3.

09
Damage and Fracture Mechanics
Techniques for Composite Structures
A. CORIGLIANO
Politecnico di Milano, Italy

3.09.1 INTRODUCTION 462

3.09.2 OVERVIEW OF THE MAIN FRACTURE AND DAMAGE MECHANISMS IN COMPOSITE

MATERIALS AND STRUCTURES 465
3.09.2.1 Rupture of Fibers and Fiber Microbuckling 466
3.09.2.2 Fiber–Matrix Debonding 469
3.09.2.3 Matrix Microcracking 470
3.09.2.4 Delamination 472
3.09.3 COMPUTATIONAL FM FOR COMPOSITES 475
3.09.3.1 Numerical Methods for the Evaluation of the ERR 477
3.09.3.1.1 Virtual crack extension method 478
3.09.3.1.2 Compliance method 479
3.09.3.1.3 Virtual crack closure method (VCC) 480
3.09.3.1.4 Displacement correlation method 481
3.09.3.2 Crack Propagation Criteria for Composites 481
3.09.3.3 Numerical Procedures for Crack Advancement in Composites 482
3.09.3.4 Basic Notions of FFM Applied to Composite Fracture 483
3.09.4 PFA FOR COMPOSITES 483
3.09.4.1 Failure Criteria for Unidirectional Fiber Composites 484
3.09.4.2 Progressive Failure Analysis (PFA) 485
3.09.5 DAMAGE MECHANICS FOR COMPOSITE MATERIALS AND STRUCTURES 485
3.09.5.1 Damage Mechanics for Heterogeneous Materials: Micro-, Meso-, and Macroscales 486
3.09.5.2 A Class of Elastic–Plastic Damage Models for Composite Materials: Formulation and Numerical
Time Integration 488
3.09.5.2.1 Model formulation 488
3.09.5.2.2 Numerical time integration 489
3.09.5.2.3 Consistent tangent matrix 490
3.09.5.2.4 1D elastic-damage model 491
3.09.5.2.5 1D elastic–plastic-damage model 492
3.09.5.3 Damage Model for a Single Lamina 492
3.09.6 THE PARTICULAR ROLE OF INTERFACES AND INTERFACE MECHANICS 495
3.09.6.1 The Importance of Interfaces in Composite Materials and Structures 496
3.09.6.2 A Class of Elastic–Plastic Damage Interface Models for Composite Materials: Formulation and
Numerical Time Integration 498
3.09.6.2.1 Model formulation 499
3.09.6.2.2 Numerical time integration 500
3.09.6.2.3 Consistent tangent matrix 501
3.09.6.3 Examples of Interface Models 501
3.09.6.3.1 Elastic-damage interface model 501
3.09.6.3.2 Elastic–plastic softening interface models 503
3.09.6.4 Interface Model Identification 503

459
460 Damage and Fracture Mechanics Techniques for Composite Structures

3.09.7 COMPUTATIONAL DAMAGE MECHANICS FOR COMPOSITES 505

3.09.7.1 FE Formulation for Mesomodeling of Composite Materials and Structures 508
3.09.7.1.1 General 3D formulation 508
3.09.7.1.2 Elastic body with nonlinear interfaces 509
3.09.7.1.3 Generalized plane strain 510
3.09.7.1.4 Boundary layer solution 511
3.09.7.2 Special FEs for the Simulation of Damage in Composites 512
3.09.7.2.1 Layered brick element 513
3.09.7.2.2 Interface element 514
3.09.7.3 Path Following Procedures 516
3.09.7.3.1 Local control with total displacement discontinuity 516
3.09.7.3.2 Local control with crack opening displacement 517
3.09.7.4 Examples of Applications 517
3.09.7.4.1 PFA of a sandwich beam 518
3.09.7.4.2 Assessment of material resistance at the micromechanical level 521
3.09.7.4.3 Delamination analysis of interlaminar fracture specimens 522
3.09.7.4.4 Simulation of GFRP reinforcement 528
3.09.7.4.5 Homogenization and parameter identification applied to the constitutive modeling of composites 529
3.09.8 DISCUSSION 532

3.09.9 REFERENCES 534

NOMENCLATURE fd damage activation function

fd vector of damage activation func-
a; w0p ; b, Yc ; Y0 ; Yc0 ; Y00 tions
material parameters in a elastic–plastic fK ; fG; f1 ; f2
damage constitutive law for a layer functions defining a fracture propaga-
A; a crack area, crack length tion criterion
b1 ; b2 ; b3 parameters of the flow rule for plastic fp plastic activation function
trial
displacement discontinuities in an fnþ1 trial plastic-damage activation func-
elastic–plastic interface law tion at time instant tn þ 1
bc bielastic constant trial
fpnþ1 trial plastic activation function at time
B width of a laminate test specimen instant tn þ 1
Bi matrix of strain interpolation func- F surface load vector
tions in a FE model Fi ; Fx ; Fy
B int matrix of shape functions defining the nodal forces in a FE model
model for displacement discontinuities FR vector of residual forces on the
in a interface FE boundary in a boundary layer sche-
c1 ; c2 parameters in a fracture propagation matization
criterion g value of the imposed quantity in a step
C compliance of a Riks’ like procedure
D vector of damage variables gp part of free energy density governing
D, Di, Dij, i, j ¼ 1, 2, 3 the hardening for a continuum or
damage variables interface constitutive law
e thickness of a fictitious layer equiva- G total energy release rate
lent to an interface GI, GII, GIII
E, Ei, i ¼ 1, 2, 3 modes I, II, and III energy release
Young’s moduli rates
ED damaged elastic strain energy for a Gc, GIc, GIIc
continuum or interface constitutive total, modes I and II critical value of
law ERR, fracture energies
E* equivalent Young’s modulus for plane Gij ; i; j ¼ 1; 2; 3
problems elastic shear stiffnesses
Ei0 ; i ¼ 1; 2; 3 H; s0 ; Yc ; Y0
undamaged Young’s moduli parameters of a 1D elastic–plastic
Eint matrix of elastic interface stiffnesses damage model
f body force vector Hv, Hint, Hs
f plastic-damage activation function in Boolean matrices for the assembling
a interface model procedure in a FE model
fcr critical value in SIF-based or ERR J flexural inertia of a beam
based fracture criteria J consistent tangent matrix
 Nomenclature 461

kint
l interface stiffness matrix in the local ttrial
inþ1 ; i ¼ 1; 2; 3
reference frame trial elastic interface tractions at time
K global stiffness matrix in a FE model instant tn þ 1
K total stress intensity factor DT temperature change with respect to a
K1, K2, K3, K3þ ; K3
reference value
elastic interface stiffnesses u displacement
KI, KII modes I and II stress intensity factor [u] displacement discontinuity
Ken kinetic energy u0 assigned displacement on the kinema-
KE elastic stiffness matrix in a FE model tically constrained surface
Ktg global tangent stiffness matrix in a FE [u]0, [u]c displacement discontinuity at activa-
model tion and at the end of damage
2l length of a linear interface element behavior
ld matrix (vector) of functions governing uboun, rboun
the flow rule for damage variables for boundary correction of the solution in
a continuum (interface) constitutive a boundary layer schematization
law udom, rdom
lup vector of functions governing the flow solution in the internal part of the
rule of plastic displacement disconti- domain in a boundary layer schemati-
nuities in a interface model zation
lep vector of functions governing the flow u* virtual displacement field
rule for plastic strains [u]e, [u]p
lZp function governing the flow rule for elastic, plastic displacement disconti-
the kinematic internal variable for a nuity
continuum or interface constitutive U class of admissible virtual displace-
law ments
DM percent of moisture (by weight) ab- U vector of global dof (nodal displace-
sorbed ments) in a FE model
Ni matrix of element displacement inter- Uen total elastic strain energy
polation (shape) functions in a FE Ui vector of element nodal dof
model Ui ; Ux ; Uy
NiC ; NiT ; i ¼ 1; 2; 3 nodal displacements in a FE model
laminate compression and tension V volume of a loaded body
strengths in directions i Wex external work
P concentrated load Y ; Yi ; Yij ; i; j ¼ 1; 2; 3
P vector of equivalent nodal loads in a variables conjugate to damage, da-
FE model mage energy release rate
PDM vector of nodal loads equivalent to Y vector of variables conjugate to da-
moisture variation in a FE model mage for a continuum or interface
PDT vector of nodal loads equivalent to constitutive law
temperature variation in a FE model Y% equivalent, cumulated in time, damage
q global internal force vector in a FE energy release rate
model Y0i ; i ¼ 1; 2; 3
qint vector of interface internal forces in parameters of an elastic-damage inter-
l
the local reference frame face model
Qi, Qij, i, j ¼ 1,y, 6 Yc, Y0 parameters of a 1D elastic-damage
parameters in Tsai–Wu failure criter- model
ion for laminates Y trial
nþ1 trial damage energy release rate
r radial coordinate from crack-tip a vector of coefficients of thermal ex-
R rotation matrix for an interface ele- pansion
ment b vector of coefficients of hygroscopic
Sij ; i; j ¼ 1; 2; 3 (or moisture) expansion
laminate shear strengths in the plane ij a; b; g parameters in a fracture propagation
Su kinematically constrained external sur- criterion or in a damage constitutive
face of a loaded body law
t interface traction G interface
t vector of interface tractions d opening displacement in a DCB speci-
t0i, i ¼ 1, 2, 3 men
tractions at the end of the elastic range e, ee, ep total, elastic, and plastic strains
462 Damage and Fracture Mechanics Techniques for Composite Structures
e; ee ; ep vectors of total, elastic, plastic strain r vector of stress components
components s, sk ; sij ; k ¼ 1; y; 6; i; j ¼ 1; 2; 3
e 0, e c strain at activation and at the end of stress components
damage behavior * s* ij ; i; j ¼ 1; 2; 3
s;
e virtual strain field effective stresses
et ; eh thermal and hygroscopic strains rtrial
nþ1 trial elastic stress at time instant tn þ 1
Zp kinematic internal variable for a con- t shear stress component
tinuum or interface constitutive law U; W vectors of shape functions for a
l plastic-damage multiplier in a inter- layered brick element
face model wp static internal variable for a conti-
ld damage multiplier nuum or interface constitutive law
kd vector of damage multipliers c phase angle or mode-mixity ratio in a
lp plastic multiplier fracture propagation criterion
m load factor C free energy density for a continuum or
n; nij ; i; j ¼ 1; 2; 3 interface constitutive law
Poisson’s ratios o mechanical dissipation
n0ij ; i ¼ 1; 2; 3 O domain of integration in a plane
undamaged Poisson’s ratios schematization
x; Z; B natural coordinates for a FE @Ou kinematically constrained boundary in
P total potential energy a plane schematization
r mass density

