
z
2l
90
0
0
c
(a)
(b)
N
xx
N
xx
x
z
y
O
Figure 4.7. Construction of a unit cell for stress analysis of a cracked cross-ply laminate:
(a) cracked laminate in tension; (b) equivalent 2-D unit cell.
v Loading:
N
xx
=Applied tensile load per unit width in x-direction
(in-plane stress resultant)
(4:50)
v Stresses and strains:
s
c
= Total applied stress for laminate along x-direction
s
0
xx
= Total x-direction stress in 0
-ply
s
90
xx
= Total x-direction stress in 90
-ply
s
0
xx0
= Initial (virgin-laminate) x-direction stress in 0
-ply
s
90
xx0
= Initial (virgin-laminate) x-direction stress in 90
-ply
(4:51)
Other notations will be dened as and when needed. The boundary value problem
can be stated as:
For a cracked cross-ply laminate loaded in tension determine the displacement and stress
elds which satisfy equilibrium and boundary conditions, and further determine its effective
stiffness properties for a xed state of damage (given crack spacing).
That is, determine s
ij
which satisfy the following conditions:
1. Force balance:
N
xx
= s
c
A: (4:52)
2. Equilibrium conditions:
s
ij; j
= 0: (4:53)
3. Boundary and continuity conditions:
Laminate mid-plane symmetry: s
90
xz
x; 0 ( ) = 0
Traction continuity across interface: s
90
xz
x; t
90
( ) = s
0
xz
x; t
90
( )
s
90
zz
x; t
90
( ) = s
0
zz
x; t
90
( )
Traction-free boundary: s
0
xz
x; h ( ) = 0
s
0
zz
x; h ( ) = 0
Traction-free crack surfaces: s
90
xz
l; z ( ) = 0; t
90
_ z _ t
90
s
90
xx
l; z ( ) = 0; t
90
_ z _ t
90
:
(4:54)
4.4 One-dimensional models: shear lag analysis
A class of one-dimensional models that has been useful in the analysis of
multiple cracking is the so-called shear lag models. Although the stress analysis
is inherently inaccurate in these models, their ability to capture the stress
73 4.4 One-dimensional models: shear lag analysis
transfer at the interface by shear stress makes them useful. Historically, the
shear lag analysis was rst used by Cox [25] to describe the stress transfer
between a ber and a matrix for discontinuous ber composites. He considered
an axisymmetrical model of a single ber embedded in the matrix. Later,
Aveston and Kelly [6] modied the model for predicting strain to initiate
multiple matrix cracking in a unidirectional ber composite with brittle matrix.
A recent work considering axisymmetrical model has shown that Coxs original
result can be derived by a series of approximations to the elasticity theory [26].
Garrett, Bailey, and Parvizi [27, 28] and Manders et al. [29] adopted Coxs
approach for the particular case of transverse cracking. Most of the early
studies on transverse cracking used shear lag analysis for the unit cell shown
in Figure 4.7(b). There have been many modications and extensions of the
same analysis. For a more detailed discussion on shear lag analysis, the reader
is referred to [3058].
All shear lag analyses are based on the following basic concept:
In the plane of a transverse crack the transverse ply does not carry the axial load,
while away from the crack a part of this load is transferred back to the transverse ply
by axial shear at the interface between the cracked transverse ply and the adjacent
uncracked ply.
The shear lag analysis is essentially a 1-D analysis and it uses the following basic
assumptions:
1. The axial shear stress, t
xy
a dn/dx, where v is the displacement in the
axial, y-direction. This violates the relationship of linear elasticity,
t
xy
= G
xy
dn
dx

du
dy
_ _
. Hence, this is equivalent to assuming that du/dy = 0 or
du/dy dn/dx.
2. The axial normal stress remains constant over the thickness of the transverse
ply after cracking. In other words, concentration of local stress near cracks is
neglected.
3. Cracks remain sufciently far apart so that their mutual interactions can be
neglected.
The shear lag analyses will be presented here with consistent notations and
expressions that may differ in form from those given in the original articles.
4.4.1 Initial shear lag analysis
Here we describe the analysis presented by Garrett, Bailey, and Parvizi [27, 43]
and modied by Manders et al. [29] to account for the presence of neighboring
cracks. For convenience, we follow a treatment similar to the one described in a
review paper by Berthelot [41].
The objective is to determine the variation of the axial (x-direction) stress
in the transverse plies on crack formation in these plies. To begin, one
assumes that the x-displacement in the 0
-ply is constant through its thick-

ness while the corresponding displacement of the transverse plies increases
linearly from the 0/90 ply interface towards the laminate mid-plane. In
accordance with the rst shear lag assumption stated above, the interfacial
shear stress is given by
t = G
90
xz
u
90
u
0
t
90
_ _
; (4:55)
where u
0
is the x-displacement in the 0
-ply, u
90
is the x-displacement at mid-plane,
and G
90
xz
is the transverse shear modulus of the 90
-ply. The equilibrium of axial

forces in an element of the 90
-ply (see Figure 4.8) gives

t = t
90
ds
90
xx
dx
; (4:56)
where s
90
xz
is the axial normal stress, assumed to be constant in the 90
-ply
across the thickness (z-direction). Substituting Eq. (4.56) into Eq. (4.55), one
obtains
ds
90
xx
dx
= G
90
xz
u
90
u
0
t
2
90
_ _
: (4:57)
The axial stresses in the 0
- and 90
-plies are related to the applied stress s

c
as
ls
0
xx
s
90
xx
= 1 l ( )s
c
; (4:58)
where the ply thickness ratio l is dened as
l =
t
0
t
90
: (4:59)
Lastly, the stressstrain relations in the 0
- and 90
-plies are:
s
0
xx
= E
0
x0
e
0
xx
; with e
0
xx
=
du
0
dx
;
s
90
xx
= E
90
x0
e
90
xx
; with e
90
xx
=
du
90
dx
;
(4:60)
90
0
0
90
xx
x
dx
90
xx
+
90
xx
d
Figure 4.8. Stresses acting on an element of 90
-ply.
where E
0
x0
= E
1
and E
90
x0
= E
2
are the initial axial Youngs moduli of the 0
- and
90
-plies, respectively. Differentiating Eq. (4.57) with respect to x and using

Eqs. (4.58)(4.60), one obtains the differential equation for axial stress in the
transverse ply as
d
2
s
90
xx
dx
2

b
2
t
2
90
s
90
xx
=
b
2
t
2
90
E
90
x0
E
x0
s
c
; (4:61)
with
b
2
= G
90
xz0
1
E
90
x0
1
lE
0
x0
_ _
; (4:62)
where G
90
xz0
is the initial in-plane shear modulus of the 90
-ply and E
x0
= E
c
is the axial modulus of the undamaged laminate given by the rule of
mixtures as
E
x0
=
lE
0
x0
E
90
x0
1 l
: (4:63)
This value can be estimated accurately by using the classical laminates plate
theory. Since the crack surfaces are traction free, the axial stress in the transverse
ply vanishes on the crack planes (x = l). Hence, the solution of Eq. (4.61) is
given by
s
90
xx
= s
c
E
90
x0
E
x0
1
cosh b
x
t
90
cosh b
l
t
90
_
_
_
_
_
_
_
_
: (4:64)
Thus b appears as a load transfer parameter, and is sometimes known as the shear
lag parameter.
A quite similar shear lag analysis was conducted by Dvorak et al. [44], except
that they included residual thermal stresses and in place of Eq. (4.55) assumed that
the shear stress is given by
t = K u
90
u
0
( ); (4:65)
where K is a shear parameter, which is to be determined from experimental data.
Dvorak et al. [44] suggested determining b from the experimentally meas-
ured stress at rst ply failure. The differential equation for the axial stress in this
case is
d
2
s
90
xx
dx
2

b
2
t
2
90
s
90
xx
=
b
2
t
2
90
s
90
xxR

E
90
x0
E
x0
s
c
_ _
; (4:66)
where the shear lag parameter is now given by
b
2
=
Kt
90
t
0
E
0
x0
t
90
E
90
x0
_ _
t
0
E
0
x0
E
90
x0
= Kt
90
1
E
90
x0
1
lE
0
x0
_ _
; (4:67)
and the solution obtained for axial stress in the transverse ply is
s
90
xx
= s
90
xxR
s
c
E
90
x0
E
x0
_ _
1
cosh b
x
t
90
cosh b
l
t
90
_
_
_
_
_
_
_
_
; (4:68)
where s
90
xxR
is the axial residual thermal stress in the 90
-ply.
4.4.2 Interlaminar shear lag analysis
Based on extensive experimental observations of cracks, Highsmith and Reifsnider
[30] found that shear deformations in a given ply were restricted to a thin region
near interfaces between adjacent plies. This region is resin rich and thus is less stiff in
its response to shear stress than the central portion of the ply. Transverse cracks
were observed to extend up to the region, but not into it. Based on these observa-
tions, interlaminar shear lag models were developed [30, 45], in which shear stresses
were assumed to develop only within this thin region, whose thickness and shear
modulus are unknown. The unit cell for such models is shown in Figure 4.9.
We present here the interlaminar shear lag analysis reported by Fukunaga et al.
[45], who incorporated the thermal residual stresses and Poissons effect into the
analysis. The displacements in the x- and y-directions of each ply are still assumed
to be constant across thickness, as above, but are expressed as
u
0
= e
c
x U
0
x ( ); v
0
=
A
12
A
22
e
c
y V
0
x ( );
u
90
= e
c
x U
90
x ( ); v
90
=
A
12
A
22
e
c
y V
90
x ( );
(4:69)
where the coefcients A
ij
are the in-plane stiffness components of half the cross-ply
laminate. These are the same as the components of the [A] matrix in classical
laminates plate theory, and can be expressed in terms of the reduced stiffness
components Q
ij
of the 0
- and 90
-plies as
A
11
= Q
11
t
0
Q
22
t
90
; A
12
= Q
12
h; A
22
= Q
22
t
0
Q
11
t
90
: (4:70)
t
0
t
90
h
x
z
2l
90
c
t
s
Figure 4.9. Unit cell for interlaminar shear lag analysis.
These relations suppose that the 0
- and 90
-plies have the same elastic properties.

The strain e
c
is the average strain of the laminate and is given by
e
c
=
s
c
E
x
; (4:71)
where the axial Youngs modulus of the laminate E
x
is given by
E
x
=
A
11
A
22
A
2
12
hA
22
: (4:72)
The equilibrium of the axial forces yields
Q
11
d
2
u
0
dx
2

G
t
0
t
s
u
90
u
0
( ) = 0;
Q
22
d
2
u
90
dx
2

G
t
90
t
s
u
90
u
0
( ) = 0;
(4:73)
where G and t
s
represent the shear modulus and the thickness of the resin-rich
layers, respectively. Next, the balance of the in-plane shear loads gives
Q
66
d
2
v
0
dx
2

G
t
0
t
s
v
90
v
0
( ) = 0;
Q
66
d
2
v
90
dx
2

G
t
90
t
s
v
90
v
0
( ) = 0:
(4:74)
The solution of the equilibrium equations, Eqs. (4.73) and (4.74), determines the
displacement eld, from which the strains are determined using the straindis-
placement relations. Finally, the stress components are found by using the consti-
tutive relations. The axial stress in the 90
-ply is obtained as
s
90
xx
= s
90
xxR
s
c
Q
22
E
0
x
_ _
1
Q
12
A
12
Q
22
A
22
_ _
1
cosh b
x
t
90
cosh b
l
t
90
_
_
_
_
_
_
_
_
; (4:75)
where the shear lag parameter b is given by
b
2
= G
t
90
t
s
1
Q
22
1
lQ
11
_ _
: (4:76)
The interlaminar shear lag models by Highsmith and Reifsnider [30] differ from
Fukunagas analysis [45] in their denition of the shear-lag parameter. For the
HighsmithReifsnider model,
b
2
= G
t
90
t
s
1
E
90
x0
1
lE
0
x0
_ _
: (4:77)
It can be seen that the expressions for the axial stress in the 90
-ply, Eqs. (4.64),

(4.68), and (4.75), are very similar in different shear lag models. Usually, the
denition of the shear lag parameter is different. The biggest limitation of the
interlaminar shear lag analysis is that the thickness of the boundary layer must be
assumed in a somewhat arbitrary fashion.
4.4.3 Extended shear lag analysis
Limand Hong [38] modied the interlaminar shear lag model of Fukunaga et al. [45]
to include the effect of crack interaction. The governing differential equations are the
the same as in Fukunagas analysis (Eqs. (4.73), (4.74)). However, the boundary
conditions are used such that they are valid for any crack spacing. With this
modication, the axial stress in the 90
-ply is given by
s
90
xx
= s
90
xx0
1 a
1
e
bx=t
90
a
2
e
bx=t
90
_ _ _ _
; (4:78)
where
a
1
=
1 e
2bl=t
90
e
2bl=t
90
e
2bl=t
90
; a
2
=
e
2bl=t
90
1
e
2bl=t
90
e
2bl=t
90
; (4:79)
and the shear lag parameter b is given by Eq. (4.76). The solution in Eq. (4.78) is
different than in Eq. (4.75) due to differences in the assumed boundary conditions.
It is noted that s
90
xx0
in Eq. (4.78) is the total stress in the transverse ply before
cracking, and thus includes thermal residual stress, if any.
To account for the progressive shear in the 90
ply, Steif [46] developed a shear

lag theory, where he used an x-displacement that was constant through the 0
layer
thickness but varied parabolically through the 90
layer thickness. This analysis

leads to an expression of the axial stress in the transverse ply similar to that given
in Eq. (4.64), with a modication in the shear lag parameter:
b
2
= 3G
90
xz0
1
E
90
x0
1
lE
0
x0
_ _
; (4:80)
which is three times the value of b
2
as given by Eq. (4.62). Later, the same
approach was applied by Ogin et al. [47, 48] to study the stiffness reduction of
glass-ber cross-ply laminates subjected to quasi-static or fatigue loading. The
transverse shear effects in both 0
- and 90
-plies have been considered by Nuismer

and Tan [49] and Lee, Daniel and Yaniv [31, 56]. These authors consider stress
approaches based on the assumption of linear shear stress variation in each layer
through the thickness, which extends the initial shear lag analysis. Lee and Daniel
[31] relate these linear variations to parabolic variations of the x-displacements
through the thicknesses of the 0
and 90
-plies. These approaches also lead to

modied expressions for the load transfer parameter. For example, the analysis by
Lee and Daniel yields the following expression for the shear lag parameter
b
2
=
3G
90
xz0
1 l
G
90
xz0
G
0
xz0
1
E
90
x0
1
lE
0
x0
_ _
: (4:81)
It can be seen that Eq. (4.81) reduces to Eq. (4.80) when G
0
xz0
G
90
xz0
.
4.4.4 2-D shear lag models
Flaggs [32], and Nuismer and Tan [33, 49] developed two-dimensional shear lag
models. Contrary to the authors claims, these models are essentially equivalent to
1-D models with a minor modication to correct for the contraction due to the
Poissons effect [42].
The analysis by Flaggs [32] accounts for both applied normal and shear load-
ings and results into a system of two coupled second-order differential equations.
For only normal applied loading, it reduces to a single ODE identical to general
1-D analysis equations, except that the shear lag parameter is given by
b
2
=
2
1
lQ
0
xx0
Q
90
yy0
Q
90
xy0
Q
0
xy0
=Q
0
xx0
_ _
Q
90
xx0
Q
90
yy0
Q
90
xy0
_ _
2
_
_
_
_
_
_
1
k
2

1
2
_ _
1
G
90
xz0
l
2G
0
xz0
; (4:82)
where k is a transverse shear correction factor and the Qs are stiffness coefcients
for the virgin laminate. The Flaggs analysis therefore uses a minor correction for
the Poissons contraction introduced by approximate inclusion of the transverse
coordinate (y).
Another 2-D elasticity analysis was performed by Nuismer and Tan [33, 49].
This analysis uses the shear lag parameter dened as
b
2
=
1
Q
90
xx0
1
lQ
0
xx0
1
3G
90
xz0
l
3G
0
xz0
: (4:83)
The analysis also yields a nonzero o (see Eq. (4.85) described in the next subsec-
tion), given by
o =
1
l
t
90
t
/
i
b
2
s
90
x0
_
; (4:84)
where t
/
i
is the slope of the interfacial shear stress at the crack location. This
analysis is also equivalent to other 1-D models with a minor correction for the
Poissons ratio.
More recently, there have been efforts to modify and extend the 1-D and 2-D
shear lag models to enable predictions for laminates of other than cross-ply
layups, such as the equivalent constraint models [13, 5153] and 2-D displacement
analysis [54, 55]. These models also suffer from deciencies that are common in the
shear lag models described above.
4.4.5 Summary of shear lag models
It can be shown [42] that all one-dimensional stress analyses discussed above can
be reduced to a generalized form of Garrett and Baileys equation [27],
d
2
s
dx
2
b
2
s = o P ( ); (4:85)
where Ds represents the total stress transferred from the 90
-layer to the 0
-layer,
and is given by
s = s
0
xx
s
0
xx0
; (4:86)
and x = x/t
90
is a dimensionless x-coordinate, b is the shear lag parameter, and
o(P) is a function which may depend on the laminate structure, crack spacing, and
the applied load (P). The corresponding boundary conditions, equivalent to
traction-free crack surfaces, are
s x = r ( ) =
s
90
xx0
l
; (4:87)
where r = l/t
90
denotes the crack spacing normalized by the cracked ply thickness.
Other than in the one-dimensional model by Nuismer and Tan [33, 49], the
function o is chosen to be zero. It can be clearly observed that all the one-
dimensional shear lag models are essentially the same; the difference being in the
choice of the shear lag parameter. The different shear lag models are summarized
in Table 4.1.
The solution of the differential equation (4.85) provides
s =
o
b
2

s
90
xx0
l

o
b
2
_ _
cosh bx
cosh br
: (4:88)
With this, the stress in the 0
-layer is given by
s
0
xx
= s
0
xx0

o
b
2

s
90
xx0
l

o
b
2
_ _
cosh bx
cosh br
; (4:89)
and in the 90
-layer, it is given by
s
90
xx
= s
90
xx0

lo
b
2
_ _
1
cosh bx
cosh br
_ _
: (4:90)
Now, the rate of load transfer from the 90
-layer to the 0
-layer, analogous to Eq.

(4.56), can be rewritten as
ds
dx
=
t
l
: (4:91)
It is noted here that the difference in Eqs. (4.56) and (4.91) is due to the denition
of Ds (see Eq. (4.86)). Differentiating Eq. (4.88) with respect to x, and comparing
the result with Eq. (4.91), gives the interfacial shear stress as
t
i
= s
0
xx0

lo
b
2
_ _
cosh bx
cosh br
: (4:92)
The stress predictions fromvarious shear lag models are compared in Figure 4.10 for
[0/90
2
]
s
carbon/epoxy (Hercules AS4/35016) laminate with lamina properties [31,
56]: t
ply
= 0.154 mm, E
1
= 130 GPa, E
2
= 9.7 GPa, G
12
= 5.0 GPa, G
13
= 3.6 GPa,
n
12
=0.3, andn
13
=0.5. The axial normal stress inthe 90
-plies increases fromzeroon

the crack to a maximum value midway between two cracks, as the load is transferred
fromthe 0
-plies back into the 90
-plies. The traction-free boundary condition at the

crack surfaces is thus fullled accurately for the normal stress in all models. However,
none of the shear lag models represents interfacial shear stress accurately. The shear
stress has a maximumon the crack surface and decays towards zero as we move away
Table 4.1 A summary of different shear lag models to analyze cracked cross-ply laminates
Shear lag model Feature Shear lag parameter (b
2
) o(P)
Garret and Bailey [27],
Manders et al. [29]
Simplest shear
lag model
G
90
xz0
1
E
90
x0
1
lE
0
x0
_ _
0
Laws and
Dvorak [36]
Use rst ply
failure data to
determine b
Kt
90
1
E
90
x0
1
lE
0
x0
_ _
0
Steif [46],
Ogin et al. [47, 48]
Parabolic
displacement
prole
3G
90
xz0
1
E
90
x0
1
lE
0
x0
_ _
0
Highsmith and
Reifsnider [30]
Use an effective
shear transfer
layer
G
t
90
t
s
1
E
90
x0
1
lE
0
x0
_ _
0
Fukunaga et al. [45] Use an effective
shear transfer
layer
G
t
90
t
s
1
Q
22
1
lQ
11
_ _
0
Limand Hong [38] Use an effective
shear transfer
layer, account
for crack
interaction
G
t
90
t
s
1
Q
22
1
lQ
11
_ _
0
Nuismer and Tan
[33, 49]
Account for
Poissons effect
1
Q
90
xx0
1
lQ
0
xx0
1
3G
90
xz0
l
3G
0
xz0
1
l
t
90
t
/
i
b
2
s
90
x0
_
Flaggs [32] 2-D shear lag
analysis to
account for
both transverse
and shear
loading
2
1
lQ
0
xx0
Q
90
yy0

Q
90
xy0
Q
0
xy0
Q
0
xx0
Q
90
xx0
Q
90
yy0
Q
90
xy0
_ _
2
_
_
_
_
_
_
_
_
_
_
1
k
2

1
2
_ _
1
G
90
xz0
l
2G
0
xz0
0
Lee and Daniel [31] Parabolic
displacement
variation,
account for
Poissons effect
3G
90
xz0
1 l
G
90
xz0
G
0
xz0
1
E
90
x0
1
lE
0
x0
_ _
0
from the crack. The nonzero shear stress on the crack surfaces is a clear violation of
the boundary condition. This makes all 1-D analyses inherently inaccurate.
Let us now turn our attention to the stiffness degradation. The treatment
provided here follows the approach covered in a book chapter by Nairn and
Hu [42]. Before cracking, both the 0
- and 90
-plies share the applied load. But

after cracking, the 90
-layers at the crack planes do not carry any load. The total
applied load at these planes is carried by the 0
-plies. Away from cracks, the

90
-plies take up load again through the shear transfer mechanism. To determine
the modulus of the cracked laminate, let us rst determine the total x-displacement
of the load bearing layer in the region between two cracks in the transverse ply.
It can be found by integrating the axial strain as
u P ( ) =
_
l
l
e
0
xx
dx = t
90
_
r
r
s
0
xx
E
0
x
n
0
xz
s
0
zz
E
0
x
n
0
xy
s
0
yy
E
0
x
_ _
dx; (4:93)
where P is the applied load. It is noted that the thermal strain is ignored because for
linear thermoelastic materials the modulus is independent of the residual stresses.
The compliance of a unit cell of the laminate with cracks is dened as
C =
u P ( ) u 0 ( )
P
: (4:94)
The effective modulus is then given by
E =
2l
hWC
: (4:95)
60
60
4 3 2 1 0
1 2
Manders et al.
Laws & Dvorak
Fukunaga et al.
Lee & Daniel
Steif
3 4
40
20
0
(
M
P
a
)
9
0
9
0
x
x
x
z
20
40
Figure 4.10. Variation of the axial normal and interfacial stresses between two cracks in a
[0/90
2
]
s
laminate from shear lag analysis assuming normalized crack spacing r = 4 and
applied stress of 50 MPa.
Assuming a plane stress state,
s
0
yy
= 0: (4:96)
Moreover there is no load applied in the thickness (z) direction, i.e.,
_
r
r
s
0
zz
dx = 0: (4:97)
Thus, for both one-dimensional analysis and two-dimensional plane stress analy-
sis the total displacement reduces to
u P ( ) = t
90
_
r
r
s
0
xx
E
0
x
dx: (4:98)
Substituting stress from Eq. (4.89) for 1-D shear lag analyses, we obtain
1
E
x
=
1
E
0
c
1
E
90
x0
lE
0
x0
tanh br
br
_ _
o P ( ) o 0 ( )
E
0
x0
b
2
s
0
1
tanh br
br
_ _
: (4:99)
Assuming o(P)o(0) = 0, as is the case for most one-dimensional analyses, the
effective axial modulus is then
1
E
x
=
1
E
0
c
1
E
90
x0
lE
0
x0
tanh br
br
_ _
: (4:100)
Thus, the axial modulus for the cracked laminate normalized with its virgin value
is given by
E
x
E
0
c
= 1
E
90
x0
lE
0
x0
tanh br
br
_ _
1
: (4:101)
The predictions for stiffness reductionusing various shear lag models as a function of
crack density for [0/90
3
]
s
glass/epoxy (Scotch Ply 1003) laminate are shown in Figure
4.11. It should be noted, however, that these predictions may vary depending upon
the parameters used in the models. It is important to note that the original shear lag
model by Garrett, Bailey, and Parvizi [27, 43], or equivalently by Manders et al. [29],
is reasonably accurate. More complicated shear lag models not only involve more
complicated analysis, and adjustable parameters, but also do not yield more accurate
predictions. The basic deciency of shear lag models lies in their one-dimensionality
of stress eld, and no signicant improvement is essentially possible.
4.5 Self-consistent scheme
The self-consistent method is widely employed for estimating properties of elastic
solids containing entities such as inclusions, voids, and cracks; see for example
Nemat-Nasser and Hori [2]. Using this method generally requires assuming
an innite solid. For composite laminates containing ply cracks, Laws and
Dvorak [59] estimated elastic properties using this method. They rst assumed that
a unidirectional ply (within a laminate) with cracks along its bers can be modeled
as an innite solid with the same ber volume fraction and crack density. Taking
then the estimated average properties of the innite solid as those of the cracked ply,
they replaced the cracked ply in a composite laminate by a homogeneous ply of the
degraded properties of the innite solid. Figure 4.12 illustrates the assumed scheme.
Elastic properties of the cracked composite laminate were then calculated by the
classical laminate plate theory. This procedure in principle allows calculating
stiffness degradation of any composite laminate with given crack density in any
ply, i.e., for any orientation of cracks. It is doubtful, however, that the assumption
of an innite solid will give accurate elastic property estimates for plies that are
typically of a fraction of a millimeter in thickness. The predicted properties of
various laminates with cracks in different plies and with different crack density
values were reported in Laws and Dvorak [59] but these properties were not
compared with experimental data or independent numerical computations.
Readers interested in details of the self-consistent method for composite mater-
ials with cracks are referred to Laws et al. [60]. We provide here a brief account
of the concepts involved. Essentially, Hill [61] and Budiansky [62] developed the
self-consistent method for composite materials, i.e., homogeneous solids with
inclusions. The idea in this method is that a single inclusion is embedded in a
homogeneous elastic solid which has the yet-unknown overall properties of the
heterogeneous solid and the local elds thus estimated are then used to obtain the
overall properties. Laws et al. [60] use a key result obtained for ellipsoidal
inclusions by Eshelby [63, 64] that the elastic eld within the inclusion is uniform.
This result was used for two models: a three-phase system and a two-phase system.
1.0
0.9
0.8
0.7
R
e
l
a
t
i
v
e

S
t
i
f
f
n
e
s
s
0.6
0.5
0.0 0.1 0.2 0.3
Hashin
Garrett & Bailey
Reifsnider
Nuismer & Tan
Glass/Epoxy
Flaggs
[0/90
3
]s
Ogin et al.
0.4 0.5
Microcrack Density (1/mm)
0.6 0.7 0.8 0.9 1.0
Figure 4.11. Variation of normalized Youngs modulus as a function of crack density for a
[0/90
3
]
s
glass/epoxy (Scotch Ply 1003) laminate. The experimental data are from [30].
Reprinted, with kind permission, from Damage Mechanics of Composite Materials,
J. A. Nairn and S. Hu, Micromechanics of damage: a case study of matrix microcracking,
85 4.5 Self-consistent scheme
In the former the bers and cracks are considered as two phases (special cases of
inclusions) embedded in an elastic matrix, while in the latter the bers and matrix are
smeared into a homogeneous solid in which cracks are embedded. Cracks in both
models are introduced as slits that are regarded as limiting cases of elliptical voids.
When the ber diameter is much smaller than the crack size, the two-phase
model is more appropriate. Thus if Q is the estimated stiffness matrix of the
cracked ply, and Q
0
is its value without the cracks present, then the two matrices
are related by [60]
Q = Q
0

1
4
bQ
0
Q; (4:102)
where b is a crack density parameter and L is a function of the aspect ratio of the
elliptical crack and the stiffness coefcient matrix Q.
It is seen from this equation that the estimation procedure for stiffness of a
cracked ply is iterative and it requires quantities L that are estimated by assuming
the homogenized composite ply to be of innite extent.
Variations of the self-consistent method described above have also appeared in
the literature. Noteworthy work is that of Hoiseth and Qu [65, 66]. They followed
the differential self-consistent method and derived an incremental differential
equation describing the effective axial modulus for the cracked laminate by
representing the change in strain energy due to an increase in terms of the number
of cracks in a lamina. If r = t
90
/l denotes the normalized crack density, then the
axial modulus is given by
(a) (b)
Figure 4.12. Self-consistent scheme: (a) cracked laminate containing bers of small
diameter and periodic cracks (b) homogenized cracked lamina.
d
E
c
dr
=
d
4t
90
2
E
c
r ( )
E
1
1 n
12
n
21
_ _
; (4:103)
subject to the initial condition
E
c
0 ( ) =

E
c0
; (4:104)
where d is the average crack opening displacement (COD), dened as
d =
1
t
90
_
t
90
0
d z ( ) dz; (4:105)
with d(z) being the COD variation across the z-direction (thickness). The authors
obtained it numerically using nite element calculations on a unit cell. For cross-
ply laminates, their predictions compared well with the independent FE simula-
tions that they carried out themselves. For multidirectional laminates, these
approaches become complex and do not yield accurate estimates.
4.6 2-D stress analysis: variational methods
4.6.1 Hashins variational analysis
An improved 2-D stress analysis can be obtained using the principle of minimum
complementary energy applied on a cracked laminate volume. Hashin [67] con-
sidered cross-ply laminates and used this principle to solve the boundary value
problem. He constructed an admissible stress eld assuming that the normal
stresses in the loading direction are constant over the ply thickness. The admissible
stress eld satises equilibrium and boundary and interface conditions. The
boundary volume problem thus solved provides the stresses and the reduced
stiffness coefcients for the cracked cross-ply laminate.
We followthe general notations for geometry, material, and stress components as
given in Section 4.3. Consider a symmetric cross-ply laminate containing a parallel
array of transverse cracks and subjected to a uniform axial load (Figure 4.13)
N
xx
per unit laminate width. Let us rst investigate the stress states before and after
cracking. Before cracking, all shear stresses and the transverse normal stress in the
entire laminate are zero, i.e.,
s
0
xx0
; s
90
xx0
,= 0; s
0
yy0
; s
90
yy0
,= 0;
s
0
zz0
= s
90
zz0
= 0; s
0
yz0
= s
90
yz0
= 0;
s
0
xz0
= s
90
xz0
= 0; s
0
xy0
= s
90
xy0
= 0;
(4:106)
where superscripts denote the layer orientations. The nonzero axial stresses in the
0
- and 90
-layers for virgin laminate are constant throughout individual layers

and can be obtained using the laminate theory as
s
0
xx0
=
E
0
x0
E
x0
s
c
; s
90
xx
=
E
90
x0
E
x0
s
c
; (4:107)
where s
c
= N
xx
/2h is the normal tensile stress on the laminate.
87 4.6 2-D stress analysis: variational methods
Now on sufcient tensile load, the 90
-ply develops continuous intralaminar

cracks in the ber direction that extend through the whole laminate width. These
cracks will cause perturbation in the stress elds in both layers. In a homogeneous
elastic material, there is theoretically a crack-tip singularity in the stress eld.
However, for transverse matrix cracking in composite laminates, the stresses at a
crack tip are nite because the bers have a nite size and will blunt the crack tip.
In Hashins analysis, he assumes the following:
1. The perturbations in axial stresses s
90
xx
and s
0
xx
are constant through the
thickness of layers. Thus,
s
90
xx
= s
90
xx
x ( ); s
0
xx
= s
0
xx
x ( ): (4:108)
2. There are no perturbations in s
yy
, s
xy
, and s
yz
, i.e.,
s
90
yy
= s
90
xy
= s
90
yz
= s
0
yy
= s
0
xy
= s
0
yz
= 0: (4:109)
The stress components for layer m (m = 0, and 90 for 0
-ply and 90
-ply,
respectively) of the cracked laminate can be expressed as
s
m
ij
= s
m
ij0
s
m
ij
(4:110)
where the subscript 0 indicates the uncracked laminate. The stress perturbations
can be expressed as
s
90
xx
= s
90
xx0
f
90
x ( );
s
0
xx
= s
0
xx0
f
0
x ( );
(4:111)
where f
90
and f
0
are unknown perturbation functions.
z
x
y
O
t
0
t
0
t
90
t
90
h h
N
xx
Figure 4.13. A cross-ply laminate loaded in axial tension.
The equilibrium equations for the plies are given by
@s
m
xx
@x

@s
m
xy
@y

@s
m
xz
@z
= 0;
@s
m
yx
@x

@s
m
yy
@y

@s
m
yz
@z
= 0;
@s
m
zx
@x

@s
m
zy
@y

@s
m
zz
@z
= 0:
(4:112)
Using Eqs. (4.108)(4.110), one obtains
@s
m
xx
@x

@s
m
xz
@z
= 0;
@s
m
zx
@x

@s
m
zz
@z
= 0:
(4:113)
Substituting Eq. (4.111) into Eq. (4.113) and integrating, one gets
s
90
xz
= s
90
xx0
f
/
90
x ( )z f
90
x ( ) [ [;
s
90
zz
= s
90
xx0
1
2
f
//
90
x ( )z
2
f
/
90
x ( )z g
90
x ( )
_ _
;
(4:114)
for m = 90 (90
-ply), and
s
0
xz
= s
0
xx0
f
/
0
x ( )z f
0
x ( ) [ [;
s
0
zz
= s
0
xx0
1
2
f
//
0
x ( )z
2
f
/
0
x ( )z g
0
x ( )[;
_
(4:115)
for m = 0 (0
-ply), where f
0
(x), f
90
(x), g
0
(x), and g
90
(x) are unknown functions
and primes denote derivatives with respect to x.
To obtain the relation between f
90
and f
0
, consider equilibrium in the
x-direction of the undamaged laminate
N
xx
=
_
h
h
s
xx0
dz = 2 s
90
xx0
t
90
s
0
xx0
t
0
_ _
: (4:116)
If the same membrane force is applied to the cracked laminate, the equilibrium in
the x-direction will give
N
xx
= 2 s
90
xx0
t
90
s
0
xx0
t
0
_ _
2 s
90
xx0
t
90
f
90
x ( ) s
0
xx0
t
0
f
0
x ( )
_
: (4:117)
From Eqs. (4.116) and (4.117), we get
s
90
xx0
t
90
f
90
x ( ) s
0
xx0
t
0
f
0
x ( ) = 0; (4:118)
i.e.,
f
0
x ( ) =
s
90
xx0
s
0
xx0
1
l
f
90
x ( ); (4:119)
where l = t
0
/t
90
is the ply thickness ratio (Eq. (4.59)).
The boundary value problem for solving the three stress components s
xx
, s
xz
,
and s
zz
is dened on the unit cell shown in Figure 4.14, in which the origin of the
coordinate system is placed at the mid-plane of the laminate, and midway between
two cracks.
Clearly z = 0 is the plane of symmetry. Hence, the shear stress s
xz
at all points
along this plane is zero. At the interface between the cracked 90
-ply and the

uncracked 0
-ply (z =t
90
), the shear stress s
xz
and normal stress s
zz
must be continu-
ous. Moreover, the surface at z = h is free from any external loading, i.e., the shear
stress s
xz
and normal stress s
zz
are zero on that surface. Finally, the crack surfaces
x = l are traction free, i.e., s
90
xz
= s
0
xz
= 0 and s
90
xx
= s
90
xx0
s
90
xx
= 0 at x = l.
Thus, all the applicable boundary conditions for the total stresses are written as
Symmetry : s
90
xz
x; 0 ( ) = 0; (a)
Interface : s
90
xz
x; t
90
( ) = s
0
xz
x; t
90
( ); (b)
s
90
zz
x; t
90
( ) = s
0
zz
x; t
90
( ); (c)
Free boundary : s
0
xz
x; h ( ) = 0; (d)
s
0
zz
x; h ( ) = 0; (e)
Traction free : s
90
xz
l; z ( ) = 0; t
90
_ z _ t
90
; (f)
s
90
xx
l; z ( ) = 0; t
90
_ z _ t
90
; (g)
(4:120)
where 2l is the distance between any two adjacent cracks. These conditions are the
same as those given earlier without derivation in Eq. (4.54)
Denoting f
90
(x) by f(x), putting boundary conditions, Eq. (4.120), into Eqs.
(4.114)(4.115), and using Eq. (4.119), after some mathematical manipulations,
the resulting stress eld in the cracked laminate is given by
x
z
l
l
h
t
0
t
90
Figure 4.14. A repeating unit cell for a cracked cross-ply laminate.
s
90
xx
= s
90
xx0
1 f x ( ) [ [;
s
90
xz
= s
90
xx0
f
/
x ( )z;
s
90
zz
= s
90
xx0
f
//
x ( )
1
2
1 l ( )t
2
90
z
2
_
;
(4:121)
in the 90
-ply, and
s
0
xx
= s
0
xx0
s
90
xx0
1
l
f x ( );
s
0
xz
= s
90
xx0
f
/
x ( )
1
l
1 l ( )t
90
z [ [;
s
0
zz
= s
90
xx0
f
//
x ( )
1
2l
1 l ( )t
90
z [ [
2
;
(4:122)
in the 0
-ply. Using the crack surface boundary conditions, Eq. (4.120fg), we get
additional conditions for the unknown perturbation function f(x)
At x = l : s
90
xx
= 0 = s
90
xx0
1 f x ( ) [ [;
s
0
xz
= 0 = s
90
xx0
f
/
x ( )z :
(4:123)
Thus,
f l ( ) = l; f
/
l ( ) = 0; t
90
_ z _ t
90
: (4:124)
This completes the description of the resulting boundary value problem. The object-
ive is to estimate f(x) in order to determine the stress eld in the cracked laminate.
The stress eld in Eqs. (4.121)(4.122) represents an admissible stress eld as it
satises all equilibrium and boundary conditions. Thus, the principle of minimum
complementary energy (see Section 2.1.5) can be applied to obtain the unknown
function f(x). The procedure for doing this, given in Hashin [67], is described next.
For the problem at hand, we can dene an admissible stress eld s
ij
within the
volume V of the cracked body with associated tractions

T
i
which satisfy only
equilibrium and the traction boundary conditions. Since the tractions are not
altered during the cracking process,
~
T
i
= T
0
i
on S = S
T
; (4:125)
where T
i
0
are the tractions applied on the uncracked body. Also, the crack surfaces
S
c
are traction free, i.e.,
~
T
i
= 0 on S = S
c
: (4:126)
If we write
~ s
ij
= s
0
ij
s
/
ij
;
~
T
i
= T
0
i
T
/
i
; (4:127)
where s
0
ij
are stresses in the uncracked laminate, s
/
ij
are the perturbation stresses
due to cracking, and T
/
is the additional traction due to cracking. Then from the
traction boundary conditions in Eqs. (4.125) and (4.126), we obtain
T
/
i
= 0 on S = S
T
;
T
/
i
= T
0
i
on S = S
c
:
(4:128)
The complementary energy functional is given by (see Eq. (2.36), Section 2.1.5)
+
0
=
1
2
_
V
S
ijkl
s
0
ij
s
0
kl
dV
_
S
u
T
0
i
^ u
i
dS (4:129)
for the uncracked body, where S
ijkl
is the compliance tensor, and
~
+
=
1
2
_
V
S
ijkl
~ s
ij
~ s
kl
dV
_
S
u
~
T
i
^ u
i
dS (4:130)
for the crackedbody. Since only tractionboundary conditions are applied, the second
term is zero (S
u
= 0; S
T
= S
t
). However, for the sake of completeness, we will follow
the proof given in Appendix 1 of [67], which applies to mixed boundary conditions.
Substituting the stress eld given in Eq. (4.127) into Eq. (4.130) and using Eq. (4.129),
the complementary energy for the cracked solid can be expressed as
~
+
=
+
0

_
V
S
ijkl
s
0
ij
s
/
kl
dV
1
2
_
V
S
ijkl
s
/
ij
s
/
kl
dV
_
S
u
T
/
i
^ u
i
dS : (4:131)
Consider the rst integral in Eq. (4.131). The stresses s
/
ij
satisfy equilibrium since
s
0
ij
and s
ij
do. Also, the strain eld in the uncracked body is given by
e
0
ij
= S
ijkl
s
0
kl
: (4:132)
Therefore, by virtual work (see Eq. (2.33), Section 2.1.5)
J =
_
V
S
ijkl
s
0
kl
s
/
ij
dV =
_
S
u
T
/
i
u
0
i
dS
_
S
c
T
/
i
u
0
i
dS
/
; (4:133)
where the tractions in the second surface integral are dened with respect to the
inward normal to S
c
. Using the boundary condition in Eq. (4.128), we have
J =
_
S
u
T
/
i
^ u
i
dS
_
S
c
T
0
i
u
0
i
dS : (4:134)
The second integral is taken over the two adjacent surfaces of each crack. Since the
normal on one surface is the negative of the normal on the other surface and since
T
0
i
and u
0
i
are continuous across the crack surface, the integrals on the two crack
surfaces cancel one another and therefore the total integral vanishes. Introducing
the remainder of Eq. (4.134) into Eq. (4.131) we obtain the complementary energy
for a cracked solid as
~
+
=
+
0

1
2
_
V
S
ijkl
s
/
ij
s
/
kl
dV =
+
0

/
(4:135)
where
/
0
is the complementary energy due to perturbation stresses. The effective
elastic compliances of the undamaged and damaged laminates can be expressed in
terms of the complementary energy functionals. To obtain the relation, consider
homogeneous boundary conditions on the cracked body
T
i
= s
ij
n
j
; s
ij
= constant; (4:136)
and let
s
ij
=
1
V
_
V
s
ij
dV: (4:137)
Then, from Eq. (4.130) we have
+
=
1
2
S
+
ijkl
s
ij
s
kl
V; (4:138)
where S
+
ijkl
is the effective elastic compliance tensor for the homogenized medium[68].
The laminate under consideration is loaded by a membrane force N
xx
on two
horizontal edges. Thus, volume and the average stress for this laminate are given by
V = 2Ah s
xx
=
N
xx
2h
= s
c
; (4:139)
where A is the area of the plane normal to the x-axis. The complementary energy
for cracked and uncracked laminates takes the form
+
0
=
1
2
s
2
c
E
x0
2Ah
+
=
1
2
s
2
c
E
x
2Ah; (4:140)
where E
x0
, E
x
are the longitudinal moduli of undamaged and damaged laminates,
respectively. Now, from the principle of minimum complementary energy, it
follows that if the admissible stress system, Eqs. (4.121)(4.122), with boundary
conditions, Eq. (4.124), is introduced into Eq. (4.135) then
~
+
_
+
: (4:141)
Using Eqs. (4.135) and (4.140) into Eq. (4.141), we obtain
1
2
s
2
c
E
x0
2Ah
/
_
1
2
s
2
c
E
x
2Ah; (4:142)
where
/
represents the perturbationincomplementary energy due tocracking. Hence,
1
E
x
_
1
E
x0

/
s
2
c
Ah
: (4:143)
Therefore, the variational boundary value problem results in the following calcu-
lus of variations problem:
Find f(x) that minimizes
/
.
The complementary energy change
/
is the sum of energies brought about by
perturbation stresses in the individual laminae, i.e.,
/
= 2
_
l
l
_
t
90
0
W
90
dz dx 2
_
l
l
_
h
t
90
W
0
dz dx; (4:144)
where W
90
and W
0
are stress energy densities due to perturbation stresses in the
90
- and 0
-layers, respectively. The stress energy density in a transversely iso-

tropic unidirectional ber composite is given by
W =
1
2
s
ij
e
ij
=
1
2
s
2
11
E
1
s
2
22
s
2
33
_ _
E
2
2n
12
s
11
s
22
s
33
( )
E
1
2n
23
s
22
s
33
E
2
s
2
23
G
23
s
2
12
s
2
13
_ _
G
12
_ _
;
(4:145)
where 1 is in the ber direction and 2, 3 are transverse directions, and the elastic
moduli are dened in Eq. (4.48). To determine the perturbation stress energy, we
can express stress energy densities in the 90
- and 0
-plies in terms of the perturb-

ation stresses in the corresponding plies. The stresses in each lamina are given in
terms of perturbation stresses by
s
22
= s
90
xx
; s
23
= s
90
xz
; s
33
= s
90
zz
; s
11
= s
12
= s
13
= 0 (4:146)
for the 90
-ply, and
s
11
= s
0
xx
; s
13
= s
0
xz
; s
33
= s
0
zz
; s
12
= s
22
= s
23
= 0 (4:147)
for the 0
-ply. Thus, the energy densities in the two sets of laminae are
W
90
=
1
2
s
90
xx
_ _
2
E
2
s
90
zz
_ _
2
E
2
2n
23
s
90
xx
s
90
zz
E
2
s
90
xz
_ _
2
G
23
_ _
;
W
0
=
1
2
s
0
xx
_ _
2
E
1
s
0
zz
_ _
2
E
2
2n
12
s
0
xx
s
0
zz
E
1
s
0
xz
_ _
2
G
12
_ _
:
(4:148)
Consider again the laminate region bounded by two transverse planes through
adjacent cracks, Figure 4.14. Due to symmetry about z = 0, it is sufcient to take
one half of the laminate l _ x _ l, 0 _ z _ h with unit width in y direction.
Further, the symmetry across x results into
f x ( ) = f x ( ); (4:149)
Substituting Eqs. (4.121), (4.122), and (4.148) into Eq. (4.144), and carrying out
the integration along z,
/
is given by
/
= s
90
xx0
_ _
2
_
l
l
t
90
C
00
f
2
t
3
90
C
02
ff
//
t
3
90
C
11
f
/
2
t
5
90
C
22
f
//
2
_ _
dx; (4:150)
where
C
00
=
1
E
2
1
lE
1
; C
02
= l
2
3
_ _
u
23
E
2
l
3
u
12
E
1
;
C
11
=
1
3
1
G
23
l
G
12
_ _
; C
22
= l 1 ( ) 3l
2
12l 8
_ _
1
60E
2
:
(4:151)
Introducing a nondimensional geometry parameter x = x/t
90
, Eq. (4.150) can be
rewritten as
/
= s
90
xx0
_ _
2
t
2
90
_
r
r
C
00
f
2
C
02
f
d
2
f
dx
2
C
11
df
dx
_ _
2
C
22
d
2
f
dx
2
_ _2
_ _
dx; (4:152)
where r = l/t
90
denotes the crack spacing normalized with the cracked ply thickness.
The EulerLagrange equation for f(x) is
d
4
f
dx
4
p
d
2
f
dx
2
qf = 0; (4:153)
where the coefcients p and q are given by
p =
C
02
C
11
C
22
; q =
C
00
C
22
: (4:154)
This is a fourth-order ODE whose characteristic equation is
r
4
pr
2
q = 0 : (4:155)
The roots of the characteristic equation are
r = a
1
ia
2
( ); i =
1
_
;
a
1
= q
1=4
cos
y
2
; a
2
= q
1=4
sin
y
2
;
y = tan
1
4q
p
2
1
;
(4:156)
provided that 4q/p
2
> 1. The general solution to Eq. (4.153) will involve terms with
e
a
1
x
cos a
2
x and e
a
1
x
sin a
2
x. However, it is more convenient to use hyperbolic
functions instead of exponentials, because of symmetry condition, Eq. (4.149).
Since only even product functions are admissible solutions, the solution of the
fourth-order ODE, Eq. (4.153), leads to the following expression for f(x)
f = A
1
cosh a
1
x cos a
2
x A
2
sinh a
1
x sin a
2
x; (4:157)
where A
1
and A
2
are constants determined from the boundary conditions as
A
1
=
2 a
1
cosh a
1
r sin a
2
r a
2
sinh a
1
rcos a
2
r ( )
a
1
sin 2a
1
r a
2
sinh 2a
2
r
;
A
2
=
2 a
2
cosh a
1
r sin a
2
r a
1
sinh a
1
rcos a
2
r ( )
a
1
sin 2a
1
r a
2
sinh 2a
2
r
;
(4:158)
when 4q/p
2
< 1, f(x) is given by [68]
f =
a
2
/
cosh a
1
/
x
sinh a
1
/
r a
2
/
coth a
1
/
r a
1
/
coth a
2
/
r ( )

a
1
/
cosh a
2
/
x
sinh a
2
/
r a
1
/
coth a
2
/
r a
2
/
coth a
1
/
r ( )
;
(4:159)
where a
/
1
; a
/
2
are dened as
a
1
/
; a
2
/
=
p
2

p
2
4
q
_
; (4:160)
assuming that p < 0, which usually holds good for typical glass/epoxy and
graphite epoxy materials [69]. Once f is known, the stress eld can be calculated
from Eqs. (4.121)(4.122). The effective stiffness coefcients for the cracked
laminate are then determined by calculating average stresses and strains.
It is important to note here that variational analysis gives lower bounds of stiffness
properties for the cracked laminates, as canbe seenfromthe inequality inEq. (4.143).
4.6.2 Effect of residual stresses
As a laminate is cooled from the cure temperature, thermal residual stresses are
generated due to differential thermal expansion of the 0
and 90
-plies. These
residual stresses can be easily incorporated by laminate theory for uncracked
laminate. The effect of thermal stresses on crack initiation was investigated in
detail by [70]. The variational stress formulation for cracked cross-ply laminate
including thermal residual stresses is covered in [69, 71]. The complementary
energy analogous to Eq. (4.152) is then given by [69]
/
=
/
0
t
2
90
_
r
r
C
00
c
2
C
02
c
d
2
c
dx
2
C
11
dc
dx
_ _
2
C
22
d
2
c
dx
2
_ _2
2a Tc C
2T
d
2
c
dx
2
_ _
dx;
(4:161)
where
c = s
90
xx0

a T
C
00
_ _
f
a T
C
00
; (a)
a = a
22
a
11
; T = T
0
T
ref
; (b)
C
2T
= a
22
T
n
23
s
c
E
c
_ _
2
3
l
_ _
a
22
T
n
12
s
c
E
c
_ _
l
2
3
; (c)
(4:162)
where a
11
and a
22
are the coefcients of thermal expansion in the longitudinal and
transverse directions, respectively, T
0
is the service temperature, and T
ref
is the
stress-free (reference/curing) temperature of the laminate. The corresponding
EulerLagrange equation for this case is
d
4
c
dx
4
p
d
2
c
dx
2
qc =
a T
C
22
: (4:163)
The solution for f remains the same as in Eqs. (4.158)(4.160), from which c can be
obtainedusingEq. (4.162a). The stresses canthenbe calculatedusingEqs. (4.121)(4.122).
4.6.3 [0
m
/90
n
]
s
vs. [90
n
/0
m
]
s
laminates
The stiffness changes due to cracking in [0
m
/90
n
]
s
laminates are different than in [90
n
/
0
m
]
s
laminates because in the former case we have cracks in an internal 90
-layer,
whereas the cracks in the latter are exposed to the free boundary. For the same area
of crack surfaces, the tractions on the crack surfaces (see Eq. (4.128)) do more work
in closing (or opening) the crack surfaces in the latter case, and, therefore, the
perturbation complementary energy has a higher value. This results in a lower
effective axial stiffness as compared to laminates with cracks in internal plies.
Nairn [69] carried out a variation of the solution for transverse cracking in
[90
n
/0
m
]
s
laminates. The procedure is the same as for [0
m
/90
n
]
s
laminates. The
obtained solution for f is also the same, except that some of the constants
appearing in Eqs. (4.150) and (4.162) are different. The newconstants are as follows
C
02
= 1
2
3
l
_ _
n
12
E
1
n
23
3E
2
;
C
22
= l 1 ( ) 3 12l 8l
2
_ _
1
60E
2
;
C
2T
=
1
3
n
23
s
c
E
c
a
22
T
_ _
n
12
s
c
E
c
a
22
T
_ _
l
2
3
l
2
_ _
:
(4:164)
A comparison of the stress proles in [0/90
2
]
s
and [90
2
/0]
s
laminates is illustrated in
Figure 4.15(b)(c). A discussion follows later.
4.6.4 Improved variational analysis
Some major improvements toHashins variational analysis were suggestedbyVarnaand
Berglund [7173] with the most rened model described in [73]. In Hashins approach,
the axial stress variation across the thickness is assumed constant, whereas the Varna
Berglund approach determines these by further application of the principle of minimum
complementary energy. The stress functions in the 90
- and 0
-plies are chosen as
90
= s
90
xx0
z
2
2
c x ( )
z
2
2
A
+
_ _ _ _
c
1
x ( )
2
z ( )
_ _
t
2
90
;
0
= s
0
xx0
z
2
2
s
90
xx0
1
z ( )c x ( )
_ _
s
90
xx0
c
1
x ( )
3
z ( )
_ _
t
2
90
;
(4:165)
where s
0
xx0
ands
90
xx0
are the axial stresses in the 0
- and 90
-plies of uncracked
laminate, respectively, x = x=t
90
and z = z=t
90
are nondimensional coordinates,
and A* is a constant. In the above expressions, the arbitrary function
2
(z) repre-
sents nonuniformity in the x-axis stress distribution in the 90
-layer close to the

cracks along z, whereas c
1
(x) characterizes this stress variation along the x-direc-
tion. For the 0
-layer, the redistribution in the stress eld is described by c

1
(x) and
3
(z) along the x- and z-directions, respectively. If we neglect
2
(z);
3
(z); c
1
(x) and
set A
+
= h=2t
90
and
1
z ( ) = (h z)
2
=2t
0
t
90
we revert back to Hashins model.
For the current model, the stress components in layer m are given by
10
20
30
40
10
20
30
40
(interfacial)
(MPa)
xz
s
90
xx
s
90
x
Crack
Crack
4
3
(a)
2 1
4
3 2 1
50
60
90
xz
90
xx
(interfacial)
(interfacial)
90
zz
60 (MPa)
40
30
20
10
40
30
20
10
3
4
Crack
Crack
2 1 2 3
4
x
50
1
(b)
70 (MPa)
(interfacial)
60
90
xx
s
90
zz
s
(interfacial)
90
xz
s
50
Crack
Crack
1
4 4
x
40
30
20
10
30
20
10
2 3 3 2 1
(c)
Figure 4.15. Prole of stresses between two adjacent ply cracks in a carbon/epoxy cross-ply
laminate: (a) using 1-D shear lag analysis for [0/90
2
]
s
; (b) using 2-D variational analysis for
[0/90
2
]
s
; and (c) using 2-D variational analysis for a [90
2
/0]
s
laminate. The normal stress s
90
xx
is same for the entire 90
-ply group, whereas shear stress s

90
xz
and transverse stress s
90
zz
are
plotted at the 0/90 ply interface x = 1. Reprinted, with kind permission, from Damage
Mechanics of Composite Materials, J. A. Nairn and S. Hu, Micromechanics of damage: a
case study of matrix microcracking, pp. 187243, copyright Elsevier (1994).
s
m
xx
=
@
2
m
@z
2
s
m
zz
=
@
2
m
@x
2
s
m
xz
=
@
2
m
@x@z
: (4:166)
Thus,
(a) in the 90
-layer
s
90
xx
= s
90
xx0
1 c x ( ) c
1
x ( )
//
2
z ( ) [ [;
s
90
xz
= s
90
xx0
c
/
x ( )z c
/
1
x ( )
/
2
z ( ) [ [;
s
90
zz
= s
90
xx0
c
//
x ( )
z
2
2
A
+
_ _
c
//
1
x ( )
2
z ( )
_ _
;
(4:167)
(b) in the 0
-layer
s
0
xx
= s
0
xx0
s
90
xx0
c x ( )
//
1
z ( ) c
1
x ( )
//
3
z ( ) [ [;
s
0
xz
= s
90
xx0
c
/
x ( )
/
1
z ( ) c
/
1
x ( )
/
3
z ( ) [ [;
s
0
zz
= s
90
xx0
c
//
x ( )
1
z ( ) c
//
1
x ( )
3
z ( ) [ [ :
(4:168)
The boundary and interface conditions remain the same as given in Eq. (4.120),
which, after using Eqs. (4.167) and (4.168), results in
/
1
1 ( ) = 0;
1
1 ( ) = A
+

1
2
;
/
1

h ( ) =
1

h ( ) = 0;
/
3

h ( ) =
3

h ( ) = 0;
/
2
1 ( ) =
/
3
1 ( ) = 1;
2
1 ( ) =
3
1 ( );
c r ( ) = 1; c
/
r ( ) = c
/
1
r ( ) = c
1
r ( ) = 0;
(4:169)
where h = h=t
90
. Clearly, this model is complex and requires determining the
constant A*, and the functions c x ( ); c
1
x ( ); and
1
z ( );
2
z ( )
3
z ( ). However, the
authors found that the following choice of functions
1
z ( );
2
z ( );
3
z ( ) is useful
1
z ( ) =
1 cosh
1
z

h ( )
1
sinh
1
l
;
2
z ( ) = A
z
2n
2n
;
3
z ( ) =
1 cosh
3
z

h ( )
3
sinh
3
l
;
(4:170)
with A a constant and D
1
, D
3
, and n arbitrary shape parameters, where n is an
integer. Using the boundary conditions described in Eq. (4.169), we obtain
A
+
=
1
2

1 cosh
1
l
1
sinh
1
l
; A =
1
2n

1 cosh
3
l
3
sinh
3
l
: (4:171)
Now, the total complementary energy of the cracked laminate system is given by
/
=
/
0

_
r
r
v c; c
/
; c
//
; c
1
; c
/
1
; c
//
1
) d x; ( (4:172)
where
/
0
is the complementary energy of the virgin laminate, and v is the comple-
mentary energy density due to perturbation stresses. Using the minimization
procedure as described above, we nally obtain the following system of ordinary
differential equations with constant coefcients
C
0
22
c
///
C
0
02
C
0
11
_ _
c
//
C
0
00
c C
01
22
c
///
1
C
R
c
//
1
C
01
00
c
1
= 0;
C
01
22
c
///
C
R
c
//
C
01
00
c C
1
22
c
///
1
C
1
02
C
1
11
_ _
c
//
1
C
1
00
c
1
= 0;
(4:173)
with the boundary conditions as in Eq. (4.169) and constants given by
C
0
00
= 1
E
2
E
1
I
1
; C
0
22
=
1
20

A
+
3
A
+
2
I
2
;
C
0
11
=
E
2
3G
23
E
2
G
12
I
3
; C
0
02
= 2n
23
A
+

1
6
_ _
2n
12
E
2
E
1
I
4
;
C
1
00
= I
T
1

E
2
E
1
I
L
1
; C
1
22
= I
T
2
I
L
2
;
C
1
11
=
E
2
G
23
I
T
3

E
2
G
12
I
L
3
; C
1
02
= 2n
23
I
T
4
2n
12
E
2
E
1
I
L
4
;
C
01
00
= 1
E
2
E
1
I
C
1
; C
01
22
= F
3
I
C
2
;
C
01
11
=
E
2
G
23
F
1

E
2
G
12
I
C
3
; C
01
02
= 2n
23
F
2
2n
12
E
2
E
1
I
C2
4
;
C
01
20
= 2n
23
F
4
2n
12
E
2
E
1
I
C1
4
; C
R
=
1
2
C
01
20
C
01
02
_ _
C
01
11
;
(4:174)
where
I
1
=
_
h
1
//
1
z ( ) [ [
2
dz; I
2
=
_
h
1
1
z ( ) [ [
2
dz; I
3
=
_
h
1
/
1
z ( ) [ [
2
dz;
I
4
=
_
h
1
1
z ( )
//
1
z ( )dz; I
T
1
=
_
1
0
//
2
z ( ) [ [
2
dz; I
T
2
=
_
1
0
2
z ( ) [ [
2
dz;
I
T
3
=
_
1
0
/
2
z ( ) [ [
2
dz; I
T
4
=
_
1
0
2
z ( )
//
2
z ( )dz; F
1
=
_
1
0
z
2
z ( )dz;
F
2
=
_
1
0
2
z ( )dz; F
3
=
_
1
0
2
z ( )
z
2
2
A
+
_ _
dz; F
4
=
1
2
A
+

2
1 ( ) F
2
;
I
L
1
=
_
h
1
//
3
z ( ) [ [
2
dz; I
L
2
=
_
h
1
3
z ( ) [ [
2
dz; I
L
3
=
_
h
1
/
3
z ( ) [ [
2
dz;
I
L
4
=
_
h
1
3
z ( )
//
3
z ( )dz; I
C
1
=
_
h
1
1
//
z ( )
//
3
z ( )dz; I
C
2
=
_
h
1
1
z ( )
3
z ( )dz;
I
C
3
=
_
h
1
1
/
z ( )
/
3
z ( )dz; I
C1
4
=
_
h
1
1
z ( )
//
3
z ( )dz; I
C2
4
=
_
h
1
//
1
z ( )
3
z ( )dz:
(4:175)
The minimum value of the complementary energy corresponding to the obtained
solution is given by [73] as
/
min
=
/
0

s
90
xx0
t
90
_ _
2
2E
2
C
01
22
///
1
r ( ) C
0
22
///
r ( )[ :
_
(4:176)
The stresses resulting from cracking can be calculated by solving the system of
coupled ODEs in Eq. (4.173) to get the perturbation functions, and then putting
them into Eqs. (4.167)(4.168). The average of the stresses will yield the average
stiffness properties of the cracked laminate. For this model, the longitudinal
Youngs modulus normalized with its virgin state value is given by [73]
E
x
E
x0
=
1
1
E
2
lE
1
f r ( )
r
; (4:177)
where
f r ( ) =
1
2
_
r
r
x ( )
1
x ( ) [ [ d x : (4:178)
The normalized effective Poissons ratio for the cracked cross-ply laminate is
given by
n
xy
n
0
xy
=
1 1
E
2
E
1
_ _
t
90
h
f r ( )
r
1
E
2
lE
1
f r ( )
r
: (4:179)
4.6.5 Related works
There have been some further developments based on the variational analysis
procedure. The analysis of transverse cracks in cross-ply laminates undergoing
shear loading was covered by Hashin in his original paper [67]. Later he utilized
the variational approach to predict the thermal expansion coefcients for the
cracked cross-ply laminate [74]; this is covered later in Section 4.11. In another
paper [75], he also analyzed the case of orthogonally cracked cross-ply lamin-
ates. Another signicant analysis of cross-ply laminates was carried out by
Kuriakose and Talreja [76] where they analyzed cross-ply laminates subjected
to bending moments.
4.6.6 Comparison between 1-D and 2-D stress-based models
Comparison of stress predictions between 1-D shear lag [27] and 2-D
variational analysis [67] for a [0/90
2
]
s
carbon epoxy laminate are shown in
Figure 4.15(a)(b). The laminate is subjected to an applied mechanical stress
of 100 MPa, and a temperature change of DT = 125
C. The normalized
crack spacing is r = 4.
As seen in Figure 4.15(b), s
90
xx
= 0 at x= r as the crack surfaces are traction
free and the transverse plies carry no axial load at the crack planes. As we move
away from the crack surfaces, s
90
xx
increases as stress is transferred back into the
90
-layer from the 0
-layer. This stress reaches a maximum midway between two

transverse cracks and its value depends on the crack spacing. When the cracks are
sufciently far apart (i.e., no interaction between adjacent cracks), s
90
xx
will reach
s
90
xx0
at this position (x = 0). The distributions of s
90
xx
are qualitatively similar for
shear lag and variational models. However, the maximum value of s
90
xx
is different
in the two cases. Also, variational predictions correctly show that this stress has
zero slope at crack surfaces (x = r), while the shear lag models display a
nonzero slope.
Shear lag analysis predicts a maximum value for the interfacial shear stress
t
i
= s
90
xz
z = t
90
( ) at the crack surfaces, in violation of the boundary conditions that
require t
i
= 0 at crack planes (Figure 4.15(a)). In addition, it cannot provide the
transverse normal stress s
90
zz
. The one-dimensional nature of the shear lag models
results in no distinction between [0/90]
s
and [90/0]
s
laminates, while experimental
observations show differences in the cracking behavior and resultant stiffness
degradation in the two cases. These differences are appropriately predicted by
the variational solution shown for a [90
2
/0]
s
laminate in Figure 4.15(c), where the
stress variations shown are different from the [0/90
2
]
s
case. The interfacial trans-
verse stress s
90
zz
near crack surfaces is compressive in [0/90
2
]
s
while it is tensile in
[90
2
/0]
s
laminate. This suggests that cracks in a [90
2
/0]
s
laminate will tend to
promote mode I delamination.
In an alternative way, Varna and coauthors [7779] investigated shear lag and
variational methods from the viewpoint of average crack opening displacement
(COD). They noted that the basic difference between different models is how they
model the average COD, which is dened as
u =
1
t
90
_
t
90
0
u z ( ) dz : (4:180)
Consider transverse cracking in a symmetric and balanced [S/90
n
]
s
laminate,
consisting of two balanced sublaminates as outer layers and 90
layers in the
middle. The longitudinal modulus and Poissons ratio for the laminate are
given by
E
xx
=
s
0
e
S
xx
; n
xy
=
e
S
yy
e
S
xx
; (4:181)
where s
0
is the applied stress, the overbars denote volume averages, and
superscript S denotes sublaminate. The average COD for this laminate is given by
u = l e
S
xx
e
90
xx
_ _
: (4:182)
Following [79], the stiffness properties for the cracked laminate can be shown to
be given by
E
xx
E
xx0
=
1
1 arR l
_ _ ;
u
xy
u
xy0
=
1 crR l
_ _
1 arR l
_ _ ; (4:183)
where r = 1/2l is the crack density; a, c, and g are known functions of laminate
material and thickness; R

l ( ) is the average stress perturbation function;
and

l = l=t
90
is the half-crack spacing normalized with the 90
-layer thickness.
Different approaches model R

l ( ) differently:
1. Shear lag models:
R

l ( ) =
2
b
tanh b
l ( ); (4:184)
where b is the shear lag parameter. Different shear lag models use different
denitions for this parameter [78].
2. Variational methods:
R

l ( ) =
4a
1
a
2
a
2
1
a
2
2
cosh 2a
1
l ( ) cos 2a
2
l ( )
a
2
sinh 2a
1
l ( ) a
1
sin 2a
2
l ( )
; (4:185)
where the constants a
1
, a
2
are dened as per model.
The comparisons of different models for [S/90
4
]
s
laminates, as performed
by Joffe and Varna [79] are shown in Figures 4.164.19 for different values
of y. For the 2-D-0 model, the expression for the stress perturbation function
is the same as for Hashins variational analysis, except that the coefcients are
now given as [79]
C
00
=
1
E
2
1
E
S
x
I
1
;
C
02
= 2
n
32
E
2
1
6
A
+
_ _
2
n
S
xz
E
S
x
I
4
;
C
11
=
1
2G
23
1
G
S
xz
I
3
:
(4:186)
This comparison conrms that variational analyses consistently provide a
lower bound to stiffness properties and are signicantly better than shear lag
models. The variational analysis predictions are also very close to FE
predictions.
4.7 Generalized plain strain analysis McCartneys model
These models may be seen as a further development of stress analysis-based
damage models for cracked laminates. The cracked cross-ply laminates, symmetric
about the laminate mid-plane, were considered in a generalized plane strain
condition, thereby reducing the directional dependence of essentially a 3-D stress
eld in the cracked laminate. Assuming a regular array of fully grown parallel
cracks in 90
-ply, McCartney [80] developed a theory of stress transfer between

0
- and 90
-plies while retaining all relevant stress and strain components.