3.09.1 INTRODUCTION instance, studying the relations between micro-

structural phenomena and macroscopic beha-
The purpose of this chapter is to discuss vior will provide a way to design the
computational methods that can be applied for microstructure of the material such that
the simulation of fracture and damage phe- specific requirements on the resulting macro-
nomena in composite materials. The need to scopic mechanical behavior can be fulfilled.
develop numerical simulation tools for the This will allow the possibility to optimize the
description of damage and fracture processes macroscopic mechanical behavior of microsco-
in composites is growing rapidly. This need is pically heterogeneous materials. In general,
for simulation at the micro-, meso-, and and this applies to all material and structural
macroscales. The reason for this can be systems, good and reliable simulation tools
understood by a series of considerations. would allow for a better understanding of
Composites usually are computed and ver- complex and difficult problems like compres-
ified in the elastic range; this is a reason for sive behavior of fibers, matrix-fiber debonding,
severe limitations of the exploitation of all their and delamination.
mechanical potentialities. In order to take Before describing possible numerical simula-
advantage of all their possibilities, there is a tion tools it should be stressed that the
strong need for advanced and reliable compu- heterogeneous nature of composites is a reason
tational methods for the nonlinear behavior of for their complicated mechanical behavior and
composites that are to take into account the this makes it very difficult to formulate
real strength limits. A classical example of the numerical models and transforms the search
way in which composite safety is assessed is for a complete simulation tool into a formid-
the so-called first-ply failure (FPF) approach; able task.
following this approach the resistance of a After some decades of intense study and an
laminate usually is strongly underestimated; abundant scientific literature, the behavior of
moreover, this is not completely sound from a composite materials and structures in the
physical point of view. It would be possible to elastic range can be described by means of
formulate reliable strength criteria by means of powerful analytical and numerical tools (see,
reliable simulation tools of damage processes; e.g., the six-volume comprehensive treatise
these are still not completely defined, in edited by Kelly and Zweben, 2000) for the
particular for single laminas. estimation of composite properties in the
Another important motivation for studying elastic range. Many methods are available;
and modeling the mechanical behavior of nevertheless, the modeling of composite beha-
composites in the nonlinear range is the vior in the elastic range and the assessment of
possibility to assist the development of new composite strength is still a subject of research
materials and to reduce the amount of quali- (see, e.g., Herakovich, 1998; Kelly and Zwe-
fication data needed for any new system. For ben, 2000).
 Introduction 463
Serious difficulties arise for the formulation considered nondeterministic due to the large
of models and computational tools that allow number of parts with mechanical and geome-
for the simulation of fracture and damage trical properties having statistical distributions.
phenomena in composites and for the assess- Therefore, at least for some kinds of composite
ment of strength properties of real composite materials, nondeterministic analysis would be
structures. Some of the difficulties coincide the most correct approach. Again, one easily
with those met in materials commonly con- realizes that this is another very difficult task.
sidered homogeneous at the macroscale like Besides the difficulties related to multiple
metals, concrete, ceramics, etc.; others are scales and nondeterministic effects, other ob-
peculiar to the nature of composite behavior. stacles must be overcome in order to correctly
In order to appreciate the difficulties related model fracture and damage phenomena in
to the simulation of damage phenomena in composites. Some of these are typical of the
composites, let us first briefly discuss the main description of all damage phenomena, not
damage mechanisms in laminates, which are necessarily in composites: size effects due to
the composite structural elements more used in crack propagation, mesh dependency in the
applications. Damage phenomena arise at the numerical simulations, and instabilities and
scale of microconstituents (microscale) where bifurcations in the overall response which
they concern fiber rupture in tension and fiber create difficulties for the control of the analysis
buckling in compression, possibly accompa- in the presence of brittle or very brittle global
nied by microcracks in the matrix; at the scale behavior thus obliging us to the use of special
of mesocomponents (mesoscale) typical da- path-following techniques. Others are more
mage processes concern fiber–matrix debond- strictly related to the composite behavior: the
ing and typical patterns of cracks in the matrix importance of initial stresses in the global
of the plies in which fibers are orthogonal to response, mixed-mode crack propagation, ani-
the loading direction (transverse matrix crack- sotropy in the response also in the inelastic
ing); finally, damage is also present at the scale range, multiple damage mechanisms with
of the whole structural element (macroscale) multiple cracks, and the presence of interfaces
where macroscopic debonding of mesoconsti- between the single constituents. The last in
tuents can arise frequently; a typical example particular is a key point that should be taken
of this phenomenon is delamination. into account in numerical simulations; fiber–
Having mentioned the main damage me- matrix interfaces, laminate–laminate inter-
chanisms in typical composite structural ele- faces, as an example, are in fact weak points
ments, the first difficulty related to the which highly influence the global strength of
numerical simulation of linear and nonlinear the composite.
composite behavior immediately arises, i.e., the From the above considerations clearly it
need to describe multiscale processes, from the appears that the numerical simulation of
micro- to the macroscale. The microscale for fracture and damage in composites is still an
modern composite materials involve dimen- open issue, as witnessed by the abundant
sions of some microns (a typical carbon fiber scientific literature on the subject (see, e.g.,
diameter can be 8 mm), while the macroscale the interesting point of view of the industry in
means dimensions of the order of meters (a Aubard, 2001). Many authors have proposed
typical laminate plate side can be 1 m); let us computational methods for composite damage
imagine a 1 m side square laminate plate, 5 mm and fracture. The majority of them can be
thick: a hypothetical three-dimensional (3D) grouped in the classes that are described briefly
finite element (FE) model of the plate built in the following.
with 10 mm side cubes would have 5  1012 Computational linear elastic and nonlinear
elements! Of course, the available computa- fracture mechanics (computational LEFM and
tional tools do not allow the simulation of the NLFM) for composites (see, e.g., Kanninen
whole composite with models which resolve at and Popelar, 1985; Argon, 2000; Dvorak, 2000;
the same time all the scales of the material; this Nairn, 2000) allow for the assessment of
main difficulty motivated the development of strength in the presence of flaws and defects,
methods that allow us to consider the compo- and for the simulation of crack propagation for
site as apparently homogeneous (homogeniza- the description of the main composite damage
tion methods) or allow us to treat the various mechanisms like matrix-cracking and delami-
scales with different numerical models which nation. The numerical procedures are in some
are related in some ways during the analysis cases well established but they suffer from the
(multiscale approaches). typical drawbacks of fracture mechanics (FM):
Another serious difficulty related to the the defect must exist from the beginning of the
modeling of composites derives from the fact analysis, therefore, it is not possible to simulate
that their behavior should, in principle, be crack initiation processes; other damage
464 Damage and Fracture Mechanics Techniques for Composite Structures
processes which involve diffuse damage due to A natural consequence of using softening
the initiation and propagation of many small constitutive laws for the simulation of damage
cracks are outside the range of analysis of FM. in composites implies the combined use of
FM theories have been expanded, particu- interface laws and what can be named con-
larly for the analysis of composite materials cisely as interface damage mechanics (IDM).
and structures, with the introduction of the so- This term in this chapter will indicate the
called finite fracture mechanics (FFM: see, e.g., possibility of modeling the damageable beha-
Hashin, 1996; Nairn, 2001). It is possible by vior of surface entities (named interfaces) that
means of this approach to study finite length separate two constituents of a composite
propagation of existing cracks directly, thus material. Suitable interface models and inter-
allowing for more direct and application- face FEs can be defined which allow for a
oriented simulations. complete description of decohesion processes
Another possibility to simulate damage like fiber–matrix debonding and delamination.
processes in composites is based on the so- The use of interface cohesive laws has become
called stiffness reduction methods or progres- more and more popular in scientific literature
sive failure analysis (PFA) (see, e.g., Spotts- (see, e.g., Bažant and Oh, 1983; Carpinteri,
wood and Palazotto, 2001); these represent a 1989; Costanzo and Allen, 1995; Maier et al.,
class of computational methodologies in which 1993; Camacho and Ortiz, 1996) and for
the stiffness of the material is locally progres- composite materials is in some cases almost
sively reduced whenever a critical situation, compulsory (Corigliano, 1993; Schellekens and
usually expressed in term of stresses, is reached. de Borst, 1993a; Crisfield et al., 1997).
The main advantage of this approach is its In spite of their potentialities, the above
simplicity that allows for the determination of mentioned simulation tools do not necessarily
reasonable estimate of the overall structural reduce the computational cost due to the
resistance with low computational cost. The multiscale physics of composite materials. In
main problems are related to the somewhat order to avoid this main obstacle, some
heuristic nature of the procedure and to its homogenization methods (see, e.g., Sanchez-
limited range of applicability. Palencia, 1980) must in any case be used. A
An alternative to the use of LEFM, NLFM, homogenization method is a procedure which
and PFA is to formulate constitutive models in substitutes the mechanical description of an
the framework of continuum mechanics able to heterogeneous material with another referred
simulate damage processes for composite to an equivalent (in some sense) homogeneous
materials at various scales. This naturally leads one. Homogenization methods are used widely
to the use of nonlinear, irreversible, constitu- in the elastic range for composite materials and
tive laws in the classes of softening plasticity structures and can bridge the gap between
(or viscoplasticity) or continuum damage different scales. Unfortunately, particularly
mechanics (CDM) (Lemaitre and Chaboche, when describing damage processes in compo-
1990). The use of CDM for the simulation of sites, it is not in generally possible to make use
composite behaviors started approximately in of one homogenization step only. The simula-
the 1980s (see, e.g., Ladevèze, 1983; Talreja, tion must in fact simultaneously describe in
1989). There are various ways to formulate detail what happens at small scales and give an
damage models for composites, depending on overall response of the composite structure.
the scale at which one wants to look at the The procedures which have been proposed in
material. Typically, it is possible to follow a the literature, therefore, make use of multistep
mesomodel approach and (for the case of homogenization (typically from micro- to
laminates) introduce models for single laminas meso- and from meso- to macroscales) thus
which take into account the main damage giving rise to multiscale approaches (see, e.g.,
mechanisms like fiber rupture, fiber–matrix Fish et al., 1999; Ghosh et al., 2000). These
debonding, and transverse matrix cracking. approaches, combined with suitable constitu-
CDM for composites are well developed and tive descriptions for damage processes, can in
can in some cases represent an ideal tool for the principle solve the problem of numerical
complete numerical simulation. There are, simulation of fracture and damage processes
nevertheless, a number of difficulties which in composites. Once again the main drawback
should be overcome in order to transform is represented by the computational effort,
CDM for composites into routinely used although this is greatly reduced with respect to
simulation tools, among them the overall that implied in an overall description.
computational cost, the difficulties in identify- The use of numerical homogenization tech-
ing material parameters which appear in niques has been combined with parameter
complicate constitutive laws, and the necessity identification procedures (Giampieretti et al.,
to obtain mesh insensitive results. 2000; Bolzon et al., 2002b) in order to obtain in
Overview of the Main Fracture and Damage Mechanisms in Composite Materials and Structures 465
a systematic way micromechanically based subsequent description of computational meth-
constitutive laws for composite materials; this ods for damage and fracture. Particular atten-
approach has opened new possibilities for the tion is dedicated to the main damage
simulation of damage and fracture in compo- mechanisms like matrix cracking at the materi-
sites and can be used in combination with al level and delamination at the structural level.
multiscale methods. In Section 3.09.3 CFM methods are described
To conclude the brief overview of damage and adapted to the mechanics of composite
and fracture methods for composites, it is materials and structures; the simulation of
important to mention the transformation field crack propagation via the virtual crack closure
methods put forward by Dvorak and co- (VCC) (or opening) methods are discussed in
workers (Dvorak, 2000; Bahei-El-Din and detail. Section 3.09.4 is devoted to the descrip-
Dvorak, 2000; Dvorak and Zhang, 2001; tion of PFA for composites, while in Sections
Suvorov and Dvorak, 2001) in which the 3.09.5 and 3.09.6 the discussion refers to CDM
development of damage is related to the and IDM for composites, respectively. In
presence of damage-equivalent eigenstrains Section 3.09.7 global strategies for numerical
generating residual stresses which in turn can FE analysis are described in which use is made
be the reason of damage. of CDM and IDM; particular attention is
After the above introductory remarks, let us focused on path following methods able to
now come back to the purpose and contents of handle sharp snap-back global behaviors due
this chapter. As mentioned in the first para- to the globally brittle character of the response.
graph, the chapter will focus on the computa- Section 3.09.8 contains a discussion of the
tional aspects relevant to fracture and damage principles covered in the present chapter.
in composites. As should be clear on the basis
of what has been discussed up to this point, 3.09.2 OVERVIEW OF THE MAIN
this subject is extremely large and difficult to FRACTURE AND DAMAGE
discuss. A choice has, therefore, been made in MECHANISMS IN COMPOSITE
order to give an overview of what can be done MATERIALS AND
with numerical simulation tools and to focus STRUCTURES
on some important procedures only: LEFM,
PFA, CDM, and IDM. Classical methods for In this section, the main damage and
the analysis of composites which can be fracture mechanisms which characterize the
applied also in the linear elastic range, e.g., mechanical behavior of composite materials
homogenization and the use of transformation and structures are discussed (see, e.g., Hull and
fields will not be discussed extensively. Clyne, 1981; Pagano, 1989; Daniel and Ishai,
The discussion will refer almost exclusively 1994; Sierakowski and Newaz, 1995; Argon,
to modeling and numerical simulation; experi- 2000; Hart-Smith, 2000).
mental results and observations are covered As already stressed, the purpose here is to
only partially in the reference list. Knowledge give basic ideas of the failure processes in
of basic notions concerning FE analysis, FM, composites in order to motivate and guide the
plasticity, viscoplasticity, and damage me- development of numerical procedures de-
chanics will be assumed. For these subjects scribed in the subsequent sections. Restrictive
the reader can usefully refer to other chapters choices are, therefore, made: attention is
of the present book, or to other books of the focused on long fiber, polymer matrix-glass
present comprehensive treatise. or carbon fibers laminates; some damage
Some limitations will be assumed through- mechanisms which are nor strictly related to
out: mainly, polymer matrix and glass or mechanical phenomena are not considered;
carbon long fiber composites (in particular these include, e.g., moisture absorption, envir-
laminates) will be considered; only determinis- onment influence, chemical reactions.
tic methods will be discussed for the analysis of The choice of laminates for the discussion on
damage and fracture mechanisms induced in damage and fracture mechanisms is due to the
composites by mechanical loading; static load- importance that they have acquired in engineer-
ing will be almost exclusively considered; the ing applications. Laminates are largely used in
small strain and displacement hypothesis will aeronautical and naval engineering. Their use is
be assumed (i.e., delamination buckling will be also spreading rapidly in other areas, such as
mentioned but not discussed analytically). mechanical engineering and civil engineering,
The chapter is organized as follows. Section particularly for reinforcement and retrofitting
3.09.2 gives an overview of fracture and (see, e.g., Karbhari and Zhao, 2000).
damage mechanisms in composite materials The basic mechanical behavior of composite
and structures with particular reference to long laminates can be better understood by first
fiber laminates, thus setting the basis for the considering the behavior of single laminas. A
466 Damage and Fracture Mechanics Techniques for Composite Structures
unidirectional single lamina loaded in tension at 901 with respect to the fiber direction. In this
in the fiber direction (longitudinally) has a case the behavior is matrix-dominated, when
behavior mainly governed by the fibers (see the fiber–matrix interface is strong, the failure
Figure 1). Therefore, the response is usually occurs due to transverse matrix cracking (see
elastic-brittle with almost no plasticity. The Figure 5). The matrix also dominates the
main collapse mechanism consists in the fiber behavior of single laminas loaded in plane
rupture, this means that the fiber strength shear, i.e., with a load direction at 451 with
governs the lamina strength in the longitudinal respect to the fiber direction. Matrix micro-
direction. Other damage mechanisms can, cracking is, therefore, another important da-
however, influence the strength and limit its mage mechanism in composite materials which
value; these are in particular related to fiber– is more fully discussed in Section 3.09.2.3.
matrix debonding, i.e., to the degradation of The basic failure modes revealed the analysis
the fiber–matrix interface. Figures 2 and 3 of single laminas are also relevant to the
show ruptures of fibers in tensile tests for glass behavior of multidirectional laminates. In
fiber reinforced plastic (GFRP) and carbon addition to them, in the case of laminates,
fiber/epoxy matrix composite specimens, re- other characteristic failure modes must be
spectively. taken into account. These last mainly are
In single unidirectional laminas loaded in related to interlaminar properties and to the
longitudinal compression the behavior is again possibility of laminas to loose their adhesion,
mainly governed by fiber properties, however, i.e., to delaminate. Hence, delamination is
the interaction with matrix behavior is stronger another important failure mode, which is
than in the case of unidirectional tension. If discussed in detail in Section 3.09.2.4.
overall buckling is excluded, and if the matrix In spite of the attempt to describe damage in
is not too brittle (see Section 3.09.2.3) the composites by means of single failure mechan-
collapse of the single lamina occurs due to local isms, it is important to observe that the failure
fiber instability (see Figure 4); the phenomenon process in composites is in general the result of
of fiber microbuckling is limited by the complicated interactions among the single
presence of the matrix and the good perfor- modes of failure. As an example, a multi-
mance of the fiber–matrix interface. In this directional laminate loaded in tension (see
respect, it can be said that compression Figure 6) can show a sequence of damage
strength of single laminas is a measure of the events prior to ultimate failure. In general,
ability of the matrix to support the fibers there are multiple matrix cracking within off-
against buckling and of the integrity of the axis plies, these can act as precursor to other
fiber–matrix interface. forms of damage like delamination, which in
From the above remarks it can be deduced turn occurs at free edges or near holes; fiber
that the main damage mechanisms which ruptures usually are the reason of final failure.
involve the single laminas loaded in the
direction of the fibers are fiber rupture, fiber- 3.09.2.1 Rupture of Fibers and Fiber
microbuckling, and fiber–matrix debonding. Microbuckling
These are the subject of more detailed discus-
sion in Sections 3.09.2.1 and 3.09.2.2. The good mechanical properties of compo-
Let us now consider a single lamina loaded sites are obtained through a suitable subdivi-
in transversal tension, i.e., with a load direction sion of the mechanical roles of matrix and

Figure 1 Single lamina loaded in longitudinal tension: rupture of fibers.

Overview of the Main Fracture and Damage Mechanisms in Composite Materials and Structures 467

Figure 2 Rupture of two different GFRP specimens under tension.

Figure 3 Rupture of a carbon fiber, epoxy matrix specimen under tension.

fibers (Figure 7), in order to minimize the therefore, high strength, high stiffness, and low
global weight. Fibers give the stiffness and density. Glass, carbon, aramid, and boron are
resistance to the material, the main load is in used.
fact taken by the fibers, their role is, therefore, Due to their low cost, glass fibers (E-glass,
of paramount importance to obtain good C-glass, and S-glass) are the most commonly
overall mechanical properties. The desirable used in low to medium performance compo-
characteristics of most reinforcing fibers are, sites. E (Electrical) glass fibers are widely used,
468 Damage and Fracture Mechanics Techniques for Composite Structures

Figure 4 Single lamina loaded in longitudinal compression: fiber microbuckling.

Figure 5 Single lamina loaded in transversal tension: matrix cracking.

Figure 6 Complex damage interactions in a holed plate loaded in tension: micrograph of a section of a
½03 =7452 =90s specimen at 92% of the rupture load (source Allix et al., 1998). Reproduced with permission
from Elsevier.

they have good strength and stiffness, good

resistance to weather effects, and good elec-
trical properties; C (Corrosion) glass fibers
have better resistance to corrosion but lower
strength; S (Strength) glass fibers have higher
strength and stiffness. In general terms, glass
fibers have good tensile strength and low cost,
but unfortunately their stiffness is not very
high and they suffer from low fatigue endur-
ance and property degradation due to severe
hygrothermal conditions. Glass fibers show
isotropic behavior, typical diameter values are
8–15 mm. It is important to remember that glass
Figure 7 Micrograph of a unidirectional laminate fibers usually are coated with an emulsified
cross-section (source Allix et al., 1998). Reproduced polymer called size, the roles of which are to
with permission from Elsevier. protect the surface of the fibers from damage
Overview of the Main Fracture and Damage Mechanisms in Composite Materials and Structures 469
Table 1 Typical fiber properties.
Density Young’s Tensile strength Failure strain
Fiber (Mg m
3) modulus (GPa) Poisson’s ratio (GPa) (%)
E-glass 2.56 76 0.22 2.0 2.6
High strength 1.75 Axial 230 0.20 3.4 1.1
Carbon Radial 20
Aramid 1.45 Axial 130 0.35 3.0 2.3
Radial 10
Source: Hull and Clyne (1996).

and to provide the possibility of forming

chemical links between the glass fiber surface
and the matrix.
Carbon fibers are used in high-performance
composites. There are many types of carbon
fibers. In general, they have high stiffness and
strength but high cost. Values of stiffnesses and
strengths vary depending on the processing
temperature. Usually, the increase in stiffness is
obtained at the expense of strength. Carbon
fibers are highly anisotropic, with carbon atom Figure 8 Fiber buckling and kink band formation.
groups aligned parallel to the fiber axis. This
means that the stiffness of the fiber in axial
direction is much higher than that in radial be observed that the fiber strength in compres-
direction, usually the transverse modulus, sion will depend on geometrical properties like
perpendicular to the fiber axis, is B3–10% of the fiber aspect ratio (L/d), where L is the fiber
the axial modulus. A typical diameter value for length and d its diameter, and mechanical
carbon fibers is B8 mm. Besides their high properties of the fiber and of the matrix which
mechanical properties, carbon fibers have high have an influence on the local stiffness of the
resistance to temperature; they can be heated fiber during bending. Usually, the phenomenon
and used at temperatures up to 2,0001, of of local fiber buckling is accompanied by the
course this relevant property cannot be formation of kink bands in the part of the fiber
exploited in polymer matrix composites, due that has compressive stresses (Figure 8), in
to the limited resistance to temperature of most particular in the case of aramid fibers. More
matrices (B2001), but it can be exploited information concerning the behavior of fibers
advantageously in carbon/carbon composites. and their damage processes can be found, e.g.,
Other kinds of fibers, less used, are quoted in Kelly and Zweben (2000).
here, such as aramid and boron. Aramid fibers
are organic, they show an anisotropic beha- 3.09.2.2 Fiber–Matrix Debonding
vior, have high stiffness and tensile strength,
but low compressive strength and high moist- Fiber–matrix interfaces play a central role in
ure absorption. Boron fibers have high stiffness the mechanical behavior of composite materi-
and high cost. Typical values of fiber mechan- als; they must transfer the load from matrix to
ical properties, taken from Hull and Clyne fibers in order to allow a correct behavior of the
(1996) are in Table 1. whole composite. Many important phenomena
Most fibers show behavior which can be may take place at the fiber–matrix interface
defined as elastic brittle, i.e., the stress–strain which tend to promote plastic deformation of
response is linear elastic until the fiber breaks the matrix and can influence the onset and
(see, e.g., Daniel and Ishai, 1994). nature of failure. The mechanical failure of the
It is of paramount importance to distinguish fiber–matrix interfaces strongly alter the beha-
between the tensile and compressive behavior vior of the composite and can be the cause of
of fibers. Fiber strength in tension can be development of serious damage processes.
considered as a real fiber property (see values In general terms, to have high modulus and
in Table 1); on the contrary, fiber strength in high strength one needs a good adhesion
compression is highly limited by the risk of between fiber and matrix. The adhesion resis-
buckling which in turn depends on the tance is mainly due to van der Waals forces
boundary conditions and on the lateral con- and can be influenced by many factors such as
finement given to the fiber by the matrix and the interdiffusion and the chemical reactions,
other geometrical properties. In general, it can the electrostatic attraction, and the possibility
470 Damage and Fracture Mechanics Techniques for Composite Structures
of mechanical keying. Moreover, the nature of between fibers, stabilize fibers when loaded in
the interfacial contact can be strongly influ- compression, increase the resistance to impact
enced by the presence of residual stresses. damage, and determine the out of plane
In the literature a number of tests have been properties of laminates.
proposed to measure the interface resistance, When the matrix is damaged, the above
among them are quoted here the single fiber tasks cannot be accomplished properly and the
pull-out, and the single fiber push-out or push- mechanical resistance of the composite materi-
down tests. The basic idea of these tests is to al can be seriously altered. Another conse-
promote a debonding process between fiber and quence of matrix damage can be the initiation
matrix by moving the fiber with respect to the of other serious damage mechanisms which
matrix. In making use of the experimental characterize, e.g., composite laminates, a typi-
results coming from these tests two main cal example of this is delamination (see Section
remarks should be taken into account: the 3.09.2.4) which can be triggered by matrix
properties of the fiber–matrix interface in real cracking.
composites could be quite different from those When discussing the matrix behavior in
measured in the tests and the interface resistance composites, it is important to have in mind
to shear and to tensile stresses are generally the fact that it is in general very difficult to find
different and uncorrelated. Typical values of in situ matrix mechanical properties. In other
debonding shear stress for polymer matrix words, the behavior of the matrix in a fiber-
composites derived from pull-out tests are in reinforced composite is not necessarily the
the range 5–100 MPa (Hull and Clyne, 1996). same as that of the pure material of which
As said before, it could appear that in order the matrix is made. This fact is particularly true
to improve the performance of a fiber-rein- for polymer matrices as will be stressed further
forced composite one should increase the in the following.
resistance to debonding of the fiber–matrix Polymer matrices can be subdivided into two
interface. This fact is not always true, in main classes: thermosetting and thermoplastic.
particular in the case of brittle matrices. A As general distinctive properties thermosetting
typical example of this situation is represented resins can be used approximately for tempera-
by first generation thermosetting matrices with tures below 1201, they have good strength, and
low toughness values and a very brittle low fracture toughness; thermoplastic resins
behavior; in this case in order to enhance the are more expensive, they can be used for
toughness of the composite as a whole, it is temperatures higher than 2001, have good
better to have a low toughness fiber–matrix strength, and high toughness.
interface; this will allow the composite to Thermosetting resins are the most used in
dissipate energy due to cracks propagating engineering applications, among them we can
along the fiber–matrix interfaces with friction find polyester and epoxy resins, the latter being
and mechanical interlocking effects. standard for aerospace applications. The so-
In order to improve the resistance to fiber– called third-generation epoxy resins are in use;
matrix debonding, surface treatment can be they have toughness which can reach quite high
applied to the fibers, and the use of silane values. Among thermoplastic resins it can be
agents can strongly improve the adhesion. For mentioned the poly-ether-ether-ketone (PEEK)
example, in Figures 9(a) and (b) are shown two which has been used since early 1980s; another
SEM micrographs of a composite made by kind is the poly-ether-imide (PEI).
empty glass spheres embedded in a polymer Typical values of the mechanical properties
matrix, called syntactic foam, previously tested of polymer matrices, taken from Hull and
in tension. Figure 9(a) concerns the syntactic Clyne (1996), are in Table 2. From the data
foam in which the microspheres have been collected in Table 2 it can be appreciated that
introduced without any surface treatment, the main difference between thermosets and
while Figure 9(b) concerns the syntactic foam thermoplastic matrices is the value of strain to
produced with microspheres treated with a failure; as already observed, thermoset resins
silane agent. From the comparison of the two are brittle materials, while thermoplastic ones
figures it can be appreciated how the rupture can undergo appreciable plastic deformation.
mechanisms have changed greatly; moreover, Another important difference between thermo-
the global resistance of the composite has setting and thermoplastic resins concerns their
almost been doubled. resistance to high temperatures, which is much
higher in thermoplastic polymers.
3.09.2.3 Matrix Microcracking Mechanical properties of polymer matrices
show in general a remarkable rate dependency,
The matrix has many important tasks in this can affect elastic and inelastic properties,
composite materials: it must transfer stresses e.g., the fracture toughness.
Overview of the Main Fracture and Damage Mechanisms in Composite Materials and Structures 471

Figure 9 Syntactic foam: epoxy resin filled with hollow glass microspheres: (a) without sylanization process,
tensile strength 26 MPa, and (b) with sylanization process, tensile strength 46 MPa (courtesy of Palumbo and
Tempesti, University of Brescia).

Table 2 Typical matrix properties.

Density Young’s Poisson’s Tensile Failure strain
Matrix (Mg/m3) modulus (GPa) ratio strength (GPa) (%)
Thermoset epoxy resin 1.1–1.4 3.0–6.0 0.38–0.40 0.035–0.1 1.0–6.0
Thermoplastic PEEK 1.26–1.32 3.6 0.30 0.17 50.0
Source: Hull and Clyne (1996).

As already observed in the introduction to vant to the matrix behavior; it is in many cases
Section 3.09.2, damage mechanisms which the first form of damage which develops in
develop in the matrix dominate the composite laminate composites. In a cross-ply laminate, it
behavior in laminas loaded in the direction generally initiates in 901 plies forming cracks in
orthogonal to the fibers or in laminas loaded in the direction orthogonal to the loading (trans-
plane shear. verse cracking), parallel to the fiber. The value
Matrix microcracking can be considered as of loading at matrix crack initiation depends
the most important damage mechanism rele- not only on the properties of 901 plies but also
472 Damage and Fracture Mechanics Techniques for Composite Structures

Figure 10 Possible patterns of matrix microcrack-

ing. (a) [01/901] periodic array, (b) [901/01] antisym-
metric or staggered, and (c) curved or oblique
microcracks.

on those of adjacent 01 plies. The importance

of matrix cracking is due to the fact that
generally it triggers other damage mechanisms,
such as delamination, which are the reason of
complete failure of the composite structural
member (Wang and Karihaloo, 1994).
Matrix cracking is influenced by the lami-
nate lay-up; in Figures 10(a) and (b) are shown Figure 11 Difference in Poisson’s ratio for two
possible patterns of transverse cracking in the lamina.
matrix, in two meaningful cases of [0/90]s
laminates and [90/0] laminate. In the first case, dropping tools and/or, runway debris, imper-
cracks are distributed in a periodic fashion in fections in the manufacturing process, rapid
the internal 901 layer, while in the second case variations in geometry, and edge effects.
they are staggered in an antisymmetric way, in Delamination generally is caused by interlami-
the external 901 layers. Figure 10(c) shows nar stresses which arises in laminates, mainly at
another phenomenon which is related to the free edges, around holes or nearby internal
matrix cracking; cracks initially develop in pre-existent flaws. Interlaminar stresses in turn
the internal 901 layer, then delamination occurs arise due to mechanical properties mismatch
between the 01 and 901 layers, this in turn is a between adjacent layers. As an example let us
reason for the development of curved cracks in consider two laminas loaded in tension, the first
the transversal direction. The usual tests for in the direction orthogonal to the fibers, the
matrix microcracking are uniaxial tension on second in the fiber direction (see Figure 11).
single laminas or laminate. Another possibility Poisson’s ratios of the two laminas are differ-
is represented by bending tests in which 901 ent, being the lateral contraction of the first
layer are on the tension side of the laminate. lamina governed by the fiber stiffness. Due to
More information concerning the behavior of Poisson’s ratio mismatch, when the laminas are
matrix materials and their damage processes glued together to form a laminate, interlaminar
can be found, e.g., in Kelly and Zweben (2000). shear stresses must arise for tension loading in
order to preserve geometrical compatibility.
3.09.2.4 Delamination In the presence of free edges, delamination
starts mainly due to tensile loading and
One of the most significant problems asso- propagates from the edge toward the interior
ciated with continuous fiber composites is their of the laminate (Figure 12(a)); internal defects
tendency to delaminate. This problem repre- can propagate due to delamination which can
sents the most prevalent type of life limiting be activated mainly by compressive loading
failure of advanced composites (see, e.g., Pipes and subsequent local buckling of the delami-
and Pagano, 1970; Pagano and Pipes, 1973; nated area (Figure 12(b)). Another important
Garg, 1988; Herakovich, 1989; Pagano, 1989; kind of loading which is responsible for
Bolotin, 1996; Pagano and Schoeppner, 2000) delamination is represented by impact of
and often is referred to as the most feared objects on the external surface of a laminates
failure mode to attack a structural composite. (Figure 12(c)).
Among the main reasons for a delamination Delamination processes all involve crack
process are low-energy impacts caused by propagation in the so-called interlaminar zone.
Overview of the Main Fracture and Damage Mechanisms in Composite Materials and Structures 473

Figure 13 Delamination crack propagating in a

(722.51) interface in the case of an edge delamina-
tion tension (EDT) test.

Frassine et al., 1993; Corigliano and Ricci,

2001)
Due to the importance of delamination in
the assessment of structural composite resis-
tance, many attempts have been made in order
to enhance interlaminar fracture properties in
laminate composites.
Figure 12 Typical delamination phenomena: (a) The first provision is to make use of matrices
free-edge delamination; (b) buckling-induced delam- with enhanced fracture toughness; as observed
ination; and (c) impact induced delamination.
in Section 3.09.2.3, thermoplastic matrices like
PEI are the ideal candidates in this respect.
Depending on the fiber lay-up of the layers High performance thermoplastic matrix com-
adjacent to the interlaminar zone, the crack posites, therefore, have been produced. The use
process can mainly interest a resin-rich region of tough matrices is in some way limited by a
or can also be influenced by the presence of still-open question related to the difficulties in
fibers which can bridge the crack or make the completely transferring the fracture toughness
crack deviate and possibly change the inter- of a very tough matrix to the composite (Hull
laminar layer. Another important feature of and Clyne, 1996). This difficulty seems to be
delamination processes is that they usually due to the limited development of the matrix
involve mixed-mode propagation in real lami- plastic process zone which contributes to the
nates. In contrast with what happens in homo- energy dissipation in ductile fracture processes
geneous materials, mode II and mixed-mode in tough matrices. This observation can explain
crack propagation are in fact possible due to the use of another, more effective, provision
the constraining effect of layers. Figures 13–16 called interleaving which consists in the intro-
show various examples of delamination. duction of layers of additional resin between
Any attempt to study and possibly model the fiber reinforced layers. The additional resin
delamination processes must start from a layers have as a main consequence the possi-
quantification of the energy necessary to make bility to completely exploit the fracture tough-
delamination cracks propagate, i.e., the inter- ness of the matrix; interleaving is, therefore,
laminar fracture energy Gc must be determined most effective in improving toughness when
experimentally. This task usually is carried out combined with a tough matrix system. Draw-
by means of interlaminar fracture tests (see, backs of the interleaving technique are related
e.g., Pagano, 1989; Williams, 1989; Allix et al., to the reduction of other mechanical properties
1995; Amrutharaj et al., 1996; Todo and Jar, of the laminate like the global stiffness.
1998; Schön et al., 2000) such as those High-performance composites have been
represented in Figure 17. The experimental produced with enhanced delamination resis-
determination of interlaminar toughness of a tance by means of the introduction of some
composite is a nontrivial task; particular devices which create a direct connection
difficulties are related to the quantification of between laminas in the transverse direction.
fracture energies in mixed-mode conditions Z-pinning and fiber stitching (Cox et al., 1996;
and to possible rate-dependence of the fracture Auricchio and Sacco, 1998) are among the
energy in polymer matrix composites (see, e.g., most effective. Additional information on the
474 Damage and Fracture Mechanics Techniques for Composite Structures

Figure 14 X-ray photographs (  1.5) of a M55J/M18 [03/7452/90]s holed-specimen (rupture load:

430 MPa), showing delamination around the hole: (a) 55% of rupture load; (b) 86% of rupture load; and
(c) 99% of rupture load.

Figure 15 Examples of delaminations in fiber-glass-syntactic foam sandwich composite beams: (a)

delamination between skin and core of a specimen tested in TPB conditions and (b) delamination between
skin and core of a specimen tested in edge-compression.
 Computational FM for Composites 475

Figure 16 Delamination and fiber rupture around connections in a fiberglass pipe.

Figure 17 Interlaminar fracture tests: (a) mode I double cantilever beam (DCB); (b) mode II end notched
flexure (ENF); and (c) mixed mode asymmetric end loaded split (aels).

phenomenology of delamination processes can As more thoroughly discussed in other