The equilibrium equations, as in Hashins formulation, are given by Eq. (4.112).
Assuming lamina properties to be transversely isotropic (or orthotropic), the
1.0
(a)
0.9
0.8
E
x
/
E
x
0
0.7
0.6
0.5
0.4
0.0 0.1
Sh.L.-2
[0
2
/90
4
]
s
Hashin
FE
2D-0
Experiment
0.2 0.3
Crack density (cr/mm)
0.4 0.5 0.6 0.7
1.0
(b)
[0
2
/90
4
]
s
0.9
0.8
0.7
V
x
y
/
V
x
y
0
0.6
0.5
0.4
0.0 0.1 0.2 0.3
0.4 0.5
Sh.L.-2
Hashin
FE
2D-0
Experiment
0.6 0.7
Figure 4.16. Variation of elastic stiffness moduli with crack density for a [0
2
/90
4
]
s
glass/epoxy
laminate: (a) normalized Youngs modulus; (b) normalized Poissons ratio versus crack
density. Reprinted, with kind permission, from Compos Sci Technol, Vol. 59, R. Joffe and
J. Varna, Analytical modeling of stiffness reduction in symmetric and balanced laminates
due to cracks in 90 degrees layers, pp. 164152, copyright Elsevier (1999).
thermomechanical constitutive relations for 0
- and 90
-plies in material (lamina)

coordinate system can be written as
e
m
11
=
s
m
11
E
m
11
n
m
21
s
m
22
E
m
22
n
m
31
s
m
33
E
m
33
a
m
11
T;
e
m
22
=
n
m
12
s
m
11
E
m
11
s
m
22
E
m
22
n
m
32
s
m
33
E
m
33
a
m
22
T;
(4:187)
e
m
33
=
n
m
13
s
m
11
E
m
11
n
m
23
s
m
22
E
m
22
s
m
33
E
m
33
a
m
33
T;
e
m
12
=
s
m
12
2G
m
12
; e
m
13
=
s
m
13
2G
m
13
; e
m
23
=
s
m
23
2G
m
23
;
1.0
(a)
0.9
0.8
0.7
E
x
/
E
x
0
0.6
0.5
0.4
0.0 0.1 0.2
Sh.L.-2
Hashin
FE
2D-0
[
+
15/90
4
]
s
Experiment
0.3
0.4 0.5 0.6 0.7
_
1.0
(b)
0.9
0.8
0.7
V
x
y
/
V
x
y
0
0.6
0.5
0.4
0.0 0.1
Sh.L.-2
Hashin
FE
2D-0
Experiment
0.2 0.3
[15/90
4
]
s
0.4 0.5 0.6 0.7
Figure 4.17. Variation of elastic stiffness moduli with crack density for a [15/90
4
]
s
glass/
epoxy laminate: (a) normalized Youngs modulus; (b) normalized Poissons ratio versus
crack density. Reprinted, with kind permission, from Compos Sci Technol, Vol. 59, R. Joffe
and J. Varna, Analytical modeling of stiffness reduction in symmetric and balanced
laminates due to cracks in 90 degrees layers, pp. 164152, copyright Elsevier (1999).
105 4.7 Generalized plain strain analysis McCartneys model
where m = 0, 90 for 0
-ply and 90
-ply, respectively, and E

ii
, G
ij
, n
ij
; i, j = 1, 2, 3
represent the Youngs modulus, shear modulus, and Poissons ratio, respectively,
in corresponding directions and planes. Following the plane strain assumption,
the displacement eld for the damaged laminate has the following form
u
m
= u
m
(x; z); n
m
= e
c
T
y; w
m
= w
m
x; z ( ); (4:188)
where e
c
T
is the uniform strain in the cracked laminate along the transverse (y)
direction. This representation of the displacement eld is valid if ply cracks form
in planes normal to the x-direction, and are well away from the laminate edges.
1.0
(a)
0.9
0.8
0.7
0.6
0.5
0.4
0.0
0.1 0.2 0.3
Shear lag model: [30/90
4
]
s
0.4 0.5
1
3
2
0.6 0.7
E
x
/
E
x
0
Experiment
1.0
0.9
(b)
0.8
0.7
V
x
y
/
V
x
y
0
0.6
0.5
0.4
0.0 0.1 0.2 0.3 0.4
0.5 0.6 0.7
Sh.L.-2
Hashin
FE
2D-0
Experiment
[30/90
4
]
s
4
]
s
glass/
epoxy laminate: (a) normalized Youngs modulus; (b) normalized Poissons ratio
versus crack density. Reprinted, with kind permission, from Compos Sci Technol, Vol. 59,
R. Joffe and J. Varna, Analytical modeling of stiffness reduction in symmetric and
balanced laminates due to cracks in 90 degrees layers, pp. 164152, copyright Elsevier
(1999).
Now, if we additionally assume that the longitudinal stress components in both
the 0
- and 90
-plies are independent of the transverse coordinate z, the relevant

stress components can be written in the following functional form
s
0
xx
= s
0
xx0
C x ( ); s
90
xx
= s
90
xx0
lC x ( ); (4:189)
where
s
0
xx0
= Q
0
11
e
c

n
0
12
E
0
22
E
0
11
a
/
0
11
T
_ _
; s
90
xx0
= Q
90
11
e
c
n
0
12
e
c
T
a
/
90
22
T
_ _
; (4:190)
where e
c
is the average longitudinal strain applied to the laminate, and Q
0
11
and Q
90
11
are the reduced stiffnesses for the 0
- and 90
-plies, respectively, given by

(a)
1.0
0.9
0.8
0.7
E
x
/
E
x
0
0.6
0.5
0.4
0.0 0.1 0.2 0.3 0.4
Sh.L.-2
Hashin
2D-0
Experiment
FE
0.5 0.6 0.7
[40/90
4
]
s
1.0
(b)
0.9
0.8
0.7
0.6
0.5
0.4
0.0 0.1 0.2 0.3
0.4 0.5 0.6 0.7
Sh.L.-2
Hashin
FE
2D-0
Experiment
[40/90
4
]
s
V
x
y
/
V
x
y
0
4
]
s
glass/
epoxy laminate: (a) normalized Youngs modulus; (b) normalized Poissons ratio versus
crack density. Reprinted, with kind permission, from Compos Sci Technol, Vol. 59, R. Joffe
and J. Varna, Analytical modeling of stiffness reduction in symmetric and balanced
laminates due to cracks in 90 degrees layers, pp. 164152, copyright Elsevier (1999).
Q
m
11
=
E
m
11
1 n
m
12
n
m
21
; m = 0; 90 (4:191)
and a
/ 0
11
and a
/ 90
22
are given as
a
/
0
11
= a
0
11
n
0
12
E
0
22
E
0
11
a
0
22
; a
/
90
22
= a
90
11
n
0
12
a
90
11
: (4:192)
To fully represent the stress state in the cracked laminate, the basic task is to
determine the unknown function C(x) appearing in Eq. (4.189). The boundary
conditions for the stress eld are: symmetry about the mid-plane, symmetry about
the plane parallel to cracks lying midway between two adjacent cracks, stress
continuity at the interface between the 0
- and 90
-plies, and zero traction at crack

surfaces and the external surfaces of the outer plies (similar to Hashins analysis,
see Eq. (4.120)). Additionally, by force balance we have
s
c
h =
_
t
90
0
s
90
xx
dz
_
h
t
90
s
0
xx
dz : (4:193)
Also, since there is no load applied on the laminate in the transverse direction, the
average transverse stress must add to zero, i.e.,
_
t
90
0
_
l
0
s
90
yy
dxdz
_
h
t
90
_
l
0
s
90
yy
dxdz = 0 : (4:194)
The stress eld in the cracked laminate can be obtained by substituting Eq. (4.189)
into the equilibrium equations, Eq. (4.112), and integrating and applying bound-
ary conditions. Thus, the transverse normal and shear stresses in the laminate are
given by
s
0
xz
= C
/
x ( ) 1 l ( )t
90
z [ [; s
90
xz
= lC
/
x ( )z;
s
0
zz
=
1
2
C
//
x ( ) 1 l ( )t
90
z [ [
2
; s
90
zz
=
1
2
C
//
x ( ) 1 l ( )t
2
90
z
2
_
:
(4:195)
It can be observed that these relations are quite similar in form to those obtained
by Hashin (Eqs. (4.121) and (4.122)). In fact, they are identical when
C(x) = (1=l)s
90
xx
f(x). The complete stress eld is given by combining Eqs.
(4.189) and (4.195).
The displacement eld, given by Eq. (4.188), can be obtained by using the
stress eld into constitutive relations, Eq. (4.187), and integrating. However, the
stressstrain relations must be satised in an average sense. Hence, the stress
and strain elds are averaged over each ply thickness. These averaged relations
lead to the following fourth-order differential equation for the unknown func-
tion C(x)
FC
IV
x ( ) GC
//
x ( ) HC x ( ) = 0; (4:196)
where the coefcients F, G, H are given by
F =
1
20E
0
22
2
15E
90
22
t
90
t
0
t
90
t
0
_ _
2
5
2
t
90
t
0
15
8
_ _
;
G =
1
3
1
G
0
12
1
G
0
23
t
90
t
0
n
0
12
E
0
11
n
90
23
E
90
22
2
t
90
t
0
3
_ _ _ _
;
H =
1
E
0
11
t
90
t
0
1
E
0
22
:
(4:197)
When the inner and outer plies are made of the same material the differential
equation, Eq. (4.196), has the same form as the EulerLagrange equation derived
by Hashin. It is worth noting that the Reissner energy functional has a stationary
value when the constitutive relations and equilibrium equations are satised in
an average sense. Thus, McCartneys approach is quite analogous to Hashins
approach: the former uses a displacement formulation (equivalent to minimiza-
tion of Reissners energy functional), while the latter uses a stress formulation
(equivalent to minimization of complementary energy functional).
For a cracked cross-ply laminate, McCartneys 2-D approach yields the
following relations for the effective longitudinal modulus
E
xx
=
E
xx0
1
t
90
l
E
0
22
E
0
11
; (4:198)
where
=
2pq
p
2
q
2
cosh
2
pl
t
0
cosh
2
ql
t
0
_ _
;
1
= q sinh
pl
t
0
cosh
pl
t
0
p sin
ql
t
0
cos
ql
t
0
;
p =
r s
2
_
; q =
r s [ [
2
_
;
r =
G
2F
> 0; s =
H
F
_
> 0 :
(4:199)
If the plies are thick, the averaging of stresses andstrains is not advisable as it canresult
in inaccurate predictions. To improve the approach, McCartney [8186] used the so-
called ply-renement procedure, in which each ply is divided thickness-wise into N
segments, and the resulting stressstrain relations are satised, in an average sense,
over each ply segment. Consequently, the unknown functions C
i
(x) for the ith ply
segment (i = 1,2, . . . , N), satisfy the following N homogeneous differential equations
N
i=1
F
ij
C
IV
i
x ( )
N
i=1
G
ij
C
//
i
x ( )
N
i=1
H
ij
C
i
x ( ) = 0; j = 1; 2; : : : ; N; (4:200)
where the coefcients F
ij
, G
ij
, and H
ij
are calculated numerically using a suitable
ODE solver. McCartney [82, 84, 86] has later generalized this approach for multi-
layered cross-ply laminates, and also for triaxial loading.
McCartney [82, 84, 86] has also shown that for cracked cross-ply laminates, the
moduli are interrelated. Similar interrelationships among the Poissons ratio,
transverse modulus, and longitudinal modulus have also been independently
derived by Nairn [87]. If we dene a damage parameter or normalized stiffness
parameter as
D =
1
E
xx
1
E
xx0
; (4:201)
then other moduli for the cracked cross-ply laminate are interrelated as
n
xy0
E
xx0
n
xy
E
xx
= k
1
D;
1
E
yy
1
E
yy0
= k
2
1
D;
n
xz0
E
xx0
n
xz
E
xx
= k
2
D;
1
E
zz
1
E
zz0
= k
2
2
D;
n
yz0
E
yy0
n
yz
E
yy
= k
1
k
2
D;
(4:202)
and
a
xx
a
xx0
= k
3
D; a
yy
a
yy0
= k
1
k
3
D; a
zz
a
zz0
= k
2
k
3
D
; (4:203)
where
k
1
=
E
xx0
E
yy0
B n
xy0
E
yy0
=E
xx0
_ _
1 n
xy0
B
;
k
2
=
E
xx0
A n
xz0
n
xz0
E
xx0
=E
yy0
_ _
B
1 n
xy0
B
;
k
3
=
E
xx0
s
xx0
Ba
yy0
C
_
1 n
xy0
B
;
(4:204)
with
A =
n
13
E
22
n
23
n
12
E
11
; B = n
12
; C = a
22
n
12
a
11
: (4:205)
These interrelationships can in principle reduce the burden of evaluating change in
each modulus subsequent to matrix cracking. Instead, the evaluation of degrad-
ation in longitudinal modulus is sufcient for predicting other moduli. However,
this result has yet to be veried experimentally.
4.8 COD-based methods
One way to view the elastic response changes caused by the presence of cracks in a
medium is to consider the additional overall (global) strain of the RVE contrib-
uted by the crack surface displacements of the individual cracks within the RVE.
Contrarily, if none of the cracks in the RVE conducts surface displacements, the
overall elastic response of the RVE will not change. This observation motivates
the focus on crack surface displacements. Although generally these displacements
can be described as crack opening displacement (COD) and crack sliding displace-
ment (CSD), one commonly refers only to COD. We will in the following discuss
stiffness relations in terms of average COD that can be calculated either analytic-
ally [8896] or numerically [97101].
4.8.1 3-D laminate theory: Gudmundsons model
Consider a general three-dimensional laminate consisting of N plies (see
Figure 4.20). Each ply is dened by its material properties, layup angle, and
thickness. For uncracked laminate the global average stresses s
ij
and strains e
ij
are dened as [102]
s
ij
=
N
k=1
V
k
s
k
ij
; e
ij
=
N
k=1
V
k
e
k
ij
; (4:206)
where s
k
ij
; e
k
ij
are the average stresses and strains, respectively, in the kth ply (k =
1, 2, . . ., N), and V
k
stands for the volume fraction of the kth ply such that
N
k=1
V
k
= 1. Note that for uncracked laminates the ply average stresses and
strains are the same as the individual (constant) ply stresses. For the sake of
simplicity, we will denote the tensors with a ~. Now, let us partition the stresses,
strains, and thermal expansion coefcients into in-plane and out-of-plane parts
s
~
=
s
~
I
s
~
O
_ _
; e
~
=
e
~
I
e
~
O
_ _
; s
~
=
s
~
I
s
~
O
_ _
; (4:207)
where
s
~
I
=
s
xx
s
yy
s
xy
_
_
_
_
; e
~
I
=
e
xx
e
yy
2e
xy
_
_
_
_
; a
~
I
=
a
xx
a
yy
2a
xy
_
_
_
_
; (4:208)
denote the in-plane stresses, strains, and thermal expansion coefcients, and
s
~
O
=
s
zz
s
xz
s
yz
_
_
_
_
; e
~
O
=
e
zz
2e
xz
2e
yz
_
_
_
_
; a
~
O
=
a
zz
2a
xz
2a
yz
_
_
_
_
; (4:209)
k
k+1
t
k
l
k
l
k
Figure 4.20. A general 3-D laminate with cracks in some off-axis layers.
111 4.8 COD-based methods
denote the out-of-plane stresses, strains, and thermal expansion coefcients. The consti-
tutive law between global average stress tensor and average strain tensor is given by
e
~
= S
~
s
~
a
~
T
=
e
~
I
e
~
O
_ _
=
S
~
II S
~
IO
S
~
IO
_ _
T
S
~
OO
_
_
_
_
s
~
I
s
~
O
_ _
a
~
I
a
~
O
_ _
T;
(4:210)
where S
~
is the compliance tensor, and the superscript T represents the trans-
pose. Similarly, the relation between ply stress tensor and ply strain tensor can
be written as
e
~
k
= S
k
~
s
k
~
s
k(r)
~
_ _
a
~
T
=
e
~
I
k
e
~
O
k
_
_
_
_
=
S
~
I
k
S
~
IO
k
S
~
IO
k
_ _
T
S
~
OO
k
_
_
_
_
s
~
I
k
s
~
I
k(r)
s
~
O
k
_
_
_
_
s
~
I
k
s
~
O
k
_
_
_
_
T;
(4:211)
where s
~
k(r)
denote residual stresses present due to reasons other than thermal
mismatch, such as chemical shrinkage during the manufacturing process. The
global average in-plane and out-of-plane residual stresses vanish due to equilib-
rium. The compatibility and equilibrium conditions in the laminate give us
e
~
I
k
= e
~
I
; s
~
O
k
= s
~
O
: (4:212)
Substituting Eq. (4.212) into Eq. (4.210), one obtains
s
~
I
k
= S
~
II
k
_ _
1
e
~
I
S
~
IO
k
s
~
O
a
~
I
k
T
_ _
s
~
I
k(r)
: (4:213)
Using Eq. (4.206), and using the fact that the volume average of residual stresses
vanish, the above equation can be rewritten as
e
~
I
= S
~
II
s
~
I
S
~
IO
s
~
O
a
~
I
T; (4:214)
where
S
~
II
=
N
k=1
V
k S
~
II
k
_ _
1
_ _
1
;
S
~
IO
= S
~
II
N
k=1
V
k S
~
II
k
_ _
1
S
~
IO
k
_ _
;
a
~
I
= S
~
II
N
k=1
V
k S
~
II
k
_ _
1
a
~
I
k
_ _
:
(4:215)
In a similar way, substituting Eq. (4.212) into Eq. (4.211), and applying Eq.
(4.206), we get
e
~
O
= S
~
IO
_ _
T
s
~
I
S
~
OO
s
~
O
a
~
O
T; (4:216)
where
S
~
OO
= S
~
IO
_ _
T
S
~
II
_ _
1
S
~
IO

N
k=1
V
k S
~
OO
k
S
~
IO
k
_ _
T
S
~
II
k
_ _
1
S
~
IO
k
_ _
;
a
~
O
= S
~
IO
_ _
T
S
~
II
_ _
1
a
~
I

N
k=1
V
k
a
~
O
k
S
~
IO
k
_ _
T
S
~
k
II
)
1
a
~
I
k
_ _
:
_
(4:217)
Eqs. (4.215) and (4.217) relate the effective laminate properties to the local
ply level properties. The above equations form the basis of a 3-D laminate
theory. The corresponding relations for a 2-D laminate theory are obtained
by considering only the in-plane tensors. To obtain the effective laminate
properties for a cracked laminate, dene the nondimensional crack density in
the kth ply as
r
k
=
t
k
l
k
; (4:218)
where t
k
and l
k
denote the thickness and average crack spacing, respectively, in the
kth ply. The process of transverse cracking reduces the elastic energy of the
laminate and the change in elastic energy is associated with the release of tractions
on the crack surfaces. It was shown by Gudmundson and Ostlund [92] that the
coupling terms between the energy of the uncracked laminate and the change in
elastic energy due to matrix cracking vanish (see also the proof by Hashin [67],
which is described earlier in the chapter, Eq. (4.135)).
It is noted that for a cracked laminate, the effective strains are different from the
average strains, whereas there is no distinction between global effective and
average stresses. The effective strains are the strains measured on a global scale,
whereas average strains come from averaging strains in individual plies over the
RVE. The difference between the effective strains and average strains is equal to
strain increment caused by crack opening displacements. Thus, the average
stresses are dened as (see Eq. (4.206))
s
(a)
ij
=
N
k=1
V
k
s
k(a)
ij
; (4:219)
where the superscript (a) denotes average variables. The global effective strains are
dened as
e
(e)
ij
=
1
2V
_
out
u
i
n
j
u
j
n
i
_ _
d; (4:220)
where u
i
, i = 1, 2, 3 indicates the displacement vector, n
i
is the unit normal vector
on the outer boundary surface G
out
of the representative volume V, and the
superscript (e) denotes the effective variables. In the same way the effective ply
strains can be dened as [102]
e
k(e)
ij
=
1
2V
k
_
kout
u
k
i
n
k
j
u
k
j
n
k
i
_ _
d; (4:221)
where V
k
is the volume of the ply k and the surface integral is performed on the
outer boundary of ply k. Obviously,
1
2V
_
out
u
i
n
j
u
j
n
i
_ _
d =
N
k=1
V
k
1
2V
k
_
kout
u
k
i
n
k
j
u
k
j
n
k
i
_ _
d
_ _
: (4:222)
Hence, from Eqs. (4.220)(4.222), one gets
e
(e)
ij
=
N
k=1
V
k
e
k(e)
ij
: (4:223)
Applying the divergence theorem on Eq. (4.221), we get
e
k(e)
ij
=
1
2V
k
_
V
k
u
k
i;j
u
k
j;i
_ _
d
1
2V
k
_
kc
u
k
i
n
k
j
u
k
j
n
k
i
_ _
d; (4:224)
where the rst integral is equal to the average ply strain, given by
e
(a)
ij
=
1
2V
k
_
V
k
u
k
i;j
u
k
j;i
_ _
d; (4:225)
and the second integral is the strain increment due to matrix cracks e
k
ij
is
given by
e
k
ij
=
1
2V
k
_
kc
u
k
i
n
k
j
u
k
j
n
k
i
_ _
d =
r
k
2t
k
u
k
i
n
k
j
u
k
i
n
k
j
_ _
; (4:226)
where
u
k
i
=
1
t
k
_
t
k
0
u
k()
i
u
k()
i
_ _
dt
k
=
1
t
k
_
t
k
0
u
k
i
dt
k
(4:227)
is the average crack opening displacement for ply k. Thus, we can write effective
ply strains as
e
k(e)
ij
= e
k(a)
ij
e
k
ij
: (4:228)
Finally, after applying the proper boundary conditions, Gudmundson and
Zang [89] arrive at the following expressions for effective properties of cracked
laminate
S
~
II(c)
= S
~
II
_ _
1
N
k=1
n
k
r
k
A
~
k
_ _
T
N
i=1
b
~
ki
A
~
k
_ _
1
;
S
~
IO(c)
=S
~
II(c)
S
~
II
_ _
1
S
~
IO
N
k=1
n
k
r
k
A
~
k
_ _
T
N
i=1
b
~
ki
B
~
i
_ _
1
;
S
~
OO(c)
= S
~
IO(c)
_ _
T
S
~
II(c)
_ _
1
S
~
IO(c)
S
~
IO
_ _
T
S
~
II
_ _
1
S
~
IO
S
~
OO
N
k=1
n
k
r
k
B
k
~
_ _
T
N
i=1
b
~
ki
B
~
i
;
a
~
I(c)
=a
~
I
S
~
II(c)
N
k=1
n
k
r
k
A
~
k
_ _
T
N
i=1
b
~
ki
C
~
i
;
a
~
O(c)
=a
~
O
S
~
IO(c)
_ _
T
S
~
II(c)
_ _
1
a
~
I(c)
a
~
I
_ _
N
k=1
n
k
r
k
B
~
k
_ _
T
N
i=1
b
~
ki
C
~
i
;
(4:229)
where matrices S
~
II
; S
~
IO
; S
~
OO
; S
~
II
; a
~
I
; and a
~
O
are given in Eqs. (4.215) and (4.217),
b
~
is the matrix containing average crack opening displacements (COD), and
matrices A
~
; B
~
; C
~
are given by
A
~
k
= N
~
I
k
S
~
II
k
_ _
1
; B
~
k
= N
~
O
k
N
~
I
k
S
~
k
II
)
1
S
~
k
IO
; C
~
k
=
A
~
k
a
~
I
a
~
k
_ _
;
_
(4:230)
where the matrices N
~
k
I
and N
~
k
O
represent the unit normal vector on the crack
surfaces in ply k, i.e.,
N
~
I
k
=
n
k
1
0 n
k
2
0 n
k
2
n
k
1
0 0 0
_
_
_
_
N
~
k
O
=
0 0 0
0 0 0
0 n
k
1
n
k
2
_
_
_
_
: (4:231)
The expressions in Eq. (4.229) for stiffness of the cracked laminates are exact
assuming the homogenization procedure. However, the main problem is the
determination of average COD for the cracked laminate. The heterogeneity in
the composite laminates and constraint effects on crack surfaces from the sur-
rounding uncracked plies make it impossible to derive exact analytical solutions.
Gudmundson and coworkers, however, made the following assumptions in order
to evaluate the average crack opening displacements:
1. The surface displacements of a ply crack in a nite-thickness laminate are
equal to those of a crack in an innite, homogeneous transversely isotropic
medium. The stress intensity factors for an innite row of equidistant cracks
in an innite homogeneous isotropic medium under the action of uniform
tractions on crack surfaces, given by Benthem and Koiter [104] and Tada et al.
[105], are assumed to hold for the current case of transversely isotropic
medium.
2. There is no effect of orientation of a cracked ply. This means that the matrices b
~
ki
in Eq. (4.229) may not be accurate for off-axis cracks in laminates.
3. The crack density is low (r 1).
4. There is no coupling between crack opening displacements of different plies.
Hence the matrices b
~
ki
are diagonal.
With these assumptions, the matrices b
~
ki
are given by
b
~
ki
=
0
~
; for all k ,= i;
or b
~
kk
=
b
k
1
0 0
0 b
k
2
0
0 0 b
k
3
_
_
_
_
;
(4:232)
where b
k
1
; b
k
2
; andb
k
3
are determined by Gudmundson using a numerical integra-
tion and are given by
b
k
1
=
4
p
g
1
ln cosh
pr
k
2
_ _ _ _
r
k
( )
2
;
b
k
2
=
p
2
g
2
10
j=1
a
j
1 r
k
( )
j
;
b
k
3
=
p
2
g
3
9
j=1
b
j
1 r
k
( )
j2
;
(4:233)
for cracks in an internal ply, where
g
1
=
1
2G
12
; g
2
= g
3
=
1 u
12
u
21
E
2
; (4:234)
and a
j
and b
j
are numerical parameters given in Table 4.2. For surface cracks (or
cracks in external plies), we have
b
k(s)
1
=
8
p
g
1
ln cosh pr
k
_ _ _
2r
k
( )
2
;
b
k(s)
2
= 2 1:12 ( )
2
p
2
g
2
10
j=1
c
j
1 r
k
( )
j
_ _
;
(4:235)
Table 4.2 Numerical parameters used in calculation of COD matrix coefcients b
k
1
; b
k
2
; and b
k
3
j a b c
1 0.63666 0.63666 0.25256
2 0.51806 0.08945 0.27079
3 0.51695 0.15653 0.49814
4 1.04897 0.13964 8.62962
5 8.95572 0.16463 51.2466
6 33.0944 0.06661 180.9631
7 74.32002 0.54819 374.298
8 103.064 1.07983 449.5947
9 73.60337 0.45704 286.51
10 20.3433 73.84223
where c
j
are another set of numerical parameters also given in Table 4.2.
It is clear despite the assumptions used in COD calculation, the above approach
is quite complex and difcult to implement in a numerical scheme. Similar
relations are also available for the combined extension and bending loading
scenario [94, 95].
4.8.2 LundmarkVarna model
Lundmark and Varna [98, 99, 106, 107] have recently derived effective properties
using a homogenization very similar to that in Gudmundsons approach. How-
ever, their relationships are simpler and they use the crack surface displacements
numerically obtained from FEM simulations, which makes their model much
more accurate. However, their model has been checked against experimental
data only for cross-ply laminates. Here, we provide the main relations derived
by them. Accordingly, the average strains in a ply, analogous to (4.228), are
dened as
e
ij
_ _
a
k
= e
ij
_ _
LAM

b
ij
_ _
k
; (4:236)
where the superscript a means average and
e
ij
_ _
a
k
=
e
11
e
22
2e
12
_
_
_
_
_
_
a
k
; e
ij
_ _
LAM
=
e
11
e
22
2e
12
_
_
_
_
_
_
LAM
;

b
ij
_ _
k
=
b
11
b
22
2
b
12
_
_
_
_
_
_
k
; (4:237)
where

b
ij
_ _
k
represents the VakulenkoKachanov tensor, dened by
b
ij
_ _
k
=
1
2V
k
_
kc
u
k
i
n
k
j
u
k
j
n
k
i
_ _
d; (4:238)
which is same as e
k
ij
in Eq. (4.226). In terms of crack surface displacements,

b
ij
_ _
k
for a cracked ply is derived as
b
_ _
k
=
r
kn
E
2
A [ [
k
U [ [
k
A [ [
k
Q
_
k
e
0

LAM
a
0

k
T
_ _
; (4:239)
where [A]
k
is the transformation matrix for ply k and [U]
k
is the displacement
matrix given by
U [ [
k
= 2 =
0 0 0
0 u
k
2an
0
0 0
E
2
G
12
u
k
1an
_
_
_
_
; (4:240)
where u
k
1an
and u
k
2an
are the normalized average crack face sliding and opening
displacements, respectively, and are given by
u
k
1an
= u
k
1a
G
12
t
k
s
k
120
; u
k
2an
= u
k
2a
E
2
t
k
s
k
20
; (4:241)
with u
k
1a
and u
k
2a
being the average crack surface displacements, dened as
u
k
1a
=
1
2t
k
_
t
k
2
t
k
2
u
1
z ( ) dz; u
k
2a
=
1
2t
k
_
t
k
2
t
k
2
u
2
z ( ) dz; (4:242)
where Du
1
and Du
2
are the relative separation of the two crack faces along and
normal to the crack surface. Now, the average stressstrain relationships for the
kth ply in the global coordinate system are
s
a
k
=

Q
_
k
e
a
k
a
k
T
_
: (4:243)
Since, the average stresses for the laminate remain equal to the applied stresses,
we have
s
LAM
= s
a
=
N
k=1
s
a
k
t
k
H
; (4:244)
where H is the total thickness of the laminate. Substituting Eqs. (4.236) and (4.239)
into Eq. (4.244), we obtain
s
LAM
= Q
0
[ [
LAM
e
LAM
e
0

LAM
th
_ _
1
H
N
k=1
t
k

Q
_
k
b
_ _
k
; (4:245)
where [Q
0
]
LAM
is the stiffness matrix for the undamaged laminate, and
e
0

LAM
th
= 1=H ( )
N
k=1
t
k
a
k
T are the thermal strains in the undamaged
laminate. Substituting Eq. (4.239) into Eq. (4.245), the average thermomechanical
response of the damaged laminate is given by
s
LAM
= Q
0
[ [
LAM
e
LAM
e
0

LAM
th
_ _
1
HE
2
N
k=1
r
kn
t
k

Q
_
k
A [ [
k
U [ [
k
A [ [
k
Q
_
k
e
0

LAM
a
0

k
T
_ _
:
(4:246)
For purely mechanical response, the stiffness matrix of the damaged laminate is
given by
Q [ [
LAM
= I [ [
1
HE
2
N
k=1
r
kn
t
k

Q
_
k
A [ [
k
U [ [
k
A [ [
k
Q
_
k
S
0
[ [
LAM
_ _
1
Q
0
[ [
LAM
; (4:247)
where [S
0
]
LAM
= ([Q
0
]
LAM
)
1
is the compliance tensor for the undamaged
laminate.
The only remaining unknown in this model is [U]
k
, Eq. (4.240). To determine the
crack surface displacements, Lundmark and Varna [98] suggest using actual FE
calculations on a unit cell of cracked laminate. For cross-ply laminates, they
carried out a set of such FE calculations for varying values of ply thickness and
stiffness, and by curve tting they provided the following power law-type expres-
sions for the normalized average crack opening displacements
u
2an
= A B
E
2
E
s
x
_ _
p
; (4:248)
where A, B, and p are constants that depend on the thickness ratio of cracked to
uncracked plies, and the type of crack (in an internal or external ply). For
example, for an internal crack in a GFRP [S
n
/90
m
]
s
laminate,
A = 0:52; B = 0:3075 0:1652
t
90
2t
s
2t
s
_ _
;
p = 0:0307
t
90
2t
s
_ _
2
0:0626
t
90
2t
s
_ _
0:7037;
(4:249)
where t
90
and t
s
are the thicknesses of cracked 90
and each uncracked sublaminate

layer. Figure 4.21 shows comparison of the model with experimental data for a
[30/90
4
]
s
laminate with cracks in the central 90
layer.
4.9 Computational methods
For simulating the effect of damage in composite laminates, in general,
many computational methods have been utilized, such as the nite element
method (FEM), the nite difference method (FDM), and the boundary element
method (BEM). For the particular problem of transverse cracking, the most
common numerical tool is FEM. Some simpler numerical tools have also been
devised, e.g., the nite strip method by Li et al. [108], and the layerwise theory of
Reddy [109]. In this subsection, we will briey describe some of these
developments.
0
Crack density (cracks/mm)
[30/30/90
4
]
s
+ experimental
model
0.2 0.4 0.6
0.6
0.7
0.8
0.9
E
x
/
E
x
0
1.0
(a) (b)
0
Crack density (cracks/mm)
[30/30/90
4
]
s
+ experimental
model
0.2 0.4 0.6
0.6
0.7
0.8
0.9
1.0
n
x
y
/
n
x
y
0
Figure 4.21. Reductions in longitudinal modulus (a) and Poissons ratio (b) for a graphite/
epoxy [30/90
4
]
s
laminate using the LundmarkVarna model [98]. Reprinted, with kind
permission, from P. Lundmark and J. Varna, Int J Damage Mech, Vol. 14, pp. 23559,
copyright # 2005 by Sage Publications.
119 4.9 Computational methods
4.9.1 Finite element method (FEM)
FEM is the most widely used numerical approach to analyze damage in composite
laminates. Its advantages are that it can model highly complex situations, and
provides quite accurate results. As can be seen from the preceding discussion, the
analytical models of cracking are severely limited in scope with respect to laminate
layup and loading scenario. Nonetheless, all numerical methods have a unique
limitation: every time the laminate geometry, loading, or material changes, simu-
lations need to be carried out afresh, which might involve new mesh generation,
and computation. Hence, FE modeling of cracked laminates is time consuming
and does not by itself provide much insight into the damage mechanisms. Even
with these limitations it can be successfully used for calibration/verication of
analytical models as well as for computation of parameters or constants useful
in analytical methods (for example, it can be used to evaluate constants used in
Talrejas continuum damage model, see the next chapter for details). FE modeling
can also be utilized to simulate complex experimental situations, a task equivalent
to carrying out numerical experiments, and eliminating the need for cumber-
some experimentation whenever possible.
The rst task in FE modeling is to dene a geometrical model of the cracked
laminate. Mostly, a representative unit cell is developed assuming periodic array
of self-similar cracks (see Figure 4.7(a)). For cross-ply laminates, a 3-D repeating
unit can be reduced to a 2-D plane stress/strain model (see Figure 4.7(b)).
Furthermore, the symmetry of the resulting boundary value problem can
reduce the size of the modeled unit cell. For example, cracking in a [0/90]
s
laminate can be modeled using a quarter of a representative unit cell as illus-
trated in Figure 4.22.
For multidirectional laminates with cracks in more than one orientation, how-
ever, a repeating unit cannot be dened uniquely because of differences in direc-
tionality and mutual spacing of cracks in different plies. A repeating unit cell can
be dened for up to two off-axis cracking modes with cracked surfaces represented
on nonorthogonal boundaries [110]. The RVE in such cases is therefore 3-D and
skewed (nonorthogonal in the xy plane; z being the thickness direction).
A reduction to two dimensions is often not possible in such cases.
As is common in micromechanics, the displacement, strain, and stress elds
must be periodic across repeating unit cells. Thus the following periodic boundary
conditions [111] are applied on the FE model
u
i
x
a
x
a
( ) = u
i
x
a
( ) x
b
@u
i
@x
b
_ _
;
e
ij
x
a
x
a
( ) = e
ij
x
a
( );
s
ij
x
a
x
a
( ) = s
ij
x
a
( );
(4:250)
where u
i
, @u
i
/@u
b
), and Dx
b
, i, b = 1,2,3 represent the displacements, the volume
averaged displacement gradients, and the vector of periodicity, respectively. For a
skewed RVE, periodic BC might be complicated to enforce. For a detailed
discussion about skewed RVE, periodic BCs, and symmetries in composites, the
reader is referred to [110, 112114].
Once the FE results are available, the overall stiffness properties of the damaged
laminate can be calculated by volume averaging of stresses and strains inside the
RVE. For example, for a cracked laminate undergoing in-plane loading, the in-
plane elasticity properties can be calculated by
E
x
=
s
xx
)
e
xx
)
; E
y
=
s
yy
_
e
yy
_ ; G
xy
=
s
xy
_
g
xy
_ ; n
xy
=
e
yy
_
e
xx
)
; (4:251)
where <g> denotes the volume average of a eld g. Important numerical
studies related to analysis of cracked composite laminates can be found in [110
113, 115, 116].
4.9.2 Finite strip method
To enable predictions for laminated composites with arbitrary layups [110,
117, 118], Li et al. [108] devised an approximate numerical approach based on
the generalized plain strain formulation. Although there can be more than
one cracked ply, all the cracked plies have to be of the same orientation.
Hence, this method cannot be used for laminates with cracks in multiple
orientations.
Consider rst an uncracked laminate. Given the generalized strains {e},
e = e
xx0
; e
yy0
; 2e
xy0
; k
xx
; k
yy
; 2k
xy
_
T
; (4:252)
the displacement eld based on the classical laminate theory without rigid body
displacements can be written as
u
0
= xe
xx0
ye
xy0
xzk
xx
yzk
xy
;
v
0
= ye
yy0
xe
xy0
yzk
yy
xzk
xy
;
w
0
=
x
2
2
k
xx

y
2
2
k
yy
xyk
xy
o z ( );
(4:253)
t
0
t
90
h
x
z
2l
90
0
0
(b)
(a)
s
c
s
c
Figure 4.22. Building FE model for a cracked cross-ply laminate: (a) 2-D representative unit
cell; (b) FE model (1/4 cell).
where o is an arbitrary integration function which depends on boundary
conditions.
When transverse cracks appear, perturbations to this displacement eld are
induced. These perturbations can simply be superimposed on the displacements due
to the linearity of the problem. Assuming sufciently long cracks, the perturbations
are independent of y. Hence, the displacement eld for a cracked laminate reads
u = u
0
U x; z ( );
v = v
0
V(x; z);
w = w
0
W(x; z);
(4:254)
where U, V, and W denote changes due to the presence of cracks. The resulting
strains are calculated using innitesimal deformation theory. Our objective then is
to solve for U, V, and W. The approach in the nite strip method is to divide the
planar region of a typical segment of the cracked laminate into a nite number of
strip elements parallel to the x-axis. In each element, M nodal lines are introduced
along which displacements are functions of x only and the displacements in the
strip elements are then interpolated by polynomials in the z-direction. For a
typical element, the displacement eld can be expressed as
U
V
o W
_
_
_
_
_
_
e
=
3
i=1
N
i
z ( )
U
i
V
i
W
i
_
_
_
_
_
_
e
; (4:255)
where the superscript e denotes the element number, and N
i
are shape functions,
the same as those used in the one-dimensional nite element analysis [119], which
are given by
N
1
=
1
2
z 1 z ( ); N
2
= 1 z
2
; N
1
=
1
2
z 1 z ( ); (4:256)
with 1 _ z _ 1 being the nondimensional coordinate along the z-axis
z =
3
i=1
N
i
z ( )z
i
; (4:257)
where z
i
is the z-coordinate of the ith nodal line in the element. Going through the
usual FEM formulation, Li et al. arrived at the following set of differential
equations for the resulting variational problem
K
2
[ [

y
_ _
K
1
[ [ K
1
[ [
T
_ _
_
y
_ _
K
0
[ [ y = F
0
; (4:258)
and a set of boundary conditions, where [K
2
], [K
1
], and [K
0
] are the matrices
dened in Li et al. [108] and
y =
U
V
W
_
_
_
_
_
_
: (4:259)
No direct solution to the differential Eq. (4.258) exists. An approximate solution
can be obtained by taking the nodal displacements as a Fourier series of unknown
constants, as
U
n
= U
h
n
x
l

K
n
k=1
U
k
n
sin
kpx
l
;
V
n
= V
h
n
x
l

K
n
k=1
V
k
n
sin
kpx
l
;
W
n
= W
h
n
x
l
_ _
2
W
0
n

K
n
k=1
U
k
n
cos
kpx
l
;
(4:260)
for n = 1,2, . . ., M, and the coefcients U
h
n
; V
h
n
; W
h
n
; U
k
n
; V
k
n
; V
k
n
; and W
0
n
are
unknown constants to be determined, K
n
is the order where the Fourier series is
truncated for the approximate solution, and M is the number of nodal lines in the
laminate. Substituting Eq. (4.260) into Eq. (4.258), and rearranging, the following
algebraic equations similar to FEM are obtained
A [ [ = : (4:261)
The solution of these algebraic equations along with the appropriate boundary
conditions gives the displacement eld, from which the strain eld can be
obtained.
4.9.3 Layerwise theory
The layerwise theory of Reddy [109] is motivated by a desire to develop a
computational model that is more efcient than the conventional 3-D nite
element models [119] and can incorporate damage effects such as transverse cracks
and delaminations in a layered medium. It is based on 3-D kinematics where the
displacement eld within each layer is expanded using the Lagrange family of
nite elements. For a comparable mesh, the layerwise theory typically takes less
computational time as compared to the conventional 3-D FEM while providing
the same level of accuracy.
In the theory, the whole laminate is divided into a number of subdivisions
across its thickness. The displacement eld in the laminate is written as
u
i
x; y; z ( ) =
N
J=1
U
J
i
x; y ( )
J
z ( ); i = 1; 2; 3; (4:262)
where N is the number of subdivisions through the thickness of the laminate and F
J
are global interpolation functions dened in terms of the Lagrange interpolation
functions associated with the layers connected to the Jth interface through the
laminate thickness, and U
J
i
are the nodal displacements. Independent interpolation
functions for u
1
, u
2
, and u
3
can also be used whenever necessary. The strain eld is
determined using the von Karman nonlinear theory. Then the governing equations
for the nodal displacements are derived using the principle of virtual displacements
[120, 121]. The resulting system of partial differential equations can be converted to
a systemof equations analogous to the FEMin a procedure similar to that followed
in previous subsection. Solution of this system gives us the nodal displacement,
from which strain and stress elds can be determined. Further details on the
approach and its implementation can be found in [122127]. Recent papers by Na
and Reddy [128, 129] provide direct implementation of the layerwise theory for
transverse cracking and delaminations, respectively, in cross-ply laminates.
4.10 Other methods
There have been other developments in analyzing cracked laminates. The detailed
treatment is not covered here and we will highlight the important aspects of these
approaches. The readers are referred to the cited articles for further details. It
should be pointed out, however, that these approaches are mostly extensions of
the previously developed ideas, and do not really improve the predictions. More-
over, many of these approaches are complex, semi-analytical and thus difcult to
implement, and may sometimes need high computational times.
Close to the development of the variational analysis by Hashin, Aboudi and
coworkers [130, 131] developed a three-dimensional semi-analytical method. In
this approach, an approximate analytical solution for the displacement eld is
sought using a series expansion in the form of Legendre polynomials.
In a somewhat similar way, Lee and Hong [132] developed a renement of
the shear lag approach using a series polynomial expansion. The approach
accounted for the crack opening displacement, thermal stresses, and Poissons
effects. However, the improvement over the traditional shear lag methods was
insignicant.
Another similar effort by Gamby and Rebiere [133] used the transverse shear
stresses in the 0
- and 90
-layers as double Fourier series in x and z (longitudinal

and transverse coordinates). Vanishing tractions at the crack surfaces were used as
boundary conditions. To obtain the stresses and strains in the cracked laminate,
the use of constitutive equations and the equilibrium equations was made. Finally,
the coefcients of the Fourier series were derived using the principle of minimum
complementary energy.
Quite recently, Zhang et al. [134, 135] have developed a state space method to
analyze cracked off-axis laminates. In this approach they use displacement elds
and the stresses to be given by the Fourier series expansions. The coefcient terms
in expansions are obtained numerically using the equilibrium equations, boundary
conditions, and ply renement.
Following on the lines of modeling of thin shells, Shoeppner and Pagano [136
138] developed a method to model the thermoelastic response of at laminated
composites. This approach employs the idea that the stress eld in an axisym-
metric cylinder approaches that in a long at coupon as the radius to thickness
ratio approaches innity. Thus, the stress eld in a at laminate is nearly the same
as in a large radius axisymmetric hollow layered cylinder model. To obtain the
stress elds, they used Reissners variational principle along with the equilibrium
equations. The predictions matched quite well with FE simulations. However, the
governing equations are very complex, need numerical tools to solve them,
and this limits the usability of the approach. The models of McCartney and
ShoeppnerPagano are compared in [139].
4.11 Changes in thermal expansion coefcients
Hashin [74] and Nairn [140] have developed variational models to predict changes
in thermal expansion coefcients due to ply cracking. They can be derived in a
relatively straightforward way following Hashins treatment (refer to [74] for
details). The longitudinal expansion coefcient for a cracked cross-ply laminate
is given by
a
90
xx
= a
c

a
11
a
22
1 l
1
B
0
C
0
_ _
k
90
x
f; (4:263)
where a
c
is the longitudinal thermal expansion coefcient for an uncracked
composite laminate, and
f =
1
2l
_
l
l
f x ( ) dx;
B
0
= 1 l ( )
n
12
E
1
;
(4:264)
Lundmark and Varna [98] also developed a model for thermal expansion coef-
cients. Following the analysis covered in Section 4.8.2 and noting that only
thermal loading is present, the strain eld in the laminate is given by (setting
mechanical loads to zero in Eq. (4.245))
e
LAM
= e
0

LAM
th
S
0
[ [
LAM
1
H
N
k=1
t
k

Q
_
k
b
_ _
k
: (4:265)
Substituting

b
_ _
k
from Eq. (4.239) and dividing by DT yields
a
LAM
= I [ [
N
k=1
t
k
r
kn
D [ [
k
_ _
a
LAM
0

1
H
N
k=1
t
k
r
kn
D [ [
k
a
k
; (4:266)
where
D [ [
k
= S
0
[ [
LAM
1
E
2
Q
_
k
T [ [
T
k
U [ [
k
T [ [
k
Q
_
k
: (4:267)
The model predictions for longitudinal thermal expansion coefcient are plotted
in Figure 4.23 for a [0/90]
s
carbon/epoxy laminate. For comparison, experimental
and model data from Kim et al. [141] are also shown.
125 4.11 Changes in thermal expansion coefcients
4.12 Summary
This chapter has provided an exposition of the main concepts and methods related to
evaluating the effects of multiple cracking in composite materials on their deform-
ational response. Beginning with the early approaches, known as shear lag methods,
where one-dimensional stress analysis is used, and progressing to computational
methods that accurately determine the local stress elds, the range is covered as much
as possible. While the eld is still evolving and more methods are appearing in the
literature, it is hoped that the treatments and discussions provided here are useful for
researchers in the eld to gain an appreciation of this class of approaches. A key
consideration in selecting an approach is the purpose at hand. For material selection
purposes a quick assessment of the elastic modulus may be needed, in which case the
shear lag approach may be adequate, while for structural analysis purposes an
evaluation of all properties and the suitability of the approach as an integral part
of a structural analysis scheme would be the factors of consideration. In any case, the
micro-damage mechanics (MIDM) treated here is one part of the total damage
mechanics picture. The next chapter on macro-damage mechanics (MADM) pro-
vides another perspective on the complex problemof damage in composite materials.
References
1. Z. Hashin, Analysis of damage in composite materials. In Yielding, Damage, and Failure
of Anisotropic Solids, ed. J. P. Boehler. (London: Mechanical Engineering Publications,
1990), pp. 331.
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Crack density (cracks / mm)
0.2
0.4
0.6
0.8
1.0
a
x
/
a
x
0
[0/90]
s
experimental, Kim et al. (2000)
model
model, Kim et al. (2000)
Figure 4.23. Variation in longitudinal thermal expansion coefcient for a carbon/epoxy
[0
2
/90
2
]
s
laminate. The LundmarkVarna model is compared with model and
experiment from Kim et al. [141]. Reprinted, with kind permission, from P. Lundmark and
J. Varna, Int J Damage Mech, Vol. 14, pp. 23559, copyright #2005 by Sage Publications.
ical Laboratory, 1971), pp. 1526.
4. A. Kelly, The 1995 Bakerian Lecture: composite materials. Phil Trans R Soc London
A, 354:1714 (1996), 184174.
5. G. A. Cooper and J. M. Sillwood, Multiple fracture in a steel reinforced epoxy resin
composite. J Mater Sci, 7:3 (1972), 32533.
8:3 (1973), 35262.
7. S. W. Wang and A. Parvizi-Majidi, Experimental characterization of the tensile behav-
iour of Nicalon bre-reinforced calcium aluminosilicate composites. J Mater Sci, 27:20
(1992), 548396.
8. J. E. Masters and K. L. Reifsnider, An investigation of cumulative damage develop-
9. J. Tong, F. J. Guild, S. L. Ogin, andP. A. Smith, Onmatrix crackgrowthinquasi-isotropic
laminates I. Experimental investigation. Compos Sci Technol, 57:11 (1997), 152735.
10. D. T. G. Katerelos, P. Lundmark, J. Varna, and C. Galiotis, Analysis of matrix
cracking in GFRP laminates using Raman spectroscopy. Compos Sci Technol, 67:9
(2007), 194654.
11. T. Yokozeki, T. Aoki, and T. Ishikawa, Consecutive matrix cracking in contiguous
plies of composite laminates. Int J Solids Struct, 42:910 (2005), 2785802.
12. 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.
Compos A, 36:9 (2005), 122935.
13. M. Kashtalyan and C. Soutis, Stiffness and fracture analysis of laminated composites
with off-axis ply matrix cracking. Compos A, 38:4 (2007), 12629.
14. M. Kashtalyan and C. Soutis, Modelling off-axis ply matrix cracking in
continuous bre-reinforced polymer matrix composite laminates. J Mater Sci, 41:20
(2006), 678999.
15. L. N. McCartney and G. A. Schoeppner, Predicting the effect of non-uniform ply
cracking on the thermoelastic properties of cross-ply laminates. Compos Sci Technol,
62:14 (2002), 184156.
16. V. V. Silberschmidt, Effect of micro-randomness on macroscopic properties and frac-
ture of laminates. J Mater Sci, 41:20 (2006), 676876.
17. V. V. Silberschmidt, Matrix cracking in cross-ply laminates: effect of randomness.
Compos A, 36 (2005), 12935.
18. L. E. Asp, L. A. Berglund, and R. Talreja, A criterion for crack initiation in glassy
polymers subjected to a composite-like stress state. Compos Sci Technol, 56:11 (1996),
1291301.
19. L. E. Asp, L. A. Berglund, andR. Talreja, Effects of ber andinterphase onmatrix-initiated
transverse failure in polymer composites. Compos Sci Technol, 56:6 (1996), 65765.
20. L. E. Asp, L. A. Berglund, and R. Talreja, Prediction of matrix-initiated transverse
failure in polymer composites. Compos Sci Technol, 56:9 (1996), 108997.
21. H. S. Huang and R. Talreja, Numerical simulation of matrix micro-cracking in short
ber reinforced polymer composites: initiation and propagation. Compos Sci Technol,
66:15 (2006), 274357.
127 References
22. H. Huang and R. Talreja, Effects of void geometry on elastic properties of
unidirectional ber reinforced composites. Compos Sci Technol, 65:13 (2005),
196481.
23. K. A. Chowdhury, R. Talreja, and A. A. Benzerga, Effects of manufacturing-induced
voids on local failure in polymer-based composites. J Eng Mater Tech, Trans ASME,
130:2 (2008), 021010.
24. K. A. Chowdhury, Damage initiation, progression and failure of polymer based
composites due to manufacturing induced defects. Ph.D. thesis, Texas A&M Univer-
sity, College Station, TX (2007).
25. H. L. Cox, The elasticity and strength of paper and other brous materials. Br J Appl
Phys, 3 (1952), 729.
26. J. A. Nairn, On the use of shear-lag methods for analysis of stress transfer unidirec-
tional composites. Mech Mater, 26:2 (1997), 6380.
laminates of a glass ber-reinforced polyester. J Mater Sci, 12:1 (1977), 15768.
28. A. Parvizi, K. W. Garrett, and J. E. Bailey, Constrained cracking in glass ber-
29. P. W. Manders, T. W. Chou, F. R. Jones, and J. W. Rock, Statistical analysis of
multiple fracture in 0
/90
/0
glass bre/epoxy resin laminates. J Mater Sci, 18:10