be obtained in Garg (1988), Pagano (1989), chapters of this series, computational FM is a
and Pagano and Schoeppner (2000). well-developed branch of FM which couples
the theory of FM with numerical simulation
3.09.3 COMPUTATIONAL FM FOR tools (usually FEs and/or boundary elements)
COMPOSITES for assessing the safety of an already existing
crack and/or simulating the process of crack
The purpose of this section is to discuss propagation. This description is consistent with
computational procedures based on the use of the following well accepted definition of FM
FM for the description of fracture processes in (Kanninen and Popelar, 1985): ‘‘Fracture
composite materials. In particular the main Mechanics is an engineering discipline that
problems relevant to the use of FM in FE quantifies the conditions under which a load-
simulations for composite materials and struc- bearing body can fail due to the enlargement of
tures are presented and discussed. a dominant crack contained in that body.’’
476 Damage and Fracture Mechanics Techniques for Composite Structures
The safety assessment of a crack can be In GFRP composites the fracture processes
carried out by identifying a critical value for are usually macroscopically brittle (or quasi-
typical quantities of FM like the stress intensity brittle); this means that, in general terms, the
factor (SIF, K), the energy release rate (ERR, tools of LEFM are well suited for fracture
G), the J integral (J); when the quantity chosen propagation studies, provided that some pecu-
for the description of the fracture process liarities are taken into account.
reaches its critical value the crack can propa- First, it must be observed that the intrinsic
gate. From a computational point of view the high heterogeneity of composite materials and
problem is in this case to correctly compute K, structures has up to now discouraged the use of
J, or G in the numerical model, given the shape boundary element methods (BEM), in spite
and position of the crack in the structural of being a very powerful tool for the simulation
component under study. Crack propagation of fracture processes in homogeneous mater-
can be simulated by coupling the use of K, J, or ials; this is the reason why in this and the
G with some procedure of mesh modification following sections the discussion is solely based
which has the task to modify the initial crack in on the FE method.
order to make it propagate. Another important point concerns the fact
Computational methods in FM have become that in fiber–matrix debonding and delamina-
very powerful and can be used to solve many tion we have in general mixed-mode crack
problems in 2D and 3D situations, for linear propagation of a crack between two dissimilar
and nonlinear material behaviors; usually they media, i.e., the presence of interface cracks.
are characterized by simplicity, robustness, and The studies of interface cracks in the frame-
computational cost-saving with respect to work of LEFM (see, e.g., the works by Rice
other methods which can be used to simulate and Sih, 1965; Rice, 1988; Suo and Hutch-
fracture processes. Nevertheless, they suffer inson, 1990) concentrated on the problem of
from the drawbacks which are intrinsic in the determining the stress state around the crack-
FM theory, the principal of which being tip of a crack between two elastic dissimilar
the fact that it is not possible to study the materials. In contrast to the homogeneous
phenomena of defect nucleation and crack case, it was found by Williams (1959) that the
initiation, i.e., the crack is supposed to be stresses possess a pronounced oscillatory char-
already existent. Other problems related to the acter, while remaining proportional to the
use of FM are relevant to the simulation of inverse square root of r, where r is a radial
mixed-mode situations, of complicated fracture coordinate from the crack-tip (Figure 18); one
pattern with diffuse and/or multiple cracks, can have, e.g., for a stress component s
particularly in 3D situations and to the
presence of diffuse material nonlinearities as 1
sEpffiffi sinðbc log rÞ ð1Þ
plasticity. r
The main alternatives to the use of computa-
tional FM are represented by CDM and by where bc (bielastic constant) is a parameter
cohesive crack models or, more generally, related to the elastic constants of the dissimilar
interface models which will be the subject of media which are on the two sides of the crack
Sections 3.09.5 and 3.09.6, respectively. These (Rice and Sih, 1965). It should be remarked
approaches can overcome, at least partially, that some authors consider the oscillatory
the main limitations related to the use of stress and displacement solutions nonphysical,
computational FM. By means of CDM it is (at as they imply interpenetration of the crack
least in principle) possible to simulate fracture surfaces.
processes from their beginning, while the use of As far as the SIF KI, KII are considered, they
interface models allow to describe in a numer- can still be evaluated in a way similar to that of
ical model the presence of discontinuities in the the homogeneous case but they lose the usual
displacement field and, therefore, to simulate simple interpretation. In fact, KI and KII will
fractures explicitly. A combined use of CDM
and interface models allow in some cases to
build the most versatile simulation tools.
Returning to the subject of this section, it
can be said generally that, among the damage
mechanisms which occur in a composite
material and described in Section 3.09.2, those
which can be studied more advantageously by
means of computational FM are fiber–matrix
debonding, matrix microcracking, and, most of
all, delamination. Figure 18 Crack tip between two dissimilar media.
 Computational FM for Composites 477
both depend on symmetric (i.e., such that mesh in order to make a crack propagate in a
mode I is induced) and skew-symmetric (re- FE or BE discretized body.
sponsible for mode II) loading components; as Procedures for the evaluation of the ERR in
observed by Rice and Sih (1965), this means a composite material or structure are discussed
that in the application of a fracture criterion it in Section 3.09.3.1; Section 3.09.3.2 is dedi-
is necessary to assume that a function of KI and cated to the discussion of fracture propagation
KII will cause the crack to grow upon reaching criteria in pure and mixed mode, while in
some critical value Section 3.09.3.3 possible procedures for mesh
modification are described.
fK ðKI ; KII Þ ¼ fcr ð2Þ Finally, Section 3.09.3 will conclude with a
brief discussion of FFM methods (Hashin,
The above-mentioned additional difficulties 1996; Nairn, 2001), which seem to be well
have in general discouraged researchers from suited for the study of matrix microcracking.
making use of SIF and fracture criteria based
on them for highly heterogeneous materials
like composites. The majority of applications 3.09.3.1 Numerical Methods for the
of Computational FM for composites, there- Evaluation of the ERR
fore, is based on the use of the ERR G which is The ERR G, whose first definition dates
more consistent with the analytical models in back to the work of Griffith, is defined from a
use than K. global energy balance as
The use of the ERR circumvents the problem
of correctly reproducing the stress state near dP dðUen
Wex Þ
the crack-tip; the safe or critical state of the G¼
¼
ð4Þ
dA dA
crack is in fact based on energetic quantities.
Mixed-mode situations, which are common, where P denotes the global potential energy of
for instance, in delamination processes, can be the system under consideration, Uen is the
governed by a fracture criterion of the kind elastic strain energy or, more generally, the
expressed by Equation (2), in terms of the thermoelastic internal energy, Wex is the
single mode contributions GI and GII to G: external work, dA is an increment in the total
cracked area. A crack can propagate when G
fG ðGI ; GII Þ ¼ fcr ð3Þ equals a critical value called fracture energy Gc.
From the definition of Equation (4) this means
Another important point should be men- that when the loss in potential energy P
tioned; this concerns the determination of the subsequent to an increment of crack propaga-
crack propagation direction, which in homo- tion dA is sufficiently high, this energy can be
geneous materials (or macroscopically homo- spent to make the crack propagate. In the case
geneous, like concrete) is a major problem. In of dynamic phenomena Equation (4) should be
these materials cracks tend to propagate in modified by adding to the elastic strain energy
pure mode I: cracks loaded in mixed mode, U the kinetic energy of the system.
after a certain amount of crack propagation When the loading is represented by a unique
deviate from their path in order to recover a force P , by applying the Clapeyron theorem it
situation of mode I. In composite materials is possible to obtain the following useful
cracks are in general constrained by physical expression for G:
interfaces, like the fiber–matrix and the lamina-
lamina ones. Hence, in composite materials it is P2 dC
possible to have crack propagation in pure G¼ ð5Þ
2 dA
mode or mixed mode but the crack path is not
completely unknown; it can in a first approx- where the compliance C ¼ u=P is given by
imation be said that cracks propagate along the ratio of the displacement u of the point
physical interfaces. of application of P, in its direction, and the
From the above considerations, it appears force P.
that, in order to simulate crack propagation Another very useful expression was obtained
processes in a composite by making use of the by Irwin by making the hypothesis that the
ERR, three main ingredients are necessary: an energy absorbed in the process of crack
efficient procedure for the numerical evalua- extension by an amount Da is equal to the
tion of the contribution to G of the single work required to close the crack to its original
modes I, II, and III; a crack propagation length. By introducing a polar coordinate
criterion of the kind in Equation (3) coupled system with the origin coinciding with the
with a criterion for the direction of propaga- extended crack-tip (Figure 19) this leads to the
tion and a numerical procedure to modify the following crack closure integral formula for G
478 Damage and Fracture Mechanics Techniques for Composite Structures
For plane strain and mode III, equations
similar to Equation (8) hold.
Equations (4)–(6) and (8) have been recalled
here because they form the basis of methods
for computing the ERR in FE computations
for composite materials and structures; these
are described below.
As in the whole chapter with reference to
other subjects, the methods briefly described
here are examples of existing methodologies
Figure 19 Definition of crack-tip coordinate system
which have gained much attention in the
and strain and displacement discontinuity distribu-
tion. literature. Other methods are not discussed
here only for the sake of clarity and concise-
ness; among these worth mentioning are the
in a plane situation:
works concerning the study of buckling in-
duced delamination processes of Storåkers and
G ¼ GI þ GII
Z Da co-workers (see, e.g., Storåkers and Andersson,
1 1988; Nilsson et al., 1993, 2001) in which the
¼ lim s3 ðDa
r; 0Þ½u3 ðr; pÞ dr
Da-0 2Da 0 ERR is derived in the context of von Karman’s
Z Da  and Mindlin’s nonlinear plate theories.
þ t13 ðDa
r; 0Þ½u1 ðr; pÞ dr ð6Þ
0
3.09.3.1.1 Virtual crack extension method
where ½u1 ðr; pÞ; ½u3 ðr; pÞ; rA½0; Da are the
relative sliding and opening displacements The virtual crack extension (VCE) method
between the crack faces behind the extended (also called stiffness derivative approach)
crack-tip and s1 ðDa
r; 0Þ; s3 ðDa
r; 0Þ; initially was proposed to calculate stress
rA½0; Da are stress distributions ahead of intensity factors and subsequently was general-
the crack-tip. ized for application in mixed-mode crack
In the case of homogeneous and isotropic propagation analysis (see, e.g., Sha, 1984;
elastic materials it is easy to correlate G to the Hwang et al., 1998; Yang et al., 2001). It
SIF K by means of a formula which can be basically consists in calculating G, starting
recovered by introducing in Equation (6) the from the modification in the stiffness of a
asymptotic stress and displacement fields; in rosette of elements around the crack-tip due to
the case of mode I it results a small amount of crack propagation. Let us
consider an elastic structural system discretized
K2 in space by means of finite elements, the
G¼ ; E n ¼ E plane stress potential function of the whole system can
En  ð7Þ
 then be expressed as
E n ¼ E= 1
n2 plane strain
P ¼ 12U T KU
PT U ð9Þ
where E and n are the Young’s modulus and
the Posson’s ratio of the material. In the case of where U is the nodal displacement vector, K is
anisotropic materials it is still possible to the stiffness matrix, and P is the vector of
obtain Equation (7), but this is more complex equivalent nodal loads, for the whole system.
and is dependent on the level of anisotropy. In In the VCE method G is computed as
particular, for orthotropic materials in plane
stress with the axes of material symmetry 1 @K @P
G ¼
UT U þ UT ð10Þ
aligned with the crack-tip coordinate axes of 2 @A @A
Figure 19, Sih et al. (1965) gave the following Equation (10) can be applied for the calcula-
expressions: tion of modes I and II strain ERR GI and GII
s ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rffiffiffiffiffiffi   separately, by distinguishing the displacements
1 E1 E1 UI which contribute to the symmetric mode I
GI ¼ KI2 pffiffiffiffiffiffiffiffiffiffiffiffiffi þ
n13
2E1 E3 E3 2G13 and those UII which contribute to the anti-
s
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rffiffiffiffiffiffi   ð8Þ symmetric mode II, respectively.
1 E1 E1 Equation (10) can be applied practically in a
GII ¼ KII2 pffiffiffi þ
n13
E1 2 E3 2G13 rather simple way by local modification of a
rosette of elements which are around the crack-
where E1, E3 are Young’s moduli for directions tip (see Figure 20). In this case only the
1, 3; n13 is the Poisson’s ratio in the plane 13, elements around the crack-tip contribute to
and G13 is the shear modulus in the plane 13. the variation of K, hence Equation (10) can be
 Computational FM for Composites 479

Figure 20 Virtual crack extension.

rewritten as (Yang et al., 2001) Figure 21 Global response of a DCB test (Equa-
tion (15)) computed via the CM (2h ¼ 6 mm; B ¼
N 20 mm; E ¼ 1  105 MPa).
1 X Nc
1 X cf

G¼
U Ti DK i U i þ U T DF j ð11Þ
2DA i¼1 DA j¼1 j
which the compliance can be calculated easily
by means of the classical beam theory. A
where Nc denotes the number of elements typical application of the CM is represented by
around the crack-tip; Ncf the number of the study of composite interlaminar fracture
elements around the crack-tip with applied specimens (see, e.g., Gillespie et al., 1986;
force; Ui is the nodal displacement vector of Williams, 1988, 1989; Allix et al., 1995; Allix
element i around the crack-tip; DKi is the and Corigliano, 1996). For example, let us
change of the ith crack-tip element stiffness consider a double cantilever beam (DCB)
matrix due to the VCE by DA; DFj is the specimen of width B (Figure 17(a)) in a plane
change of the nodal force vector of the jth schematization. By applying the classical beam
crack-tip element due to the VCE. The theory, the opening displacement d due to a
contributions of different modes of fracture load P applied at the two free ends of the
in the expression of G can be recovered by cantilever beams is given by
decomposing the displacements in symmetric
and anti-symmetric with respect to the crack 2 a3
d¼ P ð13Þ
propagation direction. This method has the 3 EJ
advantage to allow mode partitioning, which is
an important issue in mixed-mode propagation where a is the crack length and J is the second
conditions of composites. moment of area of one arm of the DCB. The
compliance C and the ERR can then be
computed from Equation (13):
3.09.3.1.2 Compliance method
d 2a3 P2 a2
Equation (5) is the basis of the compliance C¼ ¼ ; G¼ ð14Þ
P 3EJ EJB
method (CM), in which the total strain ERR
can be calculated numerically in terms of the By applying a Griffith propagation condition,
rate of change of compliance with crack from Equation (14) it is also possible to derive
extension. For a finite increment of crack the load-opening displacement plot during
surface DA, one has crack propagation
rffiffiffiffiffi
P2 DC 3=4 1=4 2
G¼ ð12Þ P ¼ ðBGIc Þ ðEJ Þ ð15Þ
2 DA 3d

Given a FE model of the system under study, where GIc is the fracture energy in mode I. An
Equation (12) can be applied by making two example of load–displacement response of a
computations for the same load level P DCB specimen as given by the Equation (15) is
corresponding to a crack with surface A and represented in Figure 21 at varying value of
an augmented crack with surface A þ DA. fracture energy GIc, for values of parameters
Equation (12) has been applied widely in the given in the figure legend. The CM in the form
analytical study of simple cracked systems, in presented here has the advantage of great
480 Damage and Fracture Mechanics Techniques for Composite Structures
simplicity. It can also be applied for the
interpretation of experimental data concerning
interlaminar fracture specimens. An obvious
limitation is represented by the single load
condition.

3.09.3.1.3 Virtual crack closure method

(VCC)
VCC is perhaps the most versatile among the Figure 22 Crack-tip in a 2D case for VCC
methods for computing the ERR; it was first technique.
adopted by Rybicki and Kanninen (1977). In
its first version, named crack closure technique
(CCT), it can be considered as a direct 1     
GI ¼ Fy5 Uy1
Uy2 þ Fy6 Uy3
Uy4
application of Equation (6); in this case two 2Da ð16Þ
FE analyses are done, the first corresponding 1
to the crack with original extension, the second GII ¼ ½Fx5 ðUx1
Ux2 Þ þ Fx6 ðUx3
Ux4 Þ
2Da
corresponding to a crack extended by an
amount Da. From the first analysis the nodal where the numbers 1–6 correspond to the
forces near the crack-tip are computed, from nodes in Figure 22, Fx ; Ux ; Fy ; and Uy are the
the second one the nodal displacements near nodal forces and displacements in directions x
the crack-tip, which allow for the computation and y, respectively. The same kind of formula
of opening and sliding, are obtained. The work can be used in 3D situations, where the ERR in
required to close the crack is then computed by mode III also can be computed (see, e.g., Shen
multiplying one half of the nodal forces with et al., 2001).
the corresponding displacements. Forces nor- By examining Equation (16) it can be
mal to the crack line, multiplied by opening observed that the application of the VCC
displacements, give the mode I ERR; forces technique to composite materials should in
parallel to the crack line, multiplied by sliding principle take into consideration the oscillatory
displacements, give mode II ERR. character of the crack-tip stresses for interface
The VCC technique has the advantage of crack in the presence of anisotropic materials.
limited computational cost with respect to Fortunately, the dominance of the oscillatory
CCT because it is based on a single FE analysis behavior is confined to a small region near the
only. It has, therefore, become more popular crack-tip, outside this region the VCC techni-
than the CCT; applications to composite que can be applied to yield reasonable values of
materials and structures can be found in G components. Moreover, there is no need to
Gillespie et al. (1986), Sukumar and Kumosa make use of special crack-tip elements, as the
(1993), Cantwell et al. (1999), and Shen et al. quarter-point with singular fields, also because
(2001). they show a square-root singularity which does
The basic assumption of the VCC method is not represent the true nature of the stress
that crack growth satisfies the hypothesis of singularity in the case, e.g., of a delamination
self-similarity; in the case of composite dela- front.
mination in 2D this can be reasonable, unless In the form briefly presented above, the VCC
fiber bridging and/or crack kinking out of the method can be applied to 2D and 3D linear
initial delamination plane occur. In the case of and nonlinear problems. It is, therefore, a
general 3D delamination the hypothesis of self- powerful tool for the study of delamination
similarity is questionable. Therefore, the appli- buckling problems in composite laminates.
cation of a VCC technique must be justified by It is interesting to observe that the computa-
showing that the values of ERR thus computed tion of the ERR via the VCC method can be
are not very different from those computed, coupled with the use of interface elements (see
e.g., via the CCT. Sections 3.09.6 and 3.09.7) (see, e.g., Allix and
Once the self-similarity hypothesis has been Corigliano (1996)). The reason of this comes
accepted, the ERR can be simply computed by directly from the Irwin’s Equation (6), in which
considering the nodal forces at the crack front G is computed from the knowledge of stresses
and the opening and sliding displacements and displacement discontinuities along the
calculated starting from the nodal displace- crack line which are computed by means of
ments behind the crack front; a direct applica- interface elements.
tion of Equation (6) gives the following As a closure of this subsection it should
equations for 2D problems: be recalled that other methods have been
 Computational FM for Composites 481
formulated and used for the analysis of elastic FE analyses. In order to be able to
fracture parameters in composite materials, assess the safety of a crack in a composite
like the comparative FE method used in Lotter structure, mixed-mode situations generally
and Santare (1991). must be considered and an empirical fracture
criterion of the kind given by Equation (3) is
3.09.3.1.4 Displacement correlation method necessary. An alternative to the general ex-
pression in Equation (3) is the following one:
A practical way to compute K for isotropic
materials is to apply a displacement correlation G ¼ G c ðc Þ ð17Þ
method (DCM) in which the results of a FE
analyis are compared to the theoretical results where G is the total ERR, Gc is the total critical
concerning the solution at the crack tip in ERR, and c is a phase angle or mode mixity
terms of displacements.
pffiffi In order to do this, the ratio which determines the different contribu-
coefficient of r term in Williams’s eigenfunc- tions of single modes to the fracture process.
tion expansion which depends on K, is Criteria of the kind in Equation (3) or (17)
compared to an analogous term obtained from generally have been proposed for the study of
a FE computation which depends on nodal delamination processes, but they could also be
displacements of nodes near the crack tip. The applied to other fracture processes like fiber–
method allows the determination of modes I matrix debonding.
and II SIF. Once the value of SIF has been A reasonable fracture criterion for compo-
determined, the ERR can be computed by sites must take into account the fact that cracks
applying Equation (7). usually initiate and also propagate in mixed-
In the case of anisotropic materials the mode conditions and that the value of fracture
DCM can still be applied by making use of energy is highly dependent on the degree of
the eigenfunction expansions for cracks given mode-mixity. In other words the fracture
by Sih et al. (1965) in which the square-root energy GIc in pure mode I is usually very
singularity is still present as in isotropic media. different from the fracture energy GIIc in pure
After having obtained the value of SIF, mode II which tends to be higher. Another
Equation (8) can be applied in order to important point to note is the fact that usually
determine the ERR. initiation and propagation values of critical
Quarter-point elements or other more refined fracture energy are different; this means that,
elements which incorporate the exact displace- in principle, two different kinds of fracture
ment field in the region surrounding the crack- criteria should be used for initiation and
tip can be used advantageously in the applica- propagation. Three examples of fracture criter-
tion of the above briefly-described DCM. ia are shown here. The first one belongs to the
In the form presented here the DCM can be category defined by Equation (3) and was
used for the study of fracture processes in proposed by Russell and Street (1985). It reads
composite materials treated as homogeneous  a  
anisotropic media, i.e., after having derived an GI GII b
þ ¼1 ð18Þ
equivalent homogeneous material by means of GIc GIIc
an homogenization procedure. It can also be
applied, as, e.g., in Gillespie et al. (1986) and In the above equations GIc and GIIc are the
Sukumar and Kumosa (1993) in the study of critical fracture energies in pure modes I and
unidirectional laminates. II, respectively, and a and b are two positive
More problematic is the application of the experimental parameters. Typical values for a
DCM to real interface crack problems like and b are: a ¼ 0:5; b ¼ 1; or a ¼ b ¼ 1: The
those met in the study of fiber–matrix debond- meaning of Equation (18) can be better
ing and general delamination processes. In appreciated on the plane ðGI =GIc Þ; ðGII =GIIc Þ;
those cases the problem of oscillatory terms as in Figure 23, where the pairs of values which
near the crack-tip, of their physical relevance satisfy Equation (18) are plotted for a ¼ b ¼
and of the consequent computational difficul- 0:5; a ¼ b ¼ 1; and a ¼ b ¼ 1:5: It can be
ties suggest to directly compute the ERR observed that for ao1 a negative interaction
without looking for the SIF value. between modes is reproduced, while for a ¼ 1
and a41 no interaction and positive interac-
3.09.3.2 Crack Propagation Criteria for tion effects are reproduced, respectively. Posi-
Composites tive and negative interactions are here meant as
increasing or decreasing the total critical ERR
Section 3.09.3.1 described briefly the various in mixed-mode situations with respect to the
methods for the determination of ERR and of values shown in pure mode in the ideal
its single mode contributions starting from situation of GIc ¼ GIIc :
482 Damage and Fracture Mechanics Techniques for Composite Structures
single mode contributions GI, GII, and GIII to
the total ERR.
An interesting comparison can be made also
in this context with the use of interface models
for the description of fracture processes in
composite materials (see Section 3.09.6). In
Allix and Corigliano (1996) and Alfano and
Crisfield (2001) it has been shown that condi-
tions of fracture propagation as governed by
interface models can be made equivalent to
failure loci of the kind given by Equation (18);
in other words an interface model can be
chosen also on the basis of the kind of mixed-
mode fracture conditions which should be
reproduced in the simulation.

3.09.3.3 Numerical Procedures for Crack

Figure 23 Mixed-mode failure locus. Advancement in Composites

The second mixed-mode fracture criterion The third ingredient necessary for the
for composite materials examined was pro- complete simulation of fracture propagation
posed by Williams (1989) and depends on a in composite materials is the choice of an
single parameter, in addition to GIc and GIIc efficient procedure for the crack advancement
in a FE mesh. This procedure must be coupled
GI GII GI GII with methods for the determination of the
þ þ ðg
1Þ ¼1 ð19Þ
GIc GIIc GIc GIIc crack driving force (Section 3.09.3.1) and crack
propagation criteria (Section 3.09.3.2).
The positive parameter g in Equation (19) As already anticipated in the introductory
governs the interaction between modes I and part of Section 3.09.3, the simulation of crack
II. It is interesting to notice that for g ¼ 0 the advancement in composite structures is in some
failure locus on the plane ðGI =GIc Þ; ðGII =GIIc Þ cases made easier by the a priori knowledge of
is a square domain with sides parallel to the the crack path. This is in particular true for 2D
axes, while for go1; g ¼ 1 and g41 the failure simulations of delamination or fiber–matrix
locus has the same kind of shape which is given debonding processes where the crack advances
by Equation (18) for a ¼ b41; a ¼ b ¼ 1; and along the interface between two layers or
a ¼ bo1; respectively. between fiber and matrix.
The third example of mixed-mode fracture When the crack path is known, usually the
criteria belongs to the class defined by Equa- more effective solution for the simulation of
tion (17), proposed by Hutchinson and Suo crack advancement is to make use of a node-
(1992), and is given by the following: release technique (see, e.g., Shen et al. (2001))
for an application to delamination growth).
G ¼ Gc ðcÞ ¼ GIc ½c1 f1 ðc; c2 Þ þ ð1
c1 Þf2 ðc; c2 Þ; This consists in releasing node pairs placed in
f1 ðc; lÞ ¼ 1 þ tanðð1
c2 ÞcÞ2 ; ð20Þ the initial mesh along the crack path and held
1 together by a multipoint-constraint or by
f2 ðc; lÞ ¼ fictitious elastic springs with a very high
1 þ ð1
c2 ÞsinðcÞ2 stiffness. This procedure can be also applied
in the case of 3D delamination processes, in
The three parameters necessary to define the which the delamination front is supposed to
above mixed-mode fracture toughness, propagate in an interface plane between two
ðGIc ; c1 ; c2 Þ; are determined by fitting Equa- adjacent layers of the laminate, an additional
tion (20) to test results. An interesting example difficulty is in this case the definition of the
of application of Equation (20) to analysis of crack front. Usually, the front is advanced by
delamination-buckling problems can be found releasing the nodes that have reached the crack
in Nilsson et al. (2001). growth criterion.
In 3D situations, equations analogous to The application of the node-release techni-
Equations (18)–(20) can be defined; these allow que is limited by the impossibility of knowing
the prediction of fracture propagation in exactly the amount of crack propagation and
general situations. The crucial point for their the shape of the crack front in 3D situations;
correct application is the determination of these in fact depend on the topology of the FE
 PFA for Composites 483
mesh in the vicinity of the crack-tip. In order to substituted by
have reliable evaluation of the ERR, the direct
use of the technique implies the definition of a DP DðUen þ Ken
Wex Þ
DG ¼
¼
ð21Þ
very refined mesh in the whole part of the DA DA
composite where crack is supposed to propa- A fracture event will occur when a critical
gate. value of DG will be reached
Alternative to the use of the node-release
technique, the moving boundary technique DðUen þ Ken
Wex Þ
DG ¼
¼ DGc ð22Þ
could be used (see, e.g., Nilsson et al. (2001)); DA
with this method the nodes along the crack
front in which the crack growth criterion is In practical applications of Equation (22) the
satisfied are advanced by moving them a small kinetic energy term is disregarded, or better
distance in the local normal direction to the becomes a part of an effective value of critical
front. Subsequently, the whole mesh is up- ERR or fracture toughness of the event
dated. This technique involves the use of a DðUen
Wex Þ DKen
remeshing procedure and could result in a DG̃ ¼
¼ þ DGc  DG̃c ð23Þ
DA DA
more time-consuming procedure.
It must be said that the study of 3D crack A typical application of FFM is the study of
propagation in composites by means of com- matrix cracking processes (see Section 3.09.2.3)
putational FM procedures usually implies the that occur in transverse layers (see Nairn,
use of very refined meshes all along the crack 2000).
front, this in turn makes the use of remeshing
procedures almost compulsory, thus increasing 3.09.4 PFA FOR COMPOSITES
a lot the computational burden.
An alternative to the use of remeshing The first attempts at simulating damage
procedures is represented by FE with em- processes in composite materials and structures
bedded cracks. This issue is further commented have been based on PFA, briefly described in
in the introduction to Section 3.09.7. this section. PFA are methods in which the
material properties of a composite, typically
elastic stiffnesses, are degraded once a failure
3.09.3.4 Basic Notions of FFM Applied to criterion such that of Tsai and Wu (1971) or,
Composite Fracture more frequently, that of Hashin (1980) is
locally satisfied. They have in general the
The term FFM was introduced by Hashin advantage of simplicity in the numerical
(1996) (see also Nairn, 2000, 2001) in an implementation and the drawback of a sort
attempt to describe some fracture processes of heuristic approach to degradation, which is
typical of composites that are not characterized governed by (as defined by Hashin (1980)) an
by continuous crack growth. Among these are appropriate stiffness change which is in fact
fiber rupture, matrix cracking, and instanta- not known. Due to the lack of information on
neous fiber–matrix debonding; these fracture the amount of stiffness reduction, the methods
events imply the formation and very rapid usually imply a sudden, brittle degradation.
dynamic propagation of a defect or a crack and This can be the reason of computational
are very difficult or impossible to observe difficulties.
experimentally. These events can all be con- In addition to the difficulty related to the
sidered instantaneous and cannot be treated definition of the appropriate stiffness change,
with the classical tool of LEFM. there is the problem related to stress concen-
In order to overcome this difficulty a trations which exist near notches or flaws and
possible alternative is to make use of the new which can diminish the validity of stress-based
concepts of FFM in which fracture events failure criteria.
associated to a finite increase in fracture area The analyses carried out by means of PFA
are studied. The basic tool of FFM is still the are, in general, mesh-dependent. This is a
ERR; two main differences with respect to common characteristic of all softening models
LEFM must be underlined. The first one is the used in FE computations without any special
fact that the quantity which is looked for is provision, in particular for those simulating a
now the total energy released per unit new very brittle behavior. Nevertheless in some
fracture area DA, where DA is a finite quantity; practical applications, like those relevant to
the second difference is the fact that now each plates and shells, the mesh-dependency is not
fracture event, even in static loading, may be critical, as observed by Crisfield et al. (1997).
associated with some kinetic energy DKen. Some of the drawbacks related to PFA can
Therefore, in FFM, Equation (4) should be be solved by making use of CDM, as discussed
484 Damage and Fracture Mechanics Techniques for Composite Structures
in Section 3.09.5; PFA remain in any case strengths in tension and compression, having 1,
methods which can be very effective in some 2, 3 as principal axes of orthotropy:
practical applications. An example of this,
based on Hashin’s failure conditions, is de- Qij si sj þ Qi si ¼ 1
scribed below. To this end, failure criteria for s1 ¼ s11 ; s2 ¼ s22 ; s3 ¼ s33 ð25Þ
unidirectional composites are presented in s4 ¼ s31 ; s5 ¼ s23 ; s6 ¼ s12
Section 3.09.4.1 and details of the procedure
are given in Section 3.09.4.2.
Parameters Qij and Qi are defined starting from
tensile and compressive strengths NT and NC in
3.09.4.1 Failure Criteria for Unidirectional directions 1, 2, 3 and from shear strengths S in
Fiber Composites the planes 12, 13, 23 (absolute values):
Failure criteria for composite materials (see 1 1 1 1 1 1
Ochoa and Reddy (1992)) usually are expressed Q1 ¼
; Q2 ¼ T
C ; Q3 ¼ T
C
N1T N1C N2 N2 N3 N3
in the form of simple polynomial combinations
of macroscopic stress components. They can, 1 1 1
Q11 ¼ T C ; Q22 ¼ T C ; Q33 ¼ T C
therefore, be used for classical yielding or N1 N1 N2 N2 N3 N3
failure criteria formulated for the failure analy- 1 1 1
Q44 ¼ 2 ; Q55 ¼ 2 ; Q66 ¼ 2
sis of homogeneous and isotropic materials. S31 S23 S12
This link is by no means casual, since the 1 1
first failure criterion formulated for composites Q12 ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi; ð26Þ
2 N TN CN TN C
by Tsai (1965) assumed the same mathematical 1 1 2 2
form as the Hill’s yield criterion for orthotropic 1 1
perfectly plastic materials (Hill (1948)) (it is Q13 ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi;
2 N TN CN TN C
therefore known as the Tsai–Hill criterion): a 1 1 3 3
quadratic polynomial, which reduces to the 1 1
Q23 ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
von Mises yield condition in the case of 2 N TN CN TN C
isotropic behavior. 2 2 3 3

The main drawback of the Tsai–Hill criter-

ion is the fact that neither isotropic stress states Notice that the strength constants Qij not
nor the sign of stresses have any effect on it. A defined in Equation (26) are zero. The Tsai–
modification of the Tsai–Hill criterion which Wu criterion represents an improvement with
accounts for different behavior in tension and respect to previous proposals and is still used.
compression has been proposed by Hoffman Hashin (1980) proposed a different approach
(1967) who added linear terms to it in order to in which the modes of failure are separately
take into account different behavior in tension treated; for each mode a polynomial stress
and compression. The Hoffman criterion is criterion is found.
expressed as Hashin’s criteria referred to a unidirectional
  fiber reinforced composite (Figure 24), where
1 1 1 1 direction 1 coincides with the fiber, direction 2
þ
ðs2
s3 Þ2
2 N2T N2C N3T N3C N1T N1C is orthogonal to the fibers, and direction 3 is
  the out-of-plane axis. Five failure modes are
1 1 1 1
þ þ
ðs3
s1 Þ2 distinguished: matrix failure in tension and
2 N3T N3C N1T N1C N2T N2C
  compression, fiber failure in tension and
1 1 1 1
þ þ
ðs1
s2 Þ2 compression, and delamination. These are
2 N1T N1C N2T N2C N3T N3C
   
1 1 1 1
þ
s1 þ
s2
N1T N1C N2T N2C
   2  2
1 1 s23 s13
þ
s3 þ þ
N3T N3C S23 S13
 2
s12
þ ¼1 ð24Þ
S12

where NiT;C denote the tension (T) and

compression (C) strenghts in direction i, while
Sij denote the shear strenghts in the plane ij.
Tsai and Wu (1971) proposed the following
general quadratic expression in the stresses,
valid for an orthotropic material with unequal Figure 24 Unidirectional lamina.
 Damage Mechanics for Composite Materials and Structures 485
governed by the following equations: composite structure which in some cases can
(i) Tensile fiber mode, s11 40: be much higher than that corresponding to
 2 FPF. Some failure criteria have been reviewed
s11 s212 þ s213 in the previous Section 3.09.4.1; it has already
þ ¼ 1 or s11 ¼ N1T ð27Þ
N1T 2
S12 been observed that Hashin’s criteria are
perhaps the most used in PFA.
The degradation rules usually are applied
(ii) Compressive fiber mode s11 o0
once a damage mode is activated on the basis
s11 ¼
N1C ð28Þ of a local failure criterion; they basically
consist in the reduction of appropriate compo-
nents of the stiffness matrix. Most frequently
(iii) Tensile matrix mode, s22 þ s33 40 sudden, brittle degradation is imposed, i.e.,
 2  stiffness components are reduced to zero. After
ðs22 þ s33 Þ2 s12 þ s213 þ s223
s22 s33
the stiffness components have been reduced,
þ ¼1 ð29Þ
N2T 2
S12 the stresses are recalculated with the new
global stiffness and then the load is again
(iv) Compressive matrix mode, s22 þ s33 o0 increased.
" # As an example the degradation model
2
1 N2C ðs22 þ s33 Þ2 adopted in Kutlu and Chang (1995a, 1995b)

1 ðs22 þ s33 Þ þ 2 is presented here. These authors apply Ha-
N2C 2S12 4S12
  shin’s failure multimode criterion; subse-
s212 þ s213 þ s223
s22 s33 quently they reduce the stiffness as follows,
þ 2
¼1 ð30Þ
S12 depending on the activated failure mode.
(i) Fiber failure in tension or compression
(v) Delamination E1 -0; G12 -0; n12 -0; n23 -0; n13 -0 ð32aÞ
 2
s33 s223
þ s213
þ ¼1 ð31Þ
N3T 2
S23
(ii) Matrix failure in tension or compression