(1983), 287689.
31. J. W. Lee and I. M. Daniel, Progressive transverse cracking of crossply composite
32. D. L. Flaggs, Prediction of tensile matrix failure in composite laminates. J Compos
Mater, 19:1 (1985), 2950.
33. S. C. Tan and R. J. Nuismer, A theory for progressive matrix cracking in composite
34. Y. M. HanandH. T. Hahn, Ply cracking and property degradations of symmetric balanced
laminates under general in-plane loading. Compos Sci Technol, 35:4 (1989), 37797.
35. Y. M. Han, H. T. Hahn, and R. B. Croman, A simplied analysis of transverse ply
cracking in cross-ply laminates. Compos Sci Technol, 31:3 (1988), 16577.
36. N. Laws and G. J. Dvorak, Progressive transverse cracking in composite laminates.
J Compos Mater, 22:10 (1988), 90016.
37. H. L. McManus and J. R. Maddocks, On microcracking in composite laminates
under thermal and mechanical loading. Polymers & Polymer Composites, 4:5 (1996),
30514.
38. S. G. Lim and C. S. Hong, Prediction of transverse cracking and stiffness reduction in
cross-ply laminate composites. J Compos Mater, 23 (1989), 695713.
39. M. Caslini, C. Zanotti, and T. K. OBrien, Fracture mechanics of matrix cracking and
delamination in glass/epoxy laminates. J Compos Tech Res, 9:4 (1987), 12130.
40. J. A. Nairn and D. A. Mendels, On the use of planar shear-lag methods for stress-
transfer analysis of multilayered composites. Mech Mater, 33:6 (2001), 33562.
41. J. M. Berthelot, Transverse cracking and delamination in cross-ply glass-ber and
carbon-ber reinforced plastic laminates: static and fatigue loading. Appl Mech Rev,
56:1 (2003), 11147.
42. J. A. Nairn and S. Hu, Micromechanics of damage: A case study of matrix micro-
cracking. In Damage Mechanics of Composite Materials, ed. R. Talreja. (Amsterdam:
Elsevier, 1994), pp. 187243.
44. G. J. Dvorak, N. Laws, and M. Heiazi, Analysis of progressive matrix cracking in
composite laminates I. Thermoelastic properties of a ply with cracks. J Compos Mater,
19 (1985), 21634.
45. H. Fukunaga, T.-W. Chou, P. W. M. Peters, and K. Schulte, Probabilistic failure strength
analysis of graphite/epoxy cross-ply laminates. J Compos Mater, 18:4 (1984), 33956.
46. P. S. Steif, Parabolic shear lag analysis of a [0/90]s laminate. Transverse ply crack growth
and associated stiffness reduction during the fatigue of a simple cross-ply laminate. In S.
L. Ogin, P. A. Smith, and P. W. R. Beaumont (eds.), Report CUED/C/MATS/TR 105,
Cambridge University, Engineering Department, UK (September 1984).
47. S. L. Ogin, P. A. Smith, and P. W. R. Beaumont, Matrix cracking and stiffness
reduction during the fatigue of [0/90]s GFRP laminate. Compos Sci Technol, 22:1
(1985), 2331.
48. S. L. Ogin, P. A. Smith, and P. W. R. Beaumont, Stress intensity factor approach to the
fatigue growth of transverse ply cracks. Compos Sci Technol, 24:1 (1985), 4759.
49. R. J. Nuismer and S. C. Tan, Constitutive relations of a cracked composite lamina.
J Compos Mater, 22:4 (1988), 30621.
50. J. Fan and J. Zhang, In-situ damage evolution and micro/macro transition for lamin-
ated composites. Compos Sci Technol, 47:2 (1993), 10718.
51. M. Kashtalyan and C. Soutis, Stiffness degradation in cross-ply laminates damaged by
transverse cracking and splitting. Compos A, 31:4 (2000), 33551.
52. J. Q. Zhang, K. P. Herrmann, and J. H. Fan, A theoretical model of matrix cracking in
composite laminates under thermomechanical loading. Acta Mech Solida Sin, 14:4
(2001), 299305.
53. M. Kashtalyan and C. Soutis, Analysis of composite laminates with intra- and inter-
laminar damage. Prog Aerosp Sci, 41:2 (2005), 15273.
54. J. M. Berthelot, Analysis of the transverse cracking of cross-ply laminates: a general-
ized approach. J Compos Mater, 31:18 (1997), 1780805.
55. J. M. Berthelot, P. Leblond, A. El Mahi, and J. F. Le Corre, Transverse cracking of
cross-ply laminates: part 1. Analysis. Compos A, 27:10 (1996), 9891001.
56. J. W. Lee, I. M. Daniel, and G. Yaniv, Fatigue life prediction of cross-ply composite
laminates. Compos Mater: Fatigue and Fracture, 2 (1989), 1928.
57. J. A. Nairn and S. F. Hu, The formation and effect of outer-ply microcracks in cross-
ply laminates a variational approach. Eng Fract Mech, 41:2 (1992), 20321.
multiple fracture in [0/90/0] glass ber/epoxy resin laminates. J Mater Sci, 18:10 (1983),
287689.
59. N. Laws and G. J. Dvorak, The loss of stiffness of cracked laminates. In Proceedings of the
IUTAM Eshelby Memorial Symposium. (Cambridge: Cambridge University Press, 1985).
60. N. Laws, G. J. Dvorak, and M. Hejazi, Stiffness changes in unidirectional composites
caused by crack systems. Mech Mater, 2:2 (1983), 12337.
61. R. Hill, A self-consistent mechanics of composite materials. J Mech Phys Solids, 13:4
(1965), 21322.
129 References
62. B. Budiansky, On elastic moduli of some heterogeneous materials. J Mech Phys Solids,
13:4 (1965), 2237.
63. J. D. Eshelby, The determination of the elastic eld of an ellipsoidal inclusion, and
related problems. Proc R Soc London A, 241:1226 (1957), 37696.
64. J. D. Eshelby, Elastic inclusion and inhomogeneities. In Progress in Solid Mechanics.
(Amsterdam: North Holland, 1961), pp. 89140.
65. K. Hoiseth, A micromechanics study of transverse matrix cracking in cross-ply
composites. M.Sc. thesis, Georgia Institute of Technology, Atlanta, GA (1995).
66. J. Qu and K. Hoiseth, Evolution of transverse matrix cracking in cross-ply laminates.
Fatigue Fract Eng Mater Struc, 21:4 (1998), 45164.
67. Z. Hashin, Analysis of cracked laminates: A variational approach. Mech Mater, 4:2
(1985), 12136.
68. Z. Hashin, Analysis of composite materials a survey. J Appl Mech, Trans ASME, 50:3
(1983), 481505.
69. J. A. Nairn, The strain energy release rate of composite microcracking: A variational
splitting behavior of glass and carbon-ber reinforced epoxy cross-ply laminates and
(1979), 599623.
72. J. Varna and L. Berglund, Two-dimensional transverse cracking in [0(m)/90(n)](s)
cross-ply laminates. Eur J Mech A Solids, 12:5 (1993), 699723.
73. J. Varna and L. A. Berglund, Thermoelastic properties of composite laminates with
transverse cracks. J Compos Tech Res, 16:1 (1994), 7787.
74. Z. Hashin, Thermal-expansion coefcients of cracked laminates. Compos Sci Technol,
31:4 (1988), 24760.
75. Z. Hashin, Analysis of orthogonally cracked laminates under tension. J Appl Mech,
Trans ASME, 54:4 (1987), 8729.
76. S. Kuriakose and R. Talreja, Variational solutions to stresses in cracked cross-ply
laminates under bending. Int J Solids Struct, 41:910 (2004), 233147.
77. J. Varna and A. Krasnikovs, Transverse cracks in cross-ply laminates 2. Stiffness
degradation. Mech Compos Mater, 34:2 (1998), 15370.
78. J. Varna, L. Berglund, A. Krasnikovs, and A. Chihalenko, Crack opening geometry in
cracked composite laminates. Int J Damage Mech, 6 (1997), 96118.
79. R. Joffe andJ. Varna, Analytical modeling of stiffness reductioninsymmetric andbalanced
laminates due to cracks in 90 degrees layers. Compos Sci Technol, 59:11 (1999), 164152.
80. L. N. McCartney, Theory of stress transfer in a 0-degrees-90-degrees-0-degrees
cross-ply laminate containing a parallel array of transverse cracks. J Mech Phys Solids,
40:1 (1992), 2768.
81. L. N. McCartney, Energy-based prediction of failure in general symmetric laminates.
Eng Fract Mech, 72:6 (2005), 90930.
82. L. N. McCartney, Physically based damage models for laminated composites. Proc Inst
Mech Engineers L J Mater Design App, 217:L3 (2003), 16399.
83. L. N. McCartney, Prediction of ply crack formation and failure in laminates. Compos
Sci Technol, 62:1213 (2002), 161931.
84. L. N. McCartney, Model to predict effects of triaxial loading on ply cracking in
general symmetric laminates. Compos Sci Technol, 60:1213 (2000), 225579.
85. L. N. McCartney, Predicting transverse crack formation in cross-ply laminates.
Compos Sci Technol, 58:7 (1998), 106981.
86. Errata to Model to predict effects of triaxial loading on ply cracking in general
symmetric laminates. [Compos Sci Technol 2000; 60(1213):22552279], Compos
Sci Technol, 62:9 (2002), 12734.
87. J. A. Nairn, Fracture mechanics of composites with residual thermal stresses. J Appl
Mech, Trans ASME, 64 (1997), 80410.
88. S. Ostlund and P. Gudmundson, Numerical analysis of matrix-crack-induced delami-
nations in [+/55-degrees] GFRP laminates. Compos Eng, 2:3 (1992), 16175.
89. P. Gudmundson and W. L. Zang, An analytic model for thermoelastic properties of
composite laminates containing transverse matrix cracks. Int J Solids Struct, 30:23
(1993), 321131.
90. P. Gudmundson and W. Zang, Thermoelastic properties of microcracked composite
laminates. Mech Compos Mater Struct, 29:2 (1993), 10714.
91. P. Gudmundson and S. Ostlund, Numerical verication of a procedure for calculation
of elastic-constants in microcracking composite laminates. J Compos Mater, 26:17
(1992), 248092.
92. P. Gudmundson and S. Ostlund, 1st order analysis of stiffness reduction due to matrix
cracking. J Compos Mater, 26:7 (1992), 100930.
93. P. Gudmundson and S. Ostlund, Prediction of thermoelastic properties of composite
laminates with matrix cracks. Compos Sci Technol, 44:2 (1992), 95105.
94. E. Adolfsson and P. Gudmundson, Matrix crack initiation and progression in
composite laminates subjected to bending and extension. Int J Solids Struct, 36:21
(1999), 313169.
95. E. Adolfsson and P. Gudmundson, Thermoelastic properties in combined bending
and extension of thin composite laminates with transverse matrix cracks. Int J Solids
Struct, 34:16 (1997), 203560.
96. E. Adolfsson and P. Gudmundson, Matrix crack induced stiffness reductions in
[(0_m/90_n/+theta_p/theta_q)(s)](m) composite laminates. Compos Eng, 5:1
(1995), 10723.
97. J. Varna, N. V. Akshantala, and R. Talreja, Crack opening displacement and the associ-
ated response of laminates with varying constraints. Int J Damage Mech, 8 (1999), 17493.
98. P. Lundmark and J. Varna, Constitutive relationships for laminates with ply cracks in
in-plane loading. Int J Damage Mech, 14:3 (2005), 23559.
99. P. Lundmark, Damage mechanics analysis of inelastic behaviour of ber composites.
Ph.D. thesis, Lulea University of Technology, Lulea , Sweden (2005), p. 175.
100. A. Krasnikovs and J. Varna, Transverse cracks in cross-ply laminates. I. Stress
analysis. Mech Compos Mater Struct, 33:6 (1997), 56582.
101. R. Joffe, A. Krasnikovs, and J. Varna, COD-based simulationof transverse cracking and
stiffness reduction in [S/90n]s laminates. Compos Sci Technol, 61:5 (2001), 63756.
102. R. Hill, Elastic properties of reinforced solids: Some theoretical principles. J Mech Phy
Solids, 11:5 (1963), 35772.
103. O. Rubenis, E. Sparnins , J. Andersons, and R. Joffe, The effect of crack spacing
distribution on stiffness reduction of cross-ply laminates. Appl Compos Mater, 14:1
(2007), 5966.
131 References
104. J. P. Benthem and W. T. Koiter, Asymptotic approximations to crack problems. In
Mechanics of Fracture I: Methods of Analysis and Solutions of Crack Problems, ed. G. C.
Sih. (Leyden: Noordhoff, 1972), pp. 13178.
105. H. Tada, P. Paris, and G. Irwin, The Stress Analysis of Cracks Handbook. (St. Louis,
MO: Del Research Corporation, 1973).
106. P. Lundmark and J. Varna, Crack face sliding effect on stiffness of laminates with ply
cracks. Compos Sci Technol, 66:10 (2006), 144454.
107. J. Varna, Physical interpretation of parameters in synergistic continuum damage
mechanics model for laminates. Compos Sci Technol, 68:13 (2008), 2592600.
108. S. Li, S. R. Reid, and P. D. Soden, A nite strip analysis of cracked laminates. Mech
Mater, 18:4 (1994), 289311.
109. J. N. Reddy, A generalization of two-dimensional theories of laminated composite
plates. Commun Appl Nume Mthds, 3:3 (1987), 17380.
110. S. Li, C. V. Singh, and R. Talreja, A representative volume element based on transla-
tional symmetries for FE analysis of cracked laminates with two arrays of cracks. Int
J Solids Struct, 46:78 (2009), 1793804.
111. J. Noh and J. Whitcomb, Effect of various parameters on the effective properties of a
cracked ply. J Compos Mater, 35 (2001), 689712.
112. S. Li, General unit cells for micromechanical analyses of unidirectional composites.
Compos A, 32:6 (2001), 81526.
113. S. Li, On the unit cell for micromechanical analysis of bre-reinforced composites.
Proc R Soc London A, 455:1983 (1999), 81538.
114. S. Li, Boundary conditions for unit cells from periodic microstructures and their
implications. Compos Sci Technol, 68:9 (2008), 196274.
115. S. G. Li and A. Wongsto, Unit cells for micromechanical analyses of particle-
reinforced composites. Mech Mater, 36:7 (2004), 54372.
116. K. Srirengan and J. D. Whitcomb, Finite element based degradation model for
composites with transverse matrix cracks. J Thermoplast Compos Mater, 11:2 (1998),
11323.
117. J. Tong, F. J. Guild, S. L. Ogin, and P. A. Smith, On matrix crack growthinquasi-isotropic
laminates II. Finite element analysis. Compos Sci Technol, 57:11 (1997), 153745.
118. F. G. Yuan and M. C. Selek, Transverse cracking and stiffness reduction in composite
laminates. J Reinf Plas Compos, 12:9 (1993), 9871015.
119. R. Cook, D. Malkus, M. Plesha, and R. Witt, Concepts and Applications of Finite
Element Analysis. (John Wiley & Sons, 2002).
120. J. Reddy, Energy Principles and Variational Methods in Applied Mechanics, 4th edn.
(Hoboken, New Jersey: John Wiley & Sons, Inc., 2002).
121. J. N. Reddy, Mechanics of Laminated Composite Plates and Shells: Theory and Analysis,
2nd edn. (Boca Raton: CRC Press, 2004).
122. D. H. Robbins and J. N. Reddy, Analysis of piezoelectrically actuated beams using a
layer-wise displacement theory. Computers & Structures, 41:2 (1991), 26579.
123. D. H. Robbins, J. N. Reddy, and F. Rostam-Abadi, An efcient continuum damage
model and its application to shear deformable laminated plates. Mech Adv Mater Struc,
12:6 (2005), 391412.
124. J. E. S. Garcao, C. M. M. Soares, C. A. M. Soares, and J. N. Reddy, Analysis of
laminated adaptive plate structures using layerwise nite element models. Computers
& Structures, 82:2326 (2004), 193959.
125. J. N. Reddy, An evaluation of equivalent-single-layer and layer-wise theories of
composite laminates. Compos Struct, 25:14 (1993), 2135.
126. Y. S. N. Reddy, C. M. D. Moorthy, and J. N. Reddy, Nonlinear progressive failure
analysis of laminated composite plates. Int J Non Linear Mech, 30:5 (1995), 62949.
127. D. H. Robbins and J. N. Reddy, Modeling of thick composites using a layerwise
laminate theory. Int J Numer Methods Eng, 36:4 (1993), 65577.
128. W. J. Na and J. N. Reddy, Multiscale analysis of transverse cracking in cross-ply
laminates beams using the layerwise theory. J Solid Mech, 2:1 (2010), 118.
129. W. J. Na and J. N. Reddy, Delamination in cross-ply laminated beams using the
layerwise theory. Asian J Civil Eng, 10:4 (2009), 45180.
130. J. Aboudi, S. W. Lee, and C. T. Herakovich, Three-dimensional analysis of laminates
with cross cracks. J Appl Mech, Trans ASME, 55:2 (1988), 38997.
131. S. W. Lee and J. Aboudi, Analysis of Composites Laminates with Matrix Cracks, 3rd edn.
Report CCMS-8803, College of Engineering, Virginia Polytechnic Institute and State
University, Blacksburg, Virginia (1988).
132. J. H. Lee and C. S. Hong, Rened two-dimensional analysis of cross-ply laminates
with transverse cracks based on the assumed crack opening deformation. Compos Sci
Technol, 46:2 (1993), 15766.
133. D. Gamby and J. L. Rebiere, A two-dimensional analysis of multiple matrix cracking
in a laminated composite close to its characteristic damage state. Compos Struc, 25:14
(1993), 32537.
134. D. Zhang, J. Ye, and D. Lam, Properties degradation induced by transverse cracks in
general symmetric laminates. Int J Solids Struct, 44:17 (2007), 5499517.
135. D. Zhang, J. Ye, and D. Lam, Ply cracking and stiffness degradation in cross-ply
laminates under biaxial extension, bending and thermal loading. Compos Struct,
75:14 (2006), 12131.
composites. Part 1. Coated-ber composites. Compos B, 29:2 (1998), 93119.
137. N. J. Pagano, G. A. Schoeppner, R. Kim, and F. L. Abrams, Steady-state cracking
and edge effects in thermo-mechanical transverse cracking of cross-ply laminates.
138. G. A. Schoeppner and N. J. Pagano, Stress elds and energy release rates in cross-ply
laminates. Int J Solids Struc, 35:11 (1998), 102555.
139. L. N. McCartney, G. A. Schoeppner, andW. Becker, Comparisonof models for transverse
ply cracks in composite laminates. Compos Sci Technol, 60:1213 (2000), 234759.
140. J. A. Nairn, The strain-energy release rate of composite microcracking a variational
141. R. Y. Kim, A. S. Crasto, and G. A. Schoeppner, Dimensional stability of composite in
a space thermal environment. Compos Sci Technol, 60:1213 (2000), 26018.
133 References
5 Macro-damage mechanics
5.1 Introduction
As described in the previous chapter, the two main approaches to treating damage
in composite materials are micro-damage mechanics (MIDM) and macro-damage
mechanics (MADM). We shall now discuss MADM, which is also commonly
known as continuum damage mechanics (CDM).
Consider an undamaged continuum (Figure 5.1, left) representing a hetero-
geneous solid undergoing deformation upon prescribed traction t and displacement
u on parts of the bounding surfaces S
t
and S
u
, respectively. Assuming the average
response of the solid to be linear elastic, the stressstrain relationships can be
written as
s
ij
= C
ijkl
e
kl
; (5:1)
where s
ij
; e
kl
; and C
ijkl
denote the tensors of stress, strain, and stiffness, respectively,
at a point of the continuum body in its undamaged state. At some loading state, the
body may initiate and evolve damage in the form of voids, microcracks, cavities, etc.
(Figure 5.1, middle). These damage entities will cause perturbations in the local
stress and strain states of the continuum body. The stressstrain response averaged
over a representative volume element (RVE) at a xed damage state is still given by
Eq. (5.1) but now C
ijkl
denotes the stiffness tensor of the homogenized continuum
body containing damage. The main objective of the CDM approach is to character-
ize C
ijkl
for the continuum body in which damage entities have been homogenized.
A concept central to CDM is the internal state representing the damage entities
that have been homogenized in the continuum (Figure 5.1, right). Historically, this
concept is credited to Kachanov [1, 2] who introduced the so-called continuity as a
varying state of a metal in an attempt to describe its degradation due to creep.
Accordingly, a single scalar variable, denoted as f, was used to describe the continu-
ity with values equal to 1 in the virgin state, and 0 at rupture. The loss of continuity
signies material deterioration, and was assumed to occur at a rate given by
df
dt
= A
s
f
_ _
m
; (5:2)
where A and m are material constants and s is the maximum principal tensile stress.
The variable f can be viewed as representing surface density of discontinuities
in the material, and can then be used to describe an effective stress. Thus in
Eq. (5.2), the effective stress is equal to s/f, which represents the current level
of stress causing further material degradation. Later, Robotnov [3] interpreted
discontinuity, dened by the parameter o = 1 f, as a measure of net area
reduction (see Figure 5.2). Accordingly, o is dened as
o =
S
D
S
=
S S
+
S
; (5:3)
where the damaged cross section S
D
= S S
*
, in which S is the cross-sectional
area in the undamaged state and S
*
is the net area of the damaged specimen which
excludes the area held by the damage entities (discontinuities). Thus the net area of
the specimen effectively carries the applied load. The damage parameter o = 0
signies an undamaged state, while o = o
c
represents a critical state corresponding
Damage
Initiation
Homogenization
Undamaged
homogenized
continuum
Continuum
with damage
Homogenized
damaged
continuum
S
u
S
t
t
i
u
i
S
u
S
t
t
i
u
i
S
u
S
t
t
i
u
i
Figure 5.1. The basic concept of internal variable damage mechanics.
Damaged state
(0<<1)
Equivalent fictitious undamaged state
(=0)
Area S
Area S*
Force, F=s.S Force, F=s
.S
Figure 5.2. The concept of effective stress for isotropic damage in uniaxial loading.
135 5.1 Introduction
to the rupture of the specimen. Following this approach, Robotnov [3] postulated
the creep strain rate as
de
dt
= B
s
1 o
_ _
n
; (5:4)
where B and n are constants. It should be noted here that Kachanovs
effective stress, as dened, increases to innity at failure, while Robotnovs effect-
ive stress, as interpreted, would increase only by a few percent since the volume
fraction of voids or discontinuities at failure is found by microstructural studies
for metals to be small.
Robotvovs concept of effective stress was also used later by Murakami and
Ohno [4] for creep damage of polycrystalline metals and by Lemaitre and co-
workers [5, 6] for effective elastic properties. Denoting the quantity in parenthesis
in Eq. (5.4) by
~ s =
s
1 o
; (5:5)
Hookes law can be expressed in two equivalent forms as
s =
~
Ee
e
; ~ s = Ee
e
; (5:6)
where e
e
is the elastic strain and
~
E is the effective Youngs modulus, which is
related to the damage parameter o as follows
~ s =
s
1 o
=
E
~
E
s = o = 1
~
E
E
: (5:7)
In a somewhat similar way, for brittle creep damage, Lemaitre and Chaboche [7]
dened the damage variable in terms of the strain rate. Using a power law, the
strain rate during secondary creep (undamaged state), _ e
s
, can be described by
_ e
s
=
s
l
_ _
N
; (5:8)
where l is a material constant and the exponent N is found from experimental
tests. The damage variable follows easily from the measurement of the strain rate
during tertiary creep, _ e, and the use of effective stress concept as
o = 1
_ e
_ e
s
_ _
1=N
: (5:9)
Similar to the effective stress concept, the strain equivalence principle has also
been proposed by Lemaitre and coworkers [6, 8], which states that any strain
constitutive equation for a damaged material may be derived in the same way as
for a virgin material except that the usual stress is replaced by the effective stress.
In creep of metals as well as in brittle cracking of ceramics, rocks, concrete, etc. the
voids and cracks usually form along some preferred orientations, e.g., on grain
boundaries and in weak planes. It is obvious that a scalar damage variable cannot
account for the directional dependence of the effects of voids and cracks.
To characterize this effect, Murakami and Ohno [4] considered an arbitrarily oriented
plane in a solid containing arbitrarily distributed voids and dened a second-order
tensor to describe the anisotropic net area reduction. This tensor was assumed to
represent the directional nature of discontinuities and was called the damage tensor.
The MurakamiOhno damage tensor V relates the elemental area dA of an
arbitrarily oriented plane with the unit normal n lying in the damaged congur-
ation to the elemental area dA* of the same plane with the unit normal n* lying in
a ctitious conguration with reduced net load carrying area. Thus,
n
+
dA
+
= I V ( )n dA; (5:10)
where I is the identity tensor of second order.
The effective stress tensor s* then is the magnied (net area) stress in the ctitious
conguration which is related to the stress in the damaged conguration by
s
+
= I V ( )
1
s: (5:11)
Following (5.7), Chaboche [911] generalized the relation between elastic con-
stants and the damage variable to the three-dimensional case, replacing o by a
damage tensor D. This damage tensor, in general, is a nonsymmetric fourth-order
tensor, and is dened by the following transformation
E
~
= I D ( ) : E; (5:12)
where I is the identity tensor of fourth order, E and E
~
are the elasticity tensor for
the undamaged and damaged materials, respectively, and (:) stands for the second-
order contraction. Thus, the damage tensor is given by
D = I E
~
E
1
; (5:13)
and the effective stress tensor is given by
s
+
= I D ( )
1
: s: (5:14)
It is worth noting that the second-order damage tensor dened by Murakami and
Ohno [4], Eq. (5.10), is based on the notion of effective net area reduction, while
the fourth-order damage tensor dened by Chaboche [911], Eq. (5.13), uses the
notion of the effective stress dened by Eq. (5.6).
The damage tensor D dened by (5.13) has been the focus of much attention in
several works on continuum damage mechanics. Ju [12], for instance, has dis-
cussed the notion of isotropic and anisotropic damage variables and has demon-
strated that isotropic damage does not necessarily imply a scalar damage variable.
From a mechanics study of changes in the elasticity compliance tensor due to
microcracks he found that the fourth-order tensor D is isotropic for the case of
isotropic damage, e.g., when the cracks are perfectly randomly distributed in all
137 5.1 Introduction
directions. However, for preferred directions in crack orientations the damage
tensor will, in general, be anisotropic.
5.2 Continuum damage mechanics (CDM) of composite materials
The damage variables dened on the basis of the effective stress concept, or
the associated notion of the reduced net area, do not contain any specic details
of the damage entities. In the LemaitreChaboche damage tensor, for instance,
the components of the damage tensor are determined from the elastic compli-
ance changes. Thus, e.g., two sets of damage entities of different characteristic
sizes and concentrations leading to the same elastic compliance changes will be
represented by the same variables. Such equivalency of different sets of damage
entities cannot, in general, be expected to hold for all effects of damage.
Furthermore, in media with, say, two levels of microstructure, e.g., in compos-
ites, the damage entities forming at the lower size-scale may be affected by the
higher size-scale microstructure as well as by the difference in the symmetry
properties of the two microstructures. To illustrate this aspect, consider an
ellipsoidal particle embedded in a compliant matrix and surrounded by a
parallel array of stiff bers (Figure 5.3). Let the particle debond from the
matrix under appropriate loading such that an ellipsoidal damage entity is
formed. The local stress perturbation induced by the surface displacements of
this damage entity, and the resulting overall elastic compliance, will depend on
the local microstructure details such as inclination of the bers with respect to
the principal planes of the damage entity and the ber stiffness. Thus the same
microstructure arranged differently can produce a different effect on the overall
elastic response. This example serves to illustrate that a characterization of
damage in terms of the elastic compliance changes will not necessarily be
unique.
Figure 5.3. Schematic illustration of anisotropic constraint effect in composites. An
ellipsoid entity is shown surrounded by bers or plies in a composite. The axes of the
ellipsoid are inclined with respect to the axes of the constraining elements. Reprinted, with
kind permission, from Damage Mechanics of Composite Materials, R. Talreja, Damage
characterization by internal variables, pp. 5378, copyright Elsevier (1994).
In the alternative characterization of damage, adopted here, the damage
entities are represented by appropriate specic variables, and averages of these
variables over an RVE then give the damage variables. The degree of charac-
terization, i.e., to what extent the details of damage entities are represented,
depends on the type of variables chosen. Thus, a scalar variable will only
represent the size but not the shape or orientation characteristics of the damage
entities. A vector variable can account for the size and orientation, but the
positive and negative senses of a vector introduce ambiguities in the damage
characterization that cause difculties in the description of damage evolution.
With two vectors simultaneously associated with a damage entity these difcul-
ties are overcome. At the same time, the resulting second-order tensor gives
mathematical convenience of formulating response functions, which are com-
monly expressed in terms of the second-order stress and strain tensors, as
discussed below.
Talrejas original paper on CDM used a vectorial description of damage [13].
Although the ambiguities concerning the sense of damage vector could be remed-
ied by imposing an additional condition, it was found that using a second-order
tensor instead avoided this step [1417]. Here we will follow the tensorial
description.
5.2.1 RVE for damage characterization
Any continuum description of a solid entails homogenization since materials
are inherently heterogeneous. For polycrystalline metals, for instance, the scale
of heterogeneity (e.g., grain size) is often small compared to the scale at which
material response characteristics (e.g., the elastic constants) are measured,
allowing the stress and strain states to be dened as continuous elds. For
commonly used ber-reinforced solids, such as glass/epoxy and carbon/epoxy,
the ber diameter of approximately 10 micrometers allows treating these
materials as a homogeneous continuum with good accuracy. When internal
surfaces in composite materials form, their characteristic dimensions and
mutual distances between them can be orders of magnitude larger than the
scale of heterogeneities underlying the homogenized pristine composite.
Furthermore, on application of external loads the internal surfaces are subject
to evolution (enlargement and multiplication), as discussed above. This war-
rants a separate (and different) homogenization of the composite with internal
surfaces (collectively called damage).
Figure 5.4 depicts a homogenization procedure for a composite solid contain-
ing damage. The heterogeneities in the pristine (undamaged) composite are
referred to as stationary microstructure and are homogenized rst. This may
also be called classical homogenization. Textbooks on mechanics of composite
materials usually begin with this homogenization. In fact, the classical laminate
theory goes one step further by developing homogeneous constitutive relations
for laminates consisting of stacked layers of homogenized unidirectionally
139 5.2 Continuum damage mechanics (CDM) of composite materials
reinforced composite (ply or lamina). Returning to Figure 5.4, the second hom-
ogenization pertains to the internal surfaces, collectively named as damage or
evolving microstructure to highlight their ability to permanently change by
processes of energy dissipation. Homogenization of the evolving microstructure
necessitates employing the notion of a representative volume element (RVE),
which will be discussed next.
A general and thorough exposition of the RVE notion in the context of micro-
mechanics is given in [18]. Here, we shall apply this notion to the particular case
of composite materials with damage. With reference to Figure 5.4 again, a generic
point P in the homogenized composite with damage has associated with it a
damage state (in addition to stress and strain states), which is given by an
appropriate volume-averaged measure of the presence of internal surfaces that
affect the constitutive behavior (stressstrain relations) at the point P. The volume
over which the averaging is performed must be representative of the neighborhood
of point P that can be associated with P. This neighborhood is the RVE, whose
volume is not xed but depends on the geometrical conguration (size, spacing,
etc.) of the internal surfaces around P. As this conguration changes under
applied loading, the RVE size changes.
With the notion of RVE at hand, the damage state at P can be dened by a
set of variables obtained by averaging over the RVE. The choice of the
variables is guided by the type of internal surfaces formed, and in this respect
P
a
i
n
j
Homogenization of
stationary microstructure
Homogenization of
evolving microstructure
Characterization of a
damage entity
RVE for damage characterization
Fully homogenized continuum
P
V
Step 2
Step 1
Evolving microstructure
Continuum after homogenizing
the stationary microstructures
Figure 5.4. Homogenization of a continuum body with heterogeneous stationary
structure and evolving damage entities. A tensorial characterization of a damage entity
is depicted on the right.
the knowledge of damage mechanisms discussed above is useful. In general,
the variables can be scalars, vectors, or tensors of second order or higher.
Settling on which variables to employ is a matter of nding a balance between
capturing sufcient physics of the damage process and usefulness of the ensu-
ing formulation of constitutive relations. In the following the second-order
characterization of damage in composite materials adopted by Talreja [14, 15,
17] is described.
5.2.2 Characterization of damage
As shown in Figure 5.4, a single internal surface within a RVE, called a damage
entity from now on, can be characterized by two vectors: a unit outward normal n
at a point on the surface, and an inuence vector a at the same point. A dyadic
product of the two vectors, integrated over the surface S, is denoted the damage
entity tensor, and given by
d
ij
=
_
S
a
i
n
j
dS; (5:15)
where the components of the vectors are with reference to a Cartesian coordinate
system. The dyadic product assures consistency of the signs of the two vectors.
A characterization of this type was rst proposed by Vakulenko and Kachanov
[19] for at cracks, where the inuence vector a represented the displacement jump
across the crack surface.
The physical signicance of this characterization is that it represents the
oriented nature of the presence of internal surfaces. As illustrated by the examples
of damage in Chapter 3, common internal surfaces are cracks (at or curved)
generated by interface debonding and matrix failure. The unit normal vector at a
point on the damage entity carries the information on orientation of the surface
(with respect to the frame of reference), while the other vector represents an
appropriate inuence induced by activation of the considered point on the surface.
This inuence is generally also directed in nature. For the case of mechanical
response, the appropriate inuence would be the displacement of the activated
point on the damage entity surface. For a nonmechanical response, such as
thermal or electrical conductivity, the perturbation induced by an internal surface
can also be cast as a vector-valued quantity.
Integrating the dyadic product in Eq. (5.15) over the damage entity surface
provides the total net effect of the entity. For example, if the entity is a at crack,
then taking a as the displacement vector in the integral gives the crack surface
separation times the crack surface area. This product may be viewed as an
affected volume associated with the crack. For a penny-shaped crack with the
two surfaces separating symmetrically about the initial crack plane, the sole
surviving term of the damage entity tensor represents an ellipsoidal-shaped
volume.
Referring once again to Figure 5.4, the RVE associated with a generic point P
carries a sufciently large number of the discrete damage entities to represent
the collective effect on the homogenized constitutive response at the point. The
number of damage entities needed for this representation, and the consequent
RVE size, depend on the distribution of the entities. For instance, if the entities
are sparsely distributed, then the RVE size would be large, while for densely
distributed case a small RVE would sufce. Furthermore, for uniformly distrib-
uted entities of the same geometry, a repeating unit cell containing a single
entity can replace the RVE, while for the cases of nonuniform distribution of
unequal entities, the RVE size will increase until a statistically homogeneous
representation is attained. This implies that further increasing the RVE size will
have no impact on the averages of the selected characteristics. As an example, if
the selected characteristic is the affected matrix volume by a damage entity, as
mentioned above, then the average value of this quantity will vary as the RVE
size increases and will approach a constant value at a certain RVE size. The
minimum RVE size beyond which no appreciable change in the considered
average is found may be taken as the needed RVE. It is apparent that the
RVE is not unique but is subject to the choice made for the particular formula-
tion of the constitutive response of a continuum with damage. Consequently,
there is no unique constitutive theory of a continuum with damage; however,
the use of the concept of an internal state in a given theory requires specifying
RVE in a consistent manner and assuring that the conditions for its existence
are present.
From the cases of damage mechanisms reviewed in Chapter 3 it can be noted
that in composite laminates the damage tends to occur as sets of parallel cracks
within the plies, each oriented along bers in the given ply. It is therefore
convenient to separate each set of ply cracks according to its orientation,
referred to a xed frame of reference, and assign it a damage mode number.
Denoting damage mode by a = 1, 2, . . ., n, a damage mode tensor can be
dened as
D
(a)
ij
=
1
V
k
a
d
ij
_ _
k
a
; (5:16)
where k
a
is the number of damage entities in the ath mode, and V is the RVE
volume. As noted above, if the ply cracks of a given orientation are uniformly
spaced, then the RVE will reduce to the unit cell containing one crack. For
nonuniform distribution of ply cracks, V must be large enough to provide a steady
average of the damage mode tensor components.
As dened by Eq. (5.16) the damage mode tensor will in general be asymmet-
rical. Decomposing the inuence vector a along directions normal and tangential
to the damage entity surface S gives,
a
i
= an
i
bm
i
; (5:17)
where n and m are unit normal and tangential vectors on S such that n
i
m
i
= 0.
Using Eq. (5.17) in Eq. (5.16) the damage entity tensor can be written in two
parts as
d
ij
= d
1
ij
d
2
ij
; (5:18)
where
d
1
ij
=
_
S
an
i
n
j
dS; d
2
ij
=
_
S
bm
i
n
j
dS: (5:19)
Here a and b are the magnitudes of the normal and tangential projections,
respectively, of vector a
i
and vectors n
i
and m
j
are unit normal and tangential
vectors, respectively. The damage mode tensor for a given mode can now be
written as
D
(a)
ij
= D
1(a)
ij
D
2(a)
ij
; (5:20)
where
D
1(a)
ij
=
1
V
k
a
d
1
ij
_ _
k
a
; D
2(a)
ij
=
1
V
k
a
d
2
ij
_ _
k
a
: (5:21)
This separation of the damage mode tensor in two parts allows the analysis to
be simplied to avoid having to deal with asymmetric tensors. For instance,
for damage entities consisting of at cracks, the two parts of the damage mode
tensor represent the two crack surface separation modes. If an assumption can
be made that only the symmetric crack surface separation (known as mode I or
crack opening mode in fracture mechanics) is signicant, then the second term
in Eq. (5.21) can be neglected. This will render the damage mode tensor
symmetrical and it can then be written as
D
(a)
ij
= D
1(a)
ij
=
1
V
k
a
_
S
an
i
n
j
dS
_ _
k
a
: (5:22)
The consequence of this assumption was examined by Varna [20] for one class of
laminates and it was found that not including the crack sliding displacement
(CSD) for ply cracks inclined to the laminate symmetry directions resulted in
errors in estimating the degradation of the average elastic properties of laminates.
However, these errors were found to be small in absolute values while being
signicant in percentages. In fact for those ply crack orientations where CSD
dominates, the cracks are difcult to initiate until high loads close to failure load
are applied.
Some damage mechanisms such as crystalline slip may require only the
tangential part. For other situations the sliding between the crack faces can
be negligible, e.g., for intralaminar cracks constrained by stiff plies, and ber/
matrix debonds. Hence D
ij
2(a)
= 0 under such conditions. For cases where the
damage entity surfaces conduct tangential displacements only (e.g., CSD by at
cracks), it is possible to formulate the damage mode tensor as a symmetric tensor.
One example of this is sliding of the ber/matrix interface in ceramic matrix
composites [16].
With stress, strain, and damage, all expressed as symmetric second-order
tensors, a constitutive theory can now be formulated to have a convenient, usable
form. Such a formulation is described next.
5.2.3 A thermodynamics framework for materials response
Referring once again to Figure 5.4, a formulation of the constitutive response
of a homogenized continuum with damage will now be discussed. In view of the
observed behavior of common composite materials such as glass/epoxy and
carbon/epoxy, only elastic response will be considered. Theoretical treatment of
elastic response of solids is classical and can be found in textbooks. Incorpor-
ating damage is, however, not a simple extension of the classical theory of
elasticity. The CDM approach to be described here is based on thermodynam-
ics and is naturally suited for thermomechanical response. It can be extended to
incorporate nonmechanical effects, such as electrical and magnetic, as well as
chemical. Every extension, however, comes with the price of having to deter-
mine associated response coefcients (material constants) by a certain identi-
cation procedure. In the treatment presented here, the task of determining
material constants is reduced by use of selected micromechanics. This way of
combining micromechanics with CDM generates useful synergism, justifying
the characterization of the combined approach as synergistic damage mechan-
ics (SDM), to be discussed later. We begin with the conventional CDM frame-
work rst.
At the foundation of CDM are the rst and second laws of thermodynamics.
Additionally, use is made of the concept of an internal state, which is identied here
as the evolving microstructure depicted in Figure 5.4. As discussed before, this
microstructure is homogenized into a damage eld characterized by the set of damage
mode tensors D
ij
(a)
. The collectionof all variables resulting fromthermodynamics with
internal state can now be placed in two categories: state variables and response
functions (see Table 5.1). The thermodynamical response of a composite body, limited
to small strains, is given by a set of ve response functions: the Cauchy stress tensor s
ij
,
Table 5.1 Thermodynamics variables for damage analysis
Thermodynamic state variables Thermodynamic response variables
1. Strain tensor e
ij
=
1
2
u
i; j
u
j;i
_ _
1. Cauchy stress tensor s
ij
2. Temperature T 2. Specic Helmholtz free energy c
3. Temperature gradient g
i
= T
,i
3. Specic entropy
4. Damage tensors D
ij
(a)
4. Heat ux vector q
i
5. Damage rate tensors
_
D
(a)
ij
the specic Helmholtz free energy c, the specic entropy , the heat ux vector q
i
, and
a set of damage rate tensors
_
D
(a)
ij
, a =1,2, ..., n. The thermodynamic state of the body is
givenby the straintensor e
ij
= 1=2 ( ) u
i; j
u
j;i
_ _
, withdisplacement vector u
i
, absolute
temperature T, temperature gradient g
i
= T
,i
, and a set of damage tensors D
ij
(a)
.
Following Truesdells principle of equipresence, which states that all state
variables should be present in all response functions unless thermodynamics or
other relevant considerations preclude their dependency, we write
s
ij
= s
ij
e
kl
; T; g
k
; D
(a)
kl
_ _
;
c = c e
kl
; T; g
k
; D
(a)
kl
_ _
;
= e
kl
; T; g
k
; D
(a)
kl
_ _
;
q = q e
kl
; T; g
k
; D
(a)
kl
_ _
;
_
D
(a)
kl
=
_
D
(a)
kl
e
kl
; T; g
k
; D
(b)
kl
_ _
:
(5:23)
The following balance laws should hold for the continuum body:
v Balance of linear momentum:
s
ij;j
rb
j
= r x
j
; (5:24)
where b
j
are the components of the body force per unit mass and r is the mass
density.
v Balance of angular momentum:
s
ij
= s
ji
: (5:25)
v Balance of energy:
r _ u s
ij
_ e
ij
q
i;i
= rr; (5:26)
where u is the specic internal energy per unit mass and r is the heat supply per
unit mass.
v Second law of thermodynamics in the form of the ClausiusDuhem
inequality:
s
ij
_ e
ij
r
_
c r
_
T
q
i
g
i
T
_ 0; (5:27)
where c = u T.
Time differentiation of c in (5.23) gives,
_
c =
@c
@e
kl
_ e
kl

@c
@T
_
T
@c
@g
i
_ g
i

@c
@D
(a)
kl
_
D
(a)
kl
: (5:28)
Substitution of Eq. (5.28) into Eq. (5.27) gives
s
ij
r
@c
@e
kl
_ _
_ e
ij
r
@c
@T
_ _
_
T r
@c
@g
i
_ g
i
r
a
@c
@D
(a)
kl
_
D
(a)
kl

q
i
g
i
T
_ 0; (5:29)
where the summation sign is shown on a to include all damage modes. Now
requiring (5.29) to hold for the independently varying strain, temperature, and
temperature gradient gives the following results
s
ij
= r
@c
@e
kl
; (5:30)
=
@c
@T
; (5:31)
@c
@g
i
= 0 : (5:32)
Equation (5.32) states that the Helmholz free energy function does not depend on
the temperature gradient and, consequently, Eqs. (5.30) and (5.31) eliminate this
dependency from stress and entropy. The last two equations in the response
function set (Eq. (5.23)) remain unaffected.
Equations (5.30)(5.32) lead to the following functional dependencies
s
ij
= s
ij
e
kl
; T; D
(a)
kl
_ _
;
c = c e
kl
; T; D
(a)
kl
_ _
;
= e
kl
; T; D
(a)
kl
_ _
;
q = q e
kl
; T; g
k
; D
(a)
kl
_ _
;
_
D
(a)
kl
=
_
D
(a)
kl
e
kl
; T; g
k
; D
(b)
kl
_ _
:
(5:33)
From Eq. (5.29) the following restriction (also known as internal dissipation
inequality) results
a
R
(a)
kl
_
D
(a)
kl

q
i
g
i
T
_ 0; (5:34)
where R
kl
(a)
are the thermodynamic forces conjugate to D
kl
(a)
and are given by
R
(a)
kl
= r
@c
@D
(a)
kl
: (5:35)
Each of these forces is analogous to the crack extension force (i.e., energy release
rate) for a single crack. As an example, for a damage mode component, say
D
11
of mode a = 1, the quantity R
11
(1)
can be interpreted as the force causing
an innitesimal change in the internal state represented by D
11
(1)
. Equation (5.34)
expresses the condition these forces must satisfy as damage evolves under thermo-
mechanical impulses.
The complete thermomechanical response for the composite body is governed
by the set of functions Eq. (5.33) and the internal dissipation inequality Eq. (5.34).
In view of the experimental data, which are mostly available for polymer matrix
composites at room temperature, the thermomechanical framework will be
developed further for the purely mechanical response. Thus, for isothermal condi-
tions (T = 0, g
i
= 0) the set of response functions is reduced to the following
s
ij
= s
ij
e
kl
; D
(a)
kl
_ _
;
c = c e
kl
; D
(a)
kl
_ _
;
_
D
(a)
kl
=
_
D
(a)
kl
e
kl
; g
k
; D
(b)
kl
_ _
;
a
R
(a)
kl
_
D
(a)
kl
_ 0 :
(5:36)
Since s
ij
are derivable fromc, according to Eq. (5.30), it sufces to formulate c and
_
D
(a)
kl
for a purely mechanical response. Dealing with this scalar-valued function (c) as
the sole response function for a given internal state of damage provides a favorable
situation for further development of the theory. The form of the Helmholz free
energy function can be chosen in different ways. A powerful way is possible by use
of the theory of invariants for polynomial functions [21]. In the following it is
illustrated for one case of orthotropic composites containing one damage mode.
To derive the rate equations for s
ij
, we differentiate Eq. (5.30) with respect to
time and use the functional dependency given in Eq. (5.36), which yields
_ s
ij
= r
@
2
c
@e
ij
@e
kl
_ e
kl
r
a
@
2
c
@e
ij
@D
(a)
mn
_
D
(a)
mn
: (5:37)
Substitution of Eq. (5.35) gives
_ s
ij
= r
@
2
c
@e
ij
@e
kl
_ e
kl

a
@R
(a)
mn
@e
ij
_
D
(a)
mn
; (5:38)
which may be rewritten as
_ s
ij
= C
ijkl
_ e
kl

a
K
(a)
ijmn
_
D
(a)
mn
; (5:39)
where
C
ijkl
= r
@
2
c
@e
ij
@e
kl
(5:40)
and
K
(a)
ijmn
=
@R
(a)
mn
@e
ij
= r
@
2
c
@e
ij
@D
(a)
mn
: (5:41)
The matrix C
ijkl
contains stiffness coefcients of the composite in its virgin mater-
ial state, whereas coefcients of K
ijmn
(a)
are functions determining the change of state
caused by the internal dissipative mechanisms. The rate shown in Eq. (5.39)
requires formulation of the scalar function c and the tensor components
_
D
(a)
kl
.
The discussion of damage evolution is left for later; the next section will treat the
stressstrain relations at a xed damage state.
5.2.4 Stiffnessdamage relationships
The stiffness coefcient matrix C
ijkl
of the composite in a given state of damage
is derivable from the Helmholtz free energy function c according to Eq. (5.40).
The scalar valued function c can be written as a polynomial in its variables, i.e.,
c = c
P
e
ij
; D
(a)
ij
_ _
; (5:42)
where c
P
stands for the polynomial function.
Let us now focus on the particular case of intralaminar cracking in composite
laminates. Figure 5.5 shows an RVE illustrating one set of intralaminar cracks in
an off-axis ply of a composite laminate. Although for clarity of illustration the
cracking is shown only in one lamina, it is understood that in general it exists in
multiple plies of the laminate. The thickness of the cracked plies is denoted by t
c
, s
is the average crack spacing, t is the total laminate thickness, and W and L
stand for the width and the length, respectively, of the RVE. The volume of the
RVE, the surface area of a crack, S, and the inuence vector magnitude, a, are
specied as
V = L:W:t;
S =
W:t
c
sin y
;
a = kt
c
;
(5:43)
where k, called the constraint parameter, is an unspecied constant of (assumed)
proportionality between a and the crack size t
c
(also the cracked-ply thickness).
Here, y is a positive quantity so that the surface area is always positive. Assuming
a to be constant over the crack surface S, one gets from Eq. (5.22)
D
(a)
ij
=
kt
2
c
st sin y
n
i
n
j
; (5:44)
where n
i
= sin y; cos y; 0 ( ).
Expansion of the polynomial function in Eq. (5.42) can in general have innite
terms, which will obviously present an impractical situation. One way to restrict
the functional form is by expanding the polynomial in terms that account for the
initial material symmetry. This is done in the polynomial invariant theory by using
the so-called integrity bases [21]. Such bases have been developed for scalar
functions of various vector and tensor variables. For the case of two symmetric
second-order tensors, such as in Eq. (5.42), the integrity bases for orthotropic
symmetry are given by Adkins [22]. Considering a single damage mode, a = 1, we
have the following set of invariant terms
e
11
; e
22
; e
33
; e
2
23
; e
2
31
; e
2
12
; e
23
e
31
e
12
;
D
11
; D
22
; D
33
; D
2
23
; D
2
31
; D
2
12
; D
23
D
31
D
12
;
e
23
D
23
; e
31
D
31
; e
12
D
12
;
e
31
e
12
D
23
; e
12
e
23
D
31
; e
23
e
31
D
12
;
e
23
D
31
D
12
; e
31
D
12
D
23
; e
12
D
23
D
31
:
(5:45)
For the sake of applying the constitutive theory to thin laminates where only
in-plane strains are of interest, and for small strains, the expansion of the function
c (Eq. (5.42)) can be restricted to no more than quadratic terms in strain compon-
ents e
11
, e
22
, and e
12
To what extent the damage tensor components are to be
taken in the expansion depends on the nature and amount of information that can
Figure 5.5. A representative volume element illustrating intralaminar multiple cracking in
a general off-axis ply of a composite laminate.
be acquired for evaluation of the material constants that will appear in the
polynomial function. This issue will be discussed later. To begin with the simplest
possible case, we will include only linear terms in D
11
, D
22
, and D
12
, which are the
nonzero components for intralaminar cracks. Thus the set of invariant terms
reduces to
e
1
; e
2
; e
2
6
;
D
1
; D
2
; D
2
6
;
e
6
D
6
;
(5:46)
where e
1
= e
11
; e
2
= e
22
; e
6
= e
12
; D
1
= D
11
; D
2
= D
22
; D
6
= D
12
. The most general
polynomial form for the Helmholtz free energy, restricted to second-order terms
in the strain components and rst-order terms in damage tensor components, is
given by
rc = P
0
c
1
e
2
1
c
2
e
2
2
c
3
e
2
6
c
4
e
1
e
2
_ _
c
5
e
2
1
D
1
c
6
e
2
1
D
2
_ _
c
7
e
2
2
D
1
c
8
e
2
2
D
2
_ _
c
9
e
2
6
D
1
c
10
e
2
6
D
2
_ _
c
11
e
1
e
2
D
1
c
12
e
1
e
2
D
2
c
13
e
1
e
6
D
6
c
14
e
2
e
6
D
6

P
1
(e
p
; D
q
) P
2
(D
q
);
(5:47)
where P
0
and c
i
, i = 1, 2, . . ., 14 are material constants, P
1
is a linear function of
strain and damage tensor components, and P
2
is a linear function of damage
tensor components. Setting the free energy to zero for unstrained and undamaged
material, we have P
0
= 0, and assuming the unstrained material of any damaged
state to be stress free, we get P
1
= 0. The stress components in the Voigt notation
are now given by (from Eq. (5.30))
s
p
= r
@c
@e
p
; (5:48)
where p = 1, 2, 6. A differential in stress can now be written as
ds
p
= r
@c
@e
p
@e
q
de
q
r
@c
@e
p
@D
r
dD
r
= C
pq
de
q
K
pr
dD
r
; (5:49)
where
C
pq
= r
@c
@e
p
@e
q
(5:50)
is the stiffness matrix when dD
r
= 0, i.e., at constant damage. This is illustrated for
the uniaxial stressstrain response in Figure 5.6. As seen there, the elastic modulus
at any point on the stressstrain curve is the secant modulus, not the tangent
modulus.
Combining Eqs. (5.47), (5.48), and (5.50), one obtains
C
pq
= C
0
pq
C
(1)
pq
(5:51)
where
C
0
pq
=
2c
1
c
4
0
2c
2
0
Symm 2c
3
_
_
_
_
=
E
0
x
1 n
0
xy
n
0
yx
n
0
xy
E
0
y
1 n
0
xy
n
0
yx
0
E
0
y
1 n
0
xy
n
0
yx
0
Symm G
0
xy
_
_
_
_
(5:52)
represents the orthotropic stiffness matrix for virgin composite material, in which
E
0
x
; E
0
y
; n
0
xy
; G
0
xy
are effective moduli for the undamaged laminate, and
C
(1)
pq
=
2c
5
D
1
2c
6
D
2
c
11
D
1
c
12
D
2
c
13
D
6
2c
7
D
1
2c
8
D
2
c
14
D
6
Symm 2c
9
D
1
2c
10
D
2
_
_
_
_; (5:53)
represents the stiffness change brought about by the damage entities of damage
mode 1.
It can be noted here that Eqs. (5.51)(5.53) show linear dependence of the
stiffness properties on damage tensor components. This is the consequence of
including only linear terms in these components in the polynomial expansion of
the free energy function, Eq. (5.47). Including higher-order terms will add additional
constants c
i
, which will need to be evaluated. The evaluation procedure is described
below, but it is remarked here that the formulation of constitutive response is in
no way restricted only to linear dependence on the chosen damage measure.
E
0
E
s
e
Figure 5.6. Stressstrain curve of a composite with damage. The secant modulus E varies
with the state of damage.
Case 1: Cracking in one off-axis orientation
From Eqs. (5.51)(5.53), it can be seen that the presence of damage entities, even
for one orientation, removes the initial orthotropic symmetry. For the case of
intralaminar cracks in one orientation, as illustrated in Figure 5.5, the nonzero
components of the damage tensor D
ij
(1)
are given from Eq. (5.44) by
D
1
= D
(1)
11
=
kt
2
c
st
sin y;
D
2
= D
(1)
22
=
kt
2
c
st
cos
2
y
sin y
;
D
6
= D
(1)
12
=
kt
2
c
st
cos y:
(5:54)
Inserting these into Eq. (5.51) and using Eq. (5.52) and Eq. (5.54) we obtain
the stiffness matrix of the damaged composite laminate for a xed state of
damage as
C
pq
=
E
0
x
1 n
0
xy
n
0
yx
n
0
xy
E
0
y
1 n
0
xy
n
0
yx
0
E
0
y
1 n
0
xy
n
0
yx
0
Symm G
0
xy
_
_
_
kt
2
c
st
sin y
2c
5
2c
6
cot
2
y c
11
c
12
cot
2
y c
13
cot y
2c
7
2c
8
cot
2
y c
14
cot y
Symm 2c
9
2c
10
cot
2
y
_
_
_
_
:
(5:55)
Case 2: Cross-ply laminates
For the special case of cross-ply laminates, y = 90
, and hence
C
pq
=
E
0
x
1 n
0
xy
n
0
yx
n
0
xy
E
0
y
1 n
0
xy
n
0
yx
0
E
0
y
1 n
0
xy
n
0
yx
0
Symm G
0
xy
_
_
_
kt
2
c
st
2a
1
a
4
0
2a
2
0
Symm 2a
3
_
_
_
_ (5:56)
where a
1
= c
5
; a
2
= c
7
; a
3
= c
9
; and a
4
= c
11
. It can be observed that the ortho-
tropic symmetry is retained for intralaminar cracking in cross-ply laminates.
The engineering moduli can be derived from the following relationships
E
x
=
C
11
C
22
C
2
12
C
22
; E
y
=
C
11
C
22
C
2
12
C
11
;
n
xy
=
C
11
C
22
; G
xy
= C
66
:
(5:57)
Thus, for cross-ply laminates with 90
-ply cracks,
E
x
=
E
0
x
1 n
0
xy
n
0
yx
2
kt
2
c
st
a
1

n
0
xy
E
0
y
1 n
0
xy
n
0
yx
kt
2
c
st
a
4
_ _
2
E
0
y
1 n
0
xy
n
0
yx
2
kt
2
c
sin y
st
a
2
;
E
y
=
E
0
y
1 n
0
xy
n
0
yx
2
kt
2
c
st
a
2

n
0
xy
E
0
y
1 n
0
xy
n
0
yx
kt
2
c
st
a
4
_ _
2
E
0
x
1 n
0
xy
n
0
yx
2
kt
2
c
st
a
1
;
n
xy
=
n
0
xy
E
0
y
1 n
0
xy
n
0
yx
kt
2
c
st
a
4
E
0
y
1 n
0
xy
n
0
yx
2
kt
2
c
st
a
2
;
G
xy
= G
0
xy
2
kt
2
c
st
a
3
:
(5:58)
Evaluation of material constants
In the damagestiffness relations Eq. (5.56) and Eq. (5.58), a
i
,i = 1,2,3,4 are a set of
four phenomenological constants, which need to be determined to predict stiffness
degradation. As seen from Eq. (5.58), the shear modulus is uncoupled from the
other three moduli and thus can be treated independently. These phenomeno-
logical constants are material and laminate conguration dependent and can be
evaluated for a selected laminate by using data generated either experimentally or
by an analytical or a computational model. As an example, let the moduli of a
given damaged cross-ply laminate at a xed state of damage, s = s
1
, be given as
E
x
; E
y
; G
xy
; and n
xy
, then by using Eq. (5.56), we obtain
a
1
=
s
1
t
2kt
2
c
E
x
1 n
xy
n
yx
E
0
x
1 n
0
xy
n
0
yx
_ _
;
a
2
=
s
1
t
2kt
2
c
E
y
1 n
xy
n
yx
E
0
y
1 n
0
xy
n
0
yx
_ _
;
a
3
=
s
1
t
2kt
2
c
G
xy
G
0
xy
_ _
;
a
4
=
s
1
t
kt
2
c
n
xy
E
y
1 n
xy
n
yx
n
0
xy
E
0
y
1 n
0
xy
n
0
yx
_ _
:
(5:59)
In the above expressions it should be noted that while the values of a
i
are xed for
a given composite laminate (that has been homogenized), the parameter k depends
on the ability of the cracks to perform surface displacements under an applied
mechanical impulse. Thus this parameter may be viewed as a measure of the
constraint to the crack surface separation imposed by the material surrounding
the crack. One way to view this is by considering a crack of a given size embedded
in an innite isotropic material, in which case the crack surface separation is
unconstrained and can be calculated by fracture mechanics methods. When the
laminate geometry is nite and its symmetry is different from isotropic, the k
parameter will take a value less than that for the innite isotropic medium. This
consideration allows us to assign k an undetermined value, say k
0
, for a reference
laminate under reference loading conditions, and evaluate a change from this
value for another crack orientation. This procedure will be discussed in more
detail later.
The CDM approach described here has been used to successfully predict
degradation in the longitudinal and transverse moduli and the Poissons ratio
for a variety of laminates, e.g., [0/90
3
]
s
, [90
3
/0]
s
, and [0/45]
s
as reported in [1315,
17, 23]. Figure 5.7 shows the degradation of the longitudinal Youngs modulus
in a [0/90
3
]
s
glass/epoxy laminate with the applied tensile stress. The predictions
agree with the observed values fairly well in the entire range of cracking.
The predictions by the ply discount method are found to overestimate the total
modulus reduction. The stressstrain curve constructed from the reduced modulus
is shown in Figure 5.8.
0
0.5
0.6
0.7
0.8
0.9
1.0
E
1
/E
1
0
100 200 300
Gl./Ep. [0/90
3
]
s
Calculated
Measured
Ply discount prediction
s
(MPa)
x
x
x
x
x
x
x
x
x x
x
x
x
x
Figure 5.7. Variation of the longitudinal Youngs modulus with applied stress for a glass/
epoxy [0/90
3
]
s
laminate. Source: [23].
5.3 Synergistic damage mechanics (SDM)
The observation that the k-parameter (hitherto referred to as constraint param-
eter) may be viewed as a carrier of the local effects on damage entities within a
RVE, while the a
i
-constants are material constants, led to a number of studies to
explore prediction of elastic property changes due to damage in different modes.
To be sure, the elastic properties are the averages over appropriate RVEs.
At rst it was found that from changes in E
x
and n
xy
due to transverse cracking
in [0/90
3
]
s
glass/epoxy laminates reported in [24] and assuming no changes in E
y
,
changes in E
x
for the same glass/epoxy of [0/90]
s
conguration could be predicted
with good accuracy. Also, in [0/45]
s
laminates of the same glass/epoxy, the
change in E
x
could be predicted by setting D
1
= D
2
(a good approximation,
supported by crack density data). These results have been reported in [14].
Later, a systematic study of the effect of constraint on the constraint parameter
was done by experimentally measuring the crack opening displacement (COD)
in [y/90
2
]
s
laminates [25] for different y-values. By relating these values to the
COD at y = 90
normalized by a unit applied strain, the predictions of E

x
and n
xy
for different y could be made. Another study of the constraint effects was made
by examining [0/y
4
/0
1/2
]
s
laminates, where the ply orientation y was varied.
Once again, using experimentally measured COD for y = 90
as the reference,
the k-parameter for other ply orientations was evaluated from the COD values
and E
x
and n
xy
for different y were then predicted [33].
While the experimental studies supported the idea of using the constraint
parameter as a carrier of local constraints, the scatter in test data and the cost of
0 0.5 1.0 1.5 2.0 2.5 e (%)
0
50
100
150
200
250
s
(MPa)
Gl./Ep. [0/90
3
]
s
Figure 5.8. Longitudinal stressstrain response for a glass/epoxy [0/90
3
]
s
laminate.
Source: [23].
155 5.3 Synergistic damage mechanics (SDM)
testing do not make the experimental approach attractive. Therefore, another
systematic study of [0
m
/y
n
/0
m/2
]
s
laminates was undertaken [28, 29] where com-
putational micromechanics was employed instead of physical testing. An elaborate
parametric study of the constraint parameter allowed developing a master curve
for elastic property predictions. The most recent study [30] examines damage
modes consisting of transverse ply cracks as well as inclined cracks of different
orientations in [0
m
/y
n
/90
r
]
s
and [0
m
/90
r
/y
n
]
s
laminates. The reference value of k
for y = 90
, as well as changes in it for other ply orientations, are computed by

nite element models of representative unit cells. Predictions made using k as a
micromechanical carrier of local damage-induced perturbations are compared
with available experimental data for [0/90/45/+45]
s
laminates. This approach
of combining MIDM (local) and MADM (global) approaches is named synergistic
damage mechanics (SDM). The following sections describe the use of SDM. First
we take multidirectional laminates with cracking in two symmetric off-axis orien-
tations, followed by the case of cracking in three orientations.
5.3.1 Two damage modes
Let us consider a case where two modes of damage are active. In this case, the
irreducible integrity bases for c
P
are given by, with a = 1 and 2,
e
11
; e
22
; e
33
; e
2
23
; e
2
31
; e
2
12
; e
23
e
31
e
12
;
D
(1)
11
; D
(1)
22
; D
(1)
33
; D
(1)
23
_ _
2
; D
(1)
31
_ _
2
; D
(1)
12
_ _
2
; D
(1)
23
D
(1)
31
D
(1)
12
;
D
(2)
11
; D
(2)
22
; D
(2)
33
; D
(2)
23
_ _
2
; D
(2)
31
_ _
2
; D
(2)
12
_ _
2
; D
(2)
23
D
(2)
31
D
(2)
12
;
e
23
D
(1)
23
; e
31
D
(1)
31
; e
12
D
(1)
12
; e
23
D
(2)
23
; e
31
D
(2)
31
; e
12
D
(2)
12
;
e
31
e
12
D
(1)
23
; e
12
e
23
D
(1)
31
; e
23
e
31
D
(1)
12
; e
31
e
12
D
(2)
23
; e
12
e
23
D
(2)
31
; e
23
e
31
D
(2)
12
;
e
23
D
(1)
31
D
(1)
12
; e
31
D
(1)
12
D
(1)
23
; e
12
D
(1)
23
D
(1)
31
; e
23
D
(2)
31
D
(2)
12
; e
31
D
(2)
12
D
(2)
23
; e
12
D
(2)
23
D
(2)
31
:
(5:60)
For a thin laminate loaded in its plane, the above set can be reduced by consider-
ing only the in-plane strain and damage tensor components. Thus, the remaining
integrity bases in the Voigt notation are given by
e
1
; e
2
; e
2
6
;
D
(1)
1
; D
(1)
2
; D
(1)
6
_ _
2
; D
(2)
1
; D
(2)
2
; D
(2)
6
_ _
2
;
e
6
D
(1)
6
; e
6
D
(2)
6
:
(5:61)
Using the above set of integrity bases, the most general polynomial form for rc,
restricted to second-order terms in the strain components (assuming small strains)
and rst-order terms in damage tensor components (assuming small volume
fraction or number density of damage entities in the RVE), is given by
rc = P
0
c
1
e
2
1
c
2
e
2
2
c
3
e
2
6
c
4
e
1
e
2
_ _
e
2
1
c
5
D
(1)
1
c
6
D
(1)
2
c
7
D
(2)
1
c
8
D
(2)
2
_ _
e
2
2
c
9
D
(1)
1
c
10
D
(1)
2
c
11
D
(2)
1
c
12
D
(2)
2
_ _
e
2
6
c
13
D
(1)
1
c
14
D
(1)
2
c
15
D
(2)
1
c
16
D
(2)
2
_ _
e
1
e
2
c
17
D
(1)
1
c
18
D
(1)
2
c
19
D
(2)
1
c
20
D
(2)
2
_ _
e
1
e
6
c
21
D
(1)
6
c
22
D
(2)
6
_ _
e
2
e
6
c
23
D
(1)
6
c
24
D
(2)
6
_ _
P
1
e
p
; D
(1)
q
_ _
P
2
e
p
; D
(2)
q
_ _
P
3
D
(1)
q
_ _
P
4
D
(2)
q
_ _
;
(5:62)
where P
0
and c
i
, i = 1, 2, . . ., 24 are material constants, P
1
and P
2
are linear
functions of strain and damage tensor components, and P
3
and P
4
are linear
functions only of the damage tensor components. Setting rc = 0 for unstrained
and undamaged material, we have P
0
= 0; and assuming the unstrained material of
any damaged state to be stress free, we get P
1
= P
2
= 0 on using Eq. (5.48).
Considering the virgin material to be orthotropic and proceeding in a similar
manner as above, we obtain the following relations for the stiffness matrix of
the damaged laminate
C
pq
= C
0
pq
C
(1)
pq
C
(2)
pq
; (5:63)
where p, q = 1, 2, 6; C
pq
0
is the stiffness coefcient matrix of the virgin laminate
given by Eq. (5.52), and the changes in stiffness brought about by the individual
damage modes are represented by C
pq
(1)
and C
pq
(2)
, which are given by
C
(1)
pq
=
2c
5
D
(1)
1
2c
6
D
(1)
2
c
17
D
(1)
1
c
18
D
(1)
2
c
21
D
(1)
6
2c
9
D
(1)
1
2c
10
D
(1)
2
c
23
D
(1)
6
Symm 2c
13
D
(1)
1
2c
14
D
(1)
2
_
_
_
_
C
(2)
pq
=
2c
7
D
(2)
1
2c
8
D
(2)
2
c
19
D
(2)
1
c
20
D
(2)
2
c
22
D
(2)
6
2c
11
D
(2)
1
2c
12
D
(2)
2
c
24
D
(2)
6
Symm 2c
15
D
(2)
1
2c
16
D
(2)
2
_
_
_
_
:
(5:64)
In general, for N damage modes, Eq. (5.63) can be written as
C
pq
= C
0
pq

N
a=1
C
(a)
pq
: (5:65)
Let us now consider a special case of a general laminate undergoing damage in two
symmetrically placed damage modes, such as 0
m
=y
n
=
p
_
s
, with restricted to
angles that do not cause ply cracking. In such laminates, an in-plane tensile
loading will produce an in-plane stress state in each off-axis ply consisting
of normal stresses along and perpendicular to bers in that ply and a shear stress
in the plane of the ply. Depending on the values of y, , and ply properties,
the stress perpendicular to the bers could be tensile or compressive. Thus,
on loading, an off-axis ply may or may not develop intralaminar cracks. When
y = 90
, the matrix will undergo multiple cracking in the transverse plies.