Hashin’s multimode failure criterion is perhaps E2 -0; n12 -0; n23 -0; n13 -0 ð32bÞ
the most used in PFA for composites. It is
important to notice that simple maximum
stress and/or maximum strain criteria can also The sudden reduction of stiffness properties
be used in PFA. An alternative to Equation can sometime give problems of convergence, in
(31) for delamination can be found in Brewer other words it is difficult to reachieve equili-
and Lagace (1988). brium after sudden degradation.
To overcome the above difficulty some
authors adopt a gradual reduction, which is
3.09.4.2 Progressive Failure Analysis (PFA) nevertheless somewhat arbitrary in the frame-
A recent review of PFA can be found in work of PFA because a mechanical rule for the
Spottswood and Palazotto (2001), while a reduction of stiffness lacks; in this case a CDM
critical discussion and comparison with CDM approach appears to be more reliable.
based procedures (see Section 3.09.5) can be A coupling of PFA and FM has also been
found in Crisfield et al. (1997), and general adopted (see, e.g., Kutlu and Chang, 1995a)
remarks can be found in Chapter 5 of Ochoa where ERR criteria for delamination have been
and Reddy (1992); other contributions in the used together with Hashin’s criteria and PFA
field are those of Vaziri et al. (1992) (coupled for single layer failure. This coupling has the
with a plasticity approach), Rizzi et al. (2000), same mechanical reasons of the mesoscale
and Zinno and Greco (2001). A review on approach for composites described in Sections
strength analysis of unidirectional composites 3.09.5 and 3.09.6 where PFA and FM are
and laminates is in Sun (2000). substituted, respectively, by the use of CDM
PFA are based on two main ingredients: and IDM models.
failure criteria to be checked locally, i.e., in
each Gauss point of the FE mesh, and a 3.09.5 DAMAGE MECHANICS FOR
degradation rule, to be applied once the failure COMPOSITE MATERIALS AND
criterion is satisfied in some points. It can be STRUCTURES
said that PFA starts usually with a FPF
verification; the main goal of PFA is to This section is dedicated to the description of
determine the final collapse load of the fundamental issues in CDM for composite
486 Damage and Fracture Mechanics Techniques for Composite Structures
materials and structures. The purpose here is to necessary to have visible cracks to study
set the basis for a discussion of computational progressive damage and fracture.
tools which are available for the simulation of The above remark is also useful to define the
damage phenomena in composite materials limits of a CDM theory, which in fact is very
and structures based on the use of CDM. A effective until a major macroscopic crack forms
comprehensive review of CDM, and of rele- and starts propagating in the body. In this
vant computational methods, can be found in situation a series of problems arise when trying
another chapter of this series. to simulate the complete phenomenon with the
The first attempts at simulating damage use of CDM only. In order to resolve these
processes in composite materials and structures difficulties many attempts have been made,
have been based on PFA, briefly discussed in among them a systematic use of mesh-remesh-
Section 3.09.4. As noticed in Section 3.09.4, ing procedures. Difficulties arise fundamentally
PFA suffers from a lack of physical basis, in because the description of a real crack cannot
particular it is not known how to define the be correctly achieved with a theory which
appropriate stiffness change which must be considers a body as a continuum. In order to
introduced after a local failure criterion is obtain realistic results, a discrete description of
satisfied. CDM models can, at least partly, cracks, i.e., displacement discontinuities, must
solve the uncertainties related to the use of be introduced in numerical simulations. This
PFA. means that FM simulation tools must again be
Usually it is recognized that CDM started used or discrete cracks should be described by
with the papers by Kachanov (1958) and means of other tools. A possibility which has
Rabotnov (1968). The basic idea of this become popular is to make use of interface
approach is to try to model the progressive models and interface FE, as discussed in
development of damage in a solid through the Section 3.09.6. The way in which coupling
decrease in its elastic stiffness, which also between continuum and discrete descriptions
involves a decrease in the resistance. For this, of damage should be carried out is still an open
a homogenized measure of damage is intro- issue. In particular the way in which the
duced, which in the simplest case is represented transition from a CDM model to a discrete
by a scalar D, called the damage variable, crack model should be attained is not still well
which has values in the range [0, 1]. The established.
description of a 1D elastic-damage behavior Other difficulties are concerned with the
then follows: practical use in numerical simulations of CDM
models. Among these is the problem of
s ¼ ð1
DÞEe; DA½0; 1 ð33Þ avoiding pathological mesh dependency (see,
e.g., the review paper by de Borst (2001)) and
When the damage variable increases from 0 to the consequent necessity of using a so-called
1, both the elastic stiffness and the resistance of regularized model.
the material decrease, until the material is In spite of the above-mentioned difficulties,
completely fractured. CDM has become a very powerful tool for the
In order to complete the formulation of an simulation of damage phenomena in materials
elastic-damage model, an equation which and structures; it has been successfully applied
governs the evolution of the damage variable to metals, concrete-like materials, and compo-
is needed. This usually is expressed in terms of sites.
a measure of the elastic strain energy The issue of damage mechanics for compo-
sites is discussed in Section 3.09.5.1. In Section
3.09.5.2 a general class of damage models is
1 1 s2
Y ¼ Ee2 ¼ ð34Þ presented, its numerical time integration is
2 2 E ð1
D Þ2 discussed and 1D specializations are shown. A
particular example of damage models suitable
The variable Y is conjugated to the damage D for the description of a single layer in a
in the sense discussed below in Section 3.09.5.1 laminate is given in Section 3.09.5.3.
and is also called the damage ERR.
From the above introductory description it
should be clear that a damage model is a 3.09.5.1 Damage Mechanics for
phenomenological constitutive law conceived Heterogeneous Materials: Micro-,
to describe from the beginning a process of Meso-, and Macroscales
material degradation. The main advantage of
this kind of description is the fact that the In general terms, a CDM based approach for
fundamental limitation of a FM approach (see the simulation of progressive composite degra-
Section 3.09.3) disappears, i.e., it is no longer dation should in principle solve the problems,
 Damage Mechanics for Composite Materials and Structures 487
identified in the Section 3.09.4, concerning the the fifth one concerns interaction between
use of PFA. The way in which degradation is layers, and involves the whole laminate. Fiber
simulated is in fact part of the model and many rupture is a very brittle phenomenon, while
practical difficulties can be avoided. fiber–matrix debonding and matrix micro-
It is important to mention that models for cracking can develop in a more progressive
the description of damage and cracking in way. In particular matrix cracking usually
composite materials have also been formu- implies the development of many small cracks
lated, and successfully applied, without making which could be considered as diffused damage,
use of a CDM theory. Noteworthy contribu- and which are sensitive to the sign of transver-
tions in this field are, e.g., those of Hashin sal stresses in a lamina, i.e., microcracks close
(1985, 1986), McCartney (1993, 1998, 2000), in the presence of a compressive stress.
and McCartney et al. (2000). Delamination is characterized by the appear-
CDM for composites has become the subject ance of macroscopic cracks. In addition to the
of an abundant literature (see, e.g., Ladevèze, above, it should be mentioned the possible
1983; Allen et al., 1987; Talreja, 1989; Allix plastic behavior of the matrix, which could be
et al., 1989; Paas and van den Eikhoff, 1992; responsible of permanent plastic strains.
Yang and Boehler, 1992; Thionnet and Re- On the basis of the above remarks, it is in
nard, 1993; Saleeb and Wilt, 1993; Matzenmil- principle possible to develop models for the
ler et al., 1995; Maire and Lesne, 1998; descriptions of damage phenomena at the
Williams et al., 1998, 1999; Derrien et al., micro-, meso-, and macroscale which take into
2000; Voyiadjis and Deliktas, 2000; Chaboche account different damage mechanisms.
and Maire, 2001; Xia and Curtin, 2001; To the author’s knowledge, little work has
Caiazzo and Costanzo, 2001a; Fish and Yu, been done up to now at the macroscale; models
2001a, 2001b; Ladevèze and Lubineau, 2001; which describe the nonlinear behavior of whole
Jansson and Larsson, 2001; and also Aubard, laminates are not very diffused. An attempt has
2001 for an interesting point of view of been made in Giampieretti et al. (2000), where
industry). The purpose of the present section these authors have used homogenization and
is to recall the main concepts which should be parameter identification techniques in order to
kept in mind when trying to formulate damage define a nonlinear model for a whole laminate
models for composites and to underline the (see also Section 3.09.7.4.3(ii)).
problems which should be solved. Micromechanical descriptions are the sub-
The first issue concerns the scale of observa- ject of abundant literature (see, e.g., Talreja,
tion and of possible description of damage 1989; Yang and Boehler, 1992; Allen et al.,
phenomena in composites. Micro-, meso- and 1994; Aghdam et al., 2000; Devries et al., 1989;
macroscales can all be used. The first, micro- Hyer and Waas, 2000; Landis et al., 2000;
scale, concerns the details of the heterogeneous Lamon, 2001), and they are very important for
composite structure: the single fiber, the fiber– a clear understanding of damage processes and
matrix interface, and so on. The third, macro- for an improvement of material design and
scale, is relevant to the structural behavior like production technologies. Unfortunately, the
that of a whole laminate. The second one, use of microscale models for structural ana-
called mesoscale, is intermediate: in laminate lyses is very difficult if not impossible; they can
structures it can be clearly identified as the nevertheless help in developing models at
scale of single layer and layer–layer interface. higher scales.
Depending on the aim of the proposed model, The mesoscale description of damage phe-
an appropriate scale of description should be nomena in composites, in particular laminates,
chosen; this in turn implies the introduction of has gained much attention in the literature,
specific properties in the model. since it appears a natural scale for the
The second important issue to be considered description of the main damage mechanisms.
carefully, which is strictly linked to the above As proposed by Ladevèze (1983, 1986), in a
mentioned choice of scales of observation, mesoscale description, a laminate is conceived
concerns the kind of damage and, more as composed by two main constituents: the
generally, dissipative mechanisms which the single layers and the interfaces which separate
proposed model should be able to describe. In the layers (see Figure 25). A constitutive model
Section 3.09.2, five basic damage mechanisms for the single layer is then introduced which
in laminate composites have been recognized: should take into account the three main
fiber rupture in tension; fiber microbuckling in damage mechanisms which occur inside the
compression; fiber–matrix debonding; matrix layers: fiber rupture, fiber–matrix debonding,
microcracking; and delamination. The first and matrix cracking. The fourth damage
four damage mechanisms are relevant to mechanism, delamination, is governed by an
phenomena which occur inside a layer, while interface law, attributed to the interface, which
488 Damage and Fracture Mechanics Techniques for Composite Structures
microcracks in the presence of tension or
compression stresses; plastic behavior of the
matrix.

3.09.5.2 A Class of Elastic–Plastic Damage

Models for Composite Materials:
Formulation and Numerical Time
Integration
3.09.5.2.1 Model formulation
A class of anisotropic damage models
Figure 25 Mesoscale approach to composites. coupled with isotropic hardening plasticity is
expressed by the following equations, which
explicitly takes into account opening and hold for isothermal conditions under the
sliding between adjacent layers. In a mesoscale hypothesis of small strain and displacements:
description of a laminate, the coupling between
e ¼ ee þ ep ð35Þ
a continuum and a discrete modeling of
damage phenomena is obtained naturally by
making use at the same time of a CDM for the 1 T

rC ¼ ED þ gp ¼ ee E ðDÞee þ gp Zp ð36Þ
layer and of interface models, which simulate 2
the presence of discrete cracks between lami-
nas. For the development of layer models two @C @C
main approaches can be followed: a micro- r ¼ r e ¼ E ðDÞee ; Y ¼
r ;
@e
 @D
mechanical and a purely phenomenological ð37Þ
one. @C @gp Zp
An example of micromechanical approach wp ¼ r ¼
@Zp @Zp
can be found in the work of Lene (1986) (see
also Talreja, 1989; Yang and Boehler, 1992;  
Allen et al., 1994); the basic idea is to couple fp ¼ fp r; wp ; D r0; fp l’ p ¼ 0; l’ p Z0 ð38Þ
homogenization theory with micromechanical
descriptions of fundamental damage phenom-
ena in a representative volume element (RVE), f d ¼ f d ðY; ld Þr0; f Td k’ d ¼ 0; k’ d Z0 ð39Þ
in order to be able to derive a complete
constitutive law in average macroscopic vari-    
e’p ¼ l ep r; wp ; D l’ p ; Z’ p ¼ lZp r; wp ; D l’ p ð40Þ
ables. The micromechanical based approach is
in principle the best one to obtain a complete
and physically sound description of composite ’ ¼ l d ðY; ld Þk’ d ;
D Di A½0; 1 ð41Þ
damage behavior at the mesoscale, nevertheless
it is very difficult to apply in a direct way, due In the above equations e and r are vectors
to the heavy computational cost it requires. which gather the six independent components
Microbased models coupled with homogeniza- of the symmetric Cauchy stress tensor and of
tion procedures can be very effectively used in the small strain tensor, respectively. r and e are
order to derive fundamental qualitative prop- defined in such a way that their scalar product
erties of a mesoscale model; they can also be rT e equals the scalar product of stress and
used in conjunction with multiscale procedures strain tensors.
for the numerical simulation of composites. In Equation (35) the vector e is expressed as
Following a phenomenological approach for the sum of an elastic part ee and a plastic one
the formulation of layer damage models, it is ep. In Equation (36) the free energy density C
important to take into consideration all the (r being the mass density) is given by the sum
observations which come from experimental of two contributions. The first one, ED, is the
results and/or from micromechanical studies. damaged elastic strain energy density, D is a
From the above discussion, and from the brief vector of damage variables which affect the
introduction on CDM made at the beginning components of the elastic stiffness, and matrix
of Section 3.09.5, it can be observed that the E in a yet unspecified manner. By distinguish-
following ingredients should be added to the ing in ED positive and negative stress contribu-
more classical description of damage me- tions, it is possible to take into account
chanics: anisotropy during the initial elastic unilateral effects in the damage law, as shown
phase and during the development of damage; in the examples presented in Section 3.09.5.2.3.
unilateral effect due to the different behavior of Notice that in some formulations the elastic
 Damage Mechanics for Composite Materials and Structures 489
strain energy is expressed as function of description of damage phenomena in compo-
stresses, i.e., site structural components, like those for a
single layer.
ED ¼ 12 rT E
1 ðDÞr ð42Þ An important point which does not emerge
from the above formulation is the way in which
The second contribution to rC is the function anisotropy of damage mechanisms is taken
gp ðZp Þ which governs the isotropic hardening into account; this is usually done by choosing
for the plastic behavior; Zp is a kinematic the elements of the elastic stiffness matrix
plastic internal variable. Equation (37) repre- which should be decreased as a consequence of
sents equations of state which define the static the damage process and by introducing for
variables conjugated to ee ; D, and Zp. The each main damage mechanism a suitable
stresses r are defined by Equation (37a) as damage variable. In Section 3.09.5.3 Equations
function of elastic strains and damage vari- (35)–(41) will be specialized to a single layer
ables; Equation (37c) defines the vector Y of model proposed in the literature.
variables conjugated to damage; in Equations
(37d) the static internal variable wp is defined as
conjugate to Zp. When the elastic energy ED is 3.09.5.2.2 Numerical time integration
expressed as function of stresses, variables Y
must be computed as Nonlinear and irreversible constitutive laws
like those presented above are commonly used
@ED ðr; DÞ in step-by-step FE analyses after numerical
Y¼ ð43Þ time integration on a time step for each Gauss
@D
point in the FE mesh. Given the situation in
Variables Y are also called damage ERRs, in terms of all static and kinematic variables at a
analogy with the ERR of FM theory, based on certain time instant tn, and given the value of
the comparison of Equation (4) with Equations the total strain enþ1 at the subsequent time
(37c) and (43). instant tn þ 1, all other variables are sought at
The scalar plastic activation function, fp, and time instant tn þ 1 by integrating the constitu-
the vector-valued damage activation function, tive law numerically over the time step.
fd, are introduced in Equations (38) and (39); Here and in the following a subscript n or
they govern the possibility to have plasticity n þ 1 means a quantity computed at time
and or damage through plastic and damage instant tn or tn þ 1, respectively; the symbol
multipliers l’ p and k’ d and Kuhn–Tucker con- DðÞ means an increment of the quantity  in
ditions. (Notice that in Equation (38a) the the time interval ½tn ; tnþ1 :
introduction of the vector of damage variables A vast literature can be found on the subject
D represents coupling between plasticity and of numerical time integration of constitutive
damage; notice also that multi-mode damage laws; here the purpose (as in the whole chapter)
behavior is allowed in the model.) The last is to highlight the main difficulties concerned
Equations (40) and (41) of the above class of with computational aspects and to present an
models represent the plastic and damage flow example of possible solution. Among the
rules, respectively; these have been given a methods which have gained much attention is
general nonassociative kind for variables e’p ; Z’ p ; the backward difference integration rule; this is
and for the damage rates D: ’ The first are an implicit scheme which belongs to the
governed by the vector-valued function category of return mapping algorithms and is
l ep ðr; wp ; DÞ and the function lZp ðr; wp ; DÞ; used commonly in nonlinear computational
respectively, while damage evolution is gov- mechanics (see, e.g., Perego, 1988; Comi et al.,
erned by the matrix of functions l d ðY; ld Þ: 1991).
The mechanical dissipation related to the The numerically integrated version of Equa-
above model is given by the following: tions (35)–(41) reads
o ¼ rT e’
rC ’
wp Z’ p
’ ¼ rT e’p þ Y T D ð44Þ enþ1 ¼ eenþ1 þ epnþ1 ð45Þ

In order to have a thermodynamically consis-

tent model the dissipation o should never be  
rnþ1 ¼ E ðDnþ1 Þeenþ1 ; Y nþ1 ¼ Y eenþ1 ; Dnþ1 ;
negative.
 ð46Þ
The class of models seen in Equations (35)– wpnþ1 ¼ wp Zpnþ1
(41) has been formulated in the framework of
the thermodynamics of irreversible processes
(see, e.g., Lemaitre and Chaboche, 1990). It is  
fpnþ1 ¼ fp rnþ1 ; wpnþ1 ; Dnþ1 r0;
general enough to include various examples of ð47Þ
anisotropic damage models suitable for the fpnþ1 Dlp ¼ 0; Dlp Z0
490 Damage and Fracture Mechanics Techniques for Composite Structures
variables lpnþ1 ; epnþ1 ; Zpnþ1 ; rnþ1 ; wpnþ1
f dnþ1 ¼ f d ðY nþ1 ; ldnþ1 Þr0; f Tdnþ1 Dkd ¼ 0; have been found they are used to find the
Dkd Z0 ð48Þ solution of the subsequent step 4.
(iv) Step 4. Trial damage ERR
  Evaluate the trial damage ERR as follows:
Dep ¼ l ep rnþ1 ; wpnþ1 ; Dnþ1 Dlp ;
  ð49Þ  
DZp ¼ lZp rnþ1 ; wpnþ1 ; Dnþ1 Dlp Y trial p
nþ1 ¼ Y enþ1
en ; Dn ð56Þ