For other cases of off-axis ply orientations, multiple cracking is typically observed
to occur for angles from 50
to 90
under an axial tensile load. However, it has

been observed that even in cases where these cracks do not initiate in the off-axis
plies, the laminate moduli change with the applied load due to shear stress-induced
damage within the plies.
The damage state subsequent to ply cracking in the +y and y plies can be
represented by two damage mode tensors. For off-axis ply cracking, it is more
convenient to rewrite the damage mode tensor dened in (5.44) in terms of the
normal crack spacing, s
y
n
= s
y
sin y, where s
y
is the crack spacing in the axial
direction (see Figure 5.9) for a ply of orientation y. Accordingly, the damage
mode tensors are given by
D
(a)
ij
=
kt
2
c
s
y
n
t
n
i
n
j
: (5:66)
With reference to Figure 5.9(b) where the orientations of the two damage modes
are shown, the damage mode elements are given by
(a)
(b)
Figure 5.9. Damage characterization for two damage modes: (a) normal crack spacing s
y
n
,
and axial crack spacing s
y
in a cracked ply; (b) directions of normal vectors for cracks
in +y- and y-plies.
a = 1 : n
(1)
i
= sin y; cos y; 0 ( ) ;
D
(1)
1
=
k
y
t
2
c
s
y
n
t
sin
2
y; D
(1)
2
=
k
y
t
2
c
s
y
n
t
cos
2
y; D
(1)
6
=
k
y
t
2
c
s
y
n
t
sin y cos y;
a = 2 : n
(2)
i
= sin y; cos y; 0 ( );
D
(2)
1
=
k
y
t
2
c
s
y
n
t
sin
2
y; D
(2)
2
=
k
y
t
2
c
s
y
n
t
cos
2
y; D
(2)
6
=
k
y
t
2
c
s
y
n
t
sin y cos y;
(5:67)
where the superscripts y
+
and y
indicate variables for the +y and y plies,

respectively. Assuming that the intensity and distribution of damage is the same
in both +y and y plies, we have
k
y
= k
y
= k
y
; s
y
n
= s
y
n
= s
y
n
: (5:68)
Substituting Eq. (5.67)(5.68) into Eq. (5.64), we obtain
C
(1)
11
C
(2)
11
= 2
k
y
t
2
c
s
y
n
t
c
5
c
7
( )sin
2
y c
6
c
8
( )cos
2
y
_
;
C
(1)
22
C
(2)
22
= 2
k
y
t
2
c
s
y
n
t
c
9
c
11
( )sin
2
y c
10
c
12
( )cos
2
y
_
;
C
(1)
66
C
(2)
66
= 2
k
y
t
2
c
s
y
n
t
c
13
c
15
( )sin
2
y c
14
c
16
( )cos
2
y
_
;
C
(1)
12
C
(2)
12
=
k
y
t
2
c
s
y
n
t
c
17
c
19
( )sin
2
y c
18
c
20
( )cos
2
y
_
;
C
(1)
16
C
(2)
16
=
k
y
t
2
c
s
y
n
t
sin y cos y c
21
c
22
[ [ = 0;
C
(1)
26
C
(2)
26
=
k
y
t
2
c
s
y
n
t
sin y cos y c
23
c
24
[ [ = 0 :
(5:69)
Thus,
C
(1)
pq
C
(2)
pq
=
2a
1
D
1
2b
1
D
2
a
4
D
1
b
4
D
2
0
2a
2
D
1
2b
2
D
2
0
Symm 2a
3
D
1
2b
3
D
2
_
_
_
_; (5:70)
where the superscripts for denoting damage mode have beendroppedfor convenience,
and a
i
and b
i
, i = 1, 2, 3, 4 are the two sets of four material constants, given by
a
1
= c
5
c
7
; a
2
= c
9
c
11
; a
3
= c
13
c
15
; a
4
= c
17
c
19
;
b
1
= c
6
c
8
; b
2
= c
10
c
12
; b
3
= c
14
c
16
; b
4
= c
18
c
20
:
(5:71)
Here, a
i
and b
i
are functions of y. Denote
a
1
y ( ) = a
1
sin
2
y b
1
cos
2
y;
a
2
y ( ) = a
2
sin
2
y b
2
cos
2
y;
a
3
y ( ) = a
3
sin
2
y b
3
cos
2
y;
a
4
y ( ) = a
4
sin
2
y b
4
cos
2
y :
(5:72)
Then,
C
(1)
pq
C
(2)
pq
=D
y
2a
1
y ( ) a
4
y ( ) 0
2a
2
y ( ) 0
Symm 2a
3
y ( )
_
_
_
_; (5:73)
where
D
y
=
k
y
t
2
c
s
y
n
t
: (5:74)
Rewrite Eq. (5.72) as
a
i
y ( ) = a
i
sin
2
y b
i
cos
2
y = a
i
sin
2
y 1
b
i
a
i
cot
2
y
_ _
: (5:75)
Consider for the moment the case when a
i
_ b
i
. Then,
b
i
a
i
cot
2
y _ 1 for
p
4
_ y _
p
2
: (5:76)
Also, it can be expected that
b
i
a
i
cot
2
y 1 for
p
3
_ y _
p
2
;
i:e:; a
i
y ( ) ~ a
i
for
p
3
_ y _
p
2
:
(5:77)
For this case, we have laminate stiffness matrix as
C
pq
=
E
0
x
1 n
0
xy
n
0
yx
n
0
xy
E
0
y
1 n
0
xy
n
0
yx
0
E
0
y
1 n
0
xy
n
0
yx
0
Symm G
0
xy
_
_
_
_
D
y
2a
1
a
4
0
2a
2
0
Symm 2a
3
_
_
_
_
; (5:78)
which is of identical form to the expression for 90
cracking in a cross-ply
laminate, see Eq. (5.56). The moduli can be nally obtained using the relations
in Eq. (5.57). The resultant expressions will be of the same form as in Eq. (5.58),
except that kt
c
2
/st is now replaced with D
y
.
The overall SDM procedure for this laminate is sketched in Figure 5.10.
As illustrated, it combines micromechanics with CDM for complete evaluation
of the structural response. Micromechanics involves analysis to determine CODs
in cracked plies within a RVE (or unit cell, if applicable) of [0
m
/y
n
/0
m/2
]
s
layup,
from which the constraint effect is evaluated for different y and/or m.
The constraint effect is carried over in the CDM formulation through the con-
straint parameter. In a separate step, the damage constants a
i
are determined
from experimental data for a reference laminate, which is chosen here to be
[0/90
8
/0
1/2
]
s
. A cross-ply laminate of the same class as the ply layup and material
in consideration is a good choice for the reference laminate because either the
experimental data are often available or can be obtained by using any of the
damage models for 90
-ply cracking as described in the previous chapter. For

evaluation of damage constants, expressions in Eq. (5.59) shall be used after
replacing kt
c
2
/st by D
y
. With the values of the damage constants and b = k
y
=k
90
known, the stiffnessdamage relations given by Eq. (5.78) are employed to predict
stiffness degradation with crack density. The overall structural behavior can be
nally analyzed using the reduced stiffness properties for the laminate.
For computation of average COD values, micromechanics can be performed
using a suitable FE analysis. The 3-D FE model for [0
m
/y
n
/0
m/2
]
s
layup, along
COMPUTATIONAL MICROMECHANICS
SYNERGISTIC DAMAGE MECHANICS
STRUCTURAL ANALYSIS
Analyze overall structural response to external
loading using the reduced stiffness properties
Use SDM to determine stiffness reduction in
present laminate configuration [0
m
/ q
n
/0
m/2
]
s
Structural scale: Meso
Structural scale: Macro
Structural scale: Micro
EXPERIMENTAL/
COMPUTATIONAL
Determine COD and constraint parameter(s)
; = = +
(
( )
(
(
)
u
2
u
2
u
2
n
=

q
90
q
n q
n +q
n
q
b
Evaluate damage constants
using available data
for reference laminate
configuration [0/90
8
/0
1/2
]
s
90
8
( )
u
2 ( )
u
2
Figure 5.10. Flowchart showing the multiscale synergistic methodology for analyzing
damage behavior in a class of symmetric laminates with layup [0
m
/y
n
/0
m/2
]
s
containing
ply cracks in the +y and y layers.
with boundary conditions and coordinate systems, is shown in Figure 5.11 [29, 31,
32]. As can be seen from Figure 5.12, the computed CODs averaged over the
thickness of the cracked plies agree quite well with the experimental data over
Figure 5.11. Representative unit cell for COD computation for laminates. Reprinted, with
kind permission, from Int J Solids Struct, Vol. 45, C.V. Singh and R. Talreja, Analysis of
multiple off-axis ply cracks in composite laminates, pp. 457489, copyright Elsevier (2008).
20 30 40 50 60 70 80 90
Ply orientation (q)
0
2
4
6
8
C
O
D

(
m
)
FEM
Experiment
Figure 5.12. Variation of average COD with change in ply orientation for a [0
m
/y
n
/0
m/2
]
s
glass/epoxy laminate. Reprinted, with kind permission, from Int J Solids Struct, Vol. 45,
C.V. Singh and R. Talreja, Analysis of multiple off-axis ply cracks in composite laminates,
the whole range of ply orientations considered. This suggests that the 3-DFEanalysis
is an accurate tool to evaluate the constraint parameter. The proles of COD nor-
malized with t
c
through the ply thickness are shown in Figure 5.13. As expected, for
cross-ply laminates, the prole is symmetric about mid-plane of the cracked ply and
consequently the maximum COD occurs at the mid-plane of the cracked layer.
However, due to the difference in constraint from the surrounding material, this
COD prole is different from an elliptical prole for a single crack in an innite
isotropic elastic medium subjected to a uniform far-eld stress (see Figure 5.13 (b)).
Thus, the magnitude of the average COD for a 90
crack is different fromthat for an

elliptical crack. For other ply orientations, the crack surface displacements are not
symmetric about the mid-plane as shown in Figure 5.13 (c)(d). This asymmetry
increases as we go away fromcross-ply laminates (y =90
) and maximumCODdoes
not occur midway through the thickness of the cracked y layers. The aspect ratio of
COD proles, g = u
2
( )
max
_
u
2
, varies from 1.33 for y = 90
to 1.40 for y = 40
(Figure 5.13 (a)), which is different from the aspect ratio of 1.273 ( = 4/p) for an
(a)
1
0
3
u
2
/
t
c
(b)
1
0
3
u
2
/
t
c
(c)
1
0
3
u
2
/
t
c
(d)
1
0
3
u
2
/
t
c
Figure 5.13. COD proles for cracked plies in [0/y
4
/0
1/2
]
s
laminates: (a) CODs averaged
over +y- and y-plies; (b) Crack prole for 90-ply crack compared with an elliptic prole
for an isotropic medium; (c), (d) COD proles for +y and y separately: (c) y = 70,
(d) y = 40. Parts (c), (d) depict the asymmetry of opening displacements for off-axis
laminates, especially at a ply orientation farther from y = 90
. Reprinted, with kind

permission, from Int J Solids Struct, Vol. 45, C.V. Singh and R. Talreja, Analysis of multiple
off-axis ply cracks in composite laminates, pp. 457489, copyright Elsevier (2008).
elliptic prole. Here (Du
2
)
max
represents the maximum COD and u
2
represents the
CODaveragedover the thickness direction. Numerically, the COD, Du
2
, is computed
by taking the difference of the node displacements on either side of the crack along a
direction transverse to the local ber direction (assuming the crack traverses parallel
to the ber).
The elastic moduli E
x
and n
xy
predicted by the SDM approach are compared
with the experimental data for y = 70
and 55
in Figure 5.14. The agreement with

data is about the same as that obtained by using experimentally measured COD
values to evaluate the constraint parameter. For y = 55
, the stiffness reduction

is caused by matrix cracking as well as shear-induced damage in off-axis plies, as
discussed in [33], where a procedure for calculating the stiffness reduction due
to shear damage was described. SDM predictions for this orientation thus
include both effects. It is interesting to note that the stiffness predictions using
experimentally measured COD values are subject to scatter, which is inherent in
testing. Furthermore, the experimental procedure requires a specialized test set-up
[33, 34], which is costly and takes a certain amount of training to operate. The 3-D
FE computations on a unit cell, on the other hand, are easy to perform and can
provide accurate values of CODs.
The SDM approach is quite handy when the relative stiffness or thickness
of cracked and supporting plies change. The difference in stiffness changes for
0
x
E
E
x
0
x
E
E
x
(a)
=70
(b)
=70
n
xy
n
xy
0
n
xy
n
xy
0
=55
(d)
(c)
=55
Figure 5.14. Variation of longitudinal modulus and Poissons ratio for [0/y
4
/0
1/2
]
s
laminates: (a), (b) for y = 70
; (c), (d) for y = 55
. Reprinted, with kind permission, from

Int J Solids Struct, Vol. 45, C.V. Singh and R. Talreja, Analysis of multiple off-axis ply
cracks in composite laminates, pp. 457489, copyright Elsevier (2008).
laminates with different ply thickness or axial stiffness are due to difference in the
relative constraint from the supporting plies, and thus can be characterized
by the changes in the constraint parameter. In [29] a parametric study was
conducted to study the effect of changes in ply thicknesses (m and n) and
axial stiffness ratio r = E
y
A
_
E
90
A
over CODs of [0
m
/y
n
/0
m/2
]
s
laminates, where E
A
y
and E
A
90
represent the axial stiffness of y-plies and the 90
-plies, respectively.
Figure 5.15 shows the variation of average COD for changes in these parameters.
The computed average CODs are given by the following parametric equation
u
2
_ _
y
n
= U:f
1
y ( ):f
2
r ( ):f
3
m ( ):f
4
n ( ); (5:79)
where U is the average COD for the reference laminate [0/90
8
/0
1/2
]
s
, and
f
1
y ( ) = sin
2
y;
f
2
r ( ) = r
c
1
;
f
3
m ( ) =
c
2
m
c
3
;
f
4
n ( ) = c
4
n
c
5
;
(5:80)
are tted functions where tting constants (unitless) c
1
c
5
are given as: c
1
=
0.0871; c
2
= 0.1038; c
3
= 0.8949; c
4
= 0.247; c
5
= 0.99 for [0
m
/y
n
/0
m/2
]
s
glass/
epoxy laminates.
The parametric study described above enables us to predict stiffness degrad-
ation in off-axis laminates with different geometry and stiffness values. For
example, one can consider a laminate with stiffer outer plies. The variation of
engineering moduli E
x
and n
xy
for different stiffness ratios (r) for a [0/70
4
/0
1/2
]
s
laminate is shown in Figure 5.16(a). As expected, stiffer outer plies cause less
severe degradation in the moduli. In contrast to changing the stiffness of outer
plies, one can vary the number of constraining plies (m) or cracked plies (n). The
effect of number of cracked plies on the change in stiffness moduli is shown in
Figure 5.16(b). These results indicate that the cracking ply thickness, i.e., crack
size, has signicant effect on stiffness degradation while the thickness of the
constraining plies as well as the change in axial stiffness ratio r have small effect.
5.3.2 Three damage modes
We will now consider a special case of cracking in three off-axis plies, of which two
are symmetrically opposite of each other, and the third is placed along the
transverse direction. Thus cracking is in the +y, y, and 90
plies. For this case

the stiffness matrix of the damaged laminate is given by
C
pq
= C
0
pq
C
(1)
pq
C
(2)
pq
C
(3)
pq
; (5:81)
where C
pq
0
is given by Eq. (5.52), and C
pq
(1)
+C
pq
(2)
is givenby Eq. (5.73). The components
of the third damage mode (a = 3) corresponding to cracking in 90
-ply are
(b)
(a)
(c)
Figure 5.15. Variation of average COD for [0
m
/y
n
/0
m/2
]
s
glass/epoxy laminates with
(a) axial stiffness ratio, r (for m = 1, n = 4); (b) number of cracked plies, n; and
(c) number of constraining plies, m. Reprinted, with kind permission, from Int J Solids
Struct, Vol. 45, C.V. Singh and R. Talreja, Analysis of multiple off-axis ply cracks in
composite laminates, pp. 457489, copyright Elsevier (2008).
D
(3)
1
=
k
90
t
2
90
s
90
t
; D
(3)
2
= D
(3)
6
= 0: (5:82)
The integrity bases (Eq. (5.61)) have additional components for D
(3)
1
. The free
energy function thus gets the following terms added to Eq. (5.62)
rc a = 3 ( ) = a
/
1
e
2
1
D
(3)
1
a
/
2
e
2
2
D
(3)
1
a
/
3
e
2
6
D
(3)
1
a
/
4
e
1
e
2
D
(3)
1
; (5:83)
where a
/
i
; i = 1; 2; 3; 4 are additional material constants. Substituting Eq. (5.83)
into Eq. (5.48), we obtain
C
(3)
pq
=D
(3)
1
2a
/
1
a
/
4
0
2a
/
2
0
Symm 2a
/
3
_
_
_
_; (5:84)
where the contribution to the shear components is zero.
It is important to emphasize here that the relative location of different
damage modes in the whole laminate will cause different losses in stiffness due to
damage in the laminate. To illustrate this let us consider two specic examples of
laminates with damage modes consideredinthe present section, viz., +y, y, and 90
.
Figure 5.17(a)(b) shows representative unit cells for laminates with [0
m
/y
n
/90
r
]
s
(a)
(c) (d)
(b)
1.2
1
0.8
0.6
0.4
0.2
0 0.2 0.4 0.6
r =0.5
r =1
r =2
n=1
n=2
n=4
n=1
n=2
n=4
r =5
r =0.5
r =1
r =2
r =5
0.8
Crack density (1/mm)
1 1.2 1.4 1.6 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
0
1.2
1
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
1.2
1
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
1.2
1
0.8
0.6
0.4
0.2
0
E
x
n
xy
n
xy
n
xy
E
x
0
E
x
E
x
0
0
n
xy
0
Figure 5.16. Effect of change in axial stiffness ratio (r): (a), (b), and cracked ply thickness (n):
(c), (d), on stiffness degradation in [0
m
/70
n
/0
m/2
]
s
glass/epoxy laminates. Reprinted, with
kind permission, from Int J Solids Struct, Vol. 45, C.V. Singh and R. Talreja, Analysis of
multiple off-axis ply cracks in composite laminates, pp. 457489, copyright Elsevier (2008).
and [0
m
/90
r
/y
n
/+y
n
]
s
congurations, respectively. The boundary conditions are
shown for the 3-D FE computation of average COD values. The global laminate
(X, Y, Z) and local crack plane (x
1
, x
2
, x
3
) coordinate systems are also shown.
Considering rst the case of a [0
m
/y
n
/90
r
]
s
laminate, we note that y modes occur
twice in the whole laminate, above and belowthe mid-plane of the laminate, whereas
90
mode occurs only once. Thus, C

pq
= C
pq
C
0
pq
=
N
a=1
C
(a)
pq
is given by
C
pq
= 2 C
(1)
pq
y ( ) C
(2)
pq
y ( )
_ _
C
(3)
pq
90 ( ) : (5:85)
Collecting terms from Eq. (5.73) and Eq. (5.84), while assuming the a
i
s to be
independent of y, we get
C
pq
= 2D
y
2a
1
a
4
0
2a
2
0
Symm 2a
3
_
_
_
_
D
90
2a
/
1
a
/
4
0
2a
/
2
0
Symm 2a
/
3
_
_
_
_
; (5:86)
where, for the laminate conguration considered,
D
y
=
k
y
2nt
0
( )
2
s
y
n
t
; D
90
=
k
90
2rt
0
( )
2
s
90
t
; (5:87)
where t
0
denotes the thickness of a single ply. The special case when y = 90
, and
DC
pq
as given by Eq. (5.86) should be equal to that given by a single 90
mode with
crack size of (4n+2r) plies. Consider, for example, the DC
11
term. If we assume
that the normal crack spacing is the same in all cracked plies, then DC
11
for a
[0
m
/y
n
/90
r
]
s
laminate at y = 90
from Eq. (5.86) is given by

C
11
= 2 C
(1)
11
C
(2)
11
_ _
C
(3)
11
= 4D
y
[
y=90
a
1
90 ( ) 2D
90
a
/
1
=
4k
y
[
y=90
2nt
0
( )
2
s
90
t
a
1

2k
90
2rt
0
( )
2
s
90
t
a
/
1
=
8t
2
0
s
90
t
2n
2
k
y
[
y=90
a
1
90 ( ) r
2
k
90
a
/
1
_
:
(5:88)
(a)
(symmetry about mid-plane)
0 layer
+q layer
(u)
x=21
=u
0
q layer
y
x
z
90 layer
(w)
z =0
=0
(u)
x=0
=0
(b)
(symmetry about mid-plane)
0 layer
+q layer
(u)
x=21
=u
0
(u)
x=0
=0
q layer
y
x
z
90 layer
(w)
z=0
=0
Figure 5.17. Representative unit cell for (a) [0
m
/y
n
/90
r
]
s
and (b) [0
m
/90
r
/y
n
]
s
laminate
congurations.
Since, for y = 90
, [0
m
/y
n
/90
r
]
s
is equivalent to [0/90
2n+r
]
s
, we can consider their
stiffness changes to be the same. Thus, DC
11
can also be written, using Eq. (5.84), as
C
11
= C
(3)
11
= 2a
/
1
D
1
=
k
90
4n2r
4n 2r ( )t
0
[ [
2
s
90
t
:2a
/
1
; (5:89)
where the sub-subscript 4n+2r in k
90
4n2r
represents the crack size for the 90
mode
in a [0/90
2n+r
]
s
laminate. Equating DC
11
from Eq. (5.88) and Eq. (5.89), we have
2n
2
k
y
[
y=90
a
1
90 ( ) r
2
k
90
a
/
1
= k
90
4n2r
2n r ( )t
0
[ [
2
a
/
1
; (5:90)
i.e.,
a
1
90 ( ) =
k
90
4n2r
2n r ( )
2
r
2
k
90
2n
2
k
y
[
y=90
a
/
1
: (5:91)
Generalizing, we can write the interrelationship between two sets of constants as
a
i
=
k
90
4n2r
2n r ( )
2
r
2
k
90
2n
2
k
y
[
y=90
a
/
i
i = 1; 2; 3; 4: (5:92)
Substituting Eq. (5.92) into Eq. (5.86), DC
pq
for a damaged [0
m
/y
n
/90
r
]
s
laminate
is given by
C
pq
=

D
2a
/
1
a
/
4
0
2a
/
2
0
Symm 2a
/
3
_
_
_
_; (5:93)
where
D =
4t
2
0
t
1
s
y
n
k
y
k
y
[
y=90
2n r ( )
2
k
90
4n2r
r
2
k
90
_ _
r
2
k
90
s
90
_ _
; (5:94)
where the constraint parameters are given by
k
y
=
u
y
_ _
y
2n
2nt
0
; k
90
4n2r
=
u
y
_ _
90
4n2r
4n 2r ( )t
0
; k
90
=
u
y
_ _
90
2r
2rt
0
: (5:95)
Consider now the case of the [0
m
/90
r
/y
n
]
s
laminate conguration. It must be
noted that, unlike [0
m
/y
n
/90
r
]
s
laminates, y damage modes in this case are
centrally placed, thereby the corresponding equivalent crack size is 4nt
0
(with
averaging over two +y and two y layers). On the other hand, the crack size for
the 90
damage mode is rt
0
. The derived stiffnessdamage relationships retain the
form of Eq. (5.93). However,

D in this case is given by
D =
2t
2
0
t
1
s
y
n
k
y
k
y
[
y=90
2 2n r ( )
2
k
90
4n2r
r
2
k
90
_ _
r
2
k
90
s
90
_ _
(5:96)
and the corresponding constraint parameters are now given by
k
y
=
u
y
_ _
y
2n
4nt
0
; k
90
4n2r
=
u
y
_ _
90
4n2r
4n 2r ( )t
0
; k
90
=
u
y
_ _
90
r
rt
0
: (5:97)
The SDM procedure for this laminate conguration is quite similar to that
described earlier for the [0
m
/y
n
/0
m/2
]
s
layup in Figure 5.10. The damage constants
are evaluated from Eq. (5.59) after replacing kt
c
2
/st by

D by using the stiffness
degradation data for the reference laminate, which is chosen as [0/90
3
]
s
in this case.
Depending on the specic 90
-ply placement, the corresponding constraint par-

ameters are calculated using expressions in Eq. (5.95) or Eq. (5.97) by using the
CODs computed from a suitable 3-D FE model. FE modeling details and COD
computations are described in detail in [30, 32]. One important consideration in
COD computation is the interaction between the families of cracks in different
orientations which can modify the crack displacements and consequently the
stiffness changes due to ply cracking [30]. This interaction can also modify the
damage evolution as illustrated in the next chapter. In Figure 5.18, SDM predic-
tions of stiffness moduli of [0/70/90]
s
and [0/55/90]
s
glass/epoxy laminates are
compared with the independent calculations from 3-D FE analysis (see [3032]
for details on procedure to calculate stiffness changes using 3-D FE analysis).
The SDM predictions for a quasi-isotropic laminate are shown in Figure 5.19.
In the experiments it was observed that the laminates failed before the cracks in
the plies could grow fully through the laminate width. The effect of partially
grown cracks on stiffness changes is accounted for in the SDM analysis by
reducing the crack density for layers by a relative density factor, dened as
r
r
=
Actual surface area for partial cracks
Surface area for full cracks
: (5:98)
To nd the actual surface area of partial cracks, the information regarding their
actual length (along lamina width) is necessary. Since such data were not reported
in the above experimental study [35], the results are presented for two assumed
values of r
r
: 0.25 and 0.5.
5.4 Viscoelastic composites with ply cracking
Polymers used as matrix materials in ber-reinforced composites usually display
viscoelastic behavior. However, the time-dependent behavior of the matrix does
not translate fully in the composite and while it is isotropic in the polymer, the
directionality of the bers makes it anisotropic in the composite. Viscoelasticity
models for composite materials, particularly for high temperature applications
where the time dependency is more pronounced, are needed. The composite time-
dependent response in most cases can be approximated by linear viscoelasticity.
When ply cracking occurs the viscoelastic response can still be linear but with
modied time-dependency.
=55 =55
=70
(a)
=70
(b)
Figure 5.18. Variation of longitudinal modulus and Poissons ratio for [0/y/90]
s
laminates:
(a) y = 70
, (b) y = 55
. The results are compared with independent FE calculations. SDM

predictions are made for two cases: no interaction between y and 90
cracks, and with

maximum interaction between them. Reprinted, with kind permission, from Mech Mater,
Vol. 41, C.V. Singh and R. Talreja, A synergistic damage mechanics approach for composite
laminates with matrix cracks in multiple orientations, pp. 95468, copyright Elsevier (2009).
(a) (b)
1.1
1
0.9
0.8
0.7
0.6
0.5
0.4
1.1
1
0.9
0.8
0.7
0.6
0.5
0.4
0 0.2 0.4 0.6 0.8 1 1.2
90-Crack density (1/mm)
0 0.2 0.4 0.6 0.8 1 1.2
90-Crack density (1/mm)
+45 cracks +45 cracks
45 cracks 45 cracks
[0/90/ 45]
s
[0/90/ 45]
s
[0/90]
s
[0/90]
s
E
x
E
0
x
Test 1: [0/90/45/+45]
s
Test 2: [0/90/45/+45]
s
Test 3: [0/90/45/+45]
s
Test 1: [0/90]
s
Test 2: [0/90]
s
Test 3: [0/90]
s
Test 4: [0/90]
s
Test 1: [0/90/45/+45]
s
Test 2: [0/90/45/+45]
s
Test 3: [0/90/45/+45]
s
Test 1: [0/90]
s
Test 2: [0/90]
s
Test 3: [0/90]
s
n
xy
n
xy
0
Figure 5.19. Stiffness reduction for quasi-isotropic [0/90/45/+45]
s
laminate compared with
experimental data [35]: (a) longitudinal modulus, (b) Poissons ratio. Solid lines represent
SDM predictions using assumed values of r
r
. Reprinted, with kind permission, from Mech
Mater, Vol. 41, C.V. Singh and R. Talreja, A synergistic damage mechanics approach for
composite laminates with matrix cracks in multiple orientations, pp. 95468, copyright
Elsevier (2009).
171 5.4 Viscoelastic composites with ply cracking
One approach to treating the coupling of viscoelasticity and damage in com-
posite materials is by internal variables based continuum models. Most models of
this nature have been applied to particulate composites. One model that
addresses composite laminates is by Schapery and Sicking [36]. This model
extends the concept of work potential developed for elastic solids to viscoelastic
composite laminates by using the pseudo variables dened by Schapery [37].
However, the SchaperySicking model makes certain assumptions that limit the
usefulness of the model. It is assumed that the shear modulus and the transverse
modulus of a ply are linearly related, and that the ply is transversely isotropic.
Furthermore, modeling is done at the ply level, neglecting the constraint effects
of the neighboring plies. The model developed later by Kumar and Talreja [38]
does not suffer from these limitations. The CDM-based modeling of linear
viscoelastic response of composites with damage in accordance with this model
will be discussed below.
Using the well-known Boltzman superposition principle the constitutive law for
a viscoelastic material can be represented as
s
ij
=
_
t
0
C
ijkl
t t ( )
@e
kl
@t
dt; (5:99)
where C
ijkl
(t) is the relaxation modulus tensor. Taking the Laplace transform of
both sides and using the convolution theorem, we obtain
s
ij
=
~
C
ijkl
e
kl
; (5:100)
where the Laplace transform

f s ( ) of a function f(t) is dened as
f s ( ) =
_
0
e
st
f t ( ) dt; (5:101)
and
~
C
ijkl
= sC
ijkl
: (5:102)
The constitutive equations (5.100) in the Laplace domain are similar to
the relationships of linear elasticity, except for the dependency on the trans-
formation parameters. Additionally, the equilibrium equations and straindis-
placement equations also retain their form in the Laplace domain. This
correspondence between linear elasticity and linear viscoelasticity is the so-
called Correspondence Principle [39], and it applies to solving boundary
value problems as long as the boundary conditions do not change in time.
Thus, if ply cracks are present in a composite laminate, then the Correspond-
ence Principle can still be applied by viewing the cracked laminate as a homo-
geneous solid in which the internal boundaries (cracks) retain the same
boundary conditions in time. This implies that the cracks do not grow or heal
in time. We shall therefore address the linear viscoelastic response of laminates
at a given (xed) damage state. Specically, we will express the relaxation
modulus tensor in Eq. (5.99) as a function of damage.
Following [38] a pseudo energy density function is dened in the Laplace
domain in terms of transformed strain and damage variables, such that
s
ij
=
@

W e
ij
;

D
(a)
ij
_ _
@e
ij
: (5:103)
It is noted that the above equation is valid for a xed ply crack density. The time-
dependent deformation of the plies themselves will result in time-varying con-
straint on the cracks, leading to the crack surface separation changing in time.
For illustration, let us consider a cross-ply laminate with transverse cracks,
Figure 5.5, with y = 90
. In this case the damage mode tensor has a single nonzero

component, which in the Laplace domain takes the form [38]
D
11
=
k s ( )t
2
90
s
t
t
T
: (5:104)
where k s ( ) represents the constraint parameter in the Laplace domain, t
90
is
the thickness of the cracked layer, t
T
is the total laminate thickness, and s
t
is the
crack spacing. Using the same procedure as in Section 5.1.4, Kumar and Talreja
[38] derived the following transformed constitutive relationships (in the Voigt
notation) as
s
1
s
2
s
6
_
_
_
_
=
~
C
11
~
C
12
0
~
C
12
~
C
22
0
0 0
~
C
66
_
_
_
_
e
1
e
2
e
6
_
_
_
_
; (5:105)
where
~
C
pq
=
~
C
0
pq

~
C
(1)
pq
: (5:106)
Here
~
C
0
pq
is the transformed relaxation modulus of the cross-ply laminate without
cracks and
~
C
(1)
pq
, given by the following, is transverse cracking (one mode of
damage, a = 1)
~
C
(1)
pq
=

D
11
2g
11
g
12
0
g
12
2g
22
0
0 2g
66
_
_
_
_; (5:107)
where g
11
, g
12
, g
22
, and g
66
are material constants that appear as coefcient terms
in the polynomial expansion of the pseudo strain energy density function W. The
damage function D
11
is given by Eq. (5.104).
The time-dependency of the crack opening displacement has been handled in
two ways. In Kumar and Talreja [38] the function W was assumed to depend on
the initial (t = 0) value of the damage tensor, thereby absorbing all time depend-
ency in the presence of damage into the coefcient terms g
11
, etc. Later Varna et al.
[41] chose to retain the time dependency in the damage tensor and dealt with the
time-varying constraint to the COD explicitly.
Taking the inverse Laplace transform of Eq. (5.106), one obtains
C
pq
t ( ) = C
0
pq
t ( )
2r
n
t
90
t
T
k
ij
t ( ); (5:108)
where
r
n
=
t
90
2s
t
(5:109)
is the normalized crack density, and
k
11
t ( ) = 2L
1
kg
11
=s ( ); k
12
t ( ) = L
1
kg
12
=s ( );
k
22
t ( ) = 2L
1
kg
22
=s ( ); k
66
t ( ) = 2L
1
kg
66
=s ( ) ;
(5:110)
where L
1
(*) represents the inverse Laplace transform. Note that k = k
0
, its value
at time t = 0, if all time dependency is assumed to be in g
11
, etc. in the above
equations.
The functions k
11
, k
12
, and k
22
in Eq. (5.110) can be determined by procedures
similar to those described above for elastic constants. Here these three unknown
functions, which are decoupled from k
66
, can be evaluated from the time-varying
differences of the axial modulus E
x
(t) and Poissons ratio n
xy
(t) from their initial
values, and by assuming no change in the transverse modulus E
y
(t), as done in
Kumar and Talreja [38]. The time-variation of the material constants can be
determined from experimental data, if available, or by a micromechanics approxi-
mation. The micromechanics can be analytical, if possible, or numerical, e.g., by a
nite element model.
In Kumar and Talreja [38] cross-ply laminates of given linear viscoelastic ply
properties were considered for validation of the CDM approach described above.
The functions k
11
(t), k
12
(t), and k
22
(t) were determined from the calculated visco-
elastic response of a [0/90
2
]
s
laminate at a xed crack density of 0.4 cracks/mm.
These functions were then used to predict the time variations of the axial modulus
and Poissons ratio at other crack densities and for other cross-ply laminate
congurations of the same material. The predictions were compared with inde-
pendently calculated time variations of the properties by a nite element model
and an analytical micromechanics model reported in Kumar and Talreja [40].
Predictions agreed well in all cases.
Varna et al. [41] demonstrated the use of SDM for linear viscoelastic response
predictions by explicitly treating the COD variations in time. The time dependency
of COD, i.e., the constraint parameter k and its counterpart in the Laplace
domain k, were calculated by a FE model. These were then inserted in
Eq. (5.110) to determine k
11
(t), k
12
(t), and k
22
(t). Predictions thus made agreed
well with independently calculated viscoelastic response at different crack densities
and different cross-ply laminate congurations. A parametric study was also
performed in Varna et al. [41] to determine a master function for COD variation.
That function has the following form
k t ( ) = a b c
t
c
2t
s
1
_ _ _ _
E
2
t ( )
E
1
t ( )
_ _
d
; (5:111)
where a, b, c, and d are constants, E
1
and E
2
are the axial and transverse values,
respectively, of the Youngs modulus of the ply material, and t
s
here is the
thickness of the 0
-plies in the cross-ply laminate. Note that if more generally a

laminate of conguration [S/90
n
]
s
is used, then t
s
will be the thickness of the
sublaminate denoted by S.
For the nonlinear viscoelastic response of composites with damage another
approach is needed since the Correspondence Principle does not apply. A CDM
approach for this case was developed by Ahci and Talreja [42]. The material
system in the experimental study performed there was a carbon ber (T-65035)
fabric (8-harness satin, 3K tow size, with UC-309 sizing) in HFPE-II polyimide
thermosetting resin. It was demonstrated that the viscoelastic response of the
virgin material was linear within a range of stress and temperature, beyond which
it became nonlinear. The microcracking within the ber bundles further enhanced
the nonlinearity of the viscoelastic response. Since the tests measured strain
response under prescribed stress, the CDM formulation was appropriately modi-
ed to consider stress as an independent variable. Thus the Gibbs free energy
function G was formulated such that the strain response can be written as
e
ij
=
@G s
kl
; D
(a)
mn
; T; g
s
_ _
@s
kl
; (5:112)
where the damage mode tensors D
ij
(a)
are as dened by Eq. (5.20) and g
s
are
the viscoelastic internal state variables. Note that the damage mode tensor is kept
time independent by taking its initial value at time t = 0, analogous to the linear
viscoelastic case in the CDM approach by Kumar and Talreja [38] treated above.
Using polynomial expansion in terms of the integrity bases of the function G,
similar to the procedures for the linear elastic and viscoelastic cases above, and by
incorporating formulation of polymer viscosities available in the literature, the
time-dependent in-plane strain response of an orthotropic composite with trans-
verse cracks is derived as [42]
e
p
= C
E
pq
C
D
pq
D C
g
pq
g C
E
pq
Dg
_ _
s
q
; (5:113)
where p, q = 1, 2, and 6, and superscripts E, D and g to the compliance matrix C
denote the elastic, damage, and viscous contributions, respectively, and the last
term in the parenthesis stands for the interactive contribution between damage
and viscosity. The damage variable D = D
11
, which is the only nonzero component
of the damage mode tensor for transverse cracks.
The material constants in the four matrices in Eq. (5.113) must be evaluated
before these relationships can be used to predict the nonlinear viscoelastic
response. The rst matrix is simply the elastic response matrix and can be found
by recording the instantaneous strain response to imposed stresses. The second
matrix can be found experimentally in conditions where the viscoelasticity is
negligible, and the procedure could be one of those discussed above for the case
of damage in elastic composites. In the absence of damage, the third matrix in
Eq. (5.113) represents the time-dependent response. Finally, the fourth matrix
needs determining to see how damage enhances the viscoelastic deformation.
A procedure for the complete evaluation of the material constants involved is
thus far from trivial. An experimental procedure aided by nite element modeling
was developed in Ahci and Talreja [42]. The resulting characterization of non-
linear viscoelasticity and damage was used to illustrate the effects of temperature
and stress levels as well as the nature of the damageviscoelasticity coupling.
5.5 Summary
This chapter has presented the basic concepts of continuum damage mechanics and
illustrated its application to damage in composite materials. The classical eld of
continuum thermodynamics combined with internal variables characterization of
damage provides a powerful tool to describe materials response affected by damage.
In its conventional form the CDM relies on material parameter identication by
experimental data. This has been a burden that has been increasingly challenging to
carry because of the multiplicity of damage modes and the ensuing complexity of
response changes. The extension of CDM developed by judiciously aiding it with
selected micromechanics solutions (analytical or numerical) has eased the burden
and thereby increased the attractiveness of CDM. This integrative approach of
combining CDM and micromechanics has been named synergistic damage
mechanics. This chapter has illustrated the application of SDM to multiple (up to
three) damage modes in laminates. Results of parametric studies have also been
presented to offer further insight into the consequences of multiple damage modes.
Although most applications of polymer-based composites have been in struc-
tures operating at temperatures safely below the glass transition temperature of
the polymer, there are instances such as in jet engine casings where high tempera-
tures are induced. Viscoelastic response of the composite in such cases takes place
and must be addressed by itself and when damage occurs. The CDM and SDM
methodologies have been presented for linear viscoelastic composites with
damage. At high temperatures and with extensive damage the viscoelastic
response can become nonlinear. This case has added complexity in identication
of material constants. A methodology that uses experimental data as well as
numerical computations has been discussed.
References
1. L. M. Kachanov, On the creep rupture time. Izv Akad Nauk SSR, Otd Tekhn Nauk, 8
(1958), 2631.
2. L. M. Kachanov, Rupture time under creep conditions. Int J Fract, 97:1 (1999), 1118.
3. Y. N. Robotnov, Creep Problems in Structural Members. (Amsterdam: North-Holland, 1969).
4. S. Murakami and N. Ohno, A continuum theory of creep and creep damage. In Creep
in Structures, 3rd IUTAM Symposium, ed. A. R. S. Ponter and D. R. Hayhurst. (Berlin,
Germany: Springer-Verlag, 1981), pp. 42244.
5. J. Lemaitre and J. L. Chaboche, Damage mechanics. Chapter 7 in Mechanics of Solid
Materials. (Cambridge: Cambridge University Press, 1990), pp. 346450.
6. J. Lemaitre, A continuous damage mechanics model of ductile fracture. J Eng Mater
Technol, Trans ASME, 107:42 (1985), 839.
7. J. Lemaitre and J. L. Chaboche, Aspect phenomenologique de la rupture par endomm-
agement. Journal de Mecanique Appiquee, 2:3 (1978), 31765.
8. J. Lemaitre and R. Desmorat, Engineering Damage Mechanics: Ductile, Creep, Fatigue
and Brittle Failures. (Berlin: Springer-Verlag, 2005).
9. J. L. Chaboche, Anisotropic creep damage in the framework of continuum damage
mechanics. Nucl Eng Des, 79 (1984), 30919.
10. J. L. Chaboche, Continuum damage mechanics: part I general concepts. J Appl Mech,
Trans ASME, 55:1 (1988), 5964.
11. J. L. Chaboche, Continuum damage mechanics: part II damage growth, crack initi-
ation, and crack growth. J Appl Mech, Trans ASME, 55:1 (1988), 6572.
12. J. W. Ju, Isotropic and anisotropic damage variables in continuum damage mechanics.
J Eng Mech, 116 (1990), 276470.
13. R. Talreja, A continuum-mechanics characterization of damage in composite materials.
Proc R Soc London A, 399:1817 (1985), 195216.
14. R. Talreja, Internal variable damage mechanics of composite materials. In Yielding,
Damage, and Failure of Anisotropic Solids, ed. J. P. Boehler. (London: Mechanical
Engineering Publications, 1990), pp. 50933.
15. R. Talreja, Damage mechanics of composite materials based on thermodynamics with
internal variables. In Polymer Based Composite Systems for Structural Applications, ed.
A. Cardon and G. Verchery. (London: Elsevier, 1991), pp. 6579.
16. R. Talreja, Continuum modeling of damage in ceramic matrix composites. Mech Mater,
12:2 (1991), 16580.
17. R. Talreja, Damage characterization by internal variables. In Damage Mechanics of
Composite Materials, ed. R. Talreja. (Amsterdam: Elsevier, 1994), pp. 5378.
19. A. A. Vakulenko and M. Kachanov, Continual theory of a medium with cracks. Izv AN
SSSR, Mekhanika Tverdogo Tela (Mech Solids), 6:4 (1971), 159.
20. J. Varna, Physical interpretation of parameters in synergistic continuum damage
mechanics model for laminates. Compos Sci Technol, 68:13 (2008), 2592600.
21. A. J. M. Spencer, Theory of invariants. In Continuum Physics, ed. C. A. Eringen.
(New York: Academic Press, 1971), pp. 239353.
22. J. Adkins, Symmetry relations for orthotropic and transversely isotropic materials.
Arch Ration Mech Anal, 4:1 (1960), 193213.
177 References
J Compos Mater, 19:4 (1985), 35575.
atedresponse of laminates withvarying constraints. Int J Damage Mech, 8:2 (1999), 17493.
26. J. Varna, R. Joffe, and R. Talreja, A synergistic damage-mechanics analysis of trans-
verse cracking in [y/90
4
]
s
laminates. Compos Sci Technol, 61:5 (2001), 65765.
27. J. Varna, R. Joffe, and R. Talreja, Mixed micromechanics and continuum damage
mechanics approach to transverse cracking in [S,90(n)](s) laminates. Mech Compos
Mater, 37:2 (2001), 11526.
28. C. V. Singh and R. Talreja, Damage mechanics of composite laminates with transverse
matrix cracks in multiple orientations. In 48th AIAA SDM Conference, Honolulu,
Hawaii, USA. (Reston, VA: AIAA, 2007).
29. C. V. Singh and R. Talreja, Analysis of multiple off-axis ply cracks in composite
laminates. Int J Solids Struct, 45:16 (2008), 457489.
30. C. V. Singh and R. Talreja, A synergistic damage mechanics approach for composite
laminates with matrix cracks in multiple orientations. Mech Mater, 41:8 (2009), 95468.
31. S. Li, C. V. Singh, and R. Talreja, A representative volume element based on transla-
tional symmetries for FE analysis of cracked laminates with two arrays of cracks. Int
J Solids Struct, 46:78 (2009), 1793804.
32. C. V. Singh, Multiscale modeling of damage in multidirectional composite laminates.
Ph.D. thesis, Texas A&M University, College Station, TX (2008).
33. J. Varna, R. Joffe, N. V. Akshantala, and R. Talreja, Damage in composite laminates
with off-axis plies. Compos Sci Technol, 59:14 (1999), 213947.
34. J. Varna, L. Berglund, R. Talreja, and A. Jakovics, A study of crack opening displace-
ment of transverse cracks in cross-ply laminates. Int J Damage Mech, 2:3 (1993), 27289.
36. Schapery, R. A. and Sicking, D. L., On nonlinear constitutive equations for elastic
and viscoelastic composites with growing damage. In Mechanical Behavior of
Materials, ed. A. Bakker. (Delft, The Netherlands: Delft UniversityPress, 1995), pp. 4576.
37. Shapery, R. A., On viscoelastic deformation and failure behavior of composite materials
with distributed aws. In Advances in Aerospace Structures and Materials, ASME-AD-01,
ed. S. S. Wang and W. J. Renton. (Philadelphia, PA: ASTM, 1981), pp. 520.
38. R. S. Kumar and R. Talreja, A continuum damage model for linear viscoelastic
composite materials. Mech Mater, 35:36 (2003), 46380.
39. R. M. Christensen, Theory of Viscoelasticity: An Introduction, 2nd edn. (New York:
Academic Press, 1982).
40. R. S. Kumar and R. Talreja, Linear viscoelastic behavior of matrix cracked cross-ply
laminates. Mech Mater, 33:3 (2001), 13954.
41. J. Varna, A. I. Krasnikovs, R. S. Kumar, and R. Talreja, A synergistic damage mech-
anics approach to viscoelastic response of cracked cross-ply laminates. Int J Damage
Mech, 13:4 (2004), 30134.
42. E. Ahci and R. Talreja, Characterization of viscoelasticity and damage in high tem-
perature polymer matrix composites. Compos Sci Technol, 66:14 (2006), 250619.
6 Damage progression
6.1 Introduction
As discussed in Chapter 1, material constitutive relationships are needed in order
to conduct failure analysis of a structure subjected to service loading. Contrary to
homogeneous materials such as metals whose stressstrain relationships can be
specied a priori, composite materials suffer damage that can alter these relation-
ships. Thus if strains are prescribed, the stress response can be expressed as
s
ij
= C
ijkl
e
kl
( )e
kl
; (6:1)
where the stiffness matrix C
ijkl
changes with strain as the composite material
suffers damage. Determination of C
ijkl
as a function of the applied strain can be
achieved by solving two related sub-problems:
1. Describe stiffness changes as a function of damage: In this step C
ijkl
is expressed
in terms of some damage characteristic, such as the ply crack density,
C
ijkl
= C
ijkl
r
(a)
_ _
; (6:2)
where r
(a)
is the crack density of damage mode a = 1, 2,. . ., N. As discussed in
Chapter 4, micromechanics solutions for multiple damage modes are not
always possible. Using the CDM formulation for multiple damage modes in
linearized form we can express Eq. (6.2) as
C
ijkl
= C
0
ijkl
N
a=1
C
(a)
ijkl
r
(a)
_ _
; (6:3)
where C
0
ijkl
is the stiffness tensor for undamaged laminate, and C
(a)
ijkl
represents
the stiffness changes brought about by damage mode a.
2. Describe the evolution of crack density as a function of applied loading:
r
(a)
= r
(a)
e
kl
( ): (6:4)
Combining the solution to the two sub-problems above, we obtain
C
ijkl
= C
0
ijkl
N
a=1
C
DAM
ijkl
r
(a)
e
mn
( )
_ _
: (6:5)
The solution to the rst sub-problem has been described in Chapters 4 and 5. This
chapter concerns the second sub-problem.
To take advantage of directional properties of plies, composite laminates are
made from a mix of longitudinal, transverse, and angle plies. Since the unidirec-
tional lamina has low strength in the transverse direction, it is prone to cracking
along bers. When the applied load is increased beyond the strain (or stress) at
which initiation of cracking occurs, new cracks form in the cracked ply in between
existing cracks. Initially these cracks are far apart and do not interact with each
other. However, quickly they form a roughly periodic array of parallel cracks.
Figure 6.1 shows the increase in density of transverse cracks (i.e., number of
cracks per unit length normal to the crack plane) for typical congurations of
cross-ply laminates when loaded under axial tension. Prediction of such curves has
been an extensive subject of study. The approaches used to model crack initiation
and progression (multiplication) can be divided into two categories: strength-
based models and energy-based models. As the names suggest, the rst category
of models involves use of strength (failure) criteria, while energy balance concepts
underlie the second category.
6.2 Experimental techniques
Before discussing experimental observations and measurements in Section 6.3
below, a brief description of each of the techniques is given next. This is to help
provide a background against which to interpret the observations. No attempt is
made here to go into any depth on these techniques, as that would distract from
the focus of this chapter. For further details references [24] are suggested.
Over the past forty years many nondestructive evaluation (NDE) techniques
have been developed to detect, monitor, and observe ply cracking damage in
40
30
20
10
0
n=4
[0/90
n
/0],T300/934
Cracks/In.
n=3
n=2
n=1
20 40 60 80 100 120 Ksi
Figure 6.1. Transverse crack density vs. applied axial stress in [0
m
/90
n
]
s
laminates [1].
Reprinted, with kind permission, from Composites Technology Review, copyright ASTM
International, 100 Barr Harbor Drive, West Conshohocken, PA 19428.
composite laminates. The quantities targeted for observations and measurements
concerning ply cracking include the following:
1. Crack initiation strains (or stresses).
2. Increase in number of cracks with applied loading or with number of cycles in
case of fatigue experiments.
3. Changes in stiffness properties with damage.
4. Crack opening displacements (COD) and crack proles.
5. Final failure strains and damage leading to the nal failure.
Brief overviews of the main techniques are provided here.
Edge replication
Although direct optical microscopy can be used [5] for surface observation to evaluate
damage, it requires in most cases specimens to be removed at regular intervals during
loading. If feasible, anoptical microscope canbe mountedonthe testing machine itself,
but it often requires the microscope to be brought close to the specimen surface for
enough resolution. This limitation can be overcome by using edge replication instead.
This is a simple and easy technique if the purpose is to monitor damage (count cracks)
in composites, as was shown early by Stalnaker, Stinchcomb, and Masters [6, 7]. The
approach involves the microscopic examination of surface replicas, which are prepared
by pressing softened rubber tape (or tape with an adhesive, e.g., acetate) against the
specimen edge. With this procedure it is possible to quickly obtain permanent records
of the specimen edge while the specimen itself is held in loading. The popularity of the
technique is evidenced by its utilization by many researchers in the eld [2, 814]. Two
examples of photomicrographs from edge replicas are shown in Figure 6.2.
Acoustic emission
In this technique, a sensor is used to monitor acoustic emission (AE) signals from
stress waves generated due to some local failure in the material, such as formation of
a ply crack. A major limitation of the approach is its inability to distinguish between
damage types and to provide information concerning crack location and orientation
[14]. However, if the damage mechanismis known based on prior experience with the
material system, this technique can indicate the applied load at which the damage
initiated. Using this to monitor damage and to take edge replicas accordingly has
been found useful. A combination of ultrasonic polar backscattering scans and AE,
which yields accurate crack location, has also been suggested [15, 16]. Figure 6.3
shows the variation of cumulative count of acoustic events and the corresponding
stressstrain response of a ceramic composite subjected to tensile loading.
X-ray radiography
X-ray radiography is a useful technique in nding internal cracks that are not visible
by optical microscopy. For clarity of observationa penetrant uidthat absorbs X-rays
is often necessary to provide sufcient contrast between exposed and unexposed
regions. Pictures of the developed X-ray lms can then be enlarged for further clarity.
181 6.2 Experimental techniques
Transverse Cracking
Transverse Cracking
Longitudinal Cracking
(a) (b)
Figure 6.2. Photomicrographs of edge replicas showing details of damage development in
fatigue loading of quasi-isotropic laminates: (a) [0/45/90]
s
and (b) [0/90/45]
s
laminate
[2]. Reprinted, with kind permission, from Damage in Composite Materials, copyright
ASTM International, 100 Barr Harbor Drive, West Conshohocken, PA 19428.
1200
(a)
900
600
300
0
0 20 40
Stress (MPa)
C
u
m
u
l
a
t
i
v
e

A
E

c
o
u
n
t
s
60 80 100
100
80
60
40
20
0
(b)
0
Strain (%)
S
t
r
e
s
s

(
M
P
a
)
0.05 0.1 0.15 0.2
Figure 6.3. Variation of cumulative acoustic event (AE) counts (a), and stressstrain curve
(b) for a ceramic composite loaded in tension [14]. Reprinted, with kind permission, from
Damage Detection in Composite Materials, copyright ASTM International, 100 Barr Harbor
Drive, West Conshohocken, PA 19428.
The pictures, however, only provide 2-D images of cracks and therefore cannot be
used to nd the depth information needed to separate cracks of one layer from
another. For that purpose penetrant enhanced X-ray stereo radiography can be
used. The standard procedure is to make two X-ray radiographs of the object at
slightly different orientations, which is usually done by rotating the X-ray source
relative to the object. The spatial locations of cracks can be observed with a
microscope but cannot be recorded. The approach has been successfully used to
detect ber fractures, delaminations and matrix cracks [1719]. To enable imaging
and ner precision detection of internal defects, the technique employed is X-ray
tomography, which uses a medical scanner and yields a three-dimensional image of
an object [20]. An X-ray radiograph of damage in a quasi-isotropic laminate is
shown in Figure 6.4.
Ultrasonic C-scan
Ultrasonic C-scan is a nondestructive inspection technique in which a short pulse
of ultrasonic energy is made incident on a sample. The attenuation of the pulse is
inuenced by voids, delaminations, state of resin cure, the ber volume fraction,
Figure 6.4. An X-ray radiograph of damage in a quasi-isotropic laminate [19]. Reprinted,
with permission, from Effects of Defects in Composite Materials, copyright ASTM
183 6.2 Experimental techniques
the condition of the ber/matrix interface, and any foreign inclusions present. The
technique is, however, limited by its inability to detect very small defects or cracks
with their planes parallel to the wave direction [14].
In some cases a damaged laminate material may contain a large number of
small and partially grown internal cracks. On increase in loading, these cracks can
grow further and coalesce to form larger cracks. The common NDE techniques
described above cannot provide accurate information on the formation, number,
size, and progression of these internal cracks. For such cases a through-transmis-
sion ultrasonic C-scan imaging with inclined focusing transducers in confocal
conguration has been recently suggested [21, 22]. Other methods based on
vibrations and lamb waves have also been suggested [23, 24].
Technique for COD measurement
In addition to the observations of damage certain measurements also need to be
made as dictated by the models. For instance, the COD of ply cracks is a quantity
that enters the SDM approach described in Chapter 5. Since there are no standard
methods for measuring this quantity, Varna et al. [2627] developed a set-up for
this purpose. To observe an individual ply crack, the set-up uses a miniature
materials tester (MINIMAT) for loading a thin strip cut out from the cracked
laminate. The open crack is observed under an optical microscope equipped with a
video camera. The video signal transmitted to a TV monitor displays the crack
prole at sufcient magnication (~ 210
3
) to measure the COD. Figure 6.5
presents COD as a function of position along the cracked 90
-ply thickness
(z-direction) and compares it with the theoretical shape prediction obtained using
Linear Elastic Fracture Mechanics (LEFM).
Raman spectroscopy
Recently Katerelos and coworkers [28] have developed an experimental technique
based on Raman spectroscopy, which uses the property that the Raman
vibrational wave numbers (frequencies) of certain chemical groups of commercial
reinforcing bers, such as aramid or carbon, are stress and strain dependent [16].
Thus, the wave number shift along an embedded ber can be utilized to determine
stress or strain. The technique can provide a high spatial resolution of ~ 1mm.
However, for the approach to work the matrix needs to be translucent. Also, the
data acquisition can take longer than milliseconds for a single measurement.
Moreover, certain amorphous bers such as glass have a weak Raman response.
Aramid bers, on the other hand, scatter the waves well. Therefore, a small
amount of aramid ber placed in strategic positions within a glass-ber laminate
can act as Raman sensors of stress and strain. The micromechanical strain
mapping results are then used to derive the properties, i.e., the longitudinal
modulus of elasticity and the magnitude of the residual strains caused by cracking
[29]. The approach has been successfully applied to evaluate ply cracking damage
evolution and resulting stiffness changes [3032].
6.3 Experimental observations
Experimental studies on initiation and growth of intralaminar cracking in composite
laminates have been extensive. Most of the work has focused on 90
-ply cracking in
cross-ply and quasi-isotropic laminates. Chapter 3 reviewed some of those observa-
tions to illustrate the nature of composite damage. A book chapter by Nairn [33] is
also recommended for a good overview of the topic. In the following we review
quantitative data related to the initiation and progression of ply cracking in laminates.
6.3.1 Initiation of ply cracking
The applied loading (stress or strain) at which ply cracking in laminates rst
occurs is of interest from materials selection as well as design points of view.
A good laminate conguration (orientation, thickness, and sequence of plies) will
delay initiation of ply cracking to as high a load as possible. Experimental
observations have indicated that all laminate conguration parameters inuence
ply crack initiation. Most early studies examined ply cracking in cross-ply lamin-
ates of glass-polyester or glass/epoxy under axial tension. The ply cracking in the
90
-plies could in most cases be observed by looking at the specimen surface under
an optical microscope at low magnication. The near-transparency of composites
25
20
15
[O2/908]s
Hybrid CF/GF
(Different sections)
LEFM theory
10
5
0
0.0 0.2 0.4 0.6 0.8 1.0
Relative position in 90-layer
C
r
a
c
k

o
p
e
n
i
n
g

d
i
s
p
l
a
c
e
m
e
n
t

u
(
m
m
)
zt
90
2t
90
Figure 6.5. Crack opening displacement measured using MINIMAT along the 90
-layer
thickness of a hybrid CF/GF cross-ply laminate at various sections through the specimen
width. The solid line depicts predictions from LEFM. Source: J. Varna, L. Berglund,
R. Talreja, and A. Jakovics, A Study of crack opening displacement of transverse cracks in
cross-ply laminates, Int J Damage Mech, Vol. 2, pp. 27289 (1993).
185 6.3 Experimental observations
of glass bers made this possible. In carbon ber composites, however, one had to
rely on edge observations until X-ray radiography allowed imaging of the interior
cracks. The axial strain values at initiation of transverse cracking in cross-ply
laminates of glass/epoxy and carbon/epoxy fall typically in the range 0.41.0%.
As the off-axis ply angles decrease from 90
the initiation strain increases. For

instance, for 45
-plies cracking barely initiates at 1.0% axial strain [34]. Figure 6.6
illustrates the effect of off-axis angle on crack initiation strain for (a) GFRP [35],
and (b) CFRP [36]. Data for notched samples are also indicated to illustrate the
effect of local stress enhancement. In Figure 6.6 the data for un-notched samples
with and without polished edges illustrate the effect of machining-induced aws in
initiating cracking. Also of note is the range of strain and the lowest strain for
GFRP versus CFRP. In the latter case, the lowest strain to crack initiation is higher
due to the higher elastic modulus of carbon bers. In both cases the initiation strain
at low off-axis angles is limited by the ber failure strain.
The thickness of the cracking ply also has a large effect on e
init
. Garret and
others [3744] performeda systematic study of thickness effect on cracking in [0
m
/90
n
]
s
laminates. They used glass-reinforced polyester [37, 38] and glass-reinforced epoxy
[3941] as the laminate materials andvariedthe thickness of the 90
-plies while keeping

the thickness of the 0
-ply constant at 0.5 mm. Figure 6.7 shows the variation of e

init
with respect to the total thickness of the 90
-plies. As the thickness of the 90
-plies
increases, e
init
decreases. When the thickness of the 90
-plies is more than 0.4 mm, the

cracks initiate instantly and span the entire cross section of the 90
-plies. At the other

extreme, if the thickness of the 90
-plies is less than 0.1 mm, cracks may be entirely

suppressed and the laminate eventually may fail due to other damage mechanisms,
e.g., delamination and ber breakage. The experiments on carbon/epoxy laminates
GFRP [0//0] laminates
notched
Angle (degrees)
40 50 60 70 80 90 100
unnotched polished unnotched
3
(a)
1.2
1
0.8
0.6
0.4
0.2
0
40
I
n
i
t
i
a
l

c
r
a
c
k
i
n
g

s
t
r
a
i
n

(
%
)
60
Off-axis angle ()
CFRP (IM7/8552) [0
2
/
4
]
s
laminates
80 100
Unnotched
Notched
(b)
2.5
2
1.5
1
0.5
0
L
o
n
g
i
t
u
d
i
n
a
l

c
r
a
c
k
i
n
g

s
t
r
a
i
n

(
%
)
Figure 6.6. Crack initiation strains for angle ply laminates as a function of ply orientation:
(a) GFRP laminates; (b) CFRP laminates. Part (a) reprinted, with kind permission, from
Composites A, Vol. 28, L.E. Crocker, S.L. Ogin, P.A. Smithand P.S. Hill, Intra-laminar fracture
in angle-ply laminates, pp. 83946, copyright Elsevier (1997). Part (b) reprinted, with kind
permission, fromSpringer Science+Business Media: J Materials Science, Intra-laminar cracking
in CFRP laminates: observations and modeling, Vol. 41, 2006, pp. 6599609, N. Balhi et al.
show similar behavior [41, 45]. In general, for [0
m
/90
n
]
s
laminates, complete crack
suppression is expected only when m / n _10 although the suppression effects may be
felt in the range 2 _ m / n < 10 [46]. This difference in initiation strain for different
thicknesses is attributed to the relative constraint of the supporting 0
-plies onto the

crack opening in the 90
-plies [47].
The actual ply layup in the laminate may also play a role in determining e
init
.
For instance, [90
2
/0]
s
laminates develop cracks sooner than [0/90
2
]
s
laminates [48].
It is because the 90
-plies in [90/0]
s
laminates are on the outer surface and may not
experience much support for crack suppression from the inner 0
-plies. For
multidirectional laminates, the situation is even more complex and e
init
will depend
on the thickness ratio and the stiffness properties of the cracking and supporting
plies. The laminate preparation and processing method also may affect crack
initiation; lament-wound laminates may be more prone to cracking than those
made from prepregs using an autoclave.
6.3.2 Crack growth and multiplication
The growth of ply cracks is usually unstable in composites such as GFRP and
CFRP. The initial transverse cracks are found to grow through the lamina
thickness quickly but are usually arrested at interfaces with adjacent plies of
different orientation, e.g., at the 90/0 interface for cross-ply laminates [38, 40]
and at the 90/45 interface in [0/90/45/+45]
s
laminates [34]. Continued loading
usually leads to formation of progressively more cracks between the already
formed cracks. Most experimental studies point out that once the ply cracks have
grown through the lamina thickness, they often grow unstably along the ber
direction through the laminate width, and are thus described as tunneling
cracks. In some cases, however, cracks in plies other than 90
may not grow

fully before the laminate fails by delamination [34, 4951]. Such partial cracks can
be observed in the 45
-plies of quasi-isotropic [0/90/45/+45]

s
laminates [34].
2
1
0
0 1
S
t
r
a
i
n

t
o

F
i
r
s
t

M
i
c
r
o
c
r
a
c
k

(
%
)
2
Total Thickness of 90 Plies (mm)
3 4 5
Figure 6.7. The strain to initiate ply cracking in [0
m
/90
n
]
s
glass/epoxy laminates as a function
of the total thickness of 90
-plies [33]. Experimental data from [38]. Reprinted, with

kind permission, from Polymer Matrix Composites, J. A. Nairn, Matrix microcracking in
composites, pp. 40332, copyright Elsevier (2000).
Once the ply cracking has initiated, more and more ply cracks start appearing in
between existing cracks, and the crack density rises quickly. As the crack spacing
between adjacent cracks decreases, the cracks start interacting. Closely interacting
cracks typically provide a shielding effect, which tends to reduce the stresses
between two adjacent cracks. Therefore, on further loading, the rate of cracking
reduces and nally approaches a saturation value. Thus a typical damage growth
curve consists of three stages: crack initiation, rapid rise in crack density by
multiplication, and reducing rate of crack density evolution until saturation (see
Figure 6.11). Reifsnider and associates [10, 52] described this microcrack satur-
ation as a material state and called it the characteristic damage state (CDS). They
proposed that CDS is a well-dened laminate property and does not depend on
load history, environment, or thermal or moisture stresses. However, later investi-
gations by Akshantala and Talreja [53] have suggested that CDS may not be a
single state under fatigue loading but may depend on the maximum stress applied.
The progression from initial rapid rise until saturation of crack density depends
on the ply material, the ply stacking sequence, and also the laminate fabrication
process. For example, well-made carbon/epoxy laminates typically have a rapid
rise in ply crack density. The saturation crack density is often found to inversely
scale with cracking ply thickness, with thin plies accumulating a large number
of cracks/mm [54, 55]. This is well illustrated in Figure 6.8 where experiments
conducted on GFRP specimens with different ply thicknesses [54] showed that thin
90
-plies can accumulate numerous cracks. The evolution of average crack density
with respect to applied load for the same specimens is shown in Figure 6.9. It can be
observed here that the saturation crack density is roughly proportional to 1/t
90
. The
damage evolution curves in Figure 6.10 also show that thicker cross-ply laminates
typically show lower crack density at saturation. The damage evolution in outer
and inner 90
-plies is also compared. The laminates with outer 90
-plies are
Figure 6.8. Transverse cracking in GFRP specimens with transverse-ply thickness of (a) 0.75
mm, (b) 1.5 mm, and (c) 2.6 mm, strained to 1.6%. Reprinted, with kind permission,
from Springer Science+Business Media: J Materials Science, Multiple transverse fracture in
90 degrees cross-ply laminates of a glass ber-reinforced polyester, Vol. 12, pp. 15768,
K. W. Garrett and J. E. Bailey.
observed to have lower saturation crack densities. A typical damage evolution
curve for ply cracking based on these studies is shown in Figure 6.11.
6.3.3 Crack shapes
When cracks are widely spaced, the maximum principal stress occurs on the plane
midway between existing cracks. Thus, at low crack density there is a tendency for
new cracks to form midway and develop into a periodic array. However, at large
crack density cracks interact causing the maximum principal stress to shift towards
the 0/90 interface close to an existing crack. This may result in curved or oblique
microcracks forming near the 0/90 interface [54, 56, 57]. These cracks make an angle
of 4060
with existing straight cracks. Lundmark and Varna [58] have recently
reported that curved microcracks form more readily in the low-temperature regime
than at room temperature. In fact, at low temperatures complex crack trajectories
are observed to form (see Figure 6.12). This may result in a highly damaged region
in the laminate encompassing multiple crack types (Figure 6.13).
6.3.4 Effect of cracking
The most direct effect of ply cracking is the reduction of the thermomechanical
properties of the laminate, including changes in the effective values of Youngs
moduli, Poissons ratios, and thermal expansion coefcients. The changes in
stiffness properties can in turn lead to change in the behavior of the whole
structure, e.g., the magnitude of its deection and vibrational frequencies, some-
times making the structure unable to carry out its intended design function. Even
if it does not cause the structure to fail, substantial ply cracking may give rise to
1.2
3.2mm
2t
90
2mm
1.5mm
0.75mm
1
0.8
0.6
0.4
0.2
0
0 50 100 150 200 250
Applied stress (MPa)
A
v
e
r
a
g
e

c
r
a
c
k

d
e
n
s
i
t
y

(
1
/
m
m
)
Figure 6.9. Average crack density as a function of applied stress for GFRP cross-ply laminates
with different transverse ply thicknesses (2t
90
). The experimental data are from [54].
Stage I
Applied load (strain or stress)
C
r
a
c
k

d
e
n
s
i
t
y

(
n
o
.

o
f

c
r
a
c
k
s
/
m
m
)
Stage II Stage III
Crack initiation &
propagation
through
laminate width
Multiple crack
formation
Saturation of
progressive cracking
Figure 6.11. A typical damage evolution curve for ply cracking in laminates.
1.4
(a)
1.2
1.0
0.8
0.6
0.4
0.2
0.0
0 200 400
[0/90
4
]
s
[0/90
2
]
s
[0/90]
s
600 800
Stress (MPa)
C
r
a
c
k

d
e
n
s
i
t
y

(
1
/
m
m
)
1.0
(b)
0.8
0.6
0.4
0.2
0.0
0 100 200 300
[90/0/90]
T
[90/0]
s
[90/0
2
]
s
[90/0
4
]
s
400 500
Stress (MPa)
C
r
a
c
k

d
e
n
s
i
t
y

(
1
/
m
m
)
600 700 800 900 1000
Figure 6.10. Damage evolution curves for (a) [0/90
m
]
s
and (b) [90/0
m
]
s
laminates. Reprinted,
with kind permission, from Polymer Matrix Composites, J. A. Nairn, Matrix microcracking
in composites, pp. 40332, copyright Elsevier (2000).
more deleterious forms of damage such as delamination and longitudinal splits, or
provide pathways for the entry of moisture and corrosive liquids.
6.3.5 Loading and environmental effects
Most experiments are performed using uniaxial tension, but ply cracks will also
form under other loading conditions, such as fatigue, biaxial, or shear loading.
Biaxial loading of [0
m
/90
n
]
s
laminates may show cracks in both 0
- and 90
-plies. If
the material response remains linearly elastic after cracking, then neglecting crack
interaction effects between plies the effect of biaxial loading of [0
m
/90
n
]
s
laminates
may be seen as equivalent to two uniaxial loading cases, one on [0
m
/90
n
]
s
and the
other on [90
m
/0
n
]
s
laminates. On thermal loading, the differential shrinkage
between the 0
- and 90
-plies may also induce biaxial loading [5965]. In general,

Normal
crack
Crack
type A
Crack
type B
Crack
type C
Crack
type D
a
a
Partial
crack
0 layer t
0
t
90
90 layer
w
w w
Figure 6.12. Different crack types observed during tensile testing of a [0
2
/90
4
]
s
CF/EP
laminate at cryogenic temperature (150
C). Reprinted, with kind permission, from Eng

Fract Mech, Vol. 75, P. Lundmark and J.Varna, Damage evolution and characterization of
crack types in CF/EP laminates loaded at low tempratures, pp. 263141, copyright (2008),
with permission from Elsevier.
Figure 6.13. A snapshot of a highly damaged region of a [0
2
/90
4
]
s
CF/EP laminate at an
applied stress of 343 MPa (0.66% strain) during tensile loading at cryogenic temperature
(150
C). Reprinted, with kind permission, fromEng Fract Mech, Vol. 75, P. Lundmark and
J.Varna, Damage evolution and characterization of crack types in CF/EP laminates loaded
at low tempratures, pp. 263141, copyright (2008), with permission from Elsevier.
thermal effects can be accounted for in the analysis and predictions of classical
laminate theory. Bailey et al. [41] studied the effect of thermal stresses and Poissons
contraction on ply cracking in CFRP and GPRP cross-ply laminates. Thermal
residual stresses typically lower crack initiation strains. Thermal effects are larger
in CFRPs than in GFRPs due to larger differences in both thermal expansion
coefcients and Youngs moduli in directions parallel and perpendicular to bers
[41]. For instance, a [0/90]
s
laminate with a ply thickness of 0.5mm shows 0.322%
thermal strains for CFRPs and 0.094% for GFRPs. The Poissons effect was found
to be greater in GFRPs because of their larger failure strains. Poissons strains can
sometimes be so high that they can induce transverse cracking of the 0
-plies.
The mismatch between the thermal expansions of different plies can also result
in a high-density form of microcracking known as stitch cracking. Lavoie and
Adolfsson [66] studied this type of cracking in [+y
n
/y
n
/90
2n
]
s
laminates (see
Figure 6.14). Stitch cracks appear to form instead of interply delamination at
the tip of a 90
crack in the adjacent constraining ply when the included angle

between the two is greater than 50
. Stitch cracking is also observed in case of

fatigue loading [19]. Variation in intralaminar fracture toughness over tempera-
ture can also change the initiation and evolution of ply cracking [67].
The residual stresses due to moisture can also affect the cracking process
[63, 68]. This could be due to degradation of the matrix during hygrothermal
aging [69, 70]. A combination of moisture and thermal residual stresses can cause
signicant stiffness reduction, especially at high crack density [71].
Figure 6.14. X-radiograph of thermal stress-induced matrix microcracks in a [+454/454/
908]
s
laminate. The 90
-ply matrix cracks run from left to right, and trigger the formation of
many short, stitch-like 45
-ply matrix cracks. Long +45
-ply matrix cracks appear.

Reprinted, with kind permission, from J. A. Lavoie and E. Adolfsson, J Compos Mater,
Vol 35, pp. 207797, copyright # 2001 by Sage Publications.
If laminates are loaded in bending, ply cracks will form on the tension side
[11, 72, 73], and for such loading analyses need to account for the resulting local
stress states [11, 7377].
6.3.6 Cracking in multidirectional laminates
Cross-ply laminates are not very common in practical applications. In fact,
efcient use of composite laminates in a wide range of applications requires
placing plies in multiple orientations. A common example is a class of laminates
known as quasi-isotropic that can be constructed in different ways; a common
form usually has a mix of plies with 0
, 45
, and 90
orientations, exemplied by
the [0/90/45]
s
conguration. In the off-axis plies of such laminates, cracks
usually initiate at much higher applied axial strains than in 90
-plies and may

sometimes show curved patterns due to inclined principal stress trajectories.
Curved cracks and cracks near free edges typically promote delamination at the
ply interface [78]. Experimental data show that the majority of off-axis cracks
form at the edges of the test coupons and often may not grow across the thickness
and width before the specimen fails by extensive delamination.
Furthermore, in some cases, other damage modes such as delamination or ber
fracture can occur even before any cracking in the off-axis plies ensues. Johnson
and Chang [79, 80] carried out extensive experiments on a variety of multidirec-
tional laminate congurations and found that for laminates having a ply angle
greater than 45
, ply cracking is a dominant damage mode, with possible delami-

nation at free edges. For instance, in [0/y/0] laminates with y _ 45
the dominant
damage was ply cracking [31, 35], while in [90/30/30]
s
a combination of edge
delamination and ber fracture led to nal failure. Crocker et al. [35] also found
that in [0/45/0] laminates, delamination ensued as soon as ply cracks initiated in
45
-plies.
In case of multidirectional laminates containing 90
-plies, an off-axis ply

adjacent to a 90
-ply shows numerous partial cracks, which may or may not

join to form through cracks on increase of loading. The cracks in the contiguous
ply almost always start from its interface with the 90
-ply. Experiments by
Yokozeki and coworkers [50, 51] on [0/y
2
/90]
s
laminates also point out that
the angle of intersection between the 90
- and y-plies and the thickness of the

y-plies may have a signicant impact on the initiation and growth of cracking
in these plies.
In more general crack systems in multiple orientations, another important
consideration is the interaction between cracks in two adjacent plies. This inter-
action can enhance cracking in a certain orientation thereby causing further
stiffness degradation. For instance, in quasi-isotropic laminates 45
cracks pro-
mote enhanced cracking in the 90
-plies. This intra-ply crack interaction is a

complex function of the relative crack positions, orientations, crack sizes (ply
thicknesses), and density of cracks in different orientations. This makes determin-
ation of stresses in cracked laminates generally impossible to solve analytically,
and numerical computations such as 3-D FEM are then necessary. Figure 6.15
depicts a representative cell of a [0/90/y
1
/y
2
]
s
laminate with cracks in multiple off-
axis orientations.
6.4 Modeling approaches
In the following we describe the main approaches to predicting the evolution of
ply cracking in composite laminates. In the models discussed the ply cracks are
assumed to be fully developed through the ply thickness as well as in the specimen
width direction. In a given array of ply cracks, all cracks are assumed to be of the
same size, shape, and orientation. The problem to address here is the crack
multiplication process as a function of applied loading. As described above, there
are two directions in which this problem can be pursued: one based on a strength
criterion (i.e., point failure), and the other based on an energy criterion (i.e.,
surface formation). This section deals exclusively with cross-ply laminates; multi-
directional laminates are considered later. The basic BVP to solve here is the same
as that approached earlier in Chapter 4, see Figure 4.7. We will use the same
symbols and notations (Section 4.4), unless specied otherwise.
6.4.1 Strength-based approaches
According to these models, microcracks form when the local stress (or strain) state
in a ply reaches a critical level [3739, 8188]. The most commonly used values of
critical level are: failure strain (e
1T
) or failure stress (s
tu
) of a ply in transverse
tension. Other lamina failure criteria such as TsaiWu and Hashins criteria,
described in Section 2.2.4, are also utilized. Since the stress state at the onset of
transverse cracking is not generally uniform [45], these models fail to account for
differences in crack initiation (as a material point failure process) and crack
progression (as a surface growth process). Consequently, the ply thickness effect
on transverse cracking cannot be properly treated by these criteria.
One problem with strength-based criteria, and more generally with all proced-
ures that require knowledge of the local stress states for failure assessment, is that
Figure 6.15. A representative cell illustrating multiple cracking systems in a [0/90/y
1
/y
2
]
s
laminate.
determining the local stresses analytically is possible only in a few cases, mostly for
cross-ply laminates. The larger problem, however, is the lack of agreement of
strength-based predictions with the experimental data. Use of statistical strength
concepts [83, 84, 86, 89, 90] may improve predictions, but require additional
material data. In the following we provide a brief overview of the strength-based
models.
A basic shear lag analysis applied to determination of crack initiation strain
and crack multiplication was carried out in [91]; it has been covered previously
in Section 4.2. It is a one-dimensional analysis, and therefore does not provide
accurate stress perturbation caused by cracking. The analysis of multiple
cracking in a unidirectional ber composite was later applied to the case of
transverse cracking in cross-ply laminates by Garrett and Bailey [38] (see Section
4.4.1). According to the analysis, the load shed by the transverse plies in the
crack plane and transferred back to the transverse plies over the distance y is
given as [38]
F = 2t
0
ws
0
1 e
by
_
; (6:6)
where t
0
is the thickness of the 0
-ply, w is the specimen width, Ds

0
is the maximum
additional stress on the longitudinal ply as a result of cracking in the transverse ply
(occurring in the plane of crack and decaying exponentially away from the crack
surface), and b
2
is the shear lag parameter, dened in Eq. (4.62), as
b
2
= G
90
xz0
1
E
90
x0
1
lE
0
x0
_ _
; (6:7)
where G
90
xz0
is the initial in-plane shear modulus of the 90
-ply, l = t
0
/ t
90
is the
ply thickness ratio, and E
0
x0
and E
90
x0
are the longitudinal moduli for the 0
- and
90
-plies, respectively. The transverse ply will fail in tension when

F = 2t
90
ws
tu
; (6:8)
where s
tu
is the transverse ply strength. The rst crack is assumed to form
in the middle of the specimen length at Ds
0
= (t
90
/ t
0
) s
tu
, i.e., at an applied
load of s
0
= E
c
e
tu
, where E
c
is the longitudinal modulus for the composite and
e
tu
is the transverse ply cracking strain. Next, the cracking process will cause
second and third cracks (one above and one below the rst crack) to form
simultaneously. From Eq. (6.6) and Eq. (6.8) with y = l, where 2l is the crack
spacing, Ds
0
will be
s
0
=
1
l
s
tu
1 e
bl
: (6:9)
These cracks will perturb the force transferred such that the new cracking process
will occur at
s
0
=
1
l
s
tu
1 e
bl
2e
bl=2
: (6:10)
195 6.4 Modeling approaches
Similarly, the (N+2)th crack formation will occur when
s
0
=
1
l
s
tu
1 e
bl
N
2e
bl
2N
: (6:11)
The evolution of crack density thus can be generated using the above iterative
scheme. Figure 6.16 shows the model prediction of crack spacing variation with
increasing loading for a glass-polyester specimen of length 130 mm and transverse
ply thickness of 3.2 mm. Although this model represents the general trends of the
experimental results it generally underestimates the average crack spacing. This is
supposedly due to the constant strength of the 90
-ply resulting in new cracks

midway between existing cracks. Later models tried to address this issue by
representing 90
-ply strength as a probabilistic function. Manders et al. [92]

considered a two-parameter Weibull distribution of the strength along the length
of the 90
-ply and represented the risk of rupture per unit volume as

p(s) =
s
s
+
_ _
v
; (6:12)
where the constant s* is the scale parameter in terms of stress, and v is the shape
parameter. The cumulative distribution function for failure is given as
5
4
3
2
1
0
0 100
Applied stress (MNm
2
)
C
r
a
c
k

s
p
a
c
i
n
g

(
m
m
)
200 300
Figure 6.16. Prediction of crack spacing as a function of applied stress using the shear
lag model [38]. Reprinted, with kind permission, from Springer Science+Business Media:
J Materials Science, Multiple transverse fracture in 90 degrees cross-ply laminates of a glass
ber-reinforced polyester, Vol. 12, pp. 15768, K. W. Garrett and J. E. Bailey.
S
V
= 1 exp
_
V
p(s) dV
_
_
_
_
= 1 exp A
_
L
p(s) dy
_
_
_
_
; (6:13)
where A is the area, and L is laminate length. Taking a logarithm of all the terms in
Eq. (6.13) gives
ln (1 S
V
) = A
_
L
p(s) dy ~ ApL: (6:14)
The quantity Ap is obtained by plotting ln(1 S
V
) against L. The Weibull param-
eters s* and v are obtained by tting the experimental crack density evolution with
the model. To describe the stress s in the laminate, Manders et al. [92] used initial
shear lag analysis, covered previously in Section 4.4.1. The model predictions of
crack spacing variation as a function of applied strain are shown in Figure 6.17.
A similar analysis was performed by Fukunaga et al. [83] who also used the
Weibull distribution for 90
-ply strength but instead used interlaminar shear lag

analysis, also covered in Section 4.4.1. The effects of thermal residual stresses and
Poissons contraction are also included in this analysis. According to their analysis
the applied axial stress and crack spacing (2l) are related as
s
c
=
E
c
Q
22
1
Q
12
A
12
Q
22
A
22
_ _ s
+
L
0
t
1
2lt
90
d
1
_ _
1=v
s
90
xx R
_ _
; (6:15)
where s
+
= s
90
xx0
ln 2=L
0
t
1
w ( )
1=v
with L
0
, w, and t
1
being the length, width, and
thickness of a single 90
-ply, respectively, and d

1
=
1
l
_
l
0
1 cosh bx= cosh bl ( ) [ [dx is
a parameter that reects the effect of stress nonuniformity on 90
-ply strength.
The corresponding strains can be calculated from the stressstrain relations for the
cracked laminate derived in [83] as
0
C
r
a
c
k

s
p
a
c
i
n
g

(
m
m
)
30
0.2 0.4 0.6 0.8 1.0
Strain (%)
1.2 1.4 1.6 1.8 2.0 2.2
20
10
0
Figure 6.17. Variation of average crack spacing for a [0/90/0] glass/epoxy composite laminate
as predicted using the probabilistic shear lag model by Manders et al. [92] for. Solid and
open circles represent experimental data from two specimens, while the solid line represents
the model prediction.
e
c
=
s
c
E
c
1
t
90
Q
22
t
0
Q
11
1
Q
12
A
12
Q
22
A
22
_ _
tanh al
al
_ _
t
90
s
90
xx R
t
0
Q
11
tanh al
al
: (6:16)
The crack initiation strain can be obtained by setting tan al / al = 0 (as l ) in
Eq. (6.16), and 2l = L
0
, d
1
= 1 in Eq. (6.15). Therefore, the applied strain at rst
cracking in the 90
-ply is given by
e
0
=
1
Q
22
1
Q
12
A
12
Q
22
A
22
_ _ s +
t
1
t
90
_ _
1=v
s
90
xx R
_ _
: (6:17)
These statistical descriptions of ply strength yielded good results for cross-ply
laminates. But these models cannot account properly for effects of changes in ply
thickness. Also, as mentioned before, shear lag analysis is a one-dimensional stress
analysis and therefore cannot be accurate. More recently, 2-D shear lag analysis
based on Steifs parabolic displacement variation in 90
-ply [93] have also been

tried [55, 9496]. A detailed discussion on the probabilistic concept will be covered
later in Section 6.5.
6.4.2 Energy-based approaches
The origin of energy-based approaches to crack extension lies in fracture mechanics
(see Section 2.4 for linear elastic fracture mechanics). In the classical version of
brittle fracture, the energy considerations lead to the condition that crack tip growth
becomes unstable when the energy release rate is equal to or greater than the fracture
toughness of the material containing the crack, i.e., G_G
c
. This material property is
obtained by an independent test, which has been standardized in some way. For
perfectly brittle fracture, i.e., when no other energy dissipating mechanism other
than that expended in the crack surface formation exists, the fracture toughness
equals twice the surface energy. In multiple cracking within a composite laminate,
however, the situation is not as assumed in the brittle crack extension of a crack tip
for which the criterion stated above applies. Here, a ply crack on extension is
arrested at the ply interfaces and any further input of energy to the laminate supplied
by external loads goes into the formation of more ply cracks. Although the individ-
ual ply crack goes through the stages of through-thickness growth and growth in the
ber direction, in most cases the ply cracks form quickly and the analysis is therefore
focused on their multiplication, i.e., an increase in crack density. Because of this, a
modication of conventional fracture mechanics, called nite fracture mechanics
[97], has been proposed. Thus, in contrast to conventional fracture mechanics,
transverse cracking comprises events that involve a nite amount of new fracture
area. For cross-ply laminates, nite fracture mechanics coupled with variational
stress analysis has been used in several works [5, 33, 46, 9799].
The criterion for nite surface formation under brittle fracture condition can be
written as
_ g A; (6:18)
where DG is the energy change (release) during crack surface increase by area
DA and g is the surface energy per unit area of the new crack formed.
In the following we describe the energy-based approaches combined with dif-
ferent approximate stress analyses used.
Shear lag analysis
Among the early energy-based analyses for ply cracking is that proposed by
Parvizi et al. [81]. They recognized the role of constraint of the outer 0
-plies to
transverse cracking in [0/90]
s
laminates and took account of it in the energy
balance during cracking. Recalling from ACK theory, Section 4.2, a crack does
not form in a specimen loaded in constant tension until
W _ U
S
U
D
2g
m
V
m
; (6:19)
where V
m
is the matrix volume fraction and, dened per unit cross-sectional area
of the composite, DW is the work done by applied stress, DU
S
is the increase in
energy stored in the composite volume, U
D
is the energy lost by some dissipative
processes during cracking (sliding friction between debonded bers and matrix),
and g
m
is the matrix surface energy per unit surface area. For the present case of
cracking in a transverse ply, the above inequality becomes
W _ U
S
2g
t
t
90
h
; (6:20)
where g
t
is the surface energy of the transverse ply for cracking in a direction
parallel to the bers, and h = t
0
+ t
90
. For a linear elastic body the work of
external forces equals half the stored strain energy, i.e.,
U
S
=
1
2
W: (6:21)
Substituting Eq. (6.21) into Eq. (6.20), the cracking will occur when
W _ 4g
t
t
90
h
: (6:22)
Oncracking anadditional stress is thrownontoouter uncrackedplies, andthe laminate
increases in length. The work done during this process can be derived as [81]
W =
2E
90
x0
E
c
e
2
tu
lE
0
x0
b
; (6:23)
where b is as dened earlier in Eq. (6.7). Combining Eqs. (6.22) and (6.23), the
strain to initiate cracking in transverse ply is given by
e
0
= e
min
tu
=
2t
0
E
0
x0
g
t
b
hE
90
x0
E
c
: (6:24)
The model predictions for various transverse ply thicknesses are shown in Figure
6.18 against experimental data for glass/epoxy cross-ply laminates.
A more sophisticated energy-based analysis of progressive ply cracking using
one-dimensional stress analysis was performed by Laws and Dvorak [100]. (Refer
to Section 4.4.1 for the shear lag analysis.) Consider a cracked cross-ply laminate
in state 1 where cracks are a distance 2l apart, and the ligament AB between
the cracks is as yet uncracked (Figure 6.19(a)). When the applied load reaches a
critical value s
c
, a new crack appears in this segment at some location C (Figure
6.19(b)). Assuming that the load is kept xed during formation of this additional
crack, Laws and Dvorak calculated the energy released during cracking of lamin-
ate of width w as
=
2t
2
90
whE
c
bt
0
E
0
x0
E
90
x0
s
90
xxR
E
90
x0
E
0
x0
s
c
_ _
2
tanh
bl
1
2t
90
tanh
bl
2
2t
90
tanh
bl
t
90
_ _
; (6:25)
0
2.5
1 2
Transverse ply thickness 2d (mm)
T
r
a
n
s
v
e
r
s
e

c
r
a
c
k
i
n
g

s
t
r
a
i
n

e
t
u

,

e
t m
u
i
n

(
%
)
3 4
2.0
1.5
1.0
0.5
0.0
Experimental e
tu
Theoretical e
tu
min
Figure 6.18. Values of crack initiation strain e
min
tu
as a function of ply thickness 2t
90
= 2d
(from Eq. (6.24)) and experimental data for various ply thicknesses in glass ber/epoxy
cross-ply laminates. The horizontal line depicts the limiting values of e
min
tu
for large inner-ply
thicknesses [81]. Reprinted, with kind permission, from Springer Science+Business
Media: J Mater Sci, Constrained cracking in glass ber-reinforced epoxy cross-ply laminates,
Vol. 13, 1978, pp. 195201, A. Parvizi, K. W. Garrett and J. E. Bailey.
A B
z
x
State 1
(a) (b)
State 2
A
s=2l l
1
l
2
B C
Figure 6.19. Progressive multiplication of ply cracks in transverse layer of cross-ply laminate:
(a) state 1 with crack spacing, s = 2l; (b) state 2 with an additional crack in the ligament
AB at location C.
where b
2
= Kt
90
1=E
90
x0
_ _
1=lE
0
x0
_ _ _
is the shear lag parameter (see Eq. (4.67)),
and s
90
xxR
is the thermal residual stress in the transverse ply. Substituting Eq. (6.25)
in Eq. (6.18), and using DA = 2wt
90
, a new crack will form if
t
90
hE
c
bt
0
E
0
x0
E
90
x0
s
90
xxR
E
90
x 0
E
0
x0
s
c
_ _
2
tanh
bl
1
2t
90
tanh
bl
2
2t
90
tanh
bl
t
90
_ _
_ g: (6:26)
The rst ply failure stress (the applied stress at crack initiation) will be given by the
limit l , i.e.,
s
fpf
c
=
bt
0
E
0
x0
E
c
g
t
90
hE
90
x0
_ _
1=2
E
c
E
90
x0
s
90
xxR
: (6:27)
The thermal residual stress, s
90
xxR
, and other parameters in the expression above
are known from property data for an undamaged laminate. Therefore,
Eq. (6.27) provides a relation among s
fpf
c
, b, and g. Now s
fpf
c
and g can
be determined from experimental data. Therefore, Laws and Dvorak regard
Eq. (6.27) as the relation that determines the shear lag parameter b. Once
b is known, the applied stress needed to cause cracking at location C is given
by
s
c
(l
1
) = s
fpf
c

E
c
E
90
x0
s
90
xxR
_ _
tanh
bl
1
2t
90
tanh
bl
2
2t
90
tanh
bl
t
90
_ _
1=2
E
c
E
90
x0
s
90
xxR
:
(6:28)
In a practical scenario, the location C is random due to spatial variation of the
resistance to crack formation. Let p be the probability density function for the
next crack to occur at a given location. In a laminate which already contains
cracks with normalized crack density, r
c
= t
90
/ l, the expected value of the applied
stress to cause additional cracking is then
E s
c
r
c
( ) [ [ =
_
2l
0
p x ( )s
c
x ( ) dx: (6:29)
Three possible choices for p(x) are:
Case 1: The next crack occurs midway, so
p x ( ) = d(x l); (6:30)
where d(x) is the Dirac delta function.
Case 2: All locations are equally likely. Therefore,
p x ( ) =
1
2l
: (6:31)
Case 3: p(x) is proportional to the stress at the location. For this case, p(x) is
given by [100] as
p x ( ) = s
90
xxR
E
90
x0
E
0
x0
s
c
_ _
1
cosh
bx
t
90
cosh
bl
t
90
_
_
_
_
: (6:32)
For case 1, the solution is explicitly given by
E s
c
r
c
( ) [ [ = s
fpf
c

E
c
E
90
x0
s
90
xxR
_ _
2 tanh
b
2r
c
tanh
b
r
c
_ _
1=2
E
c
E
90
x0
s
90
xxR
: (6:33)
For cases 2 and 3, the integral in Eq. (6.29) must be evaluated numerically.
The model predictions for these cases of p(x) are shown in Figure 6.20 (with
g = 193 J/m
2
and b = 0.9). Based on comparison with the experimental data,
Laws and Dvorak argue that fracture mechanics-based p(x) (case 3) is the most
promising choice. With this choice, model predictions also compare well with
another set of experimental data from [1] for graphite/epoxy laminates.
Variational analysis
Nairn [99] used the variational approach [101] for cracked cross-ply laminates,
including thermal residual stresses, in conjunction with the energy release rate
criterion to predict crack densities in cracked cross-ply laminates. His predictions
showed good agreement with experiments when the critical energy release rate for
matrix cracking, dened as such, was deduced from test data rather than evaluat-
ing it independently.
Another damage evolution model for cross-ply laminates is by Vinogradov and
Hashin [97, 98, 102]. It uses variational analysis [101] for stress computation and
nite fracture mechanics [97] for cracking criterion. Recalling from Section 4.6,
Figure 6.20. Crack density evolution in a glass/epoxy [0/90]
s
laminate for three choices of
probability distribution function p(x) [100]. The experimental data are from [8]. Note: [8]
has additionally three data points at the high load end that fall away from model
predictions. Reprinted, with kind permission, from N. Laws and G. J. Dvorak, J Compos
Mater, Vol. 23, pp. 90016, copyright #1988 by Sage Publications.
where the stress calculations are described in detail, complementary energy change
due to presence of N transverse cracks can be derived as [98]
+
=
N
n=1
+
n
= s
90
xx0
_ _
2
t
2
90
C
22
N
n=1
w r
n
( ); (6:34)
where C
22
= l 1 ( ) 3l
2
12l 8
_ _
1
60E
2
(see Eq. (4.151)), and
w(r
n
) =
d
3
f
n
dx
3
r
n
= 2a
1
a
2
a
1
2
a
2
2
_ _
cosh (2a
1
r
n
) cos(2a
2
r
n
)
a
1
sin (2a
2
r
n
) a
2
sinh (2a
1
r
n
)
; (6:35)
and the summation in Eq. (6.34) is over all blocks bounded by adjacent cracks. It
is noted that r
n
= l
n
/t
90
is the normalized crack spacing and should not be
confused with the crack density. For clarity of notations, the reader is referred
to Section 4.6. Initially when the cracks are far apart (r
n
), the function
achieves its asymptotic value given by
w() = 2a
1
a
1
2
a
2
2
_ _
: (6:36)
Now consider the state of a laminate with N cracks. When a new crack appears,
the energy release can be expressed as [97]
n
=
+
s
N1
_ _
+
s
N
_ _
; (6:37)
where s
N
and s
N+1
are the stress elds before and after formation of a new (N+1)th
crack. The stresses include both mechanical and thermal effects (if present).
Assuming that the new crack appears instantly, both stress elds are evaluated
at the same external load. Equation (6.37) can be rewritten as
n
=
+
s
N1
_ _
+
s
0
_ _ _

+
s
N
_ _
+
s
0
_ _ _
; (6:38)
where s
0
is the stress eld in the undamaged material at the same external load.
Assuming Eq. (6.34) is a good approximation of the energy, the energy release
due to the (N+1)th crack is given by
n
= s
90
xx0
_ _
2
t
2
90
C
22
N1
i=1
w r
N1
i
_ _
s
90
xx0
_ _
2
t
2
90
C
22
N
i=1
w r
N
i
_ _
; (6:39)
where s
90
xx0
is the stress in the undamaged 90
-ply, r
N
i
is the nondimensional crack
spacing for the ith block for the Nth cracking step (going from N to N+1 cracks).
Using the energy release rate criterion, Eq. (6.18), taking averages and express-
ing variables as continuous variables, the energy released during cracking can be
expressed as (see [98] for full derivation)
g = s
90
xx0
_ _
2
t
90
C
22
d
d r
w
r
_ _
r
2
: (6:40)
Using the lower bound of the laminate longitudinal modulus (see Section 4.6) by
Hashin [101]
1
E
+
x
=
1
E
0
k
2
1
t
90
h
C
22
w
r
; (6:41)
where k
1
= s
90
xx 0
=s
0
xx 0
if the temperature change is absent. Using A = L=2 r, we
nally obtain the cracking criterion as
g =
1
2
s
90
xx0
_ _
2 d
dA
1
E
+
x
_ _
V; (6:42)
where V is the laminate volume. Equation (6.42) represents a particular homo-
thermal case of the general fracture criterion derived by Hashin [103] given by
g =
1
2
s
@S
+
@A
s
@a
+
@A
sT
1
2
@c
+
p
@A
T
2
T
r
_ _
V; (6:43)
where S* is the effective elastic compliance tensor, a* is the effective thermal
expansion tensor, c
+
p
is the effective specic heat of a composite, and T
r
is the
reference temperature.
The damage evolution predictions using this approach are very accurate when
the probabilistic distribution of the energy release rate is utilized. This will be
discussed in the next section.
Nairns original result [48, 99] was quite similar to the fracture criterion
stated above. However, in place of the usual energy release rate (2g) he used
the matrix fracture toughness (G
m
) and suggested that it could be obtained
through tting experimental data for ply cracking. If we assume that the
new crack forms midway between the existing cracks, Nairns fracture criter-
ion is
G
m
= s
2
c
E
2
2
E
2
c
aT
2
C
2
00
_ _
t
90
C
22
2w r=2 ( ) w r ( ) [ [; (6:44)
where C
00
= (1/E
2
) + (1/lE
1
) (see Eq. (4.151)). An alternative formulation is to
allow the new crack to form anywhere between two existing cracks. If the prob-
ability of crack formation at any position is proportional to the tensile stress, the
energy release rate is given by
G
m
= s
2
c
E
2
2
E
2
c
aT
2
C
2
00
_ _
t
90
C
22
w d ( ) w r d ( ) w r ( ) [ [; (6:45)
where
w d ( ) w(r d) w(r) [ [ =
_
r=2
0
w(d) w(r d) [ [ 1 f(r 2d) [ [dd
_
r=2
0
1 f(r 2d) [ [dd
w(d): (6:46)
For cracking in [90
m
/0
n
]
s
laminates, the expressions for energy release rate remain
the same except that the constant C
22
is now given as C
22
= (l+1)(3+12l+8l
2
)
(1/60E
2
) (see Eq. (4.164)). It is noted that Nairns analysis included residual thermal
stresses, which mainly change the crack initiation strain (discussed later in detail).
The parameter G
m
is evaluated by tting the model to the experimental data. The
model predictions for a [0/90
3
]
s
glass/epoxy laminate with G
m
= 330 J/m
2
and a
thermal residual stress of 13.6 MPa are shown in Figure 6.21.
Plain-strain formulation
McCartney [104109] developed a model based on the Gibbs free energy, instead
of complementary strain energy as described above. He used a plane strain
formulation for the estimation of elastic moduli of the damaged laminate, which
has been covered in Section 4.7. Consider the damage progression from a state of
m cracks to n cracks in the 90
-ply and assume that each crack formation occurs

under conditions of xed applied tractions. Based on energy considerations, crack
formation will occur when
G _ 0; (6:47)
where DG is the change in Gibbs free energy, and DG is the energy absorbed in the
volume V of laminate due to the formation of new cracks, given by
= V o
n
( ) o
m
( ) [ [; (6:48)
where o denotes the damage parameter which characterizes the crack density
in the laminate. The corresponding change in the Gibbs free energy can be
written as
300
200
100
0
0.0 0.2 0.4 0.6 0.8 1.0
A
p
p
l
i
e
d

s
t
r
e
s
s

(
M
P
a
)
Figure 6.21. The applied stress as a function of crack density in a [0/90
3
]
s
glass/epoxy
composite. The squares are data from [8]. The solid line is the energy release rate analysis t
using Eq. (6.45) with G
m
= 330 J/m
2
and assuming that the initial level of the residual
thermal stresses in the 90
-plies was 13.6 MPa [99]. Reprinted, with kind permission,

from J. A. Nairn, J Compos Mater, Vol. 23, pp. 110629, copyright # 1989 by Sage
Publications.
G =
_
V
g o
n
( ) g o
m
( ) [ [dV; (6:49)
where g(o) represents the Gibbs free energy per unit volume. After some math-
ematical treatment (see [109] for details) the cracking criterion becomes
ê o
n
( ) ê o
m
( ) [ [
2
1
~
E o
n
( )
1
~
E o
m
( )
m
A
o
n
( ) g o
n
( ) ( )
2
m
A
o
m
( ) g o
m
( ) ( )
2
2 o
n
( ) o
m
( ) [ [ > 0;
(6:50)
where ê o ( );
~
E o ( ); m
A
o ( ); g o ( ), and G(o) denote axial strain, axial Youngs modu-
lus, in-plane axial shear modulus, applied in-plane shear strain, and the energy
absorption per unit volume for length 2L of laminate, respectively, for given
damage state o. G(o) is here given by
o ( )
h
(90)
hL
M
j=1
d
(90)
j
g
(90)
j
; (6:51)
where 2h
(90)
and 2h represent the total thickness of 90
-plies and whole laminate,

respectively; M is the number of potential cracking sites in the 90
-plies, which are

ordered from the top to bottom of the plies, taken in increasing order from the
center of the laminate to the outside; and 2g
(90)
j
is the fracture energy for the jth
potential cracking site. The expressions for other parameters can be found in [109].
The crack initiation strain can be obtained by setting o
m
= o
0
(undamaged state)
in Eq. (6.50).
COD-based models
Following the work of Parvizi et al. [81] and Wang and Crossman [110] on
energy release rates to study the formation of cracks in cross-ply laminates, Joffe
and coworkers [111114] considered fully developed cracks and developed
a methodology to predict the multiplication of transverse cracks based on the
virtual crack closure technique. The idea is to probe the region between two
existing cracks and introduce a virtual crack. For the introduced virtual
crack the work performed to close the crack surfaces is calculated and compared
with the energy needed to create a crack, the critical energy release rate (G
c
),
at this position. A crack is taken to form when the work to close the crack
exceeds G
c
.
Consider a damaged [0
m
/90
n
]
s
laminate with a periodic system of N self-
similar cracks with spacing s = 2l in the 90
-ply (Figure 6.19(a)). At applied

laminate stress s
0
(and corresponding far-eld stress s
90
x0
in the 90
-layer) a new
crack develops midway between two existing cracks and the total number of
cracks becomes 2N with spacing l (Figure 6.19(b)). According to the crack closure
concept the released energy due to these N new cracks is equal to the work needed
to close them. If we denote this work by W
2NN
and the work to close all cracks
simultaneously by W
2N0
, energy balance requires
W
2N0
= W
2NN
W
N0
; (6:52)
where the work to close N cracks with spacing s is
W
N0
= N 2
1
2
_
t
90
t
90
s
90
xx0
u(z) dz = 2Ns
90
xx 0
t
90
u
n
(l) = N
s
90
xx0
_ _
2
E
2
t
2
90
~ u
n
(l); (6:53)
where unit width is assumed and u(z), u
n
; and u
n
represent the variation of the
normal crack opening displacement (COD) along the thickness direction, its
average value, and its average value normalized with respect to the remote
stress and transverse modulus for the ply, respectively. u
n
; and u
n
are thus
dened as
u
n
=
1
t
90
_
t
90
0
u(z) dz; ~ u
n
=
u
n
s
90
xx0
=E
2
_ _
t
90
: (6:54)
Similarly, the work to close 2N cracks with spacing l is given by
W
2N0
= 4N
s
90
xx0
_ _
2
E
2
t
2
90
~ u
n
(l=2): (6:55)
Substituting Eqs. (6.53) and (6.55) into Eq. (6.52), the energy released by forma-
tion of a crack midway between two existing cracks of spacing s is
W
2NN
= 2N
s
90
xx0
_ _
2
E
2
t
2
90
2~ u(l=2) ~ u(l) [ [: (6:56)
The cracks form when this work is greater than or equal to the cumulative surface
energy of newly created surfaces, i.e.,
W
2NN
_ 2 N 2t
90
G
c
: (6:57)
From Eqs. (6.56) and (6.57), the criterion for crack formation is
s
90
xx0
_ _
2
2E
2
t
90
2~ u(s=2) ~ u(s) [ [ _ G
c
: (6:58)
To analyze cracking in an arbitrary position between two pre-existing cracks,
a new crack is introduced in an arbitrary position between the cracks (Figure 6.19
(b)), which leads to a new damage state with one crack spacing equal to s
1
and the
second one equal to s
2
= s s
1
. The cracking criterion in this case is
s
90
xx0
_ _
2
2E
2
t
90
2~ u(l
1
=2) ~ u(l
1
) 2~ u(l
2
=2) ~ u(l
2
) [ [ _ G
c
: (6:59)
The authors applied their analysis for the prediction of crack density evolution
in glass/epoxy [y/90
4
]
s
laminates for the case of varying crack spacing.
A Weibull distribution for G
c
was utilized and the CODs were calculated
using FE analysis. The predictions are shown against experimental data in
Figure 6.22.
Adolfsson and Gudmundson [11] also developed an energy-based damage
evolution approach using their stress analysis. Basic stress analysis using this
approach is covered in Section 4.9.1, although they updated their analysis to
include bending loads, which can be found in [11, 115]. The energy model for
crack density evolution is based on changes in strain energy due to cracking. From
[11], the strain energy per unit in-plane area of the damaged laminate with n plies
may be written as
w
(c)
=
1
2
e a
(c)
T e
(R)
_ _
T
C
(c)
e a
(c)
T e
(R)
_ _
n
k=1
h
k
T; s
k(R)
_ _
; (6:60)
where bold-face letters represent matrices; C
(c)
, a
(c)
, and e
(R)
are the stiffnesses,
thermal expansion coefcients and residual stress-induced eigenstrains, respect-
ively, of the cracked laminate. Expressions for these crack density quantities were
derived in [115]. In Eq. (6.60), DT is the temperature difference between the curing
and service temperature and h
k
are the functions containing energy stored in the
laminate due to interlaminar constraints and thermal residual stresses (their
expressions are given in Appendix A of [11]). From the strain energy, the energy
release rate for the ith cracked ply is derived as
0.8
[q/90
4
]
s
FEM model, energy approach
0.6
0.4
0.2
0.0
q=0
q=15
q=30
q=40
0
C
r
a
c
k

d
e
n
s
i
t
y

(
c
r
/
m
m
)
50 100
Stress s
0

(MPa)
150 200 250
Figure 6.22. Evolution of crack density as a function of applied stress using energy model
for [y/90
4
]
s
laminates. Symbols represent experimental data and lines represent the
average of four runs using the energy model [113]. Reprinted, with kind permission, from
Compos Sci Technol, Vol. 61, R. Joffe, A. Krasnikovs, and J. Varna, COD-based simulation
of transverse cracking and stiffness reduction in [S/90n]s laminates, pp. 63756, copyright
Elsevier (2001).
G
i
=
@U
@A
i
=
@ Aw
(c)
_ _
@A
i
; (6:61)
where A
i
is the crack surface area in ply i. The area A
i
is given by A
i
= Ar
i
with
normalized crack density r
i
= t
i
/ l
i
, where l
i
is the average spacing of cracks in the
ith ply and A is the laminate in-plane area. From Eqs. (6.60) and (6.61) the energy
release rate G
i
for cracking in the ith ply is given by
G
i
=
@a
(c)
@r
i
T
@e
(R)
@r
i
_ _
T
C
(c)
e a
(c)
T e
(R)
_ _
1
2
e a
(c)
T e
(R)
_ _
T
@C
(c)
@r
i
e a
(c)
T e
(R)
_ _
N
k=1
@h
k
@r
i
:
(6:62)
The above expression contains the derivatives of the effective thermoelastic
properties of the damaged laminate with respect to the ply crack densities.
Calculating these quantities is a more complex task than determining the proper-
ties themselves. For this purpose, either FEM or the approximate analytical
expressions given in Appendix A of [11] have to be utilized. The model predictions
for the stressstrain response of graphite/epoxy cross-ply laminates are compared
with experimental data in Figure 6.23.
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Strain (%)
0
100
200
300
400
500
600
S
t
r
e
s
s

(
M
P
a
)
(0/90/0)
(0/90
2
/0)
(0/90
4
/0)
(0/90
8
/0)
Simulations
Figure 6.23. Stressstrain response of graphite/epoxy cross-ply laminates. Symbols
represent experimental data whereas the solid lines represent predictions from the model by
Adolfsson and Gudmundson [11]. Reprinted, with kind permission, from Int J Solids Struct,
Vol. 36, E. Adolfsson and P. Gudmundson, Matrix crack initiation and progression
in composite laminates subjected to bending and extension, pp. 313169, copyright
Elsevier (1999).
QuHoiseth analysis
The evaluation of moduli for damaged cross-ply laminates using the approach
proposed by Qu and Hoiseth [116] is covered in Section 4.5. The cracking criterion
for this model is derived as
G
c
=
2e
c
2
t
90
E
c
E
2
(E
1
E
2
)r
exp
dr
2t
90
_ _
exp
dr
t
90
_ _ _ _
; (6:63)
where G
c
is the in-plane mode I fracture toughness of transverse ply; e
c
is the applied
strain; E
c
is the plain-strain Youngs modulus of the undamaged cross-ply laminate
in the longitudinal direction; E
1
and E
2
are longitudinal and transverse moduli of
the ply, respectively; d is the average crack opening displacement of the 90
crack;
and r = t
90
/ l is the normalized crack density. The threshold strain at which the
transverse matrix cracking initiates can be obtained by setting r 0 in Eq. (6.63) as
e
0
=
G
c
E
1
E
2
( )
dE
c
E
2
: (6:64)
The model predictions for a cross-ply laminate made of AS/350106 material for
two values of G
c
are shown in Figure 6.24. The experimental data are from [117].
6.4.3 Strength vs. energy criteria for multiple cracking
There is a fundamental difference between strength- and energy-based criteria
when applied to multiple ply cracking in laminates. Strength essentially represents
the failure at a material point when a specied stress component or function of
0.0
C
r
a
c
k

d
e
n
s
i
t
y
0.0
0.5
s
c
= E
c
e
c
Prediction with G
c
=130J/m
2
1.0
1.5
2.0
0.5
Applied stress
1.0
AS/350106 [0
2
/90
2
]
s
1.5 2.0
Experimental data
Prediction with G
c
=104J/m
2
Figure 6.24. Comparison between model predictions and experimental data for the COD
model of Qu and Hoiseth [116] for a cross-ply laminate, s
c
represents the applied stress on
the composite. Reprinted, with kind permission, from Fatigue Frac Eng Mater, Vol. 24,
J. Qu and K. Hoiseth, Evolution of transverse matrix cracking in cross-ply laminates,
pp. 451464, copyright Wiley (1998).
stress components reaches a critical value. This approach is a legacy of homo-
geneous materials such as metals and ceramics where yielding at a point (in a
metal) is assumed to occur according to, e.g., the von Mises criterion, or brittle
failure (in a ceramic) is assumed to occur when the maximum tensile principal
stress reaches a critical value. In case of yielding there is no ambiguity in terms of
the stress components attaining critical value at a point since this type of failure
can spread spatially from one point to another as indicated by the contour of the
yield criterion. However, brittle failure represents the instability of crack growth,
and unless a crack exists its growth is meaningless. This difculty has been
conventionally overcome (or bypassed) by assuming that brittle failure according
to a point-failure (strength) criterion is the concurrent formation and instability of
a crack. In an unconstrained failure case, such as the brittle failure of an unre-
inforced ceramic, this approximating assumption causes little difculty since the
initiation of cracking and its unstable growth are usually not far apart, i.e., they
occur at roughly the same applied load. However, when the constraint to crack
growth is imposed by the presence of reinforcements, or stiff elements in the matrix
generally, then the formation of a crack and the instability of its growth are
determined by different conditions. This fact was realized in [91] when it was found
that the strength criterion was inadequate to predict the formation of multiple
cracking in unidirectional brittle matrix composites. Energy considerations were
then made by recognizing the dissipation of energy in crack surface formation.
Multiple cracking in composite materials is an inherent feature of the failure
process due to the presence of directed interfaces (ber/matrix and interlami-
nar) that impart mechanisms of stress transfer from the cracking elements to
the noncracking elements, which in turn provide a constraint to the cracks.
Thus, the inadequate incorporation of constrained cracking in a multiple
cracking process is bound to induce error. One example, unfortunately not
uncommon, is using solutions of crack opening displacements in an innite
medium in models of multiple cracking.
6.5 Randomness in ply cracking
Physical observations of ply cracking indicate that in the early stages of the cracking
process randomness exists in the location of cracks, their size, and how the evolu-
tion (growth and multiplication) of cracking occurs. As the cracking process
evolves, randomness tends to decrease, and as the process approaches saturation,
uniformity in crack spacing results. The causes of randomness are many, most
induced by the manufacturing process. For instance, the ber volume fraction can
vary spatially. Image analysis reported in [118] showed that in a T300/914 carbon/
epoxy composite with an average ber volume fraction of 55.9% the local volume
fraction ranged between 15 and 85%. Other common defects are voids and inclu-
sions in the matrix, partially cured regions of matrix, broken bers, ber waviness,
unbonded regions of interfaces at ber surfaces, and between plies.
211 6.5 Randomness in ply cracking
Several attempts exist in the literature to treat random variations of microstruc-
ture. Silberschmidt [119, 120] suggested a lattice scheme which incorporates the
effects of the initial microstructural randomness as well as a dispersed evolution of
damage and its transition to spatially localized matrix cracking. The scheme
involves mapping a dynamic matrix of stress-renormalizing coefcients onto the
lattice of elements covering the cracked (90
) layer. Figure 6.25 shows the distri-

bution of ply cracks at different times in load history based on the scheme for a
[0
2
/90
4
]
s
T300934 laminate during fatigue loading.
Another example of crack distribution for [0/90]s glass/epoxy laminates, tested
under quasi-static tensile loading by Manders et al. [87], is shown in Figure 6.26,
along with the average axial stress distribution along the laminate length, esti-
mated by Berthelot and Le Corre [121]. Clearly the stress state is such that it is
difcult to predict damage evolution using deterministic approaches. Probabilistic
notions are often used to correct the cracking predictions for random effects, as
stated earlier. Although some probabilistic fracture criteria have already been
discussed, we shall focus here on the variational stress model, emphasizing details
of probabilistic considerations.
80
70
60
50
40
30
20
10
0
0 2 4 6
(cm)
(b)
(a)
8 10
s
9x
0x
(
M
P
a
)
Figure 6.26. Crack distribution (a) and corresponding variation (b) of the average
longitudinal stress in the 90
-ply along the laminate length, estimated for [0/90]

s
glass/epoxy
laminates [121]. Reprinted from Compos Sci Technol, Vol. 60, J. M. Berthelot and J. F. Le
Corre, Statistical analysis of the progression of transverse cracking and delamination in
cross-ply laminates, pp. 265969, copyright (2000), with permission from Elsevier.
Figure 6.25. Distribution of ply cracks in a [0
2
/90
4
]
s
laminate at different times in the
load history based on the lattice scheme by Silberschmidt [119, 120]: (a) 100 cycles; (b) 410
3
cycles; (c) 10
5
cycles; and (d) 210
5
cycles [120]. Reprinted, with kind permission,
from Springer Science+Business Media: J Mater Sci, Effect of micro-randomness on
macroscopic properties and fracture of laminates, Vol. 41, 2006, pp. 676876, V. V.
Silberschmidt.
A fracture mechanics-based stochastic model to predict progression of ply
cracking was initially proposed by Wang and coworkers [1, 110, 122125]. The
authors postulated that cracking in transverse plies of cross-ply laminates is
governed by a characteristic distribution of effective aws, which are essentially
inherent material microcracks that cannot be seen until grown to macroscopic
dimensions. Thus these microcracks act individually as initiators of cracks that
propagate to form fully grown transverse cracks. The distribution of aw size f(a)
and spacing f(S) along specimen length are assumed to follow the following
normal probability distributions
f (a) =
1
a
2p
_ exp
(a m
a
)
2
2s
2
a
_ _
;
f (S) =
1
S
2p
_ exp
(S m
S
)
2
2s
2
S
_ _
;
(6:65)
where 2a is the average aw size, S is the average distance between two adjacent
aws, and m
a
, m
S,
s
a
, and s
S
are tting parameters. The worst of the aws causes
the rst ply cracking. With increased loading, smaller aws cause further trans-
verse cracking. The rst transverse crack forms when
G s
c
; a
0
( ) = G
c
; (6:66)
where s
c
is the longitudinal stress applied to the composite, 2a
0
is the initial aw
size, and G
c
is the critical energy release rate, which is assumed to be constant
along the laminate length. The propagation of the aw will be stable if
G s
c
; a
0
a ( )<G
c
; (6:67)
and unstable if
G(s
c
; a
0
a) > G
c
: (6:68)
Progressive cracking will ensue when there is enough energy available for multiple
aws to propagate into fully grown transverse cracks. For crack formation after
the rst crack the energy release rate for aw propagation depends on its relative
distance S from the existing crack and can be expressed as
G(s
c
; a) = R(S)G
0
(s
c
; a); (6:69)
where G
0
is the energy release rate when no crack is present, and R(S) is the energy
retention factor, with a value between 0 and 1, accounting for the presence of a
neighboring crack. Similarly, for a aw to propagate between two existing trans-
verse cracks, the energy release rate is
G(s
c
; a) = R(S
L
)G
0
(s
c
; a)R(S
R
) (6:70)
where S
L
and S
R
are the distance of the aw from the left and right cracks,
respectively. Chou et al. [126] implemented the approach using a Monte Carlo
scheme. The results showed a fair agreement with experimental data. Essentially,
this approach predicts the event when a micro-aw develops into a fully grown
transverse crack, thereby predicting the multiplication of ply cracks. However, the
approach has not gained wide usage because it requires many unknown param-
eters which are found by tting to experimental data. As mentioned earlier,
experimental observations show that the transverse cracks usually grow quickly
through the 90
-ply thickness as well as the specimen width. Therefore, more

recent approaches do not try to predict crack propagation; rather they focus on
the multiplication of cracks, i.e., an increase in crack density.
To illustrate how a more recent fracture criterion can be modied to include
probabilistic measures we follow the treatment of Vinogradov and Hashin [98].
Accordingly, the uncertainties in the cracking process can be categorized into
two probabilistic notions: geometrical and physical. The geometrical
uncertainty refers to the probability of a crack to appear at a certain location
between two existing adjacent cracks, while the physical aspect deals with the
variation of material resistance to crack formation. The geometrical aspect of
probabilistic cracking can be introduced by considering the statistical variation of
distance between adjacent cracks, i.e.,
r =
_
0
rp r ( )dr; w =
_
0
w r ( )p r ( ) dr; (6:71)
where p(r) is the probability density function (PDF) of distances between adjacent
cracks. The criterion for rst crack formation can be found by substituting r
in Eq. (6.40) and using Eq. (6.36), to obtain
g = s
90
xx0
_ _
2
t
90
C
22
w ( ) = 2 s
90
xx0
_ _
2
t
90
C
22
a a
2
b
2
_ _
: (6:72)
In fact the criterion in Eq. (6.72) is expected to predict the initial stage of the
damage evolution curve. For any material block between two existing adjacent
cracks, the cracking criterion in Eq. (6.40) can be rewritten as [98]
g = s
90
xx0
_ _
2
t
90
C
22
w
r x
2
_ _
w
r x
2
_ _
w r ( )
_ _
; (6:73)
where x denotes the nondimensional coordinate of the new crack location between
the two existing cracks. Equation (6.73) is a local criterion for crack formation
because it deals with the location of the next crack.
The physical nature of damage evolution can be achieved by having a
probabilistic variation of material property g. Thus,
g = G(x): (6:74)
The parameter G(x) can be thought of as local toughness of the material by
arguing that it is easy to form a crack at a section which contains many aws and
has a weak interface. The variation of g is usually described using a Weibull
distribution, i.e., the PDF of g can be expressed as
p
r
(g) =

g
0
g g
min
g
0
_ _
1
exp
g g
min
g
0
_ _ _ _
; g _ g
min
; (6:75)
where g
min
is the minimum possible value of g, and and g
0
are parameters of the
distribution, usually evaluated by tting experimental data.
For different laminate systems, i.e., for different mixes of 0 and 90
plies,
e.g., [0
n1
/90
m1
]
s
and [0
n2
/90
m2
]
s
laminates, the distribution parameters
may not be the same. If the parameters for the rst laminate conguration,
1
, g
01
, are known (through tting of experimental data), the parameters
2
, g
02
for the second laminate conguration can be found from the following
relations
2
2
2
_ _
2
1
2
2
_ _
2
1
2
2
_ _ =
m
2
m
1
2
1
1
_ _
2
1
1
1
_ _
2
1
1
1
_ _ ;
g
02
= g
01
1
1
1
_ _
1
2
2
_ _;
(6:76)
where G(x) represents the standard gamma function for the random variable x.
The derivation can be found in the original article [98].
The simulation procedure for this model can be summarized as follows:
1. Choose a ply material or a laminate conguration of a ply material.
2. Distribute random points for possible crack locations along the laminate
length.
3. Generate a random value of g at each point according to the Weibull
distribution.
4. Fit the model predictions to the experimental data for crack density evolution
to deduce the parameters of the Weibull distribution.
5. Calculate Weibull parameters for other laminate systems using Eq. (6.76), and
predict the crack density evolution for these laminates.
Some examples of the numerical simulation results with the tted and calculated
parameters of the distribution are shown in Figure 6.27.
A recent strength-based analysis by Berthelot and Le Corre [121] has revealed
that the choice of probabilistic distribution should account for weakness areas in
the material. This analysis for [0/90
2
]
s
carbon/epoxy laminate shows that a prob-
abilistic distribution of strength which accounts for weakness areas properly
corrects the crack density evolution in the beginning stage (see Figure 6.28).
A divergence between models with and without consideration of weakness areas
is always observed at low crack densities. This is because, initially, cracking is
preferred at weakness areas where the fracture toughness of the material is low
(due to inherent defects) as compared to its average value in the whole laminate.
For glass/epoxy laminates, Berthelot and Le Corre found that delamination
occurs at high crack densities and is the cause of data deviating from the model
prediction.
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
(a)
200
C
r
a
c
k

d
e
n
s
i
t
y

(
1
/
m
m
)
[0/90
3
]
s
[0/90
2
]
s
400 600
Stress (MPa)
800 1000 1200 1400
Avimid K Polymer/IM6
0
1.2
(b)
100
C
r
a
c
k

d
e
n
s
i
t
y

(
1
/
m
m
)
[0/90
4
]
s
[0/90
2
]
s
[0/90]
s
200 300 400
Stress (MPa)
Hercules AS4/35016
500 600 700 800
1
0.8
0.6
0.4
0.2
0
Figure 6.27. Prediction of crack density evolution in cross-ply laminates using the
Vinogradov and Hashin model [98] for two material systems: (a) Avimid K polymer/IM6,
(b) Hercules AS4/3501-6. The experimental data are from [127]. Reprinted, with kind
permission, from Int J Solids Struct, Vol. 42, V. Vinogradov and Z. Hashin, Probabilistic
energy-based model for prediction of transverse cracking in cross-ply laminates, pp. 365
392, copyright Elsevier (2005).
6.6 Damage evolution in multidirectional laminates
Although many generic features of ply cracking are evident in cross-ply laminates,
this class of laminates is used only in limited cases. Most applications require a mix
of lamina orientations in the laminate conguration to generate properties to carry
combinations of normal loads, bending moments and torsion. In a multidirectional
laminate the ply cracking in any ply will generally take place under stresses normal
to and parallel to the bers, as well as in-plane shear. Experimental investigations
[27, 34, 35, 50, 51, 128] have claried some of the complexities relating to mode-
mixity of crack growth and interactions between cracks within and among plies.
Consider now a [0/90/y
1
/y
2
]
s
laminate where ply cracks can appear in the 90
-,
y
1
-, and y
2
-plies, assuming loading in the 0
-direction. Figure 6.29 illustrates the

development of cracking in multiple orientations in such a laminate conguration.
As indicated there, cracking initiates rst in the 90
-plies at an overall strain e