DD ¼ l d ðY nþ1 ; ldnþ1 ÞDkd ; Di A½0; 1 ð50Þ (v) Step 5. Check of damage consistency
The activation of damage behavior is
The above Equations (45)–(50) represent the checked by introducing the trial damage
nonlinear constitutive law integrated in time ERR in the damage activation function,
which relates stresses rn þ 1 to the total strain keeping all other quantities fixed at the
en þ 1. The system of equations can be solved in value in tn
various ways; a typical procedure is the  trial 
f trial
dnþ1 ¼ f d Y nþ1 ; ldn r0 ? ð57Þ
splitting algorithm suggested below

(i) Step 1. Elastic predictor (trial elastic If f trial

dnþ1 r0 for all damage modes, then the
stress) damage behavior is not activated, the trial
Evaluate the trial elastic stress as follows state is assumed as the correct solution and
  the algorithm goes to step 7; if f trial
dnþ1 40 for
rtrial p
nþ1 ¼ E ðDn Þ enþ1
en ð51Þ some mode, then the damage behavior is
activated for those modes and the elastic
damage solution is found as described in
(ii) Step 2. Check of plastic consistency step 6.
The activation of plastic behavior is checked (vi) Step 6. Damage corrector
by introducing the trial stresses in the plastic For the active modes only, the following
activation function, keeping all other quan- equations in the unknown
tities fixed to the value at tn Y nþ1 ; Dnþ1 ; kdnþ1 must be solved:
   
nþ1 ; wpn ; Dn r0 ?
trial
fpnþ1 ¼ fp rtrial ð52Þ Y nþ1 ¼ Y eenþ1 ; Dnþ1 ð58Þ

f dnþ1 ¼ f d ðY nþ1 ; ldnþ1 Þ ¼ 0 ð59Þ

trial
If fpnþ1 r0 then the plastic behavior is not
activated, the elastic trial state is assumed as
DD ¼ l d ðY nþ1 ; ldnþ1 ÞDkd ð60Þ
the correct solution and the algorithm goes
trial
directly to step 4; if fpnþ1 40 then the plastic
behavior is activated and the elastic plastic Again, depending on the kind of functions
solution is found as described in step 3. introduced in the model, the solution of the
(iii) Step 3. Plastic corrector above set of Equations (58)–(60) can be
The elastic–plastic solution at tn þ 1 is found more or less difficult.
by keeping the damage variable vector fixed (vii) Step 7. Check of convergence and
at the value at tn. The following set of iteration
nonlinear equations must then be solved The solution obtained following the above
  steps is approximated due to the splitting
rnþ1 ¼ EðDn Þ enþ1
epnþ1 methodology adopted. In order to improve
ð53Þ
wpnþ1 ¼ wp ðZpnþ1 Þ the result, steps 2–6 can be repeated
iteratively. At each iteration the final result
  in terms of variables at tn þ 1 is compared
fpnþ1 ¼ fp rnþ1 ; wpnþ1 ; Dn ¼ 0 ð54Þ with the result obtained at the previous
iteration. If a norm of the difference is
  below a predetermined tolerance, the solu-
Dep ¼ l ep rnþ1 ; wpnþ1 ; Dn Dlp
  ð55Þ tion is considered correct.
DZp ¼ l Zp rnþ1 ; wpnþ1 ; Dn Dlp

3.09.5.2.3 Consistent tangent matrix

The above system of equations can in
general be solved with an iterative proce- A numerical scheme proposed for the
dure; frequently the solution is simplified integration of a constitutive law is likely to be
and in some cases can also be found in an accompanied by the computation of the con-
explicit form (see, e.g., Perego, 1988). Once sistent tangent operator (Simo and Taylor,
 Damage Mechanics for Composite Materials and Structures 491
1985). The use of the consistent tangent described in Section 3.09.5.3 is the following:
operator is important in nonlinear analyses
when, for the solution of global equilibrium, E ðDÞ  ð1
DÞE
pffiffiffiffi pffiffiffiffiffiffi pffiffiffiffiffiffi
the equilibrium equation is linearized; as in the fd ðY ; ld Þ  Y
Yc ld
Y0 ð66Þ
case of the Newton–Raphson method for step-
ld ðY ; ld Þ  1
by-step analyses.
In the present context the consistent tangent
operator is the Jacobian of the equation where Yc, Y0 are model parameters.
connecting stresses and total strains at the time By introducing relations (66) and Equations
instant tn þ 1, as defined by the numerical (62)–(65), one obtains
integration scheme adopted, i.e., by Equations
s ¼ E hei
þð1
DÞE heiþ ; Y ¼ 12E hei2þ ð67Þ
(45)–(50):
drnþ1 ¼ Jdenþ1 ð61Þ pffiffiffiffi pffiffiffiffiffiffi pffiffiffiffiffiffi
fd ¼ Y
Yc ld
Y0 r0; fd l’ d ¼ 0;
In Equation (61) J is the consistent tangent l’ d Z0; D’ ¼ l’ d ; DA½0; 1 ð68Þ
6  6 matrix; J has to be computed from
Equations (45)–(50) by taking the derivatives The unilateral effect can now be noticed:
of each equations and by distinguishing the Young’s modulus in compression remains un-
cases of elastic unloading and plastic and/or changed, damage affects the tensile behavior
damage loading. The unilateral effect which only.
can be introduced in the model (see Sections The mechanical dissipation (Equation (44))
3.09.5.2.4, 3.09.5.2.5, and 3.09.5.3) also obliges becomes, in the present case
to distinguish cases with positive or negative
strains and/or stresses for some components. A o ¼ Y D’ ð69Þ
discussion on consistent tangent operators for
anisotropic damage models for composites can which is never negative, due to Equations (67)
be found in Voyiadjis and Deliktas (2000). and (68).
The present section closes with simple It is interesting to express the flow rule for
specializations of Equations (35)–(41) to the the damage variable in a way which is fully
cases of 1D elastic-damage and elastic–plastic- equivalent to Equation (68):
damage models; this will allow a better Dpffiffiffiffi pffiffiffiffiffiffiE
description of the model presented in Section Y%
Y0
3.09.5.3. Y%  max½Y ðtÞ; D¼ pffiffiffiffiffiffi þ
ð70Þ
t rt
0
Yc

3.09.5.2.4 1D elastic-damage model where t denotes the time variable. Equations of

An elastic-damage model for 1D behavior is the kind (69) have been used for the formula-
obtained from Equations (35)–(41) by disre- tion of the model proposed by Ladevèze and
garding all terms indexed p and by reducing the co-workers (see, e.g., Ladevèze and le Dantec,
vector-valued functions and matrices to scalars 1992), described in Section 3.09.5.3.
The response of the 1D model for an
rC ¼ 12E hei2
þ12E ðDÞhei2þ ð62Þ assigned monotonically increasing positive
strain can be computed; it reads
rffiffiffiffiffiffiffiffi
@C @C 2Y0
s¼r ¼ E hei
þE ðDÞheiþ ; Y ¼
r ð63Þ s ¼ Ee; for 0rere0 ¼
@e @D E
 rffiffiffiffiffiffi rffiffiffiffiffiffiffiffi
Y0 E 2
fd ¼ fd ðY ; ld Þr0; fd l’ d ¼ 0; l’ d Z0 ð64Þ s ¼E 1þ e
E e ð71Þ
Yc 2Yc
 rffiffiffiffiffiffi
Yc
for e0 rerec ¼ e0 1 þ
D’ ¼ ld ðY ; ld Þl’ d ; DA½0; 1 ð65Þ Y0

where E is the initial Young’s modulus. The As can be seen from the above equations, the
symbols hi
and hiþ in Equations (62) response of the model in tension after damage
and (63) mean the negative and positive part activation is in this case given by a parabolic
of ; respectively, introduced to model a function, this has been drawn in Figure 26 for
unilateral effect. In order to complete the the choice of parameters given in the figure
above elastic-damage model, functions legend; the behavior in compression is indefi-
E ðDÞ; fd ðY ; ld Þ; ld ðY ; ld Þ must be assigned. nitely elastic, due to the unilateral effect
A possible choice in analogy with the model introduced.
492 Damage and Fracture Mechanics Techniques for Composite Structures


E ðDÞ ¼ ð1
DÞE; gp Zp  12HZ2p
 
fp s; wp ; D
r
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  s 
s 2  

wp
s0 ¼  
wp
s0
1
D 1
D ð79Þ
  @fp   @fp
lep s; wp ; D  ; lZp s; wp ; D 

@s @wp
pffiffiffiffi pffiffiffiffiffiffi pffiffiffiffiffiffi
fd ðY ; ld Þ  Y
Yc ld
Y0
ld ðY ; ld Þ  1

where E is the initial Young’s modulus, H, s0,

Yc, Y0 are model parameters. The ratio s* 
s=ð1
DÞ is usually called the effective stress;
substitution of stresses with the effective
stresses is a common way to introduce
damage-plasticity coupling (see, e.g., Lemaitre
Figure 26 1D elastic-damage model (Equation and Chaboche, 1990).
(71)) (E¼ 1  105 MPa; Y0 ¼ 0.1 MPa, Yc ¼ With the above choices, the elastic–plastic-
2:0 MPa).
damage model under discussion reads
e ¼ ee þ ep ; s ¼ E hee iþ þð1
DÞE hee iþ
It is also interesting to compute the area ð80Þ
Y ¼ 12E hee i2þ ; wp ¼ HZp
under the stress–strain diagram, which coin-
cides with the dissipated energy necessary to  s 
 
completely damage the material: fp ¼  
wp
s0 r0; fp l’ p ¼ 0; l’ p Z0 ð81Þ
1
D
1Z  rffiffiffiffiffiffi 
Yc Y0 Y0
o¼ Y ðDÞ dD ¼ 1þ3 þ3 ð72Þ pffiffiffiffi pffiffiffiffiffiffi pffiffiffiffiffiffi
0 3 Yc Yc fd ¼ Y
Yc ld
Y0 r0; fd l’ d ¼ 0
ð82Þ
D’ ¼ l’ d Z0; DA½0; 1

3.09.5.2.5 1D elastic–plastic-damage model

1 s’
An elastic–plastic-damage model for 1D e’p ¼ lp ; Z’ p ¼ l’ p ð83Þ
ð1
DÞ jsj
behavior is obtained from Equations (35)–
(41) by reducing all the vector-valued functions The mechanical dissipation (Equation (44)) can
and matrices to scalars: be shown to be never negative also in the

 present case
rC ¼ 12E hee i2
þ12E ðDÞhee i2þ þgp Zp ð73Þ
’ ¼ s’ep
wp Z’ p þ Y D’
o ¼ s’e
rC
 
¼ fp þ s0 l’ p þ Y D’ ¼ s0 l’ p þ Y DZ0
’ ð84Þ
@C
e ¼ ee þ ep ; s ¼ r e ¼ E hee i
þE ðDÞhee iþ
@e
 The presence of two activation functions in the
@C @gp Zp 1D elastic–plastic-damage model (Equations
Y ¼
r ; wp ¼ rC ¼ ð74Þ (79)–(83)) makes it difficult for analytical
@D @Zp
integration. A numerical solution could be
obtained by means of the integration scheme
 
fp ¼ fp s; wp ; D r0; fp l’ p ¼ 0; l’ p Z0 ð75Þ described in Section 3.09.5.2.2.

fd ¼ fd ðY ; ld Þr0; fd l’ d ¼ 0; l’ d Z0 ð76Þ 3.09.5.3 Damage Model for a Single Lamina

One of the first proposals concerning CDM
    for composites was that put forward by
e’ ¼ lep
p
s; wp ; D l’ p ; Z’ p ¼ lZp s; wp ; D l’ p ð77Þ
Ladevèze (1983, 1986) and Ladevèze and co-
workers (see Allix et al., 1989, 1994; Ladevèze
D’ ¼ ld ðY ; ld Þl’ d ; DA½0; 1 ð78Þ and le Dantec, 1992; Ladevèze, 1995). In the
above series of papers a general theory for
The above elastic–plastic model is completed the description of anisotropic damage models
by defining functions has first been presented (Ladevèze, 1983);
 Damage Mechanics for Composite Materials and Structures 493
subsequently it has been applied to laminates The model in point is elastic–plastic damage.
and 3D composites (Ladevèze, 1986), and also It can be seen as a particular case of the general
to 4D carbon–carbon composites (Ladevèze class formulated in Section 3.09.5.2.1, and it
et al., 2000a, 2000b). Applications to impact takes into account fiber rupture, matrix micro-
problems have been described (Allix et al., cracking, and fiber–matrix debonding. Only
1997; Allix, 2001). matrix microcracking and fiber–matrix de-
The models in point are phenomenological; bonding are modeled by means of CDM. Fiber
they were formulated on the basis of a series of rupture is governed by a simple maximum
considerations coming partly from qualitative strain criterion.
micromechanical results, mostly from experi- Let us consider a unidirectional fiber-rein-
mental evidence and from a careful study of the forced lamina in the reference frame of Figure
basic damage mechanisms that develop in 24 where 1 denotes the fiber direction, 2 the in
composites. The links between the phenomen- plane direction orthogonal to fibers, and 3 the
ological models and micromechanical behavior direction normal to the layer plane. The layer is
have been discussed in Ladevèze and Lubineau assumed to be in a plane stress situation, i.e.,
(2001). only s11 ; s22 ; s12 are nonzero stresses. The
A model similar to the above referenced one functions ED ; gp ; fp ; f d ; l ep ; lZp ; l d which
is that proposed by Paas and Eikhoff (1992); completely define Equations (35)–(41) are
these authors start from a general formulation chosen as follows:
of anisotropic damage governed by a second "
order tensor and specialize it to the case of a 1 s211 n0 hs22 i2þ hs22 i2

ED  0

2 120 s11 s22 þ 0
þ
single lamina. A damage model for layers and 2 E1 E1 ð1
D2 ÞE2 E20
comparisons between micromechanical and 
s212
mesomechanical approaches have been pre- þ 0
ð85Þ
ð1
D12 ÞG12
sented in Talreja (1989).
A different path of reasoning is used by
Schipperen (2001) and de Borst and Schipperen b
aþ1
gp  Z ð86Þ
(2002), who formulated a damage model which aþ1 p
accounts for matrix cracking starting from the
transformation of the Hoffman yield failure sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2  2
criterion (Equation (24)) (Hoffman, 1967) in s12 hs22 iþ
fp  þ a2 þ hs22 i

the strain space. The anisotropic norm of 1
D12 1
D2
strains so obtained is used as the quantity
wp
w0p ð87Þ
governing the activation of damage. A damage
variable, which evolves with an exponentially " #
decaying function, is then introduced to fd2
fd ¼
modify the elastic stiffnesses. The stiffnesses fd12
in the fiber direction, in the off-ply direction " pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffi pffiffiffiffiffiffi #
and the shear stiffness for shear in the plane Y12 þ bY2
Yc0 ld2
Y00
 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffi pffiffiffiffiffiffi ð88Þ
defined by fibers and off-plane direction are Y12 þ bY2
Yc ld12
Y0
assumed not to be affected by the matrix-
cracking phenomenon. Another example of 3 2
2 30
matrix cracking modeling is that proposed by 6 7
0 6 @f 7
Crisfield et al. (1997) who introduce a damage 6 7 6 p 7 @fp
model in which the stress orthogonal to the l ep ¼ 4 lep2 5  6 @s22 7; lZp 
ð89Þ
6 7 @wp
fibers, the in plane stress and the shear stress in lep12 4 @fp 5
the plane orthogonal to fibers are involved in @s12
the damage evolution. Crisfield et al. (1997)
also compare the performance of a damage " # " #
ld2 0 1 0
matrix-cracking model with those of the PFA ld ¼  ð90Þ
adopted in Kutlu and Chang (1995a, 1995b). 0 ld12 0 1
The single-layer model, proposed by Lade-
vèze and co-workers, for the description of the In the above equations the following material
damage behavior of a unidirectional lamina parameters have been introduced: the unda-
inserted in a laminate, is discussed here below. maged elastic stiffnesses E10 ; E20 ; n012 ; G12
0
; a,
A complete and careful presentation of the and b which govern the isotropic hardening
same model can be found in the book by rule; a and w0p defining the plastic activation
Herakovich (1998), a numerical FE study of function (yielding function fp); b, Yc ; Y0 ;
composite damage based on the same model is Yc0 ; Y00 ; governing the damage activation func-
presented in Phillips et al. (2001). tions fd.
494 Damage and Fracture Mechanics Techniques for Composite Structures
Introducing the choices expressed by Equa-
tions (85)–(90) in Equations (35)–(41), (with Y%  max ½Y12 ðtÞ þ bY2 ðtÞ
t0 rt
account taken of Equations (42) and (43)), the Dpffiffiffiffi pffiffiffiffiffiffiE
following set of governing equations is ob- Y%
Y00
þ
D2 ¼ pffiffiffiffiffiffi ; if D2 o1 and Y2 oY2C
tained Yc0
otherwise D2 ¼ 1 ð96Þ
1  Dpffiffiffiffi pffiffiffiffiffiffiE
e11 ¼ 0 s11
n012 s22 Y%
Y0
E1 þ
  D12 ¼ pffiffiffiffiffiffi ; if D12 o1 and Y2 oY2C ;
1 hs22 iþ Yc
e22 ¼ 0 þ hs22 i