90
0
,
and, on increasing the load, this cracking multiplies. At the strain e
y
1
0
the y
1
-plies
1400
(a)
1200
1000
800
600
Experimental results
Without weakness areas
With weakness areas
400
C
r
a
c
k

d
e
n
s
i
t
y

(
m
1
)
200
0
250
300 350 400
Average stress (MPa)
450 500 550
1400
(b)
1200 Experimental results
Without weakness areas
With weakness areas
1000
800
C
r
a
c
k

d
e
n
s
i
t
y

(
m
1
)
600
400
200
0
250 300 350 400
Average stress (MPa)
450 500 550
Figure 6.28. Evolution of crack density as a function of applied stress for a [0/90
2
]
s
carbon/
epoxy laminate [121]. The experimental data are from [1]. Reprinted, with kind permission,
from Compos Sci Technol, Vol. 60, J. M. Berthelot and J. F. Le Corre, Statistical analysis
of the progression of transverse cracking and delamination in cross-ply laminates,
217 6.6 Damage evolution in multidirectional laminates
begin cracking (assuming y
1
>y
2
), and witha further increase in the imposed load, an
interactive cracking process continues inthe 90
- andy
1
-ply orientations. At straine
y
2
0
cracking initiates in the y
2
-plies, and eventually all off-axis plies conduct interactive
crack multiplication. The crack initiation strains and crack multiplication rates
dependonthe constraint imposedby the neighboring plies tothe cracks ina givenply.
For predicting the evolution of ply cracking the authors of this book have
developed an energy-based approach, which is capable of dealing with cracking
in off-axis plies of orthotropic laminates. In [129] the approach is described and
applied to several ply cracking cases. A brief description of the approach follows.
As illustratedinFigure 6.30, twodamage states are considered: state 1 withNparallel
off-axis cracks spaced at distance s, and state 2 where the cracks have multiplied to 2N
and attained the spacing s/2. Evolution of cracking damage is assumed when the work
requiredingoing fromstate 1 tostate 2 (whichis the same as the workneededtoclose N
cracks in going from state 2 to state 1) exceeds a critical value, i.e., if
W
2NN
_ N:G
c
:
1
sin y
t
c
; (6:77)
where y is the off-axis angle and G
c
is the critical (threshold) value of the energy
required for multiple ply crack formation within the given laminate (more discus-
sion about this later). The work required to form N additional cracks in going from
state 1 to state 2 (the same as the work required to close those cracks) is given by
W
2NN
= W
2N0
W
N0
; (6:78)
90 q
1
q
2
0
0 0
Figure 6.29. Cracking process in a [0/90/y
1
/y
2
]
s
half-laminate.
State 1: N cracks, crack spacing = s State 2: 2N cracks, crack spacing = s/2
t
q
s
s/2
x
3
x
2
Figure 6.30. Progressive multiplication of off-axis ply cracks in a multidirectional laminate
[129]. Reprinted, with kind permission, from Int J Solids Struct, Vol. 47, C. V. Singh
and R. Talreja, Evolution of ply cracks in multidirectional composite laminates,
where W
N0
and W
2N0
represent the work required to close N cracks in state 1,
and 2N cracks in state 2, respectively, and the two quantities are calculated as (see
[130] for detailed derivation)
W
N0
= N
1
sin y
t
c
( )
2
1
E
2
s
y
20
_ _
2
:~ u
y
n
s ( ) s
y
120
_ _
2
~ u
y
t
s ( )
_ _
; (6:79)
W
2N0
= 2N
1
sin y
t
c
( )
2
1
E
2
s
y
20
_ _
2
:~ u
y
n
s
2
_ _
s
y
120
_ _
2
~ u
y
t
s
2
_ _ _ _
; (6:80)
where ~ u
y
n
; ~ u
y
t
are the normalized average crack opening and sliding displacements
(COD and CSD), respectively. These are given by
~ u
y
n
=
u
y
n
t
c
s
y
20
=E
2
_ _ =
1
t
c
s
y
20
=E
2
_ _
_
t
y
=2
t
y
=2
u
n
(z)dz;
~ u
y
t
=
u
y
t
t
c
s
y
120
=E
2
_ _ =
1
t
c
s
y
120
=E
2
_ _
_
t
y
=2
t
y
=2
u
t
(z) dz;
(6:81)
where u
n
and u
t
represent the relative opening and sliding displacement of the
cracked surfaces, respectively, and overbars represent averages. For the special
case of cracking in the 90
-ply only, the sliding displacement is zero and hence the

criterion for ply crack multiplication is written as
t
c
:
s
y
20
_ _
2
E
2
2:~ u
y
n
s
2
_ _
~ u
y
n
s ( )
_ _
_ G
Ic
; (6:82)
where G
Ic
is the critical energy release rate in mode I (crack opening mode).
This is the same relation as that derived for cracking in cross-ply laminates by
Joffe et al. [113] except that they consider centrally placed cracked 90
-plies in
their model and normalize the average COD with half the ply thickness (t
c
/2).
For cracking in a general off-axis ply, one can use a multi-mode criterion
given as
w
I
G
Ic
_ _
M
w
II
G
IIc
_ _
N
_ 1; (6:83)
where
w
I
=
s
y
20
_ _
2
t
c
E
2
2:~ u
y
n
s
2
_ _
~ u
y
n
s ( )
_ _
; w
II
=
s
y
120
_ _
2
t
c
E
2
2:~ u
y
t
s
2
_ _
~ u
y
t
s ( )
_ _
; (6:84)
where G
IIc
is the critical energy release rate in mode II (crack sliding mode), and
the exponents M and N depend on the material system, e.g, for a glass/epoxy
system, M = 1, N = 2 [130].
In our work [129] we interpret the critical material parameters G
Ic
and G
IIc
not
in the usual linear elastic fracture mechanics sense where they are dened as the
resistance to advancement of the crack front at the point of unstable crack growth.
Instead, we postulate that the work required to go from state 1 to state 2 involves a
range of dissipative processes that all depend on the material condition in a
cracking ply within the given laminate. The material parameter representing the
dissipated energy per unit of ply crack surface is, therefore, not what is obtained in
a standard fracture toughness test for determining G
Ic
or G
IIc
. To emphasize that
the critical energy terms used here are not the usual fracture toughness values G
Ic
or G
IIc
, we will henceforth use the symbols W
Ic
and W
IIc
instead. These new
quantities are not to be obtained by independent tests, but are to be evaluated
by tting model predictions (Eq. (6.83)) to the experimental data for a reference
laminate. This way the values obtained will be representative of the energies
associated with multiple cracking within a laminate. A reference laminate is
chosen from the class of laminates (material etc.) for which predictions are to be
made, and for which experimental data are readily available [129, 131]. Further-
more, as described in [129], it is argued that a ply crack within a laminate cannot
be formed unless sufcient energy is available to open its surfaces (i.e., in mode I
cracking). In other words, a pure sliding action will not by itself generate the set of
parallel cracks illustrated in Figure 6.30. This will imply that the second term in
Eq. (6.84) is negligible in comparison to the rst term. With these assumptions and
approximations the predictions of crack density evolution agree well with experi-
mental data [129].
The complete procedure to implement the described energy model for micro-
crack initiation and evolution in an off-axis ply of a general symmetric laminate is
outlined below. The procedure is in two parts:
Part I: Estimate W
Ic
1. From FE simulations, determine the variation of normalized COD and CSD
(Eq. (6.81)) with crack spacing.
2. Assume a value for W
Ic
. Plot the damage evolution for the reference laminate as
follows:
(a) Divide the specimen length into small intervals of length, dX = t
y
/10 is
chosen here.
(b) Find the multiple crack initiation strain, Eq. (6.82) with the COD value
calculated with a very large spacing (s ).
(c) Assume a small initial crack density, e.g., r
initial
= 1/50t
y
is chosen here.
(d) Choose a random length interval and check for cracking. A new crack forms
when the criterion set in Eq. (6.82) is satised. Increase the crack density and
eliminate the crackedlengthinterval fromfurther considerationfor plycracking.
(e) Choose another length interval and repeat the previous step until the
fracture criterion is satised.
(f) Increase the applied strain. Repeat steps (d) and (e) using this strain value.
3. Iterate step 2 by varying W
Ic
so that the resulting evolution curve ts the
experimental data for the reference laminate. For example, for predicting the
damage evolution in [0/y
4
/0
1/2
]
s
laminates, we chose [0/90
8
/0
1/2
]
s
as the refer-
ence laminate.
Part II: Predict the damage evolution for other off-axis plies:
1. From FE simulations, determine the variation of COD and CSD (Eq. (6.81))
with crack spacing for a given off-axis laminate.
2. Using the value for W
Ic
obtained above, predict the damage evolution by
following steps 2(a)(f) described in Part I.
The above semi-analytical model is coded in a MATLAB program. The input
data include the following laminate properties: ply material (elastic moduli), ply
thicknesses and orientations (i.e., laminate layup), and variation of COD with
respect to crack density (which can be obtained from independent 3-D FE
analysis).
The energy model described above was applied to predict damage evolution
in glass/epoxy [0/y
4
/0
1/2
]
s
, quasi-isotropic ([0/90/45]
s
), and [0
m
/90
n
/y
p
]
s
laminates [129]. Figure 6.31 shows the variation of crack initiation strains with
off-axis ply orientation (y) for [0/y
4
/0
1/2
]
s
laminates. As expected, the crack
initiation strain increases as y decreases and it may exceed 1.5% if y < 45
. In
fact, the experiments by Varna et al. [27] for this laminate revealed that ply
cracks did not form fully for y < 40
. Using the procedure described above, W

Ic
is
evaluated by tting model predictions with experimental data for a chosen
reference laminate [0/90
8
/0
1/2
]
s
. The evolution of crack density against applied
strain for these laminates are shown in Figure 6.32 and Figure 6.33 for y = 70
and 55
, respectively. For these laminates, direct application of Eq. (6.84)

does not yield accurate predictions. The reason is that at low crack densities,
the work term w
I
is almost constant (i.e., independent of crack spacing). How-
ever, on development of sufcient cracks with a distribution in inter-crack
spacing, w
I
depends on the crack spacing due to interactions between adjacent
cracks. To account for this behavior, Liu and Nairn [127] suggested that the
effective crack spacing be used in place of the average crack spacing s. Thus, w
I
is
modied as
40 50 60 70
Ply Orientation (deg)
80 90
0
0.5
1
1.5
2
2.5
3
3.5
4
C
r
a
c
k

i
n
i
t
i
a
t
i
o
n

s
t
r
a
i
n

(
%
)
Model
Experimental Data
Figure 6.31. Variation of crack initiation strain with ply orientation for a glass/epoxy
[0/y
4
/0
1/2
]
s
laminate. The experimental data are from [27].
w
I
=
s
y
20
_ _
2
t
c
E
2
2:~ u
y
n
fs
2
_ _
~ u
y
n
fs ( )
_ _
; (6:85)
where the parameter f is the average ratio of the crack interval in which a
microcrack forms to the average crack spacing. For [0/y
4
/0
1/2
]
s
laminates, the
predictions shown in Figure 6.32 and Figure 6.33 are made with f = 0.8.
The same model was also used to predict damage evolution in a quasi-isotropic
laminate. The value of W
Ic
for this case was obtained by tting model predictions
with experimental data for reference [0/90]
s
laminate. Since 45
-plies in the quasi-

isotropic laminates contained partially grown cracks, this was accounted for in the
analysis by considering cracks in multiple orientations while calculating CODs
from FE analysis. This showed that COD for the 90
-ply increased resulting in

Figure 6.32. Damage evolution in [0/70
4
/0
1/2
]
s
laminates. The experimental data are from
[27]. The crack density is average of crack densities in +70
and 70
-plies. Reprinted, with

kind permission, from Int J Solids Struct, Vol. 47, C. V. Singh and R. Talreja, Evolution of
ply cracks in multidirectional composite laminates, pp. 133849, copyright Elsevier (2010).
Figure 6.33. Damage evolution in [0/55
4
/0
1/2
]
s
laminates. The experimental data are from
[27]. The crack density is the average of crack densities in +55
and 55
-plies. Reprinted,
with kind permission, from Int J Solids Struct, Vol. 47, C. V. Singh and R. Talreja, Evolution
of ply cracks in multidirectional composite laminates, pp. 133849, copyright Elsevier (2010).
enhanced cracking in the transverse ply. The model predictions for crack density
evolution in this case are compared with experimental data in Figure 6.34. No
modier for crack spacing was needed for this case, i.e., f = 1.
A parametric study performed for [0
m
/90
n
/y
p
]
s
laminates reveals that inter-
actions between the crack systems of different orientations may have a signicant
effect on damage evolution. If y is close to 90
, this intra-mode interaction is higher

because of the close proximity of the 90
and y-crack planes. Ply cracks usually

initiate in the 90
-plies rst and then in the 60
-layers. Thus, initial simulation

assumes only 90
-cracks; whereas after the initiation of 60
-cracks, a multi-mode
scenario is used in FE modeling. The 60
-cracks inuence the damage progression

in the 90
-layer. Model predictions for y =60
for different values of m, n, and p are

shown in Figure 6.35(a)(c) for the 90
-, 60
-, and +60
-layers, respectively. For p

= 2, the model predicts that cracks in the 60
- and +60
-layers will initiate earlier

than in the 90
-layer. If y < 45
, this intra-mode interaction is not appreciable, and

for this case the damage evolution in the 90
-layer may not be affected at all.

6.7 Damage evolution under cyclic loading
While the fatigue process in composite materials is treated in the next chapter, here
we shall describe a modeling approach for the evolution of transverse cracking in
cross-ply laminates under cyclic axial tension. This case because of its simplicity
of geometry serves as a good illustration of the fundamental ideas in the fatigue of
composites. The reader is urged to refer to two papers [53, 132] for full details of
the treatment described below.
Figure 6.34. Evolution of 90
-crack density in [0/90]

s
and [0/90/45]
s
laminates. The
experimental data are from [34]. Reprinted, with kind permission, from Int J Solids Struct,
Vol. 47, C. V. Singh and R. Talreja, Evolution of ply cracks in multidirectional composite
laminates, pp. 133849, copyright Elsevier (2010).
223 6.7 Damage evolution under cyclic loading
(a)
(b)
(c)
Figure 6.35. Evolution of crack density in a [0
m
/90
n
/60
p
]
s
laminate for varying ply
thicknesses in (a) 90
-layer, (b) 60
-layer, and (c) +60
-layer. Reprinted, with kind

permission, from Int J Solids Struct, Vol. 47, C. V. Singh and R. Talreja, Evolution
of ply cracks in multidirectional composite laminates, pp. 133849, copyright
Elsevier (2010).
The guiding principle in any fatigue analysis must be to address the question:
what is the mechanism of irreversibility that causes the accumulation of damage
from one load cycle to another? The common energy-dissipating mechanisms are
plasticity, friction, and surface formation. For composite laminates that are
modeled as layered elastic solids plasticity is not admissible. Frictional processes
within the volume of such composites are possible between crack surfaces if the
surfaces are in contact. Finally, new surface formation without plasticity is pos-
sible by brittle fracture.
As a case for illustration we shall consider a cross-ply laminate with ply cracks
in the central 90
-plies that formed under the rst application of an axial tensile

load. The problem posed is: if the load is removed and repeatedly reapplied to the
previous maximum value, when would new cracks form between the pre-existing
cracks? To begin the analysis we note that for the cracked cross-ply laminate a
solution of good accuracy for the stress eld in the region between cracks is
available (e.g., [101]). This solution is valid for perfectly and linearly elastic
(no plasticity) laminates. Also, the solution does not apply to partial cracks, i.e.,
cracks that are not fully extended in the thickness and width directions of the 90
-
plies. Thus to use this solution we must retain the symmetry and periodicity of the
cracks assumed in obtaining the solution. This condition eliminates analysis of the
cyclic growth of transverse cracks from partial to full extent. Obviously, a numer-
ical analysis of this case is possible, but that would not provide the analytic fatigue
damage model we intend to develop.
With the analytical stress solution to the cracked cross-ply laminate at hand
we see that unless some irreversible mechanism is included no change in the
crack density can be predicted since any repetition of load in an elastic solid
cannot change the stress eld. Therefore, in order to have an analytical stress
solution, i.e., keeping the symmetry of laminate geometry and periodicity of
cracks, and to incorporate irreversibility, a novel idea was proposed in [132].
According to that, all irreversibility leading to damage accumulation is lumped
into delamination surfaces emanating from the transverse crack fronts. Figure
6.36(a) illustrates the resulting model geometry of the cracked laminate while
Figure 6.36(b) shows the repeating unit cell. As shown, the pre-existing trans-
verse cracks are spaced at distance 2a and the delamination on either side of the
cracks is of distance d. The idea behind the model is that the delamination grows
under applied cyclic loading, the same way as a crack does, imparting changes to
the stress eld in the region between transverse ply cracks. In this way the model
captures cycle-dependent irreversibility, thereby allowing the fatigue-induced
multiplication of transverse ply cracks to be modeled. Although the formation
of new cracks between pre-existing cracks can be modeled by different criteria, in
[132] a maximum stress criterion for cracking is used, supported by previous
work [133]. It can, however, be shown that if the delamination surfaces are
traction free, then the maximum axial stress between the ply cracks goes down
as the delamination length d increases. This suggests that the irreversibility
captured in such delamination growth is inadequate for the purpose. As argued
in [132], the delamination surfaces are indeed under compressive stress [134],
making it plausible that frictional contact exists between those surfaces. The
ensuing frictional sliding was modeled by an interfacial shear stress, which
provided the needed increase in the axial stress for crack formation.
A description of the model now follows.
Referring to Figure 6.36(a) and (b), the stress analysis is performed by a
variational approach along the lines in [101], conducting the minimization of
complementary energy separately for region I and region II. First, an admissible
stress system in the xz plane of region II is expressed as
s
L(m)
ij
= s
0(m)
ij
s
(m)
ij
; (6:86)
where s
L(m)
ij
and s
0(m)
ij
are the stress components in the cracked laminate and in the
virgin laminate, respectively, and s
(m)
ij
are the perturbations; m = 1 and 2 indicate
the 90
- and 0
-plies, respectively. The axial perturbation stresses in the plies are

assumed to have the following form
s
(1)
xx
= s
1
f
1
x ( ); s
(2)
xx
s
2
f
2
x ( ) A x ( ) z [ [; (6:87)
where s
0(m)
xx
= s
m
has been used; f
1
(x), f
2
(x), and A(x) are unknown functions.
Applying the equilibrium in the x direction and the interface iso-strain condition
at z = t
1
, f
2
(x) and A(x) in Eq. (6.87) are eliminated, so the axial perturbation
stresses can be expressed by the only unknown function, f
1
(x). After integrating
equilibrium equations and using Eq. (6.87), all perturbation stress components
Z
90 0
X
N
xx
Y
X
d
a
0
90
Z
Region I
Region II
t
1
t
2
h
(a) (b)
Figure 6.36. Schematic of a cracked cross-ply laminate: (a) uniformly distributed transverse
matrix cracks in 90
-plies with the associated delamination of 0
/90
interface; (b) a unit

cell between two matrix cracks with delaminated region (region I) and perfectly bonded
region (region II).
are expressed as functions of f
1
(x) by applying interface continuity conditions at
z = t
1
and traction-free boundary conditions at z = h. The admissible stress
system for a cracked laminate is then established based on the only unknown
function, f
1
(x). The corresponding complementary energy functional for linear
elastic materials in a volume V, with only traction boundary conditions in region
II, can be written as [101]:
U
c
= U
0
c
U
/
c
=
1
2
_
V
s
ijkl
s
0(m)
ij
s
0(m)
kl
dV
1
2
_
V
s
ijkl
s
(m)
ij
s
(m)
kl
dV; (6:88)
where S
ijkl
are the components of the compliance tensor. Since the virgin laminate
stresses are constant, U
0
c
is not of importance to the analysis. Substituting all
perturbation stress components into Eq. (6.88) gives
U
/
c
= (s
1
)
2
_
(ad)
(ad)
t
1
C
00
f
2
1
t
3
1
C
11
(f
/
1
)
2
t
5
1
C
22
(f
//
1
)
2
t
3
1
C
02
f
1
f
//
1
_ _
dx; (6:89)
where C
ij
are constants determined by the elastic constants and thickness of each
layer, a is half the crack spacing, and d is the delamination length. Minimizing U
/
c
after introducing the nondimensional variable x = x/t
1
, the following Euler
Lagrange differential equation in f
1
is obtained
d
4
f
1
dx
4
p
d
2
f
1
dx
2
qf
1
= 0; (6:90)
where p and q are constants determined by C
ij
. Dependent on the material elastic
property and geometry of the given laminate, two solutions for Eq. (6.90) exist
f
1
x ( ) =
A
1
cosh ax ( ) cos bx ( ) A
2
sinh ax ( ) sin bx ( );
p
2
4
q
_ _
<0
A
1
cosh ax ( ) A
2
cosh bx ( );
p
2
4
q
_ _
> 0
_
_
_
_
_
_
(6:91)
The axial normal stress of interest at the mid-plane in the 90
plies is obtained as
s
xx
0; z ( ) = s
1
1 f
1
0 ( ) ( ); t
1
<z<t
1
: (6:92)
The two constants, A
1
and A
2
, are found using the traction continuity conditions at
x =(a d) if the stress in region I is known.
To obtain the stress state in region I, a very similar variational approach is
applied. For an admissible stress system, the perturbation stresses are assumed
rst as
s
(1)
xx
= s
1
c
1
(x); s
(2)
xx
= s
2
c
2
x ( ) A
z [ [: (6:93)
A cubic variation of shear stress along the interface (z = t
1
) is enforced to account
for the effect of frictional sliding along delamination
s
(1)
xx
(x; t
1
) =
t
a
3
(x a)[x (a d)[
2
; (6:94)
where t is an unknown. After integrating the equilibrium equation, and applying
continuity of stress across the interface at z = t
1
, as well as traction-free boundary
conditions at the crack surfaces, the admissible stress system is built up as
functions of the only unknown, t. By minimizing the corresponding complemen-
tary energy functional,
dU
c
dt
= 0; (6:95)
the only unknown t is solved, so the stress eld in region I is obtained. By applying
the traction continuity conditions at the boundary between region I and region II,
A
1
and A
2
in Eq. (6.91) are determined. Substituting A
1
and A
2
into Eq. (6.92),
nally, the axial normal stress at the mid-plane in 90
plies is obtained for the

given unit cell with specied crack spacing and delamination length. A typical
variation of the axial normal stress with delamination at a xed crack density is
shown in Figure 6.37.
The following power-law relation is assumed to describe the growth of delami-
nation under cyclic loading
dl
dN
= B
t
l
_ _
m
; (6:96)
where t = t
max
t
min
( )=s
max
and l = l=t
1
, and where l is the delamination
length, denoted as d in the above stress analysis.
Integration of Eq. (6.96) yields the relationship between delamination length
and the number of cycles, N. From Figure 6.37 we see that after the initial
drop the axial normal stress increases as delamination grows. When the
maximum stress criterion s
xx
(x = 0, t
1
< z < t
1
) = s
c
is satised, a new
crack forms midway between the pre-existing cracks, and the corresponding
crack spacing (or crack density) is updated. In this way, the multiplication of
matrix crack under fatigue loading is actually controlled by the growth of
delamination, and therefore determined by the number of cycles through
integration of Eq. (6.96). The damage evolution as a function of the number
0.0004 0.0002
s
x
x
(
x

=

0
,

z
=
0
)

(
P
a
)
3.7510
7
4.2510
7
4.510
7
4.7510
7
510
7
5.2510
7
0.0006
d (m)
Figure 6.37. Typical variationof the axial normal stress with delamination at a xed crack density.
of cycles in cross-ply laminates is nally quantitatively described in the model.
Figure 6.38 shows the variation of crack density as a function of the number of
cycles under a given cyclic tension.
6.8 Summary
Initiation of cracks within the plies of a laminate, and their growth and multipli-
cation, are part of the eld of damage evolution in composites that constitutes a
key element in the performance assessment of structures made of these materials.
This chapter has focused on various stress and failure analysis methods associ-
ated with the prediction of initiation and progression of ply cracks. Since cracks
form from material defects that can be considered random in their size and
spatial distributions, statistical considerations have been included in the
analyses.
For the formation of cracks the criteria used are based either on strength or on
the energy associated with fracture. Both criteria have been treated and compared.
Experimental data, wherever available, have been used to assess the predictions.
Most of the damage evolution work in the literature has been for transverse
cracking in cross-ply laminates. While this cracking mode has been amply dealt
with, more recent work on oblique cracks, i.e., cracks in off-axis plies of multi-
directional laminates, has also been treated.
While the fatigue of composite laminates is the focus of a separate chapter,
where a broad treatment of the subject is given, a section on ply cracking under
cyclic loading has been included here to illustrate how damage accumulation
under repeated loads is modeled.
2.0
1.8
1.6
1.4
C
r
a
c
k

d
e
n
s
i
t
y

(
/
m
m
)
1.2
1.0
0 200 000 400 000 600 000
Model prediction
Experiment
Number of cycles
800 000 1000 000
Figure 6.38. Transverse crack evolution with cycles for a [0/90
2
]
s
carbon/epoxy laminate at a
maximum stress of 482.633 MPa and a stress ratio of 0.1.
229 6.8 Summary
References
1. A. S. D. Wang, Fracture mechanics of sublaminate cracks in composite materials.
In Composites Technology Review. (Philadelphia, PA: ASTM, 1984), pp. 4562.
2. J. M. Masters and K. L. Reifsnider, An investigation of cumulative damage develop-
3. K. L. Reifsnider, ed. Damage in Composite Materials, ASTM STP 775, ed. K. L.
Reifsnider. (Philadelphia, PA: ASTM, 1982).
4. J. E. Masters, ed. Damage Detection in Composite Materials, ASTM STP 1128.
(Philadelphia, PA: ASTM, 1992).
5. J. A. Nairn, S. F. Hu, and J. S. Bark, A critical-evaluation of theories for predicting
microcracking in composite laminates. J Materials Sci, 28:18 (1993), 5099111.
6. D. O. Stalnaker and W. W. Stinchcomb, Load historyedge damage studies in
two quasi-isotropic graphite epoxy laminates. In Composite Materials: Testing and
Design (Proc. 5th Conference), ASTM STP 674. (Philadelphia, PA: ASTM, 1979), pp.
62041.
7. J. E. Masters, An experimental investigation of cumulative damage development in
graphite epoxy laminates. Ph.D. dissertation, Materials Engineering Science, Depart-
ment, Virginia Polytechnic Institute, Blacksburg, VA, 1980, p. 4062.
9. K. L. Reifsnider and R. Jamison, Fracture of fatigue-loaded composite laminates. Int
J Fatigue, 4:4 (1982), 18797.
10. K. L. Reifsnider and A. Talug, Analysis of fatigue damage in composite laminates. Int
J Fatigue, 2:1 (1980), 311.
11. E. Adolfsson and P. Gudmundson, Matrix crack initiation and progression in composite
laminates subjected to bending and extension. Int J Solids Struct, 36:21 (1999), 313169.
12. B. F. Sorensen and R. Talreja, Analysis of damage in a ceramic matrix composite. Int
J Damage Mech, 2:3 (1993), 24671.
13. B. F. Sorensen, R. Talreja, and O. T. Sorensen, Micromechanical analysis of damage
mechanisms in ceramic-matrix composites during mechanical and thermal cycling.
Composites, 24:2 (1993), 12940.
14. T. J. Dunyak, W. W. Stinchcomb, and K. L. Reifsnider, Examination of selected NDE
techniques for ceramic composite components. In Damage Detection in Composite
Materials, ASTM STP 1128, ed. J. E. Masters. (Philadelphia, PA: ASTM, 1992), pp.
324.
15. K. V. Steiner, Defect classications in composites using ultrasonic nondestructive
evaluation techniques. In Damage Detection in Composite Materials, ASTM STP 1128,
ed. J. E. Masters. (Philadelphia, PA: ASTM, 1992), pp. 7284.
16. K. V. Steiner, R. F. Eduljee, X. Huang, and J. W. Gillespie, Ultrasonic NDE tech-
niques for the evaluation of matrix cracking in composite laminates. Compos Sci
Technol, 53:2 (1995), 193198.
17. G. E. Maddux and G. P. Sendeckyj, Holographic techniques for defect detection in
composite materials. In Nondestructive Evaluation and Flaw Criticality for Composite
Materials, ASTM STP 696. (Philadelphia, PA: ASTM, 1979), pp. 2644.
18. G. P. Sendeckyj, G. E. Maddux, and E. Porter, Damage documentation in composites
by stereo radiography. In Damage in Composite Materials, ASTM STP 775, ed. K. L.
Reifsnider. (Philadelphia, PA: ASTM, 1982), pp. 1626.
PA: ASTM, 1984), pp. 2155.
20. C. Bathias and A. Cagnasso, Application of x-ray tomography to the non-destructive
testing of high-performance polymer composites. In Damage Detection in Composite Mater-
ials, ASTM STP 1128, ed. J. E. Masters. (Philadelphia, PA: ASTM, 1992), pp. 3554.
21. K. Maslov, R. Y. Kim, V. K. Kinra, and N. J. Pagano, A new technique for the
ultrasonic detection of internal transverse cracks in carbon-bre/bismaleimide compos-
ite laminates. Compos Sci Technol, 60:1213 (2000), 218590.
22. V. K. Kinra, A. S. Ganpatye, and K. Maslov, Ultrasonic ply-by-ply detection of matrix
cracks in laminated composites. J Nondestr Eval, 25:1 (2006), 3951.
23. Y. Zou, L. Tong, and G. P. Steven, Vibration-based model-dependent damage (dela-
mination) identication and health monitoring for composite structures a review.
J Sound Vib, 230:2 (2000), 35778.
24. S. K. Seth, et al., Damage detection in composite materials using lamb wave methods.
Smart Mater Struct, 11:2 (2002), 269.
J Compos Mater, 19:4 (1985), 35575.
ated response of laminates with varying constraints. Int J Damage Mech, 8 (1999), 17493.
with off-axis plies. Compos Sci Technol, 59:14 (1999), 213947.
28. D. G. Katerelos, L. N. McCartney, and C. Galiotis, Local strain re-distribution and
stiffness degradation in cross-ply polymer composites under tension. Acta Mater, 53:12
(2005), 333543.
29. D. T. G. Katerelos, P. Lundmark, J. Varna, and C. Galiotis, Analysis of matrix
cracking in GFRP laminates using Raman spectroscopy. Compos Sci Technol, 67:9
(2007), 194654.
30. D. G. Katerelos, M. Kashtalyan, C. Soutis, and C. Galiotis, Matrix cracking in
polymeric composite laminates: Modelling and experiments. Compos Sci Technol,
68:12 (2008), 231017.
31. D. G. Katerelos, L. N. McCartney, and C. Galiotis, Effect of off-axis matrix cracking
on stiffness of symmetric angle-ply composite laminates. Int J Fract, 139:34 (2006),
52936.
32. D. T. G. Katerelos, J. Varna, and C. Galiotis, Energy criterion for modelling
damage evolution in cross-ply composite laminates. Compos Sci Technol, 68:12 (2008),
231824.
33. J. A. Nairn, Matrix microcracking in composites. In Polymer Matrix Composites, ed. R.
Talreja and J. A. E. Manson. (Amsterdam: Elsevier Science, 2000), pp. 40332.
34. J. Tong, F. J. Guild, S. L. Ogin, and P. A. Smith, On matrix crack growthinquasi-isotropic
35. L. E. Crocker, S. L. Ogin, P. A. Smith, and P. S. Hill, Intra-laminar fracture in angle-
ply laminates. Compos A, 28:910 (1997), 83946.
231 References
36. N. Balhi, et al., Intra-laminar cracking in CFRP laminates: observations and model-
ling. J Mater Sci, 41:20 (2006), 6599609.
37. K. W. Garrett and J. E. Bailey, Effect of resin failure strain on tensile properties of glass
ber-reinforced polyester cross-ply laminates. J Mater Sci, 12:11 (1977), 218994.
38. K. W. Garrett and J. E. Bailey, Multiple transverse fracture in 90
cross-ply laminates
of a glass bre-reinforced polyester. J Mater Sci, 12:1 (1977), 15768.
40. A. Parvizi, K. W. Garrett, and J. E. Bailey, Constrained cracking in glass bre-
splitting behavior of glass and carbon-ber reinforced epoxy cross ply laminates and
(1979), 599623.
42. M. G. Bader, J. E. Bailey, P. T. Curtis, and A. Parvizi, eds., The mechanisms of
initiation and development of damage in multi-axial bre-reinforced plastic laminates.
In Proceedings of the Third International Conference on the Mechancial Behaviour in
Materials (ICM3), Cambridge, UK, Vol. 3. (1979), pp. 22739.
43. J. E. Bailey and A. Parvizi, On ber debonding effects and the mechanism of trans-
verse-ply failure in cross-ply laminates of glass ber-thermoset composites. J Mater Sci,
16:3 (1981), 64959.
44. F. R. Jones, A. R. Wheatley, and J. E. Bailey, The effect of thermal strains on the
microcracking and stress corrosion behaviour of GRP. In Composite Structures 1st
International Conference, ed. I. H. Marshall. (Barking, UK: Applied Science Publishers,
1981), pp. 41529.
45. D. L. Flaggs and M. H. Kural, Experimental-determination of the insitu transverse
lamina strength in graphite epoxy laminates. J Compos Mater, 16 (1982), 10316.
46. J. A. Nairn and S. Hu, Micromechanics of damage: a case study of matrix micro-
cracking. In Damage Mechanics of Composite Materials, ed. R. Talreja. (Amsterdam:
Elsevier, 1994), pp. 187243.
J Compos Mater, 19:4 (1985), 35575.
48. J. A. Nairn and S. F. Hu, The formation and effect of outer-ply microcracks in cross-
ply laminates a variational approach. Eng Fract Mech, 41:2 (1992), 20321.
laminates II. Finite element analysis. Compos Sci Technol, 57:11 (1997), 153745.
50. 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.
Compos A, 36:9 (2005), 122935.
51. T. Yokozeki, T. Aoki, and T. Ishikawa, Consecutive matrix cracking in contiguous
plies of composite laminates. Int J Solids Struct, 42:910 (2005), 2785802.
52. J. E. Masters and K. L. Reifsnider, An investigation of cumulative damage deve-
lopment in quasi-isotropic graphite/epoxy laminates. In Damage in Composite
Materials, ASTM STP 775, ed. K. L. Reifsnider. (Philadelphia, PA: ASTM, 1982),
pp. 4062.
life of composite laminates. Mater Sci Eng A, 285:12 (2000), 30313.
laminates of a glass ber-reinforced polyester. J Materials Sci, 12:1 (1977), 15768.
55. J. M. Berthelot, Transverse cracking and delamination in cross-ply glass-ber and
carbon-ber reinforced plastic laminates: static and fatigue loading. Appl Mech Rev,
56 (2003), 11147.
56. S. E. Groves, et al., An experimental and analytical treatment of matrix cracking in
cross-ply laminates. Exp Mech, 27:1 (1987), 739.
57. S. F. Hu, J. S. Bark, and J. A. Nairn, On the phenomenon of curved microcracks in
[(S)/90(n)](s) laminates their shapes, initiation angles and locations. Compos Sci
Technol, 47:4 (1993), 3219.
58. P. Lundmark and J. Varna, Damage evolution and characterisation of crack types in
CF/EP laminates loaded at low temperatures. Eng Fract Mech, 75:9 (2008), 263141.
59. H. L. McManus, D. E. Bowles, and S. S. Tompkins, Prediction of thermal cycling
induced matrix cracking. J Reinf Plast Compos, 15:2 (1996), 12440.
60. H. L. McManus and J. R. Maddocks, On microcracking in composite laminates under
thermal and mechanical loading. Polym Compos, 4:5 (1996), 30514.
61. C. H. Park and H. L. McManus, Thermally induced damage in composite laminates:
predictive methodology and experimental investigation. Compos Sci Technol, 56:10
(1996), 120917.
62. T. G. Reynolds and H. L. McManus, Understanding and accelerating environmentally-
induced degradation and microcracking. In Proceedings of 39th AIAA Structures, Struc-
tural Dynamics and Materials Conference. (Reston, VA: AIAA, 1998, pp. 210316.
63. D. S. Adams, D. E. Bowles, and C. T. Herakovich, Thermally induced transverse
cracking in graphite-epoxy cross-ply laminates. J Reinf Plast Compos, 5:3 (1986),
15269.
64. C. T. Herakovich and M. W. Hyer, Damage-induced property changes in composites
subjected to cyclic thermal loading. Eng Fract Mech, 25:56 (1986), 77991.
65. R. G. Spain, Thermal microcracking of carbon bre/resin composites. Compos, 2:1
(1971), 337.
66. J. A. Lavoie and E. Adolfsson, Stitch cracks in constraint plies adjacent to a cracked
ply. J Compos Mater, 35:23 (2001), 207797.
67. X. Huang, J. W. Gillespie, and R. F. Eduljee, Effect of temperature on the transverse
cracking behavior of cross-ply composite laminates. Compos B, 28:4 (1997), 41924.
68. M. H. Han and J. A. Nairn, Hygrothermal aging of polyimide matrix composite
laminates. Compos A, 34:10 (2003), 97986.
69. J. E. Lundgren and P. Gudmundson, Moisture absorption in glass-bre/epoxy lamin-
ates with transverse matrix cracks. Compos Sci Technol, 59:13 (1999), 198391.
70. E. Vauthier, J. C. Abry, T. Bailliez, and A. Chateauminois, Interactions between
hygrothermal ageing and fatigue damage in unidirectional glass/epoxy composites.
71. A. Tounsi, K. H. Amara, and E. A. Adda-Bedia, Analysis of transverse cracking and
stiffness loss in cross-ply laminates with hygrothermal conditions. Comput Mater Sci,
32:2 (2005), 16774.
72. P. A. Smith and S. L. Ogin, On transverse matrix cracking in cross-ply laminates loaded
in simple bending. Compos A, 30:8 (1999), 10038.
73. P. A. Smith and S. L. Ogin, Characterization and modelling of matrix cracking in a (0/90)
2s GFRP laminate loaded in exure. Proc R Soc London A, 456:2003 (2000), 275570.
233 References
74. S.-R. Kimand J. A. Nairn, Fracture mechanics analysis of coating/substrate systems: part
I: analysis of tensile and bending experiments. Eng Fract Mech, 65:5 (2000), 57393.
75. S.-R. Kim and J. A. Nairn, Fracture mechanics analysis of coating/substrate systems:
part II: experiments in bending. Eng Fract Mech, 65:5 (2000), 595607.
76. L. N. McCartney and C. Pierse, Stress Transfer Mechanics for Multiple-ply Laminates
Subject to Bending. NPL Report CMMT(A) 55, February 1997.
77. L. N. McCartney and C. Pierse, Stress transfer mechanics for multiple ply laminates for
axial loading and bending. Proc ICCM-11, 5 (1997), 66271.
78. S. A. Salpekar and T. K. Obrien, Analysis of matrix cracking and local delamination in
(0/y/y)s graphite-epoxy laminates under tensile load. J Compos Tech Res, 15:2 (1993),
95100.
79. P. Johnson and F. K. Chang, Characterization of matrix crack-induced laminate
failure part I: experiments. J Compos Mater, 35:22 (2001), 200935.
80. P. Johnson and F. K. Chang, Characterization of matrix crack-induced laminate
failure part II: analysis and verications. J Compos Mater, 35:22 (2001), 203774.
81. A. Parvizi, K. W. Garrett, and J. E. Bailey, Constrained cracking in glass ber-
82. H. T. Hahn and S. W. Tsai, Behavior of composite laminates after initial failures.
J Compos Mater, 8 (1974), 288305.
83. H. Fukunaga, T. W. Chou, P. W. M. Peters, and K. Schulte, Probabilistic failure
strength analysis of graphite epoxy cross-ply laminates. J Compos Mater, 18:4 (1984),
33956.
84. H. Fukunaga, T. W. Chou, K. Schulte, and P. W. M. Peters, Probabilistic initial failure
strength of hybrid and non-hybrid laminates. J Mater Sci, 19:11 (1984), 354653.
85. P. W. M. Peters, The strength distribution of 90-deg plies in 0/90/0 graphite-epoxy
laminates. J Compos Mater, 18 (1984), 54556.
86. P. W. M. Peters, The ber/matrix bond strength of CFRP deduced from the strength
transverse to the bers. J Adhes, 53 (1995), 79101.
87. P. W. Manders, T. W. Chou, F. R. Jones, and J. W. Rock, Statistical analysis of multiple
fracture in [0/90/0] glass ber/epoxy resin laminates. J Mater Sci, 19 (1983), 287689.
88. F. W. Crossman, W. J. Warren, A. S. D. Wang, and J. G. E. Law, Initiation and
growth of transverse cracks and edge delamination in composite laminates: part 2.
Experimental correlation. J Compos Mater Suppl, 14 (1980), 89108.
89. P. W. M. Peters, Strength distribution of 90 degree plies in 0/90/0 graphite-epoxy
90. N. Takeda and S. Ogihara, In-situ observation and probabilistic prediction of micros-
copic failure processes in CFRP cross-ply laminates. Compos Sci Technol, 52:2 (1994),
18395.
8:3 (1973), 35262.
multiple fracture in 0
/90
/0
glass bre/epoxy resin laminates. J Mater Sci, 18:10

(1983), 287689.
93. P. S. Steif, Parabolic shear lag analysis of a [0/90]s laminate. In Transverse Ply Crack
Growth and Associated Stiffness Reduction during the Fatigue of a Simple Cross-ply
Laminate, ed. S. L. Ogin, P. A. Smith and P. W. R. Beaumont, Report CUED/C/
MATS/TR 105, Cambridge University, September 1984, pp. 401.
94. J. M. Berthelot, P. Leblond, A. El Mahi, and J. F. Le Corre, Transverse cracking of
cross-ply laminates: part 1. Analysis. Compos A, 27:10 (1996), 9891001.
95. J. M. Berthelot, A. El Mahi, and P. Leblond, Transverse cracking of cross-
ply laminates: part 2. Progressive widthwise cracking. Compos A, 27:10 (1996), 100310.
96. J. M. Berthelot, Analysis of the transverse cracking of cross-ply laminates: a general-
ized approach. J Compos Mater, 31:18 (1997), 1780805.
97. Z. Hashin, Finite thermoelastic fracture criterion with application to laminate
cracking analysis. J Mech Phys Solids, 44:7 (1996), 112945.
98. V. Vinogradov and Z. Hashin, Probabilistic energy based model for prediction of
transverse cracking in cross-ply laminates. Int J Solids Struct, 42:2 (2005), 36592.
99. J. A. Nairn, The strain-energy release rate of composite microcracking a variational
100. N. Laws and G. J. Dvorak, Progressive transverse cracking in composite laminates.
J Compos Mater, 22:10 (1988), 90016.
101. Z. Hashin, Analysis of cracked laminates: A variational approach. Mech Mater, 4:2
(1985), 12136.
102. Z. Hashin, Micromechanics of cracking in composite materials. In Continuum Models
and Discrete Systems. Proc 9th Int Symposium, ed. E. Inan and K. Markov. (Singapore:
World Scientic Publishing, 1998), pp. 7029.
103. Z. Hashin, Thermal-expansion coefcients of cracked laminates. Compos Sci Technol,
31:4 (1988), 24760.
104. L. N. McCartney, Predicting transverse crack formation in cross-ply laminates.
105. L. N. McCartney, Energy-based prediction of progressive ply cracking and strength of
general symmetric laminates using an homogenisation method. Compos A, 36:2 (2005),
11928.
106. L. N. McCartney, Energy-based prediction of failure in general symmetric laminates.
Eng Fract Mech, 72:6 (2005), 90930.
107. L. N. McCartney, Physically based damage models for laminated composites. Proc
Inst Mech Engineers L J Mater Design Appl, 217:L3 (2003), 16399.
108. L. N. McCartney, Prediction of ply crack formation and failure in laminates. Compos
Sci Technol, 62:1213 (2002), 161931.
109. L. N. McCartney, Model to predict effects of triaxial loading on ply cracking in
general symmetric laminates. Compos Sci Technol, 60:1213 (2000), 225579.
110. A. S. D. Wang and F. W. Crossman, Initiation and growth of transverse cracks and
edge delamination in composite laminates part 1. An energy method. J Compos Mater,
14:Suppl. (1980), 7187.
111. R. Joffe and J. Varna, Damage evolution modeling in multidirectional laminates and
the resulting nonlinear response. In Proceeedings of International Conference on Com-
posite Materials 12, Paris, France, July 59, 1999.
112. J. Varna, R. Joffe, and R. Talreja, A synergistic damage-mechanics analysis of
transverse cracking in [y/90
4
]
s
laminates. Compos Sci Technol, 61:5 (2001), 65765.
113. R. Joffe, A. Krasnikovs, and J. Varna, COD-based simulation of transverse cracking
and stiffness reduction in [s/90n]s laminates. Compos Sci Technol, 61:5 (2001), 63756.
114. J. Varna, R. Joffe, and R. Talreja, Mixed micromechanics and continuum damage
mechanics approach to transverse cracking in [S,90(n)](s) laminates. Mech Compos
Mater, 37:2 (2001), 11526.
235 References
115. E. Adolfsson and P. Gudmundson, Thermoelastic properties in combined bending
and extension of thin composite laminates with transverse matrix cracks. Int J Solids
Struct, 34:16 (1997), 203560.
116. J. Qu and K. Hoiseth, Evolution of transverse matrix cracking in cross-ply laminates.
Fatigue Frac Eng M, 21:4 (1998), 45164.
118. C. Baxevanakis, D. Jeulin, and J. Renard, Fracture statistics of a unidirectional
composite. Int J Fract, 73:2 (1995), 14981.
119. V. V. Silberschmidt, Matrix cracking in cross-ply laminates: effect of randomness.
Compos A, 36:2 (special issue) (2005), 12935.
120. V. V. Silberschmidt, Effect of micro-randomness on macroscopic properties and
fracture of laminates. J Mater Sci, 41:20 (2006), 676876.
121. J. M. Berthelot and J. F. Le Corre, Statistical analysis of the progression of transverse
cracking and delamination in cross-ply laminates. Compos Sci Technol, 60:14 (2000),
265969.
122. 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, part 2: experi-
mental correlation. J Compos Mater, 14:Suppl. (1980), 88108.
123. F. W. Crossman and A. S. D. Wang, The dependence of transverse cracking and delamina-
tion on ply thickness in graphite/epoxy laminates. In Damage in Composite Materials,
124. A. S. D. Wang, P. C. Chou, and S. C. Lei, Stochastic model for the growth of matrix
cracks in composite laminates. J Compos Mater, 18:3 (1984), 23954.
125. A. S. D. Wang, N. N. Kishore, and C. A. Li, Crack development in graphite-epoxy
laminates under uniaxial tension. Compos Sci Technol, 24:1 (1985), 131.
126. P. C. Chou, A. S. D. Wang, and H. Miller, Cumulative Damage Model for Advanced
Composite Eaterials, Final report by Dyna East Corp, Philadelphia, PA, September 1982.
127. S. L. Liu and J. A. Nairn, The formation and propagation of matrix microcracks in
cross-ply laminates during static loading. J Reinf Plast Compos, 11:2 (1992), 15878.
128. T. Yokozeki, T. Ogasawara, and T. Ishikawa, Evaluation of gas leakage through
composite laminates with multilayer matrix cracks: cracking angle effects. Compos Sci
Technol, 66:15 (2006), 281524.
129. C. V. Singh and R. Talreja, Evolution of ply cracks in multidirectional composite
laminates. Int J Solids Struct, 47:10 (2010), 133849.
130. M. Kashtalyan and C. Soutis, Stiffness and fracture analysis of laminated composites
with off-axis ply matrix cracking. Compos A, 38:4 (2007), 12629.
131. C. V. Singh, Multiscale modeling of damage in multidirectional composite laminates.
Ph.D. thesis, Department of Aerospace Engineering, Texas A&M University, College
Station, TX (2008).
132. N. V. Akshantala and R. Talreja, A mechanistic model for fatigue damage evolution
in composite laminates. Mech Mater, 29:2 (1998), 12340.
133. R. Talreja and N. V. Akshantala, An inadequacy in a common micromechanics
approach to analysis of damage evolution in composites. Int J Damage Mech, 7:3
(1998), 23849.
134. T. W. Kim, H. J. Kim, and S. Im, Delamination crack originating from transverse
cracking in cross-ply laminates under extension. Int J Solids Struct, 27:15 (1991), 192541.
7 Damage mechanisms
and fatigue-life diagrams
7.1 Introduction
The fatigue of composite materials presents a tremendous challenge when one
considers the number and variety of parameters that can possibly affect the
governing mechanisms. There is a considerable risk of the fatigue design becoming
empirically based, and quite cost-ineffective, if rational guidelines based on phys-
ical models cannot be developed. To help alleviate this problem, we will in this
chapter develop a mechanisms-based framework for interpretating the fatigue
behavior of composites, beginning with the baseline conguration of unidirec-
tional ber-reinforced plies and proceeding later to laminate congurations and
other ber architectures. The framework in the form of fatigue-life diagrams will
allow assessment of the effects of constituent properties, such as ber stiffness and
matrix ductility, and provide guidelines for fatigue design as well as for developing
mechanism-based life prediction models.
After a review of the fatigue-life diagrams and their utility, we shall discuss the
fatigue design methodologies, taking the examples of aircraft components and
wind turbine blades. Finally, a mechanisms-based modeling of multi-axial fatigue
will be discussed.
7.2 Fatigue-life diagrams
The S-N, or Wo hler diagram, originating from metal fatigue, is a familiar way to
represent the resistance of a given material to the cyclic application of loads.
It describes the observed fact that the material strength, given by the maximum
stress sustained in the rst application of load, reduces with repeated application
of load, and is inversely dependent on the number of cycles applied. The strength
value corresponding to a pre-selected large number of cycles, e.g., 10
6
, is custom-
arily taken as the fatigue limit. In some cases, a true fatigue limit exists,
representing the stress value below which a fatigue mechanism cannot be initiated,
but in most cases, one uses the operational denition of no failure until the
selected high number of cycles.
In a composite material with two constituents bers and matrix the useful-
ness of the Wo hler diagram may be questioned. The obvious problem is: how do
we represent the roles of the two constituents in determining the fatigue-life, and
what roles do the constituents play in determining it? In order to address these
issues, a framework for conceptual interpretation of fatigue of composite materials
was proposed by Talreja [1]. That framework will be discussed below, beginning
with the case of on-axis tensiontension fatigue of unidirectional composites.
7.3 On-axis fatigue of unidirectional composites
Consider a unidirectional composite that is subjected to cyclic tension in the ber
direction. Let the specimens of the composite be tested under load control, i.e., the
load varies between set limits of maximum and minimum. We shall examine the
mechanisms operating and the failure resulting under this condition. To do this,
let us construct a plot with the horizontal axis as a logarithm of the number of
cycles applied and the vertical axis as the maximum strain attained in the rst load
cycle as seen in Figure 7.1. The reasons for plotting strain and not stress are
motivated by the following considerations. (1) The failure in the rst cycle occurs
when the composite strain equals the failure strain of bers irrespective of the ber
volume fraction. A point with coordinates (log 1, e
c
), where e
c
is the composite
failure strain, can thus be plotted in the diagram. (2) The composite fatigue limit is
governed by the matrix fatigue limit, as will be discussed later. The matrix within
the composite is subjected to strain controlled fatigue, due to the ber constraint,
although the composite is undergoing load controlled testing. Thus, the composite
fatigue limit is expressible in terms of strain.
Since the two extreme values of the fatigue-life are given in terms of strain, there
appears to be no compelling reason not to plot the strain between these limits.
We note that for a given ber type (e.g., glass or carbon) and matrix material
the stressstrain relationship changes when the ber volume fraction or ber
stiffness change. The signicance of plotting the maximum strain in the rst cycle
lies in the fact that this strain value provides a good reference to the damage
attained in the rst cycle and that the subsequent damage and fatigue-life are likely
to depend on this damage state.
Figure 7.1. Schematic illustration of fatigue-life diagram for a unidirectional composite,
showing three regions with different damage mechanisms.
238 Damage mechanisms and fatigue-life diagrams
As rst described in Talreja [1], the test data plotted on the axes (log N, e
max
) can
be viewed as falling in three regions. Region I is the horizontally extending scatter
band (e.g., between 5% and 95% probability of failure) of the composite failure
strain. This region represents a lack of degradation in strength (failure strain),
i.e., the underlying predominant mechanism of ber failure is nonprogressive.
The arguments to support this postulate will be described below. Region II is
the fatigue-life scatter band, which deviates from region I at a certain number of
cycles and extends down to the fatigue limit. This region is governed by the
progressive mechanism of ber-bridged matrix cracking, as described below.
Finally, region III is the region of no fatigue failure (in a selected large number
of cycles, say 10
6
) lying below the fatigue limit.
Figure 7.2 depicts three scenarios, the rst of which shows ber breaks
resulting from the rst application of a high load that produces a maximum
strain lying within the scatter band of composite failure. These ber breaks are
caused by the local ber stress (or strain) exceeding the ber strength at those
(weak) points during the rst application of load. Unloading and reapplying
the load would change the local ber stresses only if an irreversible (inelastic)
deformation occurs. Assume now that the matrix surrounding a ber break
undergoes only small inelastic deformation due to the constraint of the stiff
bers. This is likely if the matrix is relatively brittle, such as an epoxy. The case
of a matrix of high ductility (i.e., ability to ow easily) will be considered later.
For small inelastic deformation in the matrix the repeated application of load
would cause little change in the stresses on bers surrounding a broken ber
site. The next bers to break in a subsequent load cycle could appear near any
of the previously broken bers but not necessarily near all previously broken
bers when the local stresses exceed their local strengths, see the illustration
at N = N
1
in Figure 7.2. The important thing to note is that, due to the small
cycle-to-cycle changes in the local ber stresses and randomness of the ber
strength, it is unlikely that the ber breakage will be a progressive mechanism,
in the sense that the number of ber failures in a given location increases
monotonically. Thus the nal failure, which results from a core of (a few) ber
failures growing unstably (see illustration in Figure 7.2), could occur in any of
Figure 7.2. Nonprogressive ber breakage in region I in fatigue of unidirectional composites.
239 7.3 On-axis fatigue of unidirectional composites
the potential failure sites, without preference. The consequence of this is that
the composite failure under the specied loading condition can occur at any
number of load cycles. This can be described as a nonprogressive failure
mechanism, with no associated strength degradation. Thus the scatter band in
region I is horizontal.
The mechanism in region II is illustrated in Figure 7.3. At applied maximum
load levels below the lower bound of the region I scatter band, the composite
failure resulting from a cluster of neighboring ber breaks is unlikely. Instead,
with increasing load cycles, the matrix will undergo fatigue cracking. The matrix
cracks will progress by failing bers or by debonding and going around them.
A typical ber-bridged crack would then appear as illustrated in Figure 7.3.
Among several such cracks the one to cause failure earliest would be the one that
grows unstably rst.
Region III of the fatigue-life diagram can be viewed as the domain of no fatigue
failure in a pre-selected large (> 10
6
) number of load cycles. One possible scenario
in this region is illustrated in Figure 7.4. As shown in the gure, fatigue cracks
in the matrix are likely to develop, but they remain conned to the parts of the
cross sections between bers. The driving force for the cracks is insufcient to
advance them by failing and/or debonding bers. Another likely scenario in this
region is ber-bridged matrix cracking progressing at rates too low to cause
failure in the pre-selected large number of cycles. A true fatigue will exist only
if a mechanism of effectively arresting matrix crack growth is available.
Fiber-bridged matrix crack
Crack front
(N=N
1
)
Crack front
(N=N
f
)
Figure 7.3. Fiber bridged matrix cracking in region II in fatigue of unidirectional
composites.
Figure 7.4. Matrix cracking between bers in region III in fatigue of unidirectional
composites.
7.4 Effects of constituent properties
The baseline fatigue-life diagram provides a good starting point for making an
assessment of the roles of constituent properties in determining the fatigue
response of composite materials. In the following we shall discuss some expected
trends in the fatigue-life caused by changes to constituent properties.
Fibers of different materials, e.g., glass, carbon, and SiC, have different axial
moduli and fail at different axial tensile strains. Carbon bers of different stiffness
are commercially available and these bers can fail at strains as low as 0.5%, or at
strains exceeding 1.8%, depending on processing and surface treatment. The com-
posite stiffness and failure strain in the ber direction are mainly determined by the
ber properties in most composites of practical interest. The failure of bers is
subject to a signicant scatter due to imperfections, in particular the surface defects.
The composite failure is therefore also a statistical process, with added complexities
of stress transfer between matrix and bers, ber debonding, ber misalignment,
stress concentration in bers near a broken ber, etc. For our considerations
regarding the effect on composite fatigue of changing bers and/or matrix proper-
ties we can viewthe fatigue-life diagram and identify the following trends, as shown
in Figure 7.5, where ber-bridged cracking is the progressive mechanism.
1. As explained above, region I consists of the horizontally extended scatter band
of composite failure strain, e
c
. This strain can be raised or lowered by the ber
failure strain. Thus, the placement of region I with respect to the fatigue limit
e
m
(which is primarily a matrix property) is determined by the choice of bers.
This in turn means that the extent of region II (e
c
e
m
) is also determined
by bers. This provides a remarkable exibility in view of the wide variety of
bers that are commercially available and are being developed. Note also that
region II can be made to vanish (i.e., no fatigue!) by selecting bers such that e
c
is less than or equal to e
m
, as we shall demonstrate later.
2. The progressive fatigue mechanism in region II of the fatigue-life diagram is
inuenced signicantly by bers. The closing tractions on the matrix cracks are
Figure 7.5. Trends in fatigue-life diagrams due to constituent properties.
241 7.4 Effects of constituent properties
supplied by the bridging bers as they are strained. For the same stretching
strain on the bers, stiffer bers would supply greater forces than would more
compliant bers. Thus, lower fatigue degradation rates would result in com-
posites with stiffer bers, leading to a rightward displacement of region II with
increasing ber stiffness, as indicated in Figure 7.5.
3. Consider a ductile matrix versus a brittle matrix, in relative terms. A crack in a
ductile matrix would cause higher strains at the crack tips than in a brittle
matrix. Thus, higher crack opening displacement will result in a ductile matrix
than in a brittle one for the same crack length. This would mean higher strains
in the bridging bers for cracks in composites with more ductile matrices,
resulting in earlier ber failures than in the case of brittle matrices. In terms
of fatigue-life this would translate into lower fatigue-life for a more ductile
matrix, as indicated by the trend line in Figure 7.5.
4. The effect of constituent properties on the fatigue limit can be viewed in the
following way. First, fatigue in the composite cannot occur unless cracks can
be initiated in the matrix. Thus, the matrix fatigue limit in strain cycling is the
baseline limit (as described above). This limit can be enhanced by bers that
provide a means of arresting (obstructing) matrix crack growth. However,
stiffer bers would be more effective in suppressing the crack growth, than
would be more compliant bers, by supplying higher closing pressure on the
matrix crack planes. This would result in the trend in the fatigue limit indicated
in Figure 7.5.
The interpretation of fatigue behavior of unidirectional composites under cyclic
on-axis tension will now be illustrated by considering test data.
7.5 Unidirectional composites loaded parallel to the bers
The fatigue-life diagram, described above, is developed by generic considerations
of damage mechanisms. In what follows we will illustrate the diagram for specic
material systems.
7.5.1 Polymer matrix composites (PMCs)
The fatigue-life diagram will be used as a baseline diagram for interpretation of
fatigue of polymer matrix composites. Fatigue of other material systems metal
matrix composites and ceramic matrix composites then becomes easy to inter-
pret by way of emphasizing the differences and similarities with respect to polymer
matrix composites.
To begin, consider the fatigue-life data shown in Figure 7.6 for a glass/epoxy
composite of three different ber volume fractions indicated in the gure. These
data are plotted in the conventional manner with the applied cyclic stress
amplitude as the vertical axis. The fatigue-life curves appear as distinct lines for
each case of the ber volume fraction. The data are replotted in Figure 7.7 using
the maximum strain applied in the rst cycle as the vertical axis, in accordance
with the fatigue-life diagram. The data for the three volume fractions are found to
fall together, as seen in the gure. The conceived fatigue-life diagram with the
three regions has been superposed on the data. The horizontal scatter band of
composite failure strain is drawn as region I and fatigue limit of the epoxy resin in
strain control testing, reported by Dharan [2] to be at 0.6% strain, has been drawn
as the assumed fatigue limit of the glass/epoxy composites. A scatter band has
been placed about the fatigue-life data to indicate region II. This plot suggests that
plotting fatigue-life against strain allows more meaningful interpretation, as
argued above. It also provides conrmation of the proposition that the fatigue
Figure 7.6. Stress/life data of a unidirectional glass/epoxy composite loaded in tension
parallel to the bers [2]. Reprinted, with kind permission, from Fatigue in Composite
Materials, copyright ASTM International, 100 Barr Harbor Drive, West Conshohocken,
PA 19428.
0 2 4 6 8
log N
f
V
f
0.008
0.016
0.024
e
c
e
m
e
m
a
x
0.50
0.33
0.16
Figure 7.7. Fatigue-life diagram of a unidirectional glass/epoxy composite loaded in tension
parallel to the bers [2]. Reprinted, with kind permission, from Fatigue in Composite
Materials, copyright ASTM International, 100 Barr Harbor Drive,West Conshohocken,
PA 19428.
243 7.5 Unidirectional composites loaded parallel to the bers
limit of the matrix is a good indication of the composite fatigue limit, when plotted
in the strain coordinate. The existence of region I is not strongly indicated by the
glass/epoxy data, for reasons that will become clear as we examine other data.
Consider now a set of unidirectional carbon/epoxy composites where each com-
posite has different type of carbon bers and the same epoxy matrix. Figure 7.8
illustrates two examples of stressstrain behavior, one with relatively low stiffness
bers (Figure 7.8(a)) and the other with relatively high stiffness bers (Figure 7.8(b)).
The composite failure strain e
c
, assumed equal to the ber failure strain, has been
marked on the strain axis for both cases. Note that the failure strain is lower for
higher ber stiffness in accordance with observed behavior of carbon bers. Also
marked is the fatigue limit of epoxy, at the same value in both cases, assuming for
simplicity that it is a matrix property not affected by bers.
The consequence of changing carbon ber stiffness is thus essentially to increase or
decrease the extent of region II. Imagine now that a certain carbon ber is used that
has such high stiffness that the composite failure strain e
c
<e
m
, the fatigue limit. This
can also be caused by defective bers of low strength. The consequence of composite
failing at strains below the fatigue limit is that no fatigue progression would be
possible. In other words, region II would not exist, as the fatigue limit would lie
above region I. Thus, only region I (a horizontal scatter band) will appear in the
fatigue-life diagram. Figure 7.9 illustrates this phenomenon by using fatigue-life data
gathered by Sturgeon [3] for a carbon/epoxy composite with high stiffness bers. The
fatigue limit, assumed for epoxy to be 0.6% strain, has been marked in the gure
along with the average composite failure strain at approximately 0.48% strain.
The next set of test data, plotted in Figure 7.10, is from Awerbuch and Hahn [4]
for carbon/epoxy with bers of stiffness lower than those in the composite used by
Sturgeon [3] just discussed. Here the data suggest no fatigue degradation when
viewed without the fatigue-life diagram. However, when the scatter band of region
s
s
e
m
e
m
e
c
e
c
e
e
Fiber
(a) (b)
Fiber
Composite
Composite
Matrix
Matrix
Figure 7.8. Effect of ber stiffness and strain to failure on composite stiffness in a
unidirectional composite in longitudinal loading, where e
c
is the composite failure strain
and e
m
is the matrix fatigue limit. (a) Composite with low stiffness bers; (b) composite with
high stiffness bers.
I and the fatigue limit of epoxy are placed in the diagram, the test data suggest a
different interpretation. The progressive fatigue mechanism of region II is now
seen to exist, albeit in a narrow range of strain.
Another case of differing interpretation with and without the fatigue-life
diagram is shown in Figure 7.11, where data from Sturgeon [5] for a lower stiffness
carbon/epoxy are plotted. Contrary to Sturgeons assertion that no fatigue deg-
radation exists for this material, the fatigue-life diagram suggests that, depending
on where the fatigue limit lies, there would be a region II of progressive fatigue
damage. In the gure the fatigue limit is indicated at 0.6% strain as a reference
for comparison with other cases of carbon/epoxy composites.
As the last case of carbon/epoxy fatigue we take the data supplied by
P. T. Curtis (personal communication) plotted in Figure 7.12. Here the regions
of the fatigue-life diagram appear distinctly, illustrating the power of interpret-
ation offered by the diagram. The value of recognizing the existence of region I is
0
0.002
0.004
0.006
4 8 log N
f
e
c
e
m
e
max
Figure 7.9. Fatigue-life diagram of unidirectional composite reinforced by stiff bers with
low strain to failure (data from [3]). It is notable that there is no fatigue degradation, since
the mechanism is nonprogressive ber breakage only.
0 2 4 6 log N
f
0.002
0.006
0.010
e
max
e
c
e
m
Figure 7.10. Fatigue-life diagram of unidirectional composite reinforced by medium stiffness
bers, showing a narrow range of strain where fatigue occurs [4]. Reprinted, with kind
permission, from Fatigue of Filamentary Composite Materials, copyright ASTM
particularly evident. Without this, the error introduced in assessing the fatigue-life
would be signicant.
As further illustration of the fatigue-life diagram, Figures 7.13 and 7.14 show
test data for Kevlar
-epoxy and Kevlar
-J-2 polymer, respectively. The J-2

0
0.01
0.02
4 8 log N
e
m
e
c
e
max
Figure 7.11. Fatigue-life diagram of a unidirectional composite with low stiffness bers
(data from [5]), with a relatively wide range of strain where fatigue occurs.
0
0.6
0.9
1.2
1.5
1 2 3 4 5 6 7
log N
f
c
e
max
(%)
e
m
Figure 7.12. Fatigue-life diagram of a unidirectional carbon ber/epoxy composite with
a distinct region I (data courtesy of P. T. Curtis, RAE, UK).
0
0.0
0.5
1.0
1.5
2.0
1 2 4 3 5 6 7
log N
f
e
max
(%)
e
c
e
m
Figure 7.13. Fatigue-life diagramof a unidirectional Kevlar ber/epoxy composite (data from[6]).
polymer is an amorphous polyamide thermoplastic [6]. The generality of the
fatigue-life diagrams is that they lend themselves to description and interpretation
of any kind of fatigue-life data for composites. A comparison of fatigue perform-
ances between different material candidates can then be made. Commercial mater-
ial systems of unknown constituent composition can thus be compared, and the
best contender for an intended application can be identied.
7.5.1.1 Experimental studies of mechanisms
The fatigue-life diagrams developed in [1] and described above for PMCs were
based on deductive reasoning in the absence of any systematic investigation
of such mechanisms. Since then an investigation was conducted by Gamstedt
and Talreja [7], which, through direct observations of on-axis tensiontension
fatigue of unidirectional carbon/epoxy, gathered evidence to support the diagrams.
For instance, the presence of ber-bridged cracking postulated in region II was
0 1 2 3 4 5 6 7
log N
f
0.0
0.5
1.0
2.0
1.5
e
max
(%)
Figure 7.14. Fatigue-life diagram of a unidirectional Kevlar ber/J2 polymer composite
(data from [6]).
Figure 7.15. A crack on the surface of a carbon/epoxy unidirectional composite subjected to
tensiontension fatigue. The crack tips are squeezed by the bridging bers, as illustrated
schematically by the bottom gure. Reprinted, with kind permission, from Springer Science
+Business Media: J Mater Sci, Fatigue damage mechanism in unidirectional carbon
bre-reinforced plastics, Vol. 34, 1999, pp. 253546, E.K. Gamstedt and R. Talerja.
found as exemplied by an image of a surface replica shown in Figure 7.15.
The crack opening prole shows squeezing of the crack tips by bers.
A mechanism for arresting matrix crack growth in region III was also dis-
covered in [7]. Its evidence is displayed in Figure 7.16. As seen there, a matrix
crack growing transverse to the applied load is arrested by extensive ber/matrix
debonding. Without the energy dissipation in the debonding process the crack will
have a greater driving force for extension.
In [8] the fatigue propagation of ber-bridged cracks in unidirectional PMCs was
analyzed. As expected, in comparison to an unreinforced matrix crack, the fatigue
growth rate of ber-bridged cracks was found to be lower, and even showing
decelerating growth. In another study, Gamstedt [9] examined the role of ber/
matrix interfaces prone to debonding on the composite fatigue under on-axis
tensiontension loading. If the interfaces resist debonding, which may be viewed
as the baseline (normal) case, the fatigue mechanisms described above in developing
the fatigue-life diagram would apply. As debonding increases, the ber strength
variability becomes increasingly important in determining the damage progression,
and in signicant debonding cases, the ber-bridged crack growth gives way to
debonding-assisted ber breakage as the progressive mechanism. The slope and
extent of region II of the fatigue-life diagram is then determined by the debonding-
assisted ber breakage. The at region I, which is a consequence of the unconnected
ber breakage, then diminishes, and can eventually become part of region II.
Figure 7.17 illustrates the cases of little or no debonding vs. extensive debonding.
The former case is typical of relatively strong interfaces in carbon/epoxy composites.
When ber breaks occur, initially at randomly located weak points, they remain
isolated and unconnected due to the lack of debonding. As discussed above in
Fiber/matrix debonding
Matrix crack
Figure 7.16. A surface replica image showing a matrix crack arrested by ber/matrix
debonding. Reprinted, with kind permission, from Springer Science+Business Media:
J Mater Sci, Fatigue damage mechanism in unidirectional carbon bre-reinforced plastics,
Vol. 34, 1999, pp. 253546, E.K. Gamstedt and R. Talerja.
describing region I, any of the initial breaks are likely to form a critical-size crack that
could grow unstably, leading to failure. Thus there is no progressive mechanism and
composite failure results essentially at any number of cycles, resulting in a horizontal
scatter band. In the case of interfaces that are prone to debond, the inuence of an
initial ber break extends to other bers, depending on the rate at which the debond
crack elongates, and a progressive ber breakage results. The rate of this damage
progression depends on the resistance to the debond crack growth given by the
interface region and the likelihood of the debond cracks connecting with other cracks.
The overall path to nal failure may be described in terms of an average progression
rate. Modeling of this progression is far from an easy task. Attempts at this were
made by Gamstedt [9], who conducted numerical simulation of the ber breakage
process and claried the roles of interfacial debonding and ber strength variability.
Figure 7.18 compares the fatigue-life diagrams of unidirectional carbon/epoxy and
carbon/PEEK (polyetheretherketone) composites loaded in tensiontension loading.
The former material typies the strong ber/matrix bonding case (left in Figure 7.17)
while the latter is prone to debonding. The PEEK matrix is a semi-crystalline thermo-
plastic that develops a so-called trans-crystalline structure extending from the ber
surface. This structure provides weak planes parallel to bers for debonding to occur.
Althoughotherwise a toughmaterial, PEEKbecomes a source of brittle cracking of the
interfaces. The extensive debonding in carbon/PEEK removes region I of the fatigue-
life diagram by inducing progressive ber breakage, as explained above.
(a)
(b)
Figure 7.17. Two cases of the role of ber/matrix debonding in fatigue damage progression
under on-axis tensiontension loading of unidirectional ber-reinforced PMCs. Case 1 (a)
illustrates unconnected ber breaks when little or no debonding occurs while case 2 (b) is
for extensive debonding that connects ber breaks and causes the ber breakage to be
progressive. Reprinted, with kind permission, from Springer Science+Business Media:
J Mater Sci, Fatigue damage mechanism in unidirectional carbon bre-reinforced plastics,
Vol. 34, 1999, pp. 253546, E.K. Gamstedt and R. Talerja.
7.5.2 Metal matrix composites (MMCs)
The fatigue-life diagram discussed above should be viewed as a baseline diagram
for polymer matrix composites, as the mechanisms considered for its construction
were motivated by observations and conjectures related to this material system.
We shall now discuss what changes in the diagram are plausible when the matrix
material is a metal. Note that the role of bers is in modifying the fatigue
mechanisms taking place in the matrix, which is the material conducting irrevers-
ible deformation (plasticity). The three regions of the fatigue-life diagram for
MMCs treated by Talreja [10] will be discussed next.
Region I: As discussed earlier, this region is manifested by ber failures.
For polymer matrix composites we argued that the ber failures occurred in a
manner that did not have signicant progressiveness (accumulation of ber
failures in a localized zone), leading to the nal (composite) failure from
random sites at random number of cycles. An important factor in this deduction
was insufcient irreversible deformation in the matrix to allow cycle-dependent
stress enhancement and failure of bers. This scenario will change when a
relatively brittle polymer matrix is replaced by a more ductile metal matrix.
The cyclic plastic deformation of a metal matrix around a broken ber will
redistribute stresses in the neighboring bers, allowing an accumulative ber
failure process to occur. This will introduce a localized progressive degradation,
which on reaching a critical level will cause composite failure. Thus, the hori-
zontal scatter band of region I, characteristic of polymer matrix composites, will
now have a downward slope.
Region II: The mechanism of ber-bridged matrix cracking is also expected to
be the primary progressive mechanism for metal matrix composites. The
0
0.6
0.9
1.2
1.5
M
a
x
.

s
t
r
a
i
n

(
%
)
1 2 3 4 5 6 7
Log cycles
Carbon/PEEK
Carbon/epoxy
Figure 7.18. Comparison of the fatigue-life diagram for unidirectional carbon/epoxy and
carbon/PEEK composites loaded in tensiontension cycles along bers. Note the absence
of region I in carbon/PEEK.
difference with respect to polymer matrix composites lies in the role of ber/
matrix debonding. The ber/matrix bond in metal matrix composites is generally
stronger, leading to shorter extent of the interface failure. Furthermore,
increased ductility of the matrix is expected to provide more crack opening
displacement via crack tip blunting, causing increased failure of the bridging
bers. The matrix ductility effect will thus be as conjectured above in describing
the trend in region II of the fatigue-life diagram as illustrated in Figure 7.5.
Region III: This region is expected to be the same as in polymer matrix
composites. The composite fatigue limit is also expected to be related to the
matrix fatigue limit.
Let us now examine some test data for metal matrix composites. Figure 7.19
plots fatigue-life at room temperature for a SCS6/Ti-153 unidirectional composite
under cyclic tension at two R-ratios. Static failure strain is plotted on the vertical
axis at the rst cycle. As discussed above, region I displays some progressiveness by
deviating from a horizontal scatter band. Region II shows greater degradation
rates and tends to the fatigue limit gradually. Note that the matrix fatigue curve,
plotted as a broken line, tends to approximately the same fatigue limit as the
composite. The data for the same composite tested at a high temperature are
shown in Figure 7.20. The general features of the fatigue-life diagram at room
temperature are retained with different quantitative characteristics. Comparisons
Figure 7.19. Fatigue-life diagram of SCS6/Ti-153 at room temperature.
Figure 7.20. Fatigue-life diagram of SCS6/Ti-153 at high temperature (540550
C).
of the two regions are displayed in Figures 7.21 and 7.22. Figure 7.21 shows the
data in region I at room and high temperatures on loglog scales. The best-t lines
for the data show that higher fatigue degradation (more progressiveness in the
ber failure mechanism) occurs at high temperature than at room temperature.
Similar region II comparison at the two temperatures is shown in Figure 7.22.
A leftward shift of region II, indicative of more fatigue degradation, is seen with
an increase in temperature. This is likely due to higher matrix ductility (lower yield
stress) at the high temperature, in accordance with the trends in the fatigue-life
diagram described above.
7.5.3 Ceramic matrix composites (CMCs)
In comparison with polymers and metals, ceramics have insignicant irreversi-
bility of deformation. This suggests that there should be little incentive for fatigue
mechanisms to occur in ceramics. However, when ceramics are reinforced by
Figure 7.21. Region I for SCS6/Ti-153 at high (540550
C) and low temperature (20
C).
Figure 7.22. Region II for SCS6/Ti-153 at high (540550
C) and low temperature (20
C).
bers with which they bond weakly, a dissipative mechanism becomes available
from reversed (i.e., back-and-forth, not reversible) frictional sliding at interfaces.
This mechanism is believed to play a major role in causing fatigue of CMCs. In the
following a discussion of the fatigue-life diagram for unidirectional CMCs under
cyclic axial tension is presented.
Before getting into a discussion of fatigue of CMCs it would be useful to briey
reviewthe damage and accompanying stressstrain response observed under mono-
tonic axial tension. Several works in the literature have reported such data, and we
shall mainly draw upon Srensen and Talreja [11] to illustrate some characteristics
of interest. Figure 7.23 shows three surface replicas of a SiC/CAS (calcium alu-
minosilicate) glass specimen taken at three applied axial strains at room tempera-
ture. For reference, the failure strain is approximately 1.0%. Note that at 0.15%
strain a few irregularly spaced cracks transverse to bers are seen. At this low strain
level the cracks are partially developed, i.e., they do not extend across the entire
width of the specimen. As the applied strain increases, the cracks span the width
fully and acquire more regular spacing. At some point the crack spacing approaches
a minimum value, usually described as saturation, as seen in the replica at 0.8%
strain. The stressstrain response in the axial as well as transverse directions is
shown in Figure 7.24, alongside a plot of the AE(acoustic emission) events recorded
during testing. The double stressstrain curves are due to strain gages used on both
sides of the specimen to monitor axiality of loading. Figure 7.25 summarizes the
different stages of damage progression, indicating the ranges in which they operate,
based on microscopy observations and AE event counts.
When considering fatigue, the rst question to ask is: what happens in the
material in the second and subsequent load cycles that is different from the
Fiber-bridged
matrix crack
Fibers
500 m
e =0.15% e =0.5% e =0.8%
Figure 7.23. Micrographs of surface replicas of a unidirectional CMC with increasing strain:
015%, left; e = 05%, middle; e = 08% right. Source: [11].
damage caused, if any, in the rst load cycle? The answer to this question lies in
knowing the source of irreversibility in the material. For instance, if a material
is cracked (singly or multiply) in the rst load cycle, it would not be cracked
further in the next and subsequent cycles unless an irreversible mechanism such as
plasticity or friction is present. As stated earlier, ceramic materials have insigni-
cant irreversible (plastic) deformation to enable crack growth in cyclic loading.
The likely source of irreversibility is frictional sliding at the interface. Keeping this
in mind we shall examine below the existence and nature of the three regions of the
fatigue-life diagram for CMCs (see Figure 7.26).
0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
Strain (%)
0
100
200
300
400
S
t
r
e
s
s

(
M
P
a
)
500
600
Monotonic tension
IV Fiber
fracture
III Fiber
bridging
(Matrix fully
cracked)
IV
III
II
I
I No
damage
II Matrix crack
development
(initiation of
matrix crack)
D
a
m
a
g
e

d
e
v
e
l
o
p
m
e
n
t
SiC
f
/CAS-II
RT, 100 MPa/s
s
pl
=285 MPa
s
mc
=120 MPa
Figure 7.25. Schematic drawing of damage mechanisms at different stages in monotonic
tensile loading of a unidirectional CMC loaded in tension parallel to the bers.
2.0 1.5 1.0 0.0
0
100
200
300
400
500
600
0.0 0.25 0.50 0.75 1.00
AE events (10
6
) e (%)
e
L
e
T
s

(
M
P
a
)
Figure 7.24. Measured acoustic emission events and stressstrain response. Source: [11].
Region I: When a rst load is applied up to a maximum value within the scatter
band of the composite failure strain, the matrix cracks develop beyond their
saturation stage (see Figure 7.25). This means that a set of fully grown ber-
bridged cracks exist at the end of the rst load cycle. It is likely that a signicant
portion of the bers will be broken due to the high load applied. When the
second and subsequent cycles are applied, the ber stress will increase due to the
increased debond length caused by the irreversible frictional sliding, supplying
the increased shear stress at the interfaces. Also, the cyclic grinding of the ber
surfaces by the debris generated by the interfacial wear will damage bers and
degrade their strength. All this will fail the bers and when any of the cross
sections going through the crack planes has insufcient number of intact bers
to bear the applied load, composite failure will occur. Now the critical question
is whether this failure mechanism is progressive or nonprogressive. If failure
were to come from only one cross section, i.e., if only one crack existed, or if one
crack was somehow the favored one, then the accumulative ber breakage will
result in a progressive failure. However, that is not characteristic of the damage
here. We have multiple cracks, and failure could potentially result from any one
of those cracks. Although each ber-bridged crack has progressive (accumula-
tive) ber failure, the rate of progression would likely be discontinuous and
differ from that of other cracks. Thus the nal failure would not necessarily
result from the weakest of the cross sections produced by the rst load. We
postulate, therefore, that the number of cycles for composite failure in this case
cannot be derived from a single rate equation, and consequently, the failure
process is nonprogressive.
Region II: Consider now the case of loading where the rst cycle maximum load
is in a range such that the matrix cracks are either partially developed or are
Figure 7.26. Mechanisms in the three different regions of the fatigue-life diagram for
unidirectional CMCs loaded in tension parallel to the bers.
fully developed but the bridging bers are not broken extensively. The partial
cracks will grow in subsequent cycles by debonding and/or breaking bers at the
crack tips (or fronts). The cyclic frictional sliding at the interfaces will debond
the bridging bers, causing crack surface separation to increase as the debond
length increases and as more bridging bers break. With cyclic stressing, there-
fore, partial cracks will grow to their full extent. As more debonding occurs,
stress will be transferred from bers to the matrix increasingly, resulting in
matrix cracking between two pre-existing cracks. A state of crack saturation
will be approached eventually, unless composite failure results from failure of
all bers bridging any of the cracks. The damage development in this region of
the fatigue-life diagram has well-dened progressiveness from partial cracking
to full cracking until crack saturation. The terminal point is when at least one
crack loses all its bridging bers. This point may occur before crack saturation
(toward the lower end of region II) or beyond crack saturation (toward the
higher end of region II). The main difference between this region and region I is
that in the latter saturation cracking with extensive ber breakage already
occurs in the rst load application.
Region III: We have dened this region to be a range of loading (maximum
strain in the rst cycle) in which damage may develop but does not lead to
a critical state (failure) in a pre-selected large number of cycles (typically 10
6
).
In a unidirectional composite, failure (separation) will occur when all the
bridging bers of at least one crack fail, or, alternatively, broken ber ends
are interlinked by interfacial debonds and matrix cracks to form two separating
surfaces. In either case, interfacial frictional sliding plays an important role
by increasing debonding and thereby stressing bers to failure. What has been
found, however, is that the wear of the debonded interfaces (breakage of
asperities resulting in smoothing) can make the frictional sliding less effective.
This could slow down the cyclic development of damage to the extent of not
reaching failure in the pre-specied number of cycles. Srensen et al. [12] found
that the temperature rise of SiC/CAS specimens caused by frictional heating at
debonded interfaces increased sharply just before failure, while for specimens
that did not fail in 10
8
cycles the temperature rise stopped at some cycles and
then took a downward turn (Figure 7.27). These authors have challenged the
notion of a true fatigue limit in CMCs, arguing that damage continues to
develop even after 10
8
cycles, as long as interfacial debonding occurred in the
rst cycle. They contend that the condition for no damage development is no
initiation of ber-bridged crack, providing the true fatigue limit as the stress
(or strain) at which such crack forms.
The discussion concerning fatigue limit in CMCs is somewhat academic because
the energy dissipating mechanism of frictional sliding at debonded interfaces
depends on the loading frequency and specimen geometry. This is because the
time rate at which conduction of the heat generated by frictional sliding occurs
affects the rate per cycle of energy dissipation by frictional sliding.
The fatigue data on CMCs reported in the literature are inadequate for a good
illustration of the usefulness of the fatigue-life diagram. Since the testing was
done with a traditional S-N curve in mind the data generated were too few in
region I to validate its existence. Also, because of the material cost, the data
generated were at too few load levels to have all three regions of the diagram
appear with clarity.
7.6 Unidirectional composites loaded inclined to the bers
We consider the simple case of a single ber orientation in a composite placed at
an angle to the applied cyclic tension load as shown in Figure 7.28. As illustrated
in the gure, cracks initiate along the bers, either at interfaces or in the matrix.
At rst the cracks may initiate from defects in the matrix or at interfaces and grow
with cycles through the thickness as well as along the bers. At some stage in the
evolution of this damage, a single crack may grow to the extent of attaining
unstable growth in the next application of load and separate the composite in
two pieces, as illustrated in the gure.
In comparison to the on-axis loading case, i.e., when the bers are aligned with
the loading direction, the damage mechanism of the off-axis loading case is
drastically different and dramatically simple. The single progressive mechanism
depicted in Figure 7.28 will produce a continuous curve (and the associated scatter
band) in the fatigue-life diagram starting at the rst-cycle failure strain and ending
asymptotically in the fatigue limit. For each off-axis angle the curve will be different.
A schematic depiction of the fatigue-life diagram is shown in Figure 7.29, which
illustrates the fatigue-life dependence on the off-axis angle. The on-axis fatigue-life
diagram is shown for reference in the gure in broken lines.
1
MPa
0
10
20
30
40
10
2
10
4
10
6
10
8
Cycles N
T
e
m
p
e
r
a
t
u
r
e

r
i
s
e

(
K
)
X
X
220 MPa
212 MPa
Run
out
s
max
=260 MPa
s
min
=10
[0
16
] SiC
f
/ CAS II
200 Hz, RT
Figure 7.27. Temperature increase due to frictional heating during cyclic loading of
a unidirectional SIC/CAS II composite.
257 7.6 Unidirectional composites loaded inclined to the bers
As Figure 7.29 illustrates, region I (horizontal scatter band) of the fatigue-life
diagram does not exist when the applied cyclic load is inclined to the ber
direction. The single scatter band starts at the composite failure strain and
asymptotically ends at the fatigue limit. The failure strain as well as the fatigue
limit strain depend on the off-axis angle, y. From the S-N data reported for glass/
epoxy in [13] the fatigue-life diagrams for a few angles are shown in Figure 7.30.
The curves drawn by visual t to the data show the trends depicted in Figure 7.29.
The fatigue limits extracted from the data are shown plotted in Figure 7.31 where
the fatigue limits deduced from [14] for angle-ply laminates of glass/epoxy
are also plotted. The signicant improvement in the angle-ply laminates over
unidirectional composites for the same off-axis angle is discussed in the next
section on fatigue of laminates.
e
max
e
c
e
c
(0<q<90)
(q=0)
logN
q
Figure 7.29. Fatigue-life diagram for unidirectional composites under off-axis loading.
The diagram in broken lines is for the on-axis loading case.
Figure 7.28. Cracking in a unidirectional composite under off-axis cyclic tension (left)
and failure from growth of a crack (right).
7.7 Fatigue of laminates
Composite structures are built by placing bers in different orientations to
effectively carry multi-axial loading. We will here consider how the mechanisms
of fatigue damage are affected by multi-directional ber placement in a
0
q(degrees)
0.004
0.002
0.006
30 60 90
e
f
l
Figure 7.31. The off-axis fatigue limit from data in Figure 7.30 (dashed line) is shown
for comparison with the fatigue limit of angle-ply laminates of glass/epoxy taken from data
reported in [14], plotted against the off-axis ply angle.
0 2 4 6
lg N
e
db
e
c
e
c
e
c
e
m
e
m
a
x
e
c
0.008
0.006
0.004
0.002
(q=60)
(q=30)
(q=10)
q/deg
(q=5)
5
10
30
60
Figure 7.30. Fatigue-life data for glass/epoxy unidirectional composites from [13] plotted
as fatigue-life diagrams for different off-axis angles of tensiontension loading. The fatigue
limit of on-axis loading is indicated at 0.6% strain for reference. Reprinted, with kind
permission, from Z. Hashin and A. Rotem, J Compos Mater, Vol. 7, pp. 44864, copyright
# 1973 by Sage Publications.
259 7.7 Fatigue of laminates
laminate subjected to a cyclic tensile load and how that translates into changing
the basic fatigue-life diagram described above.
7.7.1 Angle-ply laminates
These laminates have two ber orientations placed symmetrically about a princi-
pal direction, e.g., [y
n
]
s
illustrated in Figure 7.32. When loaded in the axial,
i.e., 0
direction, a y-ply in the laminate behaves differently from a unidirectional

composite loaded off-axis at the same y-angle. First, for the same applied axial
strain, the stress state in the ply within the laminate differs from that in the
unidirectional composite. Second, the unidirectional composite fails when a crack
appearing along its bers grows unstably, while in the laminate the unstable
growth of a ply crack does not cause laminate failure. Instead, the presence of
the interface to the ply (or plies) in the other off-axis orientation leads to multiple
ply cracking. The so-called shear lag process underlying the crack multiplication
has been treated in detail in Chapter 4. Figure 7.32 depicts schematically the
multiple cracking and the consequent (associated) delamination. As a ply crack
front encounters the ply interface, the intense stress state ahead of the crack front
debonds the interface, causing local ply separation (delamination). This delamina-
tion accompanies each ply crack and under cyclic loading undergoes growth along
the interfacial plane. While this occurs, the multiple cracking process within the
plies can continue due to stress redistribution in the plies, if the renewed stress
state is high enough. Under what conditions would the renewed stress state be
critical for producing new cracks is a subject of damage evolution modeling and
will be discussed in a subsequent section of this chapter.
The key elements of the fatigue failure process in laminates are illustrated by the
angle-ply fatigue, and may be summarized as: multiple ply cracking, delamination
formation, and growth, possibly followed by more ply cracking, and delamination
merger in ply interfaces, leading to separation of plies, which in turn can overload
Matrix
crack
Delamination
q q
Figure 7.32. Multiple matrix cracks in off-axis ply of a laminate (left) and subsequent
delamination caused by fatigue (right). For clarity, cracks and delamination are shown
for one ply only; the other ply is indicated in broken lines.
individual plies resulting in ply failure (separation). The laminate failure (separ-
ation in parts) comes from the nal event of ber failures.
Thus while an off-axis loaded unidirectional composite has essentially a two-
stage fatigue process consisting of ply crack initiation and growth until instability,
in laminates the fatigue process generally has several interactive cracking
processes. One can lump them in two parts as the critical and sub-critical failures.
The critical failure is the ber breakage process, which itself can have its progres-
sion. The sub-critical failure process consists of all events until ber failures.
The enhancement of the off-axis fatigue limit in angle-ply laminates shown
by data in Figure 7.31 can now be explained by the presence of the sub-critical
failure processes in laminates and its absence in unidirectional composites loaded
at off-axis angles.
The fatigue-life diagram for angle ply laminates is expected to lose region I,
except at the y-angle approaching 0
. The fatigue limit will not be adversely

affected until the sub-critical damage processes become insignicant, such as at
y > 30
for the data in Figure 7.31, and at y = 90
the fatigue limit will be the

same as for a unidirectional composite under transverse loading. It is possible that
at small y-angles some enhancement in the fatigue limit results from the lamin-
ation effect of inducing multiple cracking. Experimental data does not seem to be
available, however, to conrm this.
7.7.2 Cross-ply laminates
Cross-ply laminates, [0
n
/90
m
]
s
, are orthogonal angle-ply laminates, on which
loading is usually applied in one of the two ber directions. Historically, this
laminate was the rst class of laminates for which multiple ply cracking was
experimentally observed and reported in the literature in the late 1970s (see
Chapter 3 for details). In those studies, monotonically increasing tensile load
was applied along the 0
-direction, resulting in multiple ply cracks in the 90
-plies.
Since multiple cracking is a basic feature of damage in composite materials, this
conguration and loading combination has been, and continues to be, a test-bed
for modeling studies of multiple cracking and its consequences on the material
response.
The observations of multiple cracking in cross-ply laminates under monotonic-
ally increasing load do not, however, reveal the key feature underlying progres-
siveness of damage under cyclic loads. Since most studies typically observed the
cracking process on the free edges of a at specimen, the details of damage
accumulation were missed. The rst study to examine the interior of a laminate
with painstaking patience and using X-ray radiography, combined with stereo
radiography, was by Jamison et al. [15]. The key X-ray picture, along with a
schematic to depict interior details, is shown in Figure 7.33.
Other than the transverse cracks, the details seen in Figure 7.33 are not found to
develop sufciently under monotonic loading. However, these details hold the key
to the further progression of transverse cracking under cyclic loading. As we shall
discuss below in the section on modeling, the irreversible changes needed from one
load cycle to the next can be lumped into frictional sliding between the surfaces of
the propagating delamination.
In constructing the fatigue-life diagram of a cross-ply laminate under cyclic
axial tension, one needs to ask certain basic questions. First, are conditions
available for region I to be present? As discussed above in explaining the reasoning
behind this region, a statistical nonprogressive ber breakage mechanism must be
available for this region to exist. The presence of 0
-plies in the laminate makes

this mechanism plausible at applied axial strains near the average ber breakage
strain that leave the laminate intact at the rst application of load.
Second, where is the fatigue limit? To determine the fatigue limit, if not
available from test data, a good estimate will be given by the strain at which
transverse cracking initiates. The reasoning is simply that if no cracks initiate in
the transverse plies, no progression of crack multiplication and subsequent
damage are possible. The models for crack initiation strain have been discussed
in Chapter 6.
What remains to complete the fatigue-life diagram for cross-ply laminates is the
location of region II. If the scatter-band of this region is assumed to be straight
and sloping, a good approximation to begin with, then by xing its lower end at
the fatigue limit (at, say, 10
6
cycles), its slope or the point of its deviation from
region I remain to be determined. A model that has been successfully determined
will be discussed below in the section on fatigue-life prediction.
Figure 7.34 shows data reported by Grimes [16] for a carbon/epoxy cross-ply
laminate, plotted on the strain scale. The anticipated fatigue-life diagram, as
discussed above, is superimposed on the data. The scatter-bands drawn are guided
by the data, as insufcient data have been reported to estimate these bands from
probability distributions. The location of the fatigue limit strain has been taken at
the value of stress (converted to strain) reported by the author as the value where
the rst transverse cracking was found.
Transverse ply
cracks
Transverse
cracks
Axial
splits Axial
splits
Interior
delaminations
Interior
delaminations
Figure 7.33. X-ray radiograph of a carbon/epoxy cross-ply laminate taken after tension
tension fatigue shows transverse ply cracks, interior delamination, and axial splits. The
accompanying schematic claries the interior details [15]. Reprinted, with kind permission,
from Effects of Defects in Composite Materials, copyright ASTM International, 100 Barr
Harbor Drive, West Conshohocken, PA 19428.
7.7.3 General multidirectional laminates
For practical structures the composite architecture is designed to satisfy multiple
requirements, such as resistance to bending and torsion as well as thermal expan-
sion. A common composite architecture is a laminate consisting of plies, each with
unidirectional bers, stacked in a sequence such that the resulting structure has the
required combination of properties. An example is [0/45/90]
s
laminate, which is
quasi-isotropic, i.e., it has directionally independent average elastic moduli in its
mid-plane. Many other laminate congurations are possible, but often the number
of ply orientations are kept to three, and the most used ones are 0
, 45
, 45
, and
90
. From a fatigue point of view the assessment of laminates is remarkably

simple. First, irrespective of other ply orientations, the presence of 0
-plies
provides region I, which lies as a scatter-band about the ber failure strain, as in
the unidirectional on-axis loading case as well as in the cross-ply laminate case.
The fatigue limit for any laminate is determined by the rst cracking mechanism.
Therefore, if the 90
-ply orientation is present in a laminate, the strain at which the

rst transverse cracking occurs will determine the fatigue limit. This strain value is
affected by the so-called ply constraint, i.e., the ratio of the transverse ply modulus
to the axial modulus of the constraint-providing plies as well as the thickness
ratio of the cracking plies to the constraining plies. For further discussion of the
constraint effect see [17] and Chapter 3. The progressive fatigue damage, repre-
sented by region II, appears as a sloping scatter band, starting at a low number of
cycles (10
2
10
3
) and asymptotically approaching the fatigue limit at a high number
of cycles (10
6
10
7
). What remains to determine now is the slope of the region II
band (or line). Obviously, a life prediction model accounting for the (sub-critical)
progressive fatigue damage would be needed to predict this slope. Let us examine
some test data to get some insight into this slope.
Figure 7.35 shows the fatigue-life diagram for a glass/epoxy [0/45/90]
s
lamin-
ate under tensiontension loading along the 0
-direction with data from Hahn and

Kim [18]. To construct the diagram, the failure strain of bers (same as that of the
0
0.004
0.008
2 4 6
log N
e
m
a
x
e
c
e
dl
Debonding in 90 plies,
delamination
Figure 7.34. Experimental data verifying the anticipated fatigue-life diagram of cross-ply
laminates, with data from [16]. Reprinted, with kind permission, from Composite Materials:
Testing and Design, copyright ASTM International, 100 Barr Harbor Drive, West
Conshohocken, PA 19428.
laminate) is taken to place the scatter-band of region I. Since the failure strain data
were not available to determine the failure probabilities, the scatter-band was
drawn based on other similar data regarding ber failure strain. The fatigue limit
was placed at 0.46% strain based on information concerning the strain at which
transverse cracking was observed. Region II has been placed around the fatigue-
life test data. It is noted that the scatter-band of region II is approximately straight
(or negligibly curved) and meets the region I band at 10
2
10
3
cycles. The lower end
of the region II band is at approximately 10
7
cycles.
Another data set to consider is from [19] for a carbon/epoxy [0/45/90]
s
laminate under tensiontension loading along the 0
-direction (see Figure 7.36).

The procedure for constructing the fatigue-life diagram is the same as that just
described related to Figure 7.35.
From the two examples of fatigue-life diagrams in Figures 7.35 and 7.36 it is
remarkable that the conceptual framework these diagrams represent is a powerful
means of interpreting and estimating fatigue-life of composite laminates. Under
the restriction of tensiontension cycling along a principal direction, usually a
e
m
a
x
e
c
0.0100
0.0075
0.0050
0.0025
0 2 4 6
lg N
e
d.1.
=0.0046
Figure 7.36. Fatigue-life data for a carbon/epoxy [0/45/90/45
2
/90/45/0]
s
laminate
under tensiontension loading along the 0-direction [19]. The fatigue-life diagram is
superimposed on the data. Reprinted, with kind permission, from Fatigue of Filamentary
Composite Materials, copyright ASTM International, 100 Barr Harbor Drive, West
Conshohocken, PA 19428.
0
0.005
0.010
0.015
2 4 6
lg N
e
c
e
d.1.
=0.0046
e
m
a
x
Figure 7.35. Fatigue-life data for a glass/epoxy [0/45/90]
s
laminate under tensiontension
loading along the 0
-direction (data from [18]). The fatigue-life diagram is superimposed

on the data.
symmetry axis of the given laminate, the diagram can be constructed by looking
for simple indicators of its characteristic features. If, for example, the laminate has
a 0
-ply along which the cyclic tension load is applied, then region I will exist,
providing a at scatter-band around the failure strain of the composite (which
equals the failure strain of bers). The next thing to look for is the ply orientation
in the laminate that makes the largest acute angle with the loading axis. In the two
examples above, that angle was 90
. The ply with this orientation starts the fatigue

process by initiating cracks along its bers. A good approximation of the fatigue
limit is the axial strain at which this cracking occurs. Methods for estimating this
strain have been discussed in Chapter 6. Once the upper limit (failure strain) and
the lower (fatigue) limit of the laminate have been marked in the fatigue-life
diagram, the progressive fatigue mechanism is captured between those limits.
One can model the progressive mechanism and predict fatigue-life, as we shall
discuss in the next section, or simply get a good approximation to it by drawing a
straight line (or scatter-band) going from the point on the fatigue limit line at
selected number of cycles (10
6
10
7
) to the region line (or scatter-band) at a low
number of cycles. A good guideline for taking the low cycle number is 10
2
for
glass/epoxy and 10
3
for carbon/epoxy.
7.8 Fatigue-life prediction
Fatigue-life prediction of composite materials has suffered from concepts and
methodologies developed for metal fatigue that are not quite relevant. Except in
a few cases, such as delamination growth, the analysis of single crack initiation,
propagation, and unstable growth to failure does not apply to composite fatigue.
Thus, fracture mechanics by itself has little use in composite damage. The associ-
ated methodologies of stress intensity factor threshold, the so-called Paris Law,
residual strength, etc. do not help in assessing fatigue damage tolerance and
durability of composite structures.
The classical (pre-fracture mechanics) approaches to metal fatigue, typied by
fatigue-life evaluation based on empirical S-N curves, are just as lacking of useful-
ness for composite fatigue. The obvious fact that fatigue mechanisms are local,
and therefore dependent on microstructure, should suggest that homogenizing a
composite and describing fatigue in terms of average stresses cannot correctly
describe the driving impetus for composite damage. As argued above in explaining
the construction of the fatigue-life diagram, strain correctly describes the limiting
conditions for fatigue. The remarkable success of the fatigue-life diagram
in evaluating the roles of constituents and interfaces between them, as well as in
clarifying the impact of material properties, should justify moving away from the
pure empiricism of the S-N diagrams. The misconception generated by the plot-
ting of strain in the fatigue-life diagram that it requires strain-controlled testing is
indeed unfortunate. The test data are required to be generated in the completely
traditional way, i.e., in load control, while conversion of the average stress to
265 7.8 Fatigue-life prediction
the average strain is to be done for plotting the data in the fatigue-life diagram.
The result measured in strain (such as strain corresponding to a given number of
cycles) is to be converted back to stress for use in design.
The power of the conceptual framework of the fatigue-life diagram also lies in
guiding mechanisms-based modeling of life prediction. The diagram assures that
the right mechanisms are addressed in modeling. Needless to say, mechanisms-
based modeling is the only rational approach to fatigue in composites. The
alternative of empiricism is unreliable and costly in the long run.
In the following we shall illustrate mechanisms-based fatigue-life modeling for
cross-ply laminates. So far this laminate conguration is the only one treated by
the approach. Its success, as we shall see below, should provide incentive for
treating more general laminates.
7.8.1 Cross-ply laminates
In Section 7.7.2 above, we discussed the fatigue damage mechanisms and con-
struction of the fatigue-life diagram for cross-ply laminates. To model the pro-
gressive damage in region II let us begin by reviewing the quantication of the
transverse cracking mechanism. Figure 7.37 shows the variation of the transverse
crack density (number of cracks per unit axial length) with the number of cycles
(on logarithm scale) for different levels of maximum load. The characteristic
feature of the damage evolution is that the crack density increases exponentially
at rst, followed by the rate of increase decreasing and approaching the saturation
state of zero rate. The level to which the crack density saturates appears to depend
on the load level.
The rst step in the life prediction procedure would be to predict the observed
transverse cracking behavior under fatigue. This requires predicting the crack
density at the rst application of the maximum load and, next, determining
the increase in crack density with the number of cycles having the same load
excursion. Transverse cracking under monotonic loading was treated in Chapter 6.
Here we will focus on crack multiplication in repeated loading.
Figure 7.38 depicts the crack multiplication process. On rst application of the
cyclic load, transverse cracks of a certain spacing form. On next and subsequent
applications of the same load, another crack between the previously formed
cracks would be possible for one of two (assumed) reasons: (1) a aw exists in
the transverse plies that initiates a new crack, which grows and becomes identical
to the previous cracks, and (2) the stress state between the pre-existing cracks
changes with load cycling and becomes favorable at certain cycles to produce a
new crack. In each case we must identify the mechanism of irreversibility from
one load cycle to the next in order to model damage accumulation leading to
formation of a new crack.
In the rst assumed case, we would need some knowledge about the aws,
possibly induced by the manufacturing process. The size and spatial distributions
of the aws are likely to be random, requiring probability distributions for their
description. The initiation of cracks from aws and their propagation with load
cycling need the presence of irreversibility, which is likely to come from the
inelasticity in the matrix (bers are usually elastic) and/or microcracking in the
crack-tip region. In this scenario of transverse crack progression the modeling
effort will involve statistical and numerical simulation.
The second assumed case relies on stress enhancement in transverse plies in the
region between cracks without resorting to aws. In this case, the initiation of new
transverse cracks can be assumed to occur when the axial normal stress in the
transverse plies reaches a critical value. Figure 7.39 shows the distribution of the
three stress components in the longitudinal section of the transverse plies as
0
0
10
20
30
40
50
1 2 3 4 5 6 7
Cycles (Log n)
C
r
a
c
k

d
e
n
s
i
t
y

l
(
i
n
1
)
85%
66%
53%
35%
28%
Figure 7.37. Density of transverse cracks in a [0/90
2
]
s
laminate plotted against the log number
of tensiontension cycles in the axial direction. The data points are for different maximum
load values indicated as percentages of the ultimate tensile strength of the composite [20].
Reprinted, with kind permission, from Springer Science+Business Media: Appl Compos
Mater, Vol. 3, 1996, pp. 391406, X.X. Diao, L. Ye and Y.W. Mai.
Figure 7.38. Transverse cracks produced by the rst application of load (left); generation
of a new crack midway between the pre-existing cracks on repeated application of the load
(middle); and completely grown crack at certain number of load applications (right).
calculated by variational analysis [21, 22]. These stresses result from interaction
between cracks. As seen in Figure 7.39, the axial normal stress attains a maximum
midway between cracks and its value reduces from its pre-crack (constant) value.
Thus, in the absence of aws, a new crack can form midway between cracks if the
maximum stress there exceeds a critical value. However, since this stress is lower
than the critical value at which the previous cracks formed, new cracks can only
form if the applied load is increased. In a cyclic load of constant amplitude, new
cracks are therefore not possible, if the assumption of no aws still holds.
The conclusion has to be that in this scenario of transverse cracks in an elastic
composite, damage progression under cyclic loads cannot be achieved.
If the modeling of damage progression in cross-ply laminates is pursued without
entering aws in the model, then the only plausible place for irreversible mechan-
isms is the interface between the 90
-plies and the 0
-plies. It can be argued that a

transverse crack approaching this interface is bound to cause damage to the
interface in some form or another because of the intense stress eld that accom-
panies the crack front. The most likely damage is cracking in the interfacial plane
(delamination). Figure 7.40 depicts this scenario, showing delamination of length
2l at the crack fronts that are 2s distance apart. Two possible sub-scenarios are
40
30
20
10
10
20
Crack
Crack
30
40
60
4
3
(MPa)
0
9
0
2 1 2 3
4
x
s
xz
(interfacial)
(1)
s
zz
(interfacial)
(1)
s
xx
(1)
z
x
50
1
Figure 7.39. Axial distribution of stresses in the 90
-plies under an axial tensile load

(from [22]). Of interest is the axial normal stress, which is assumed to be constant in the
z-direction. Reprinted, with kind permission, from Damage Mechanics of Composite Materials,
J.A. Nairn and S. Hu, Matrix microcracking, pp. 187243, copyright Elsevier (1994).
now possible, one where the delamination surfaces are traction free, and the other
where frictional sliding between the delamination surfaces can take place. In [23]
stress analysis was conducted for transverse cracks and delamination with a
variational mechanics method to estimate the stresses in the 90
-plies between
the transverse cracks. A cubic variation of the shear stress along the delamination
length 2l was assumed and the calculated axial normal stress then showed an
increase in its maximum value with l. Figure 7.41 shows the axial stress maximum
value for traction-free delamination and when a shear stress acts between the
delamination surfaces.
As seen in Figure 7.41, if delamination surfaces are assumed traction free,
then the right conditions would not be present for new cracks to form between
cracks formed in the rst load cycle. The incentive for crack formation would
indeed decline as the delamination crack propagates under cyclic loading. On
2s
2l
Figure 7.40. Transverse cracks of spacing 2s in a cross-ply laminate with delamination
emanating from crack fronts and extending a distance l on either side of the transverse
crack. The delamination growth is assumed to be caused by the cyclic axial tension applied
to the laminate.
0
0
5
10
15
20
25
30
35
M
a
x
.

s
t
r
e
s
s

i
n

9
0

d
e
g
r
e
e

p
l
y

(
M
P
a
)
0.2 0.4 0.6 0.8 1
l / s
s l l
s
s l l
s
0
0
50
100
150
200
250
300
M
a
x
.

s
t
r
e
s
s

i
n

9
0

d
e
g
r
e
e

p
l
y

(
M
P
a
)
0.2 0.4 0.6 0.8 1
l / s
Figure 7.41. The variation of the maximum axial normal stress between two transverse cracks
is shown when delamination at crack fronts exists. The stress reduces with increase in the
delamination half-length l when the delamination surfaces are traction free (left) and it
increases when a shear stress acts between the delamination surfaces (right).
the contrary, if delamination surfaces are engaged by asperities and/or by
compressive normal stress on the surfaces, then frictional sliding between the
surfaces will result, giving rise to a shear stress. This stress alters the redistri-
bution of stresses between the axial and transverse plies, resulting in the
increase of the maximum axial normal stress as indicated in Figure 7.41. The
number of cycles beyond the rst cycle that will elevate this stress to a critical
value for crack formation is then the cycles needed to increase the delamination
length by fatigue growth.
Based on the assumption of frictional sliding of the delamination surfaces,
Akshantala and Talreja [24] developed a procedure by which the transverse crack
density under cyclic loading could be determined. Figure 7.42 shows examples of
the predicted crack density variation with the number of cycles. Note the crack
density variation displays saturation to different levels depending on the maximum
stress in accordance with the experimental data in Figure 7.37, except the initial
exponential rise in the crack density. Since the stress analysis model is for inter-
acting transverse cracks the initial exponential increase in noninteracting cracks
cannot be predicted by the model.
Having predicted the crack density increase with load cycles, Akshantala
and Talreja [24] proceeded to use these data to predict fatigue-life. The
assumption was made that in the progressive damage represented by region
II of the fatigue-life diagram for cross-ply laminates (see Figure 7.34), a certain
maximum crack density is attained at failure of the laminate. This crack
density denoted
f
lies between the maximum achievable crack density
c
under
static loading and the minimum crack density
fpf
at the initiation of multiple
cracking, commonly called the rst ply failure (a misnomer since the ply cracks
0.0
0.0
0.5
1.0
1.5
2.0
0.2 0.4 0.6 0.8
Number of cycles (10
6
)
C
r
a
c
k

d
e
n
s
i
t
y

(
/
m
m
)
350 MPa
400 MPa
500 MPa
Figure 7.42. Transverse crack density in a cross-ply laminate predicted by Akshantala and
Talreja [24] at different cyclic load levels. Note the tendency for saturation at different levels
depending on the load level. Reprinted, with kind permission, from Mater Sci Eng A,
Vol. 285, A micromechanics-based model for predicting fatigue-life of composite laminates,
but does not fail). The variation of
f
with failure load cycles was assumed to
be as depicted in Figure 7.43. The equation describing this variation,
f
= A
log N
f
+ B, has empirical constants A and B, which are determined by using
the minimum and maximum values of crack densities and their corresponding
number of cycles. The extreme values of the fatigue cycles are the beginning
and end of the progressive damage, i.e., region II. In Figure 7.43, these values
are shown as 10
2
and 10
6
cycles for illustration. As suggested at the end of
Section 7.7, a good approximation of the beginning of region II is 10
2
cycles
for glass/epoxy and 10
3
for carbon/epoxy composites. Note that the assumed
process of fatigue failure is not that it comes from the transverse cracks of a
certain density, but that the crack density increases from its rst-load value to
that crack density. The failure of the laminate must involve delamination
growth and ber failures. The transverse crack density is simply assumed to
scale with fatigue-life in the assumed way.
Figure 7.44 depicts the procedure for fatigue-life prediction using the calcu-
lated crack density from the stress analysis model [23] and the assumed
relationship of the crack density to failure cycles as displayed in Figure 7.43.
At a given cyclic load, the fatigue-life is given by the minimum number of
cycles until the crack density equals the limiting value. Thus, the graphical
method for determining this value is to nd the point of intersection of the
crack density increase curve and the straight line describing the limiting
crack density (see Figure 7.44). The number of cycles corresponding to the
intersection point is then the fatigue-life at that load level. The fatigue-life
prediction thus obtained is compared with the test data for a cross-ply lamin-
ate in Figure 7.45.
Cycles to failure (log N
f
)
10
2
10
6
C
r
a
c
k

d
e
n
s
i
t
y

(
h
f
)
h
c
h
fpf
h
f
=A logN
f
+B
A
1
e
e
c
e
fpf
Figure 7.43. The assumed variation of the transverse crack density at failure in fatigue plotted
against the fatigue-life of a cross-ply laminate. The upper limit to this crack density is the
maximum saturation crack density under static load and the lower limit is the minimum
crack density at initiation of multiple cracking. The composite strains corresponding to the
two limits are also indicated. Reprinted, with kind permission, from Mater Sci Eng A,
Vol. 285, A micromechanics-based model for predicting fatigue-life of composite laminates,
10
0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
10
1
10
2
10
3
10
4
10
5
10
6
C
r
a
c
k

d
e
n
s
i
t
y

(
/
m
m
)
0.6%
0.65%
0.7%
0.76%
0.9%
1.0%
1.2%
h
fpf
h
c
h
f
= 0.425 N
f
+3.35
Number of cycles
Figure 7.44. The crack density data points represent the initial (rst load) values and their
increase with cyclic loading at different load levels (indicated by the corresponding rst-
cycle peak strains). The crack density relationship to failure cycles, schematically described
in Figure 7.43, is also shown with the calculated end-values of crack density. Fatigue-life
at a given load level is given by the cycles at intersection of the calculated crack density
with its failure value.
Cycles to failure
10
0
10
1
10
2
10
3
10
4
10
5
10
7
10
6
0.0
0.5
1.0
1.5
2.0
2.5
M
a
x
i
m
u
m

a
p
p
l
i
e
d

s
t
r
a
i
n

(
%
)
Fatigue limit
Experiment data
Model
Region II
Region 1
Figure 7.45. The predicted fatigue-life of a glass/epoxy [0/90]
s
laminate compared with test
data from [25].
7.8.2 General laminates
In Chapter 6 we treated the damage progression of a broad class of laminates
consisting of plies in multiple orientations subjected to axial tensile loads. From
the models described there one can calculate the load at initiation of multiple
cracking and crack densities resulting at the rst application of the maximum load
in the given cyclic load. For further progression of damage under repeated
application of that load one must ask: what is the mechanism responsible for
irreversibility that accumulates damage with each load cycle? For cross-ply lamin-
ates we addressed this question and described the model that incorporates the
answer to the question. There is every reason to believe that in laminates with
multiple off-axis ply orientations the delamination occurring at the off-axis crack
fronts conducts frictional sliding and thereby provides the needed irreversibility
(energy dissipation). For cross-ply laminates the stress analysis conducted in [21]
for transverse cracks with delamination cannot simply be extended to multiple
crack orientations. In fact even without delamination the stress analysis of cross-
ply laminates is not easily extended to the multiple crack orientation case.
7.9 Summary
This chapter has provided a systematic conceptual framework for interpretation
of the fatigue process in composite materials. No effort has been made to give
a comprehensive exposition of the vast literature on the subject. Instead, emphasis
has been placed on understanding of the physical mechanisms underlying fatigue
and incorporating this in a systematic way in the representation called fatigue-
life diagrams. These diagrams provide a healthy departure from the empiricism
dominant in the fatigue literature. They also generate bases for material selection
and give useful guidelines for mechanisms-based modeling.
The eld of fatigue damage modeling and life prediction needs a great deal of
further work. The challenges of addressing the mechanisms of fatigue damage
accumulation have discouraged most research efforts that have instead taken the
path of empirical approach. In fact even in the century old eld of metal fatigue
empirical approaches are common. The so-called Paris Law of cyclic crack growth
is basically a curve t to the observed data. However, its simplicity has attracted
practical use but has impeded advances in fundamental understanding. It is
therefore not surprising that the composites community has so far also opted
for empiricism. It is worth noting that because of the large number of parameters
in composite materials (constituent properties, ply orientations, woven and other
complex architectures, etc.) the empirical path is highly inefcient. As demon-
strated by the fatigue-life diagrams, focusing on the essential mechanisms allows
one to transcend the apparent complexity of the composite microstructure and
ber architecture.
273 7.9 Summary
In the next chapter we shall outline the areas in composite fatigue that must
receive attention by the composites community and must be supported by govern-
ment and industry.
References
1. R. Talreja, Fatigue of composite materials: damage mechanisms and fatigue-life
diagrams, Proc R Soc London A, 378 (1981), 46175.
2. C. K. H. Dharan, Fatigue failure mechanisms in a unidirectionally reinforced compos-
ite material. In Fatigue in Composite Materials, ASTM STP 569. (Philadelphia, PA:
ASTM, 1975), pp. 17188.
3. J. B. Sturgeon, Fatigue and creep testing of unidirectional carbon ber
reinforced plastics. In Proceedings of the 28th Annual Technical Conference of the
Society of the Plastics Industry. (Washington, DC: Reinforced Plastics Division, 1973),
pp. 1213.
4. J. Awerbuch and H. T. Hahn, Fatigue and proof-testing of unidirectional graphite/
epoxy composite. In Fatigue of Filamentary Composite Materials, ASTM STP 636, ed.
K. L. Reifsnider and K. L. Lauraitis. (Philadelphia, PA: ASTM, 1977), pp. 24866.
5. J. B. Sturgeon, Fatigue Testing of Carbon Fibre Reinforced Plastics, Technical Report,
Royal Aircraft Establishment, Farnborough (1975).
6. R. B. Croman, Tensile fatigue performance of thermoplastic resin composites
reinforced with ordered Kevlar aramid staple. In Proceedings of the Seventh Inter-
national Conference on Composite Materials, vol. 2, ed., Y. Wu, Z. Gu, and R. Wu.
(Oxford: International Academic Publishers, 1989), pp. 5727.
7. E. K. Gamstedt and R. Talreja, Fatigue damage mechanisms in unidirectional carbon
bre-reinforced plastics. J Mater Sci, 34 (1999), 253546.
8. E. K. Gamstedt and S. O
stlund, Fatigue propagation in bre-bridged cracks in unidir-

ectional polymer-matrix composites. Appl Compos Mater, 8 (2001), 385410.
9. E. K. Gamstedt, Effects of debonding and ber strength distribution on fatigue-
damage propagation in carbon bre-reinforced epoxy. J Appl Polym Sci, 76 (2000),
45774.
10. R. Talreja, A conceptual framework for interpretation of MMC fatigue. Mater Sci Eng
A, 200 (1995), 218.
11. B. F. Srensen and R. Talreja, Analysis of damage in a ceramic matrix composite.
Int J Damage Mech, 2 (1993), 24671.
12. B. F. Srensen, J. W. Holmes, and E. L. Vanswijgenhoven, Does a true fatigue limit
exist for continuous ber-reinforced ceramic matrix composites? J Amer Chem Soc, 85
(2002), 35965.
13. Z. Hashin and A. Rotem, A fatigue failure criterion for ber reinforced materials.
J Compos Mater, 7 (1973), 44864.
14. A. Rotemand Z. Hashin, Fatigue failure of angle ply laminates. AIAAJ, 14 (1976), 86872.
PA: ASTM, 1984), pp. 2155.
16. G. C. Grimes, Structural design signicance of tensiontension fatigue data on com-
posites. In Composite Materials: Testing and Design (Proc. 4th Conference), ASTM STP
617. (Philadelphia, PA: ASTM, 1977), pp. 10619.
J Compos Mater, 19 (1985), 35575.
18. H. T. Hahn and Y. Kim, Fatigue behavior of composite laminate. J Compos Mater,
10 (1976), 15680.
19. J. T. Ryder and E. K. Walker, Effect of compression on fatigue properties of a quasi-
isotropic graphite/epoxy system. In Fatigue of Filamentary Composite Materials, ASTM
STP 636. (Philadelphia, PA: ASTM, 1977), pp. 326.
20. X. X. Diao, L. Ye, and Y. W. Mai, Simulation of fatigue performance of cross-ply
composite laminates. Appl Compos Mater, 3 (1996), 391406.
21. Z. Hashin, Analysis of cracked laminates: a variational approach. Mech Mater, 4,
(1985), 12136.
22. J. A. Nairn and S. Hu, Matrix microcracking. In Damage Mechanics of Composite
Materials, ed. R. Talreja. (Amsterdam: Elsevier Science, 1994), pp. 187243.
23. N. V. Akshantala and R. Talreja, A mechanistic model for fatigue damage evolution in
composite laminates. Mech Mater, 29 (1998), 12340.
24. N. V. Akshantala and R. Talreja, A micromechanics based model for predicting
fatigue-life of composite laminates. Mater Sci Eng A, 285 (2000), 30313.
25. C. J. Jones, R. F. Dickson, T. Adam, H. Reiter, and B. Harris, The environmental
fatigue behaviour of reinforced plastic. Proc R Soc London A, 396 (1984), 31538.
275 References
8 Future directions
In Chapter 1 we discussed the durability assessment of composite structures, the
overall goal for the subject of this book. As outlined there in Figure 1.1, the
mechanisms of damage and their effects on deformational response constitute
the main thrust of the eld of damage mechanics, which is at the core of durability
assessment. After discussing the physical nature of damage observed experimen-
tally in Chapter 3, the next two chapters treated the two main approaches in
damage mechanics micro-damage mechanics (MIDM) and macro-damage
mechanics (MADM), both aimed at predicting deformational response at xed
damage. Damage evolution was treated in Chapter 6, while Chapter 7 was devoted
to fatigue, a subject that requires special attention due to the conceptual difcul-
ties it poses.
In closing the book we wish in this chapter to review what has been
achieved and what directions the eld of damage and failure of composite
materials should pursue to further advance toward durability assessment and
beyond.
8.1 Computational structural analysis
Obviously, complex structural geometries require computational structural
analysis. The analytical modeling of damage initiation and evolution, and its
effects on deformational response of composite laminates, discussed in previous
chapters, were developed for idealized simple cases. Direct application of these
models is limited to structures with simple geometry and loading conditions. For
complex geometries, such as an airplane wing or a wind turbine blade, usually
subjected to multi-axial mechanical loads, and possibly combined with thermal
and moisture environments as well as manufacturing-induced residual stresses,
computational approaches are inevitable. In industry, one often uses commercial
software, e.g., ANSYS, ABAQUS, and NASTRAN, and the obvious need is to
integrate damage and failure analyses into these codes. Efforts have been made
to attempt some simple test cases where FE analysis of composites is combined
with damage using failure criteria [1]. A series of World Wide Failure Exercises
(WWFE) [24] have been conducted to compare several composite failure models
with experimental data and provide guidance for their usage in composite design.
While previous exercises focused on the ultimate failure, the ongoing WWFE-III
offers opportunities to examine the initiation and progression of sub-critical events
and their effect on the mechanical response. Yet no comprehensive code exists
which can combine stress analysis with damage prediction. The current codes
predict structural failure based on lamina-based failure criteria and not on the basis
of sub-critical damage such as ply cracking. To improve this situation, the authors
are currently developing a user subroutine (UMAT) in ABAQUS with an ability to
dynamically update the stiffness properties of a region using the energy-based
fracture criterion described in Chapter 6. The aim is to predict the initiation of
ply cracking and its evolution, as well as the resultant stiffness properties, using the
synergistic damage mechanics approach described in Chapter 5.
While Figure 1.1 outlined the overall methodology for durability assessment,
Figure 8.1 here illustrates the procedure for performing computational stress
analysis of a composite structure in combination with damage analysis. At rst,
stress analysis is conducted assuming no damage is present in the whole structure.
At this point the necessary data input includes the geometry of the structure, the
conguration (layup, thicknesses of plies, etc.) and material of the composite
PRELIMINARY STRESS ANALYSIS
INPUT
Geometry, laminate configuration, material
properties and loading conditions
Create geometric
model
Meshing
Apply loading &
service conditions
Stress analysis
DAMAGE ANALYSIS
Identify regions where
damage might have
developed
Predict damage
initiation &
evolution
Evaluate stiffness
properties of
damaged regions
UPDATED STRESS ANALYSIS
OUTPUT
Failure characteristics, stress-strain response,
deformation behavior, life and durability
Update stiffness of damaged regions Perform stress analysis of whole structure again
Figure 8.1. Flowchart illustrating computational stress analysis of a composite structure in
combination with damage analysis.
277 8.1 Computational structural analysis
laminate, and the loading and service conditions. Then, based on the stress
pattern, regions are identied which are potential sites for damage under the
prescribed loading and service conditions. These regions include any stress con-
centrators and could be based on previous design experience. In this step, possible
damage mechanisms are also identied. For instance, the region close to a hole or
cut-out in the structure is expected to develop damage in the form of ply cracks
and delamination. Next, using the preliminary stress states, damage initiation and
its evolution are predicted with the methods and analyses treated in earlier
chapters. The damage prediction step may involve a substructural analysis and
be carried out over multiple length scales. The multiscale modeling aspect will be
discussed later in the present chapter. The stiffness properties of damaged regions
are then updated using the analytical predictions, and a new stress analysis is
performed at the same applied load levels. On increased loading, the same iterative
process is repeated until the analysis predicts failure based on the design criteria.
Depending on the purpose and need, the output of the computational structural
analysis may include detailed failure characteristics, overall stressstrain response,
deformation characteristics (deections, vibrational frequencies, etc.), as well as
damage tolerance and life prediction of the structure.
The main challenge in any computational scheme is to properly incorporate
damage and its effects. It is not uncommon to nd works where damage with all its
complexities is reduced to a parameter D, which is then arbitrarily assigned a value
between 0 and 1, often based on computational convenience and seldom account-
ing for the physics behind damage. Much progress is still needed in taking the
theoretical advances in damage mechanics into computational design procedures.
One of the main hurdles is properly conducting multiscale computational analysis
of damage. In the following we shall address issues in multiscale modeling of
damage, which present some fundamental challenges arising in the computational
treatments of composite failure.
8.2 Multiscale modeling of damage
The computational power available today has given impetus to the so-called multi-
scale modeling, which follows an intuitive notion that the physical phenomena
occurring at lower length scales determine the material response at higher scales.
Fiber-reinforced composites and, more generally, heterogeneous solids appear to
be natural candidates for this modeling idea. Thus, numerous studies have resulted
to address deformation, damage, and failure of these materials and to some extent
also their nonmechanical properties. Avast literature on multiscale modeling exists,
and a recent collection of works [5] covers a wide range of such treatments.
Since in a heterogeneous solid the microstructural entities (heterogeneities)
are embedded in the matrix, a given property (e.g., mechanical or thermal
response characteristic) must be dened at a scale much larger than the charac-
teristic size of the heterogeneities. This leads to the suggestion that, at a chosen
278 Future directions
scale, a representative volume element (RVE) exists such that by appropriate
averaging over this volume the property of interest can be computed. The RVE
idea is a necessary fundamental concept in any proper multiscale treatment. The
reader is referred to texts in micromechanics such as that by Nemat-Nasser and
Hori [6] for in-depth treatment of the RVE concept and the associated aver-
aging schemes.
Denoting the RVE scale as mesoscale, and the scale of microstructural entities
as microscale, the three-level micro-meso-macro hierarchy forms the basis of a
multiscale modeling treatment. In fact if the microstructure in a heterogeneous
material contains nano-scale reinforcements (e.g., particles, bers or tubes), then
further differentiation of the lower-end scale to nano- and micro-levels can be
made. While the hierarchical multiscale modeling is a viable approach for micro-
structures that are stationary (Figure 8.3), i.e., they do not change their charac-
teristic size under a loading impulse, it is not a given that the approach will remain
valid when energy dissipative mechanisms permanently alter the microstructure.
Our concern here is multiscale analysis of the energy dissipative mechanisms
underlying damage in composite materials for the purpose of assessing structural
integrity and durability, and for this case further examination of the multiscale
approach seems warranted.
In the context of computational structural analysis described above in
Section 8.1 and in a previous review of this subject [7] we refer to Figure 8.2,
Figure 8.2. Durability analysis procedure for composite structures.
u
1 u
2
t
t t
u
Figure 8.3. Heterogeneous solid in pristine state (left) and two damage states, type 1
(middle) and type 2 (right).
279 8.2 Multiscale modeling of damage
which was also discussed in Chapter 1. Here we draw attention to the damage
mechanics part of the scheme. As indicated, micro-meso-macro multiscale model-
ing is at the core of this analysis. The three-scale differentiation when damage
exists does not necessarily coincide with the multiscale differentiation adopted
before damage. This critical point was treated in [8], from which the following
discussion is drawn.
8.2.1 Length scales of damage
Let us consider three RVEs depicted in Figure 8.3 to discuss characteristic scales
of damage and their relation to the scales of microstructure. The RVE to the left in
the gure is for pristine (undamaged) composite. When the bounding surface of
this RVE is subjected to a prescribed traction t, the combined deformation of the
matrix and the heterogeneities (shown symbolically as lled circles) contained
within the RVE produces a displacement u of points on the bounding surface.
This displacement eld can be written as u = u
0
+ du
0
, where u
0
represents the
deformation of the matrix in the absence of bers and du
0
is the perturbation in it
caused by the heterogeneities in the absence of damage. The middle RVE in
Figure 8.3 illustrates a damage scenario let us call it type 1 damage where
some of the heterogeneities (e.g., inclusions) have partially or fully separated from
the matrix. The interfacial cracks, thus formed, then perturb further the deform-
ation eld within the RVE if the traction t on the RVE surface is sufcient to
activate the cracks (i.e., displace the crack surfaces). The displacement of points on
the bounding surface is now u
1
. Finally, the RVE to the right in the gure depicts a
damage scenario type 2 damage where the cracks within the RVE are restricted
to the matrix and are geometrically unconnected to the heterogeneities. The
displacement response to surface traction t on the RVE in this case is
denoted u
2
. The perturbation eld du
0
can, in principle, be determined by know-
ledge of the properties of matrix and heterogeneities, as well as conguration
variables such as size, shape, and spacing of heterogeneities. Depending on the
model employed, one may use limited conguration information such as volume
fraction of heterogeneities, or more enriched information such as statistical cor-
relation functions that describe their relative size and placement. A thorough
treatment of the morphological characterization of microstructures is found
in [9], while [10] gives an extensive review of models that aim at carrying the
morphological information to RVE level averages.
The models in general have multi-level features and often rely on homogeniza-
tion concepts. In the absence of damage, the models may be characterized as
stationary microstructure models, i.e., the microstructure conguration remains
unchanged under application of RVE surface traction t. In calculating the RVE
surface displacement response to the prescribed traction t in the presence of
damage, one can take two alternative approaches. One approach is to homogenize
the matrix and stationary microstructure and embed the damage entities (e.g.,
cracks) in the homogenized composite. The displacement response can then be
written as u
i
= u + d
i
u, where d
i
u results from the perturbation in the displace-
ment eld of the homogenized composite caused by damage (i = 1 or 2 for
damage type 1 or type 2, respectively). This approach is common and has many
versions, a familiar one being the MoriTanaka estimation procedure [11]. It is
noted that the MoriTanaka procedure is a homogenization method for station-
ary microstructures, i.e., for undamaged composites. For further reference, we
shall call it the homogenized microstructure (HM) approach. The other approach
is to retain the discrete nature of the stationary microstructure in estimating the
RVE displacement response. In this approach, called here the discrete microstruc-
ture (DM) approach, the displacement response can be written as u
i
= u
0
+ d
i
u
0
,
where d
i
u
0
results from combined perturbation in the displacement response of the
matrix caused by the stationary microstructure and damage, the subscript index i
referring still to the damage type.
The two approaches to estimating the RVE surface displacement, just
described, are approximate and will generally yield different results. In the HM
approach, the explicit association of a damage entity to the microstructure is lost,
since the microstructure in which the damage entity resides has been homogenized,
while in the DM approach the heterogeneities as well as damage entities are
explicitly present. Thus, in the HM approach, damage entities of type 1 and type 2
are both surrounded by the homogenized microstructure, thereby their association
with the microstructure (as to how the microstructure affects their initiation) is
lost. In this sense, the HM approach is insensitive to which of the two types of
damage is treated. On the other hand, in conducting the DM approach a signi-
cant difference exists depending on whether damage of type 1 or type 2 is
considered. Since the type 1 damage is geometrically tied to the stationary micro-
structure, the perturbation caused by it in the local elds can be analyzed by
viewing it as a modication to the perturbation induced by the heterogeneities.
This would not be the case for type 2 damage, as it is unconnected to the
heterogeneities but affected by them. In fact, for this reason, the HM approach
would be preferable for type 2 damage.
The considerations described above have been made for a specic purpose: to
clarify the characteristic scales associated with the micro-level (heterogeneities)
and their relation to the meso-level (RVE) scale. As noted, the micromeso
bridging in the absence of damage is clear and unambiguous. In fact, if the matrix
has a heterogeneous structure itself, then knowing the characteristic scales of that
substructure it would be possible to homogenize it with a multiscale (sub-micro to
micro) approach. Thus, in general for stationary heterogeneities, the multiscale
modeling is plausible and systematic, at least conceptually. This is far from the
case when damage in the form of distributed internal surfaces exists. When these
internal surfaces are geometrically connected to the microstructure, such as in
type 1 damage, their characteristic micro-level scales can be deduced from those of
the heterogeneities and the micromeso bridging is then relatively simple. How-
ever, few cases of damage in composite materials belong to this type, i.e., where
the damage entities remain connected to the heterogeneities. Although this might
occur in the initial stages of damage, the damage entities formed grow away from
the heterogeneities. In the case of type 2 damage, where the damage entities have
scales unrelated to those of the heterogeneities, the resulting complexities of scales
question the viability of the hierarchical multiscale modeling, as discussed in more
detail in the following.
8.2.2 Hierarchical multiscale modeling
We shall now focus on the specic case of composite laminates in which individual
plies are of a matrix material reinforced with unidirectional bers. Thus the
stationary microstructure here consists of bers distributed in the matrix and
corresponds to the picture on the left in Figure 8.3 for a heterogeneous solid
in pristine state. The microstructural length scale, or micro-scale, is the size of a ber
(radius or diameter). The next scale in the hierarchy, the meso-scale, is the RVEsize,
which depends on the distribution of bers. Assuming all bers are straight and
parallel, then for the case of uniformly distributed bers in the cross section a
repeating unit of a ber embedded in a surrounding matrix, i.e., a unit cell, replaces
RVE. For nonuniform distribution of bers, reference is made to treatments in [9]
and [10]. Simpler estimates for ply properties can be found in any of the common
texts on composite mechanics, e.g., [11]. Finally, the macro-scale for a composite
laminate is a structural scale that depends on the structural geometry.
When a composite laminate suffers damage under loading, a variety of length
scales develop that may or may not be connected with the initial length scale of
undamaged composite. Although this issue has been discussed above in general
terms, we will treat it now specically for composite laminates.
8.2.2.1 Damage in unidirectional composites: transverse loading case
Figure 8.4 (left) shows a typical view of damage observed in the cross section of a
unidirectional ber composite of a polymeric matrix loaded in tension normal to
Debonding induces matrix cracking
Matrix cracking causes debonding
(b)
(a)
Figure 8.4. A typically observed micrograph (left) of a cross section of a unidirectional ber
composite loaded in tension normal to bers. To the right are depicted two plausible
scenarios underlying the picture on the left.
bers. The damage appears to be a mix of ber/matrix debonding (type 1 damage)
and matrix cracking (type 2 damage) described above. However, a closer study of
ber/matrix debonding [1214] suggested that this failure mechanism could be a
consequence of cavitation-induced brittle cracking in the matrix. Type 2 damage,
on the other hand, is matrix ow-induced ductile cracking in this case. Thus, in
polymer matrix composites, damage of type 1 and type 2 could both be different
realizations of matrix failure with drastically different governing scales, as dis-
cussed below.
In Figure 8.4 (right) are shown two plausible scenarios of cracking that can
underlie the observed damage in Figure 8.4 (left). The rst one, marked (a), is for
ber/matrix debonding that progresses out of the ber surface into the matrix.
Such matrix cracks then coalesce forming a continuous ber-to-ber crack. The
second scenario assumes formation of matrix cracks rst, which on growing
towards bers induce ber/matrix debonding. This sequence of cracking would
also produce a continuous ber-to-ber crack. Which of the two scenarios holds
depends on the transverse loading-induced local stress states, which in turn
depend on the microstructure conguration, i.e., ber volume fraction and distri-
butions of ber diameter, inter-ber spacing, etc. A discussion of the effect of local
stress states on matrix damage follows.
A cross-sectional region of a transversely loaded composite is illustrated in
Figure 8.5. For nonuniformly distributed bers in the cross section, stress analysis
studies conducted in conjunction with the work reported in [12] suggested that for
points in the matrix close to the ber surface, such as that indicated in the gure,
the deformation of the matrix is nearly or fully dilatational, while for points in the
matrix away from bers, such as the other point indicated in the gure, the matrix
deformation has a signicant distortional component. The mix of dilatational and
distortional deformation will depend on the constraint to deformation imposed by
the presence of bers. For instance, for points at nearly-touching ber surfaces
and in matrix regions that are squeezed between three bers, the deformation will
approach the dilatational state, while at points in the matrix sufciently away
from bers such that the local stress perturbation induced by bers is negligible,
the deformation state will have a high degree of distortion. For points near a ber
surface where conditions for dilatational deformation are favorable, the same
studies [12, 13] proposed that cavitation of the polymer matrix is induced by the
hydrostatic tension (the insert in Figure 8.5(b)). The cavities thus formed expand
stably at rst and become unstable when the dilatation energy reaches a critical
value. The unstable cavity growth in a region with lack of sufcient distortional
energy results in brittle cracking, which nds its way into the ber/matrix inter-
face. The consequent debonding grows as an arc-shaped crack along the ber
surface (Figure 8.5(b)) and subsequently diverts into the matrix along a direction
normal to the local maximum tensile stress (Figure 8.4(a)). Further progression of
damage produces the view seen in Figure 8.4 (left).
An alternative damage scenario is illustrated in Figure 8.5(c), where a point
away from bers undergoes signicant distortional deformation, which eventually
localizes in shear-intensive bands, leading to ductile cracking. As the crack
growth advances to brittle regions near bers, ber/matrix debonding occurs
(Figure 8.4(b)). As in the other damage scenario, further progression of damage
can produce the same view seen in Figure 8.4 (left).
Modeling of the two damage scenarios depicted above involves analyses at two
different sets of characteristic scales. For the cavitation-induced brittle cracking
resulting in ber/matrix debonding (the rst of the damage scenarios, Figure 8.5
(b)), the characteristic scale at which cavitation begins is the molecular scale of the
polymer matrix. Once debonding occurs, the scale of damage is the diameter of the
ber on whose surface the arc-shaped debond crack grows. For the second
damage scenario (Figure 8.5(c)), the sequence of deformation and failure mechan-
isms can be long and complex depending on the polymer morphology and the
stress triaxiality (dened in an appropriate way to describe the mix of deviatoric
and hydrostatic stress). These mechanisms can be classied as brittle, quasi-brittle,
or ductile to characterize the relative degree of material ow involved in the
cracking process. The characteristic scales of damage will then vary accordingly.
A large body of literature exists on models that address the deformation and
(a)
s
Dilatational
Distortional
(c)
Distortional
s s
s
s
s
(b)
Figure 8.5. (a) Points in a transversely loaded heterogeneous solid with dilatational and
distortional deformation; (b) cavitation and subsequent cracking induced by hydrostatic
stress state; and (c) cracking in matrix caused by distortional ow.
failure of polymers ranging from molecular level treatments to continuum mech-
anics analyses incorporating the details of the mechanisms in various explicit and
implicit ways. Review of this literature is not the purpose here, but in the context
of the multiscale damage in heterogeneous solids it is important to note that the
presence of heterogeneities alters the nature of the problem. While for an unre-
inforced polymer a multiscale modeling effort must address all scales of morph-
ology that are activated by the loading, in the case of a polymer composite,
activities are enhanced at certain scales and subdued at others, depending on the
local stress perturbation caused by the heterogeneities. Thus, the hierarchy of
scales and their relative roles are generally different in unreinforced and in
heterogeneous polymers.
Although the discussion here has been focused on polymers as matrix materials,
the inferences drawn will largely apply to reinforced ductile metals as well.
8.2.2.2 Cross-ply laminate: transverse ply cracking
Transverse ply cracking in a cross-ply laminate is an extensively studied area and
has been reviewed in previous chapters; here the focus of discussion will be the
characteristic scales of damage. When a cross-ply laminate is loaded in tension
along the longitudinal plies, the rst event of damage is the transverse crack
formation in one of the two ways discussed above. The crack grows across bers
rst, and then along bers, eventually spanning the thickness and width of the
transverse ply. Up to this point, the scales associated with the damage are as
discussed above for transverse loading-induced damage in unidirectional compos-
ites. On encountering the ply interfaces the transverse crack fronts bring about
interfacial stress perturbation, which traverses a certain distance along the longi-
tudinal plies, whence the stresses return to the pre-cracking state, unless a perturb-
ation by another crack intervenes, in which case another equilibrium stress state
results. This so-called shear lag distance is the distance on either side of the
transverse crack where reduction of the axial stress in the transverse ply prevents
another transverse crack from forming, and increased loading is needed to pro-
duce such a crack. This phenomenon of stress transfer from cracked transverse
plies to the longitudinal plies is responsible for multiple cracking in the transverse
plies. The conditions for single crack formation versus multiple cracking were rst
explained in the landmark paper now commonly known as the ACK theory [15].
From the viewpoint of the characteristic scales of damage, which is what is of
concern here, the situation changes drastically when damage evolves from the
phase of single transverse crack formation to multiple transverse cracking. The
scales associated with single crack formation were discussed above. In multiple
cracking the governing scale is that of the shear lag distance along the 0/90 ply
interface. This distance is determined by the ply properties and the thickness of
plies in the two orthogonal directions. In other words, it is the homogenized ply
properties and laminate conguration that determine the scale of multiple
cracking, while it was the ply constituent properties and reinforcement morph-
ology (ber volume fraction, ber diameter, inter-ber spacing, etc.) that governed
formation of a single transverse crack. This fact does not seem to be fully
appreciated by most multiscale modeling efforts, which tend to treat transverse
cracking with the scales of heterogeneities (ber size and spacing, and associated
distributions).
8.2.3 Implication on multiscale modeling: Synergistic damage mechanics
The two cases discussed above are intended to illustrate the complexity and
richness of the damage phenomena in composite materials. More cases have been
treated elsewhere [7, 8, 16, 17]. It seems clear from these studies that a hierarchical
multiscale approach for undamaged composites cannot generally be extended to
cover all cases of damage. This may suggest that multiscale modeling for the
purpose of structural integrity and durability should be approached on its own
rather than tying it to the hierarchical approach for undamaged heterogeneous
solids. Efforts in this direction were proposed and labeled as synergistic damage
mechanics (SDM) [18]. Since the publication of that work, a systematic demon-
stration of the viability of the approach has been made [1923].
In Section 5.2 the SDM approach was described in detail. Here, for the sake of
completeness of the multiscale modeling discussion we include two gures from
Chapter 5. Figure 8.6 (same as Figure 5.4) shows the two-step homogenization
procedure involved in the characterization of damage. As depicted in the gure,
the stationary microstructure is homogenized rst and represented by appropriate
constitutive relations. The evolving microstructure, consisting of damage entities,
Evolving microstructure
RVE for damage characterization
Step 2
V
a
i
n
j
Fully homogenized continuum
Characterization of a
damage entity
P
P
Step 1
Homogenization of
evolving microstructure
Homogenization of
stationary microstructure
Continuum after homogenizing
the stationary microstructures
Figure 8.6. Two-step homogenization of a composite body with damage is depicted. The
characterization of damage and the associated RVE are also illustrated.
is homogenized next and regarded as an internal structure embedded in the
homogenized solid of stationary microstructure. The new continuum is now
represented by a thermodynamics framework in which internal state variables
characterize damage by a set of second-order tensors. The internal variables are
dened by descriptors that require averaging over a RVE. The RVE size is then
the meso scale, while the characteristic size of the damage entities is the micro
scale. The size-scales of the stationary microstructure enter separately in the step 1
homogenization.
In this scheme the micro-scale level is the single damage entity size and the
meso-scale level is the RVE size, as described above. The micro-level descriptor
is the damage entity tensor d
ij
described in Chapter 5, while its average over the
RVE, denoted D
a
ij
for a selected damage mode a is the meso-scale descriptor,
also described in Chapter 5. Because of the way the damage entity tensor is
constructed, it is possible to explicitly incorporate micro-level information into
this descriptor. In this way the inuence of the surrounding heterogeneous
solid, i.e., the microstructure, on the damage entity can be analyzed by a
convenient means and thereby transmitted to the meso level. A parameter,
called the constraint parameter, has been devised to effectively accomplish
this task, either experimentally or by computational micromechanics. This has
all been described in detail in Chapter 5 for single and multiple modes of
damage. Figure 5.10 from Chapter 5 is reproduced here as Figure 8.7 for a
convenient recollection of the SDM procedure.
The multiscale modeling of damage and the accompanying SDM methodology
are the appropriate way to treat the effect of damage on the composite material
response and by extension to assess structural durability. The hierarchical multi-
scale treatment of heterogeneous solids, although suited for estimating their
overall response, cannot generally be extended to account for damage.
8.3 Cost-effective manufacturing and defect damage mechanics
A composite structure, constructed by any practical manufacturing process, is
rarely perfect. The defects induced during manufacturing can be in the ber
architecture, e.g., ber misalignment, irregular ber distribution in the cross
section, and broken bers; in the matrix, e.g., voids; and in the interfacial regions,
e.g., debonding and delaminations. Such defects must be analyzed to determine
their effects on the composite structural integrity and durability. The results of
such analyses can be used in two ways: (a) to develop acceptance/rejection criteria
for the manufactured part; and (b) to design the part to meet performance
requirements accounting for the defects. The former is what industry largely
practices today, if at all. A host of nondestructive inspection (NDI) techniques
are potentially possible to detect defects. These techniques can be applied to
implement thresholds for product quality, e.g., a maximum void volume fraction
or a maximum delamination surface area.
287 8.3 Cost-effective manufacturing and defect damage mechanics
The eld of designing a composite part with known defects is far from mature.
What need to be developed are accurate analyses of effects of real-life defects as
well as a strategy for incorporating the results in a cost-effectiveness assessment of
the manufacturing process. As is the case, most of the cost of a part lies in the
manufacturing process. In the following, we shall rst review a cost-effectiveness
assessment procedure, clarifying the different elements that make up the proced-
ure. We shall then focus on the mechanics of materials approaches for analysis of
real-life defects. For illustration we shall review some recent results on (i) elastic
property changes due to matrix voids, and (ii) effects on the propensity for crack
extension induced by interlaminar voids. In closing we shall make recommenda-
tion for future work.
8.3.1 Cost-effective manufacturing
Figure 8.8 describes the interrelated elements involved in the process of assessing
the cost-effectiveness of a composite structure with respect to its long-term per-
formance. To begin, the manufacturing process selected for a given composite
structure is described by material and geometry parameters, processing param-
eters and their time variations, as well as descriptors of machining, tooling,
joining, assembly, etc., as needed. The product resulting from the manufacturing
process is characterized in terms of the material state and its corresponding
COMPUTATIONAL MICROMECHANICS
SYNERGISTIC DAMAGE MECHANICS
STRUCTURAL ANALYSIS
Analyze overall structural response to external
loading using the reduced stiffness properties
Use SDM to determine stiffness reduction in
present laminate configuration [0
m
/ q
n
/0
m/2
]
s
Structural scale: Meso
Structural scale: Macro
Structural scale: Micro
EXPERIMENTAL/
COMPUTATIONAL
Determine COD and constraint parameter(s)
; = = +
(
( )
(
(
)
u
2
u
2
u
2
n
=

q
90
q
n q
n +q
n
q
b
Evaluate damage constants
using available data
for reference laminate
configuration [0/90
8
/0
1/2
]
s
90
8
( )
u
2 ( )
u
2
Figure 8.7. Flowchart showing the multiscale synergistic methodology for analyzing
damage behavior in a class of symmetric laminates with layup [0
m
/y
n
/0
m/2
]
s
containing ply
cracks in the +y and y layers.
properties. The conventional material state description consists of constituent
properties and their relative proportions, e.g., volume fractions, and of the ber
architecture, e.g., ply thickness, orientation, and stacking sequence in a laminate,
or fabric type (e.g., 5- or 8-harness satin), thickness, and layup in a woven fabric
composite, etc. In addition to this, the material state needs to be described by
certain defect descriptors. As we shall see, homogenized descriptions of the
constituents and defects do not sufce for the cost-effectiveness assessment of a
manufacturing process.
The defect descriptors needed would depend on the manufacturing process.
Examples of such descriptors are distributions (or other statistics) of ber mis-
alignment angles, of void size and location, of ber/matrix interfacial disbonds, of
delamination size and location, etc. The appropriate set of defect descriptors,
along with the conventional material state descriptors, makes up a complete
characterization of the manufactured composite material. Depending on the service
environment in which the composite structure is to function, i.e., the design
requirements imposed on the structure, the cost-effectiveness assessment will
consider the necessary material properties and their relationships with the material
and defect descriptors. A cost/performance trade-off exercise, and any iteration on
it, will result in an optimized cost-effective product. In most applications, where
long-term performance is the critical design consideration, one needs to look at
the degradation of initial (end-of-manufacturing) properties of interest under the
service environment. Thus the cost/performance trade-off will consider the
residual properties. A common approach is to consider only the initial (i.e., pre-
service) properties even for the long-term case, assuming implicitly that the
residual properties will relate to the defects (and the cost) the same way as the
initial properties do. This assumption is in fact questionable since the initial
properties may not show sensitivity to some of the material defects that may turn
out to be signicant in governing the long-term properties.
There is a variety of manufacturing processes used for composite structures,
e.g., autoclave molding, liquid compression molding, resin transfer molding,
Figure 8.8. Procedure for the cost-effectiveness assessment of composite structures.
lament winding, chemical vapor deposition, etc. Each of these processes pro-
duces defects in the manufactured part that are usually characteristic of that
process. The machining, joining, and assembly methods used for composite struc-
tures produce defects that are generally different from those produced during
molding, winding, and vapor deposition. For instance, the defects in the inter-
facial region between two parts will be different depending on whether the parts
were co-cured or adhesively bonded. Signicant differences in the fatigue lives of
co-cured vs. bonded joints have been reported [24].
In recent years, many methods have been developed to observe manufacturing
defects in composite materials and structures by nondestructive evaluation, based
mainly on ultrasound and radiography [25] and to some extent on thermal wave
imaging [26]. In the conventional approach these methods are utilized primarily for
quality control of the manufactured product. The premise of the quality control is
often that the presence of defects is undesirable. If defects of more than certain
threshold values are found, then the part is rejected and one strives to improve the
manufacturing process. This inevitably increases the cost and can result in the
composite part not being competitive with a metallic alternative. It is important to
realize that the presence of defects is not undesirable in all cases. In fact, if some defects
are allowed, the part can be produced at a lower cost. Figure 8.9 illustrates the
dependence of strength per dollar of manufacturing cost on the defect density. As
seen in the gure, the strength decreases gradually with defect density for low
densities, and drops rapidly at high densities, while the manufacturing cost increases
rapidly at low defect densities and falls off as more defects are allowed. Thus the
strength of the part achieved per dollar of manufacturing cost increases with defect
density, up toa point, beyond which the benet of allowing more defects decreases. It
must be remembered, however, that this situation is typical of the static strength. The
dependence of residual strength in long-termloading on the initial defect density may
show different behavior. This aspect has not been investigated sufciently.
Figure 8.9 also suggests that we should get away from the accept/reject
approach and advance to what may be called defect engineering. More specically,
we should engineer the components to have a certain amount of defects in order to
bring down the manufacturing cost while still having the performance require-
ments satised. To achieve this higher level of engineering we need certain cap-
abilities to be in place. Referring again to Figure 8.8, the connection between
manufacturing (box at top) and the material state achieved (box at left) requires a
capability to predict the defect structure along with the material composition
attained from the employed manufacturing process. Some attempts in this direc-
tion have been made. As an example, see references [2731] for prediction of voids
in a liquid compression molding process.
Our efforts are focused on the connection between defect structures and the
mechanical properties, as well as their degradation in service. This type of activity
may be viewed as an extension of damage mechanics, which in its conventional
form deals with initiation and evolution of damage and the consequent changes in
mechanical properties. Thus our starting point in the extended damage mechanics
is not a homogenized continuum, but a composite with bers and matrix as
constituents, and in addition, defects. The defects in our analyses are real-life
defects with their geometry and distribution as given by actual observations.
A broader strategy for durability assessment that includes analysis of defects
was coined as defect damage mechanics [32]. It is discussed next.
8.3.2 Defect damage mechanics
To illustrate the mechanics of damage incorporating defects, we take two
examples in the following.
8.3.2.1 Autoclave processing voids
The rst example deals with voids in composite laminates manufactured by
autoclave molding where we describe the observed characteristics of voids and
their incorporation in the modeling (for more details, see [33]).
Figure 8.10 (upper part) shows two cross-sectional views, parallel and normal to
bers, of a unidirectional carbon/epoxy composite made by the autoclave process.
The voids seen are generally not spherical and are largely trapped between the
prepreg layers. In the lower part of the gure are two cross sections, a short
distance (1.2 mm) apart, showing the voids more closely.
Figure 8.11 summarizes numerous observations and measurements reported in
the literature [3537] concerning voids in composites made by autoclave molding.
The shape can be described as elongated cylinders of elliptical cross section capped
at the ends. The volume fraction of the voids is found to be less than 3% for a well-
controlled autoclave process, which is safely below the 5% value taken for rejec-
tion of parts in the aerospace industry.
The process by which voids form suggests that the voids must displace the bers
around them as they settle down in their equilibrium positions. Most continuum
models that homogenize the composite and embed voids do not account for this
fact. Such models essentially replace bers, not displace them. Reference [33]
accounted for the ber displacement as schematically illustrated by Figure 8.12.
Strength/dollar
Strength
Manufacturing
cost
Defect density
Figure 8.9. Illustration of the dependence on defect density of strength, manufacturing
cost, and strength per unit cost.
The predictions of the elastic moduli by the HuangTalreja procedure [33] are
compared with experimental data in Figure 8.13. Note the large change in the
through-thickness modulus (E
zz
) due to the voids.
8.3.2.2 Interlaminar voids
We now consider the effect of the presence of voids in an interlaminar plane
(layer) on the growth of a crack in that plane. As described above, most voids in
1.2 mm
2.5 mm
2.5 mm
Figure 8.10. Observed voids in unidirectional carbon/epoxy composites made by autoclave
process. Upper pictures [34]: cross section parallel to bers (left) and across bers (right).
Lower pictures [35]: two cross sections 1.2 mm apart showing voids. Upper pictures
reprinted, with kind permission, from K. J. Bowles and S. Frimpong, J Compos Mater,
Vol. 26, pp. 1487509, copyright # 1992 by Sage Publications. Lower pictures reprinted,
with kind permission, from Springer Science+Business Media: Review of Progress in
Quantitative Nondestructive Evaluation, A morphological study of porosity defects in
graphite-epoxy composite, Vol. 6B, 1986, pp. 117584, D. K. Hsu and K. M. Uhl.
Figure 8.11. Observed characteristics of voids in carbon/epoxy composites made by autoclave
molding.
Figure 8.12. Modeling of voids accounting for displacement of bers. Reprinted, with
kind permission, from Compos Sci Technol, Vol. 65, H. Huang and R. Talreja, Effects
of void geometry on elastic properties of unidirectional ber-reinforced composites,
Figure 8.13. Elastic moduli predicted by [33] compared with experimental results. Data for
E
x
and E
y
are from [37] and for E
z
from [36]. Reprinted, with kind permission, from
Compos Sci Technol, Vol. 57, C. A. Wood and W. L. Bradley, Determination of the effect
of seawater on the interfacial strength of an interlayer E-glass/graphite/epoxy composite
by in situ observation of transverse cracking in an environmental SEM, pp. 103343,
layered composites tend to place themselves between layers when manufactured
with a compression molding process. These voids can have different shapes, sizes,
and spacing in the interlaminar plane. This plane is also a plane that is prone to
cracking under service or may have pre-existing aws due to insufcient adhesion.
A convenient way to make assessment of the effects of voids on interlaminar
fracture is to consider the geometry used for evaluation of interlaminar fracture
toughness. Ricotta et al. [38] conducted a systematic study of the voids inuence
on crack growth by considering this geometry. In the following some results from
that study are discussed to illustrate the effects.
Figure 8.14 shows a woven fabric composite where voids are found in the resin-
rich regions between the ber bundles. These regions are likely to develop cracks
under service environment such as fatigue or fail under in-plane compression,
leading to delamination. A representation of the effect of such voids on crack
growth is illustrated in the gure where a double cantilever beam (DCB) specimen
with voids ahead of the crack tip is shown. This geometry for mode-1 crack
growth has been systematically analyzed in [38] considering various parameters
such as void shape (circular and elliptical), void size, and distance of void from of
the crack tip.
The approach described in detail in [38] consists essentially of rst validating
an analytical method by nite element analysis and then using the method to
conduct a parametric study of the effects of voids. The method uses a beam-on-
elastic-foundation analysis accounting for shear compliance and material
orthotropic symmetry. The voids are simulated as regions without support
from the elastic foundation. The strain energy release rate with voids present
(G
I,v
) is calculated for different cases. Figure 8.15 shows the effect of a single
circular void of different radius R placed at different distance D from the crack
tip. The G
I,v
normalized by the value without void (G
I
) shows increasing
enhancement as the void approaches the crack tip, and this enhancement
increases with increasing void radius. Figure 8.16 shows a similar effect for
elliptical voids.
The effects of multiple circular voids on the energy release rate are shown in
Figures 8.17 and 8.18. Figure 8.17 shows the effects of two and three voids of xed
radius and xed mutual spacing located at different distances from the crack tip.
In Figure 8.18 an interesting effect of void interaction is shown. As seen there, for
multiple circular voids where the nearest void is kept at a xed distance from the
crack tip, while the mutual void spacing is varied, the energy release rate does not
show a monotonic dependence on the void spacing. Instead, the void interaction
increases the energy release rate up to a certain void spacing, beyond which the
effect of having multiple voids decreases.
Finally, Figure 8.19 illustrates how the energy release rate increases with
crack propagation when a void exists ahead of the crack tip. The increase of the
energy release rate with crack length when no void exists is plotted for refer-
ence. Thus the enhancement of the energy release rate is seen as the crack tip
approaches the void.
Figure 8.14. Voids in resin-rich areas between bundles in a woven fabric laminate and a DCB
specimen representation of crack growth in the presence of the voids.
1.3
1.2
1.1
G
I
,
v
/
G
I
1
0 5 10
R=0.2 mm
R=0.1 mm
R=0.08 mm
R=0.05 mm
Distance from crack tip D (mm)
15
D
20
Figure 8.15. Effects of a circular void of radius R and of distance D from crack tip on the
energy release rate.
1.9
1.8
1.7
1.6
1.5
1.4
1.3
1.2
1.1
1
0 2
a/b=1
a/b=1.5
a/b=2
a/b=4
b=0.1 mm
D
2b
2a
4 6
G
I
,
v
/
G
I
Figure 8.16. Effects of an elliptical void of different aspect ratio a/b, b = 0.1 mm, and of
distance D from crack tip on the energy release rate [38]. Reprinted, with kind permission,
from Compos Sci Technol, Vol. 68, M. Ricotta, M. Quaresimin and R. Talreja,
Mode-I strain energy release rate in composite laminates in the presence of voids,
8.4 Final remarks
The analyses and methods presented in this book have been mostly directed at
composite materials having continuous ber reinforcement in individual layers
that are stacked to form laminates. These material systems with polymers as
matrix materials and with stiff bers such as carbon have spurred the development
of lightweight structures in the aeronautics industry. Today, new aircraft such as
Boeing 787 and Airbus 380 are products of those developments. It is arguable,
however, how much of the advancement in damage modeling presented in this
book is embedded in the design of these aircraft. While this situation is under-
standable due to the stringent and costly airworthiness certication requirements,
it is hoped that eventually the output of research efforts in damage and failure will
transfer to designing safer and more cost-effective structures.
1
0 1
3 voids R =0.1 mm
2 voids R =0.1 mm
1 void R =0.1 mm
2
3 4
1.1
1.2
1.3
G
I
,
v
/
G
I
Figure 8.17. Effect on the energy release rate of circular voids of radius R = 0.1 mm and of
2.0 mm mutual spacing placed ahead of the crack tip for varying distance D from the
crack tip.
1.11
1.10
1.09
1.08
0 2 4
3 voids R = 0.1 mm
2 voids R = 0.1 mm
6
Distance from first void c (mm)
8 10
G
I
,
v
/
G
I
Figure 8.18. Effect on the energy release rate of multiple voids of xed radius R = 0.1 mm of
varying mutual spacing c placed at a xed distance from the crack tip.
The use of composites has over the years expanded beyond the aerospace
applications to other areas of structures. Carbon ber composites have experi-
enced an explosive growth in recent decades with an annual growth rate ranging
from 10 to 15%. The emerging applications of composite materials in automotive
and wind energy sectors place different challenges on design of these materials
than what has been the case in the aerospace industry. Although the affordability
of aerospace vehicles, even in the defense industry, has been a consideration, cost-
effectiveness is a prime factor in design of automotive and wind turbine structures.
The defect damage mechanics discussed above is bound to be an integral part of
the design process for these structures in the future. Incorporating this approach
in computational design methodologies will be a crucial next step.
For wind energy applications the key factor is long-term durability, other than
low cost. The design life of these structures is currently at 20 years (earlier it was
30 years!). For fatigue this translates to 10 million load cycles, or more. Most
composites fatigue testing has traditionally been done until 10
6
cycles, motivated
by metal fatigue where steels typically have a fatigue limit, which is revealed by the
S-N curve attening out before this number of load cycles. For composite mater-
ials the fatigue limit is not as easily determined. As discussed in Chapter 6,
considerations of damage mechanisms are necessary to deduce this property. This
is a signicant challenge for a highly complex composite construction in wind
turbine rotor blades. A much greater challenge is to determine the fatigue life at
a large number of cycles under multiaxial loading conditions typical for these
structures.
The eld of multiaxial fatigue in composites must be given the support it
deserves. A large-scale activity that is comprehensive in its approach is needed.
It must include testing and evaluation at scales where damage initiates, to
scales of damage progression (crack multiplication), and failure criteria that
properly represent the mechanisms. The activity so far has been limited in
scope and mostly focused on empirical and semi-empirical approaches. Most
540
520
500
480
G
I

(
J
/
m
2
)
460
440
420
0 1 2
Da
Circular void, R=0.1 mm
Without void
Crack propagation (mm)
3 4
Figure 8.19. Increase in the energy release rate as the crack tip approaches the void is
illustrated. The lower curve shows the energy release rate for comparison when no void
exists.
297 8.4 Final remarks
work continues to emulate metal fatigue despite fundamental differences in the
underlying mechanisms [39].
This book has not specically dealt with nanoscale reinforcements and compos-
ite systems. Although many advances have recently occurred, it is not yet clear
what advantages this area holds for improving durability at low cost in load-
bearing composite structures. The area of multifunctionality of composites, such
as conducting polymers and their composites, has on the other hand shown clear
promise for applications in structural health monitoring.
References
1. O. O. Ochoa and J. N. Reddy, Finite Element Analysis of Composite Laminates.
(Dordretchet, The Netherlands: Kluwer Academic Publishers, 1992).
2. M. J. Hinton and P. D. Soden, Predicting failure in composite laminates: the back-
ground to the exercise. Compos Sci Technol, 58:7 (1998), 100110.
3. M. J. Hinton, A. S. Kaddour, and P. D. Soden, Evaluation of failure prediction in
composite laminates: background to part B of the exercise. Compos Sci Technol,
62:1213 (2002), 14818.
4. M. J. Hinton, A. S. Kaddour, and P. D. Soden, Evaluation of failure prediction
in composite laminates: background to part C of the exercise. Compos Sci Technol,
64:34 (2004), 3217.
5. Y. W. Kwon, D. H. Allen, and R. Talreja, eds., Multiscale Modeling and Simulation of
Composite Materials and Structures. (New York: Springer, 2008).
7. R. Talreja, Multi-scale modeling in damage mechanics of composite materials. J Mater
Sci, 41:20 (2006), 680012.
8. R. Talreja, On multiscale approaches to composites and heterogeneous solids with
damage. Philos Mag, 90:3132 (2010), 433348.
9. R. Pyrz, K. Anthony, and Z. Carl, Morphological characterization of microstructures.
In Comprehensive Composite Materials. (Oxford: Pergamon, 2000), pp. 46578.
10. S. Ghosh, Adaptive concurrent multilevel model for multiscale analysis of composite
materials including damage. In Multiscale Modeling and Simulation of Composite Mater-
ials and Structures, ed. Y. W. Kwon, D. H. Allen, and R. Talreja. (New York: Springer,
2008), pp. 83163.
11. R. M. Jones, Mechanics of Composite Materials, 2nd edn. (Philadelphia, PA: Taylor &
Francis, 1999).
12. L. E. Asp, L. A. Berglund, and R. Talreja, Effects of ber and interphase on matrix-
initiated transverse failure in polymer composites. Compos Sci Technol, 56:6 (1996),
65765.
13. L. E. Asp, L. A. Berglund, and R. Talreja, Prediction of matrix-initiated transverse
failure in polymer composites. Compos Sci Technol, 56:9 (1996), 108997.
14. L. E. Asp, L. A. Berglund, and R. Talreja, A criterion for crack initiation in glassy
polymers subjected to a composite-like stress state. Compos Sci Technol, 56:11 (1996),
1291301.
of Fibre Composites. (Guildford, Surrey: IPC Science and Technology Press, 1971),
pp. 1526.
16. R. Talreja, Damage analysis for structural integrity and durability of composite mater-
ials. Fatigue Frac Engng Mater Struc, 29 (2006), 481506.
17. R. Talreja and C. V. Singh, Multiscale modeling for damage analysis. In Multiscale
Modeling and Simulation of Composite Materials and Structures, eds. Y. W. Kwon,
D. H. Allen, and R. Talreja. (New York: Springer, 2008), pp. 52978.
18. R. Talreja, A synergistic damage mechanics approach to durability of composite
systems. In Progress in Durability Analysis of Composite Systems, eds. A. M. Cardon
et al. (Rotterdam: A.A. Balkema, 1996), pp. 11729.
with off-axis plies. Compos Sci Technol, 59 (1999), 213947.
20. J. Varna, R. Joffe, and R. Talreja, A synergistic damage mechanics analysis of trans-
verse cracking in [y/90
4
]
s
laminates. Compos Sci Technol, 61 (2001), 65765.
21. J. Varna, A. Krasnikovs, R. S. Kumar, and R. Talreja, A synergistic damage mechanics
approach to viscoelastic response of cracked cross ply laminates. Int J Damage Mech,
13 (2004), 30134.
22. C. V. Singh and R. Talreja, Analysis of multiple off-axis cracks in composite laminates.
Int J Solids Struct, 45 (2008), 457489.
23. C. V. Singh and R. Talreja, A synergistic damage mechanics approach for composite
laminates with matrix cracks in multiple orientations. Mech Mater, 41 (2009), 95468.
24. M. Quaresimi and M. Ricotta, Fatigue behaviour of bonded and co-cured joints in
composite materials. In Experimental Techniques and Design in Composite Materials 6,
Extended Abstracts, ed. M. Quaresimin. (Vicenza, Italy: University of Padova, 2003),
pp. 312.
25. B. R. Tittmann and R. L. Crane, Ultrasonic inspection of composites. In Comprehen-
sive Composite Materials, Vol. 5, eds. L. Carlsson, R. L. Crane, and K. Uchino; eds.-in-
chief A. Kelly and C. Zweben. (Amsterdam: Elsevier, 2000), pp. 259320.
26. R. L. Thomas, L. D. Favro, X. Hanand, and Z. Ouyang, Thermal methods used in
composite inspection. In Comprehensive Composite Materials, Vol. 5, eds. L. Carlsson,
R. L. Crane, and K. Uchino; eds.-in-chief A. Kelly and C. Zweben. (Amsterdam:
Elsevier, 2000), pp. 42746.
27. A. D. Mahale, R. K. PrudHomme, and L. Rebenfeld, Quantitative measurement of
voids formed during liquid impregnation of nonwoven multilament glass networks
using an optical visualization technique. Polym Eng Sci, 32 (1992), 31926.
28. N. Patel and J. L. Lee, Effect of ber mat architecture on void formation and removal
in liquid composite molding. Polym Composites, 16 (1995), 38699.
29. N. Patel, V. Rohatgi, and J. L. Lee, Micro scale ow behavior and void formation
mechanism during impregnation through a unidirectional stitched berglass mat.
Polym Eng Sci, 35 (1995), 83751.
30. V. Rohatgi, N. Patel, and J. L. Lee, Experimental investigation of ow induced
microvoids during impregnation of unidirectional stitched berglass mat. Polym Com-
posites, 17 (1996), 16170.
31. S. Roychowdhury, J. W. Gillespie, Jr., and S. G. Advani, Volatile-induced void
formation in amorphous thermoplastic polymeric materials: I. Modeling and paramet-
ric studies. J Compos Mater, 35 (2001), 34066.
299 References
32. R. Talreja, Defect damage mechanics: broader strategy for performance evaluation of
composites, Plastics Rubber Composites, 38 (2009), 4954.
33. H. Huang and R. Talreja, Effects of void geometry on elastic properties of unidirec-
tional ber reinforced composites. Compos Sci Technol, 65 (2005), 196481.
34. K. J. Bowles and S. Frimpong, Void effect on the interlaminar shear strength of
unidirectional graphite ber-reinforced composites. J Compos Mater, 26 (1992),
1487509.
35. D. K. Hsu and K. M. Uhl, A morphological study of porosity defects in graphite-epoxy
composite. In Review of Progress in Quantitative Nondestructive Evaluation, Vol. 6B, eds.
D. O. Thomson and D. E. Chimenti. (New York: Plenum Press, 1986), pp. 117584.
36. Z. Guerdal, A. P. Tamasino, and S. B. Biggers, Effects of processing induced defects on
laminate response: interlaminar tensile strength. SAMPE J, 27 (1991), 349.
37. P. Olivier, J. P. Cottu, and B. Ferret, Effects of cure cycle pressure and voids on some
mechanical properties of carbon/epoxy laminates. Composites, 26 (1995), 50915.
38. M. Ricotta, M. Quaresimin, and R. Talreja, Mode-I strain energy release rate in
composite laminates in the presence of voids, Special issue in honor of R. Talrejas
60th birthday. Compos Sci Technol, 68:13 (2008), 261623.
39. M. Quaresimin, L. Susmel, and R. Talreja, Fatigue behaviour and life assessment of
composite laminates under multiaxial loadings. Int J Fatigue, 32 (2010), 216.
Author index
Aboudi 124
Adkins 149
Adolfsson 192, 208, 209
Ahci 175, 176
Akshantala 188, 270
Aoki 70
Aveston 3, 4, 49, 58, 66, 74
Awerbuch 244
Bailey 4, 49, 74, 80, 82, 84, 188, 192, 195,
196, 200
Balhi 186
Benthem 115
Berglund 97
Berthelot 74, 212, 215, 216, 217
Bowles 292
Bradley 40, 293
Broutman 49
Budiansky 44, 85, 249
Chaboche 4, 136, 137, 138
Chang 193
Cooper 49, 58
Cox 74
Crocker 186, 193
Crossman 206
Curtis 245
Daniel 79, 82
Dharan 239
Diao 276
Dvorak 76, 82, 85, 200, 201, 202
Eshelby 85
Fenner 46
Flaggs 80, 82
Frimpong 292
Fukunaga 77, 78, 79, 82, 197
Galiotis 40, 46
Gamby 124
Gamstedt 38, 247, 248, 249
Garrett 4, 49, 80, 82, 84, 186, 188, 195, 200
Grimes 262
Gudmundson 111, 113, 114, 117, 155, 208, 209
Hahn 244, 263
Halpin 19
Hashin 19, 23, 57, 87, 91, 97, 101, 104, 113, 124,
194, 202, 204, 214, 216, 259
Highsmith 77, 78, 82
Hill 85, 186
Hoiseth 86, 210
Hong 79, 82, 124
Hori 3, 84, 279
Hsu 292
Hu 85, 98, 268
Huang 292
Ishikawa 70
Jamison 261
Joffe 104, 208, 219
Johnson 193
Ju 137
Kachanov 4, 114, 117, 134, 141
Katerelos 40, 184
Kelly 4, 49, 58, 66, 74
Kim 125, 263
Koiter 115
Krajcinovic 4, 208
Kumar 172, 173, 174, 175
Kuriakose 101
Lavoie 192
Laws 76, 82, 85, 200, 201, 202
Le Corre 212, 215, 216, 217
Lee 79, 82, 124
Lemaitre 4, 136, 138
Li 119, 121, 122
Lim 79, 82
Liu 47, 190, 221
Lundmark 117, 118, 119, 125, 189, 191
Mai 276
Manders 74, 82, 84, 196, 197, 212
McCartney 104, 109, 125, 205
Mori 19, 44, 281
Murakami 136, 137
Na 124
Nairn 83, 85, 97, 98, 110, 125, 185, 187, 202, 204,
205, 221, 268
Nath 46
Nemat-Nasser 3, 84, 279
Niu 43
Nuismer 79, 80, 81, 82
Ogin 79, 82, 186
Ohno 136, 137
Ostlund 113
Pagano 124
Parvizi 49, 74, 199
Parvizi-Majidi 67
Qu 86, 210
Queresimin 295
Ravichandran 47, 190
Rebiere 124
Reddy 119, 123
Reifsnider 77, 78, 82, 188
Rice 45
Ricotta 294, 295
Robotnov 135
Rotem 259
Sahu 49
Schapery 172
Shoeppner 124
Shtrikman 19
Sicking 172
Silberschmidt 212
Silwood 58
Singh 162, 171, 218, 222, 223, 224
Sirivedin 46
Sjogren 38
Smith 186
Srensen 253, 256
Steif 79, 82, 198
Sturgeon 244, 245
Tada 115
Talreja 4, 6, 43, 50, 101, 141, 162, 171, 172, 173,
174, 175, 188, 218, 222, 223, 224, 238, 239, 247,
250, 253, 270, 292, 295
Tan 79, 80, 81, 82
Tanaka 19, 44, 281
Timoshenko 43
Tracey 45
Tsai 19, 23, 194
Uhl 292
Vakulenko 117, 141
Varna 40, 97, 102, 104, 117, 118, 119, 125, 143,
174, 189, 191, 208, 221
Vinogradov 202, 214, 216
Wang 67, 206, 213
Weibull 196, 208, 215
Wood 40, 293
Wu 23, 194
Yaniv 79
Ye 276
Yokozeki 70, 193
Zhang 124
302 Author index
Subject index
ACK theory 4, 58, 199
Acoustic emission 181, 253
Characteristic damage state (CDS) 47, 48, 188
Classical laminate plate theory (CLPT) 24
COD based methods 206
Complementary energy 12, 92, 93, 94, 101, 203,
227
Computational methods 119
Computational structural analysis 276
Constitutive relations/Constitutive response/
Constitutive relationships/Constitutive
equations 1115, 20, 26, 172
Constraint factor. See Constraint parameter
Constraint parameter 148, 155, 163, 165, 169,
174, 287
Constraint. See ply constraint
Continuum damage mechanics (CDM) 4, 57, 134,
137, 144, 154, 161, 174, 176
Correspondence Principle 172
Cost-effective manufacturing 288
Crack density 86, 173, 192, 202, 208, 214, 266,
270
Normalized crack density 201
Crack density evolution. See damage progression
Crack initiation 180, 185, 201, 273, 276, 278
Crack initiation strain 198, 205, 221
Crack opening displacement (COD) 2, 3, 32, 102,
111, 117, 155, 161, 163, 164, 174, 181, 207,
219
Crack progression. See Damage progression
Crack sliding displacement (CSD) 3, 143,
219
Crack spacing 148, 195
Normalized crack spacing 203
Crack size 148
Critical energy release rate. See energy release
rate: critical
Cyclic loading. See Fatigue
Damage 36
Damage characterization 139
Damage development. See damage progression
Damage entity tensor 141
Damage evolution. See damage progression
Damage evolution curve. See damage progression
curve
Damage initiation. See Crack initiation
Damage mechanics 2, 33, 36, 48, 50, 57, 276, 280,
290
Damage mechanisms 36, 37, 142, 143, 181,
242, 278
Damage modes 45, 46, 142143, 156, 158, 165,
167, 175, 179, 193
Damage mode tensor 142143, 158
Damage progression 46, 139, 179, 180, 184, 187,
188, 192, 194, 196, 204, 208, 212, 215, 217,
221, 222, 229, 253, 266, 268, 270, 273, 276,
283, 284, 294
Curve 189, 190
Energy based approaches 198, 210, 277
Strength based approaches 180, 194, 210
Damage state 173
Damage tensor 151, 157
Damage tolerance 265
Debonding (interfacial) 3738, 48, 248, 249, 251,
256, 283, 284, 287
Defect damage mechanics 287, 291, 297
Delamination 33, 3941, 48, 183, 187, 193, 216,
225, 260, 271, 273, 287
Distortional energy (Tsai-Hill) criterion 23
Durability 1, 37, 265, 279, 287
Assessment 277
Edge replication 181
Energy release rate 31, 203, 204, 208, 294
Critical 31, 206, 213, 219
Failure 36
Failure criteria 21, 180
Fatigue 223, 265, 276
Fatigue-life 246, 271
Fatigue-life prediction 265271
Crossply laminates 266270, 273
General laminates 273
Fatigue-life diagrams 67, 237265, 273
Polymer matrix composites (PMC) 242, 250,
251, 283
Fatigue-life diagrams (cont.)
Angleply laminates 260, 261
Ceramic matrix composites (CMC) 252
Crossply laminates 261262
Metal matrix composites (MMC) 250, 251
Multidirectional laminates 263
Quasi-isotropic laminates 263
Unidirectional composite 257
Fatigue limit 237
Fiber breakage 42, 48, 183, 239, 249, 255
Fiber failure. See ber breakage
Fiber microbuckling 42, 43
Finite element method (FEM) 117, 119, 209, 276
Finite strip method 121
Fracture 36, 48
Frictional sliding 256, 268 See also Interfacial
sliding
Generalized Hookes law 12
Generalized plain strain analysis 104, 205
Hashin criterion 23
Interfacial sliding 39, 41
Interlaminar cracking. See delamination
Intralaminar cracking. See ply cracking
Laminate 18
Balanced 29
Crossply 29, 285
Multidirectional 193, 217
Quasi-isotropic 29, 193
Symmetric 29
Layerwise theory 123
Length scales of damage 280
Linear elastic fracture mechanics (LEFM) 29, 184
Macro damage mechanics (MADM) 5, 51, 57,
126, 134, 276
Matrix cracking. See ply cracking
Maximum stress theory 22
Maximum strain theory 22
Mechanisms of damage. See damage mechanisms
Microbuckling. See Fiber microbuckling
Microcracking. See ply cracking
Micro damage mechanics (MIDM) 4, 51, 57, 126,
134, 276
Micromechanics 3, 17, 44, 57, 144, 161, 176
Multiple cracking 63
Multiple matrix cracking. See ply cracking
Multiscale modeling 278
Hierarchical 279, 282, 286
Particle cleavage 44
Periodic boundary conditions 120
Ply constraint 49, 187
Ply cracking 39, 48, 61, 64, 170, 183, 194, 267, 283
Principle of minimum potential energy 16
Principle of minimum complementary energy
17, 91
Principle of virtual work 16, 92
Raman spectroscopy 69, 184
Randomness in ply cracking 70, 210
Randomness. See randomness in ply cracking
Reference Laminate 220, 221
Representative volume element RVE 2, 5, 69,
110, 121, 134, 139148, 161, 278, 280, 281,
282, 287
Residual stresses 96, 192, 201, 208
RVE. See representative volume element
Self-consistent method. See Self-consistent scheme
Self-consistent scheme 8486
Shear lag analysis 198
Shear lag methods. See Shear lag models
Shear lag models 65, 7384, 101104
Shear lag parameter 7682, 103, 195, 197, 201
Shear lag. See Shear lag models
Shear lag theory. See Shear lag models
Stiffness changes. See Stiffness degradation,
Stiffness damage relationships
Stiffness degradation 83, 85, 102, 154, 170, 193
Stiffness-damage relationships 148, 152, 157, 169,
179, 184
Stiffness reduction 192
Strain energy density 12
Strength criteria. See Failure criteria and Damage
progression: strength based approaches
Structural integrity 1, 37
Synergistic damage mechanics (SDM) 5, 155, 156,
164, 170, 171, 174, 176, 184, 286
Thermal expansion coefcient 125
Thermal residual stress 202, 205
Thermal stress. See Thermal residual stress
Tsai-Wu criterion 23
Ultrasonic C-scan 183
Unidirectional lamina (UDL) 18
Variational analysis. See variational methods
Variational methods 87, 97, 202, 226
Virtual work. See Principle of virtual work
Viscoelastic composites 170
Viscoelastic response 173, 174
Viscoelasticity 170
Void growth 44
X-ray radiography 181, 261
304 Subject index

Damage and Failure of Composite Materials

Diunggah oleh

Informasi Dokumen

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Damage and Failure of Composite Materials

Diunggah oleh

Hak Cipta:

Format Tersedia

Damage and Failure of Composite Materials

Understanding damage and failure of composite materials is critical for reliable

) and transverse (90

) directions, the laminate thus built is

-plies were made thinner, the strain to initiate

-plies (< 0.1 mm),

-plies of [0/90/ 45]

-plies decreases leading to ply

-plies exert a larger

-plies, thereby delaying the crack initiation

Figure 4.4. Mechanism of load transfer at the ber/matrix interface.

-ply behaves differently from a similar crack in the 90

crack were different. For the laminates containing

-plies. As described in Chapter

-plies and span the

-ply (undamaged laminate)

-ply (undamaged laminate)

-ply is constant through its thick-

-ply. The equilibrium of axial

-ply (see Figure 4.8) gives

-plies are related to the applied stress s

-plies, respectively. Differentiating Eq. (4.57) with respect to x and using

-plies have the same elastic properties.

-ply, Eqs. (4.64),

ply, Steif [46] developed a shear

layer thickness. This analysis

-plies have been considered by Nuismer

-plies. These approaches also lead to

-layer, analogous to Eq.

-plies increases fromzeroon

-plies back into the 90

-plies. The traction-free boundary condition at the

-plies share the applied load. But

-plies. Away from cracks, the

-layers for virgin laminate are constant throughout individual layers

-ply develops continuous intralaminar

-ply and the

-layers, respectively. The stress energy density in a transversely iso-

-plies in terms of the perturb-

-plies are chosen as

-layer close to the

-layer, the redistribution in the stress eld is described by c

-ply group, whereas shear stress s

-layer from the 0

-layer. This stress reaches a maximum midway between two

-ply, McCartney [80] developed a theory of stress transfer between

-plies while retaining all relevant stress and strain components.

-plies in material (lamina)

-ply, respectively, and E

-plies are independent of the transverse coordinate z, the relevant

-plies, respectively, given by

-plies, and zero traction at crack

and each uncracked sublaminate

-layers as double Fourier series in x and z (longitudinal

glass bre/epoxy resin laminates. J Mater Sci, 18:10

normalized by a unit applied strain, the predictions of E

, as well as changes in it for other ply orientations, are computed by

, the matrix will undergo multiple cracking in the transverse plies.

under an axial tensile load. However, it has

indicate variables for the +y and y plies,

-ply cracking as described in the previous chapter. For

crack is different fromthat for an

. Reprinted, with kind

in Figure 5.14. The agreement with