n012 s11 ð91Þ
E 2 ð1
D 2 Þ otherwise D12 ¼ 1
s12
2e12 ¼ 0
ð1
D12 ÞG12 Let us now comment the choices made by
the authors in formulating the above model.
The two damage variables D2 and D12 govern
the progressive damage which develops in the
hs22 i2þ s212 unidirectional layer due to matrix microcrack-
Y2 ¼ 2
; Y12 ¼ ; ing, mainly in the direction parallel to the
2ð1
D2 Þ E2 0
2ð1
D12 Þ2 G12
0
ð92Þ

a fibers; only transversal tension s22 and shear
wp ¼ b Zp stresses s12 can cause damage. Hence a
unilateral effect has been introduced in the
model by means of the distinction of positive
and negative stresses s22; this explains the form
e’p11 ¼ 0 given to the damaged elastic strain energy ED
  in Equation (85).
hs22 iþ
a2 þ hs22 i
In other versions of the model the effect of
1
D2
e’p22 ¼ l’ p fiber microbuckling has also been introduced in
wp þ w0p
8 a phenomenological way, by considering the
< 1 ; if s 40 ð93Þ Young’s modulus E1 as a nonlinear, decreasing
22
 1
D2 function of the stress s11 (see Allix et al., 1994).
:
1; if s22 o0 The matrix can yield during loading, hence a
’lp ðs12 =1
D12 Þ plastic activation function and a hardening rule
2’ep12 ¼ have been introduced; the hardening rule is
1
D12 wp þ w0p
exponential, in view of experimental results,
while the plastic activation function (Equation
(87)) does not depend on stresses s11 in the
Z’ p ¼ l’ p ; D’ 2 ¼ l’ d2 ; D’ 12 ¼ l’ d12 ; D2 ; D12 A½0; 1 ð94Þ
fiber direction. The plastic-damage coupling
has been introduced by means of the following
effective stresses
As already observed when describing the 1D
elastic-damage model in Section 3.09.5.2.3 hs22 iþ s12
s* 22  þ hs22 i
; s* 12  ð97Þ
(Equation (70)), the damage evolution law 1
D2 1
D12
can be expressed in a compact way which is
fully equivalent to the set of Equations (39), The activation of damage and its evolution is
(41), (88), (90) and is the one used by Ladevèze governed by the square root of a linear
and co-workers combination of the two conjugate damage
variables Y2 and Y12. This choice comes from
direct experimental evidence.
Y%  max ½Y12 ðtÞ þ bY2 ðtÞ; It is interesting to compare the first activa-
t0 rt
Dpffiffiffiffi pffiffiffiffiffiffiE Dpffiffiffiffi pffiffiffiffiffiffiE tion of damage with the plane stress failure
Y%
Y00 Y%
Y0 ð95Þ
D2 ¼ pffiffiffiffiffiffi þ
; D ¼ pffiffiffiffiffi
ffi þ conditions proposed by Hashin (1980) and
12
Yc0 Yc recalled in Section 3.09.4. On the basis of
Equations (88) and (92(a) and (b)) damage is
first activated when the following equation in
In addition to the above equations, typical of a terms of stresses holds
CDM approach, the model is completed by a
brittle failure criterion which takes into ac- s212 s2
count failure of the fiber–matrix interface in Y12 þ bY2 ¼ 0
þ b 220 ¼ Y0 ; if s22 40
2G12 2E2
transverse tension; this is governed by a limit ð98Þ
value Y2c for the variable Y2. The damage s212
Y12 þ bY2 ¼ 0
¼ Y0 ; if s22 o0
evolution law can therefore be rewritten as 2G12
 The Particular Role of Interfaces and Interface Mechanics 495
By comparing the above Equation (98) with some fracture phenomena in composites, in
the Hashin’s criteria for matrix failure in plane particular of delamination. A practical way to
stress (obtained by Equations (29) and (30)), it obtain this is based on the use of interface
can be recognized that the first coincides with models or generalized cohesive crack models in
the tensile matrix mode, while the second is a which an equation between cohesive tractions
simplified version of the compressive matrix and the separation of the two parts of a crack is
mode. established. Interface laws are, therefore, con-
The identification of the above model stitutive equations between tractions and dis-
implies the determination of the following placement discontinuities along the crack line.
parameters: elastic E10 ; E20 ; n012 ; G12
0
; hardening A more precise definition of an interface law is
law a, and b; plastic activation a and w0p ; given at the beginning of Section 3.09.6.2.
damage activation and evolution b, The use of cohesive models for the descrip-
Yc ; Y0 ; Yc0 ; Y00 : A complete identification proce- tion of fracture processes initiated with the
dure and comparison with experimental results works by Barrenblatt (1959) and Dugdale
have been presented in Ladevèze and le Dantec (1960), and subsequently was applied mainly
(1992); a discussion on parameter identification for the solution of crack problems in concrete
can also be found in Herakovich (1998). (see, e.g., Hillerborg et al., 1976; Bažant and
In order to avoid spurious mesh dependency Oh, 1983; Carpinteri, 1989) and quasibrittle
in numerical analyses the model under discus- materials in general. A simple example of
sion has been implemented by the authors with cohesive law is shown in Figure 27 where it
the additional constraint of uniform damage has been assumed that the cohesive traction t
along the layer thickness imposed by comput- on the crack surfaces progressively decreases as
ing the conjugate damage variables from the a function of the crack separation [u] until a
average of stresses along the layer thickness. critical distance [u]c is reached and the traction
Rate-dependent versions of the model have goes to zero leaving the fracture surfaces
also been proposed, suitable for the simulation completely free. It is noteworthy the fact that
of dynamic phenomena (see, e.g., Ladevèze the area under the traction-opening displace-
et al., 2000). ment plot coincides with the fracture energy
Gc, i.e.,
3.09.6 THE PARTICULAR ROLE OF Z ½u c
INTERFACES AND INTERFACE t d½u ¼ Gc ð99Þ
0
MECHANICS
The difficulties related to the simulation of Two important consequences of the use of
damage processes in composite materials have cohesive laws must be underlined; first, the
been highlighted in various parts of the singularity of stresses typical of LEFM dis-
previous sections; for the simulation of dela- appears, second is that due to the progressive
mination, in particular, it appears necessary to decrease of cohesive tractions, a process zone
make use of a mesoscale approach, in which forms behind the crack-tip, which can be
delamination cracks are taken explicitly into considered as the zone where dissipative
account and not smeared. phenomena due to fracture are concentrated.
In Section 3.09.4, on PFA, it was observed The use of interface models is not limited to
that an efficient procedure could be based on quasibrittle materials; they are more and more
PFA methods for the single layer damage,
coupled with computational FM methods for
the study of delamination cracks (see Kutlu
and Chang, 1995a, 1995b); this is equivalent to
the application of a mesoscale approach in
which the main ingredients are the single layer
and the delaminated zone.
In Section 3.09.5 the mesoscale approach for
the study of laminate composites has been
discussed in the framework of CDM, in this
case the use of a CDM model for the single
layer and an interface model for the study of
delamination were proposed as a possible
optimal compromise for the study of compo-
site damage.
The above remarks all underline the neces-
sity of making use of a discrete description for Figure 27 Cohesive law.
496 Damage and Fracture Mechanics Techniques for Composite Structures
applied for the simulation of fracture processes the nature of a composite to be built by
in rocks, concrete, ductile metals, polymers, assembling different components or phases;
and composites; both in static and complicated the surfaces of separation of the various phases
dynamic conditions as impact and fragmenta- are therefore natural interfaces along which
tion (see Pandolfi et al., 1999). Interface models small defects frequently can be the reason of
are used in numerical simulations coupled with crack initiation. The basic idea of IDM applied
FE or BE procedures (see Maier et al., 1993; to composites is to give to the interface the role
Maier and Frangi, 1998), and also with new of an independent entity to which a particular
approaches like the embedded-discontinuity constitutive law can be attributed.
FE (see, e.g., Bolzon and Corigliano, 1997b, In the framework of the mesoscale approach
2000); or the partition of unity method (see, to damage of composites presented in Section
e.g., Möes et al., 1999; Wells and Sluys, 2001; 3.09.5, interfaces are mainly used for the
Mariani and Perego, 2001; Wells et al., 2002). simulation of delamination processes. Other
The generality and importance of applica- applications are nevertheless possible. These
tions of interface models for the simulation of concern all phenomena of debonding which
decohesion processes allow for the introduc- can arise in a multiphase material at different
tion of the term IDM which is being used scales of study. In all these cases an interface
throughout the chapter. law is used to simulate a decohesion process; as
The popularity of the use of IDM for the in cohesive crack models, it must simulate the
simulation of fracture processes comes essen- progressive decreasing of cohesive forces at
tially from the possibility of introducing increasing value of displacement discontinu-
displacement discontinuities in numerical mod- ities on the crack surface.
els and of governing the crack initiation and The study of delamination processes by
advancement by means of a constitutive law; means of IDM started with the works of
apparently the use of IDM gives the advan- Ladevèze (1986), Allix and Ladevèze (1992),
tages of FM and of CDM altogether. Corigliano (1993), and Schellekens and de
Apart from the above-mentioned advan- Borst (1993a), and subsequently was pursued
tages, IDM has of course some drawbacks also by others like Hellweg et al. (1994) and
which stimulate further research in the field. Crisfield et al. (1997). In the works by Back-
Among the critical issues which must still be lund (1981) and Bruno and Grimaldi (1990) an
solved are: the correct formulation of interface idea similar to that of interfaces had also been
models; the problem of parameter identifica- applied previously for the study of delamina-
tion of interface models; the necessity to use tion. The application of IDM to composites
very refined meshes in the crack-tip region (the must take into consideration some basic
same difficulty characterizes computational aspects which on one side simplify and on the
methods in FM); the introduction of additional other side complicate the numerical implemen-
dof in the numerical models; and the global tation.
strategy of the numerical analysis, which in The first peculiar aspect is the fact that in
many cases needs complicated remeshing pro- composite materials and structures cracks
cedures. Some of these issues will be discussed amenable to an IDM modeling follow a path
below and in Section 3.09.7. which is known a priori; the crack is in fact
The use of IDM for composite materials is constrained in most case to propagate at the
discussed in Section 3.09.6.1. In Section interface between two phases (as in fiber–
3.09.6.2 a general class of interface models is matrix debonding) or between two mesocon-
presented, together with a possible numerical stituents (as in delamination). The a priori
time integration procedure. Section 3.09.6.3 is knowledge of the crack path greatly simplifies
dedicated to the presentation of some interface the computational aspects; interface elements
models which could be used for the simulation can in fact be introduced in the initial FE
of delamination and debonding in composites. model thus eliminating the need to make use of
Finally, the issue of parameter identification of remeshing procedures or embedded crack FE
interface models is discussed briefly in Section (see Section 3.09.7).
3.09.6.4. A second important feature which must be
taken into account when developing an IDM
3.09.6.1 The Importance of Interfaces in model for composites, strictly related to the
Composite Materials and Structures first mentioned one, is the fact that, due to the
constraint on the crack path; fracture can
The popularity that IDM has gained in the propagate also in mixed-mode conditions. As
simulation of fracture processes finds addi- already remarked in Section 3.09.3, the mode-
tional special motivations in the case of mixity conditions of fracture propagation
composite materials and structures. It is in complicate the modeling of the phenomenon;
 The Particular Role of Interfaces and Interface Mechanics 497
in particular the interface laws must be able to damage interface models have been formulated
reproduce the separation of energies in single for application to the study of composite
modes as obtained from experimental evidence. delamination. The same models have been
Due to the fact that the fracture energies for improved subsequently and applied in Daude-
the various modes are usually different, the ville and Ladevèze (1993), Daudeville et al.
simulation of mixed-mode conditions men- (1995), and Allix et al. (1998). The issue of
tioned above is related to a third peculiar parameter identification has been in particular
aspect of IDM for composites: the interface discussed in Allix et al. (1998). Rate dependent
models must in general be anisotropic. Parti- elastic-damage interface models for the simula-
cularly in delamination cracks, it is observed tion of impact in composites have been
experimentally that fracture energies and the formulated and used in Allix et al. (1997) and
whole phenomenology of crack propagation Ladevèze et al. (2000b). In Corigliano (1993) a
depends on the layer lay up; in addition to this, general class of interface models has been
it is the direction of fibers in the adjacent layers proposed and examples of elastic damage,
which greatly influences the properties of the elastic–plastic softening, and elastic–plastic-
interface, e.g., phenomena like fiber-bridging damage models have been presented. Applica-
are observed more frequently in 01–01 inter- tions of elastic–damage interface models for
faces (see, e.g., S
rensen and Jacobsen, 1998, the study of pure-mode interlaminar fracture
2000). This suggests that a fourth feature to specimens have been presented in Allix et al.
take into consideration is the fact that inter- (1995), while the study of mixed-mode inter-
faces used for the simulation of delamination laminar fracture specimens has been discussed
depend on the angle formed by directions of in Allix and Corigliano (1996). In Bolzon and
the fibers of adjacent layers. Corigliano (1997a) a class of interface models
Other important issues must also be con- which satisfy the maximum dissipation princi-
sidered, among them the need to distinguish ple has been proposed and the consequences of
tension and compression for mode I fracture interface softening on stability and bifurcations
(unilateral effect) and the possibility to make at the structural level have been studied. In
use of time-dependent interface models for the Corigliano and Allix (2000) various aspects of
simulation of rate dependent delamination the simulation of interlaminar degradation in
(see, e.g., Frassine et al., 1993, 1996; Frassine composites by means of interface elements
and Pavan, 1995; Corigliano et al., 1997; have been analysed. Rate-dependent interface
Corigliano and Ricci, 2001). models have been formulated in Corigliano
The interface models which are proposed in and Ricci (2001). The issue of parameter
order to take into consideration all the above identification of interface models by means of
peculiarities frequently depend on material indirect identification procedures has been
parameters which are not easy to find. As studied by Corigliano and Mariani (2001,
discussed in the subsequent Section 3.09.6.4, 2002).
the issue of parameter identification for inter- Interface models for the study of composite
face models is particularly important because delamination have also been proposed by de
no direct experimental tests can be done on Borst and co-workers in a series of papers,
interfaces; apart from the fracture energies in starting with Schellekens and de Borst (1993a)
different modes which can be considered as where the authors proposed a decomposition
fundamental data for the identification of approach inspired by the concrete literature in
interface models, other data are difficult to which the total incremental relative displace-
find or can be obtained only through indirect ment vector was given by an elastic part and an
procedures. inelastic one, called crack relative displace-
In general, it can be said that the simulation ment. Subsequently, Schellekens and de Borst
of debonding and delamination in composites (1994) proposed an orthotropic softening
by means of interface models can be seen as plasticity model with orthotropic hardening.
one of the most natural applications of IDM; it A similar model, inspired by the Hoffman
extends the applicability of CDM to discrete failure criterion for composites (Equation
crack phenomena, maintain the energetic (24)), was used subsequently in Hashagen
aspects of a FM simulation based on the et al. (1999) and in Hashagen and de Borst
ERR, eliminates the problem of stress based (2000) for the study of delamination in fiber
computations and related failure criteria. metal laminates. Schipperen and de Borst
The contributions in the literature to IDM (2001) and de Borst and Schipperen (2002)
are numerous and it is not possible to mention have also proposed elastic-damage interface
all of them here. Therefore, a brief review is models.
given here. In Ladevèze (1986), Allix et al. Composite delamination has been studied by
(1989), and Allix and Ladevèze (1992), elastic means of interface models by Crisfield et al.
498 Damage and Fracture Mechanics Techniques for Composite Structures
(1997), Mi et al. (1998), and Chen et al. (1999), nation and debonding in composites, it is
where elastic-damage interface models have worth mentioning a number of contributions
been proposed. Alfano and Crisfield (2001) in the wider literature of interface and cohesive
discuss the issue of mixed-mode delamination models used for the simulation of general
and suitable formulations of interface models; fracture processes. Lemaitre (1992) formulated
in particular they compare their elastic-damage the damage behavior of the interface in a
interface model with the one proposed in Allix consistent thermodynamic way. In Costanzo
and Corigliano (1996). Bao and Song (1993) and Allen (1995) an interesting thermodynamic
studied crack-bridging models for fiber com- analysis has been proposed concerning the
posites with slip-dependent interfaces and study of crack modeled with cohesive laws. Xu
applied them for the study of fiber pull-out and Needleman (1995) made a systematic use
and fatigue cracking of fiber composites. The of nonlinear elastic interface laws for the
effect of fiber bridging and R-curve was taken simulation of dynamic crack processes. Cama-
into account in the interface formulation cho and Ortiz (1996) formulated a general
proposed by Yan et al. (2001). cohesive law suitable for mixed-mode condi-
Bridging laws for the study of delamination tions and applied it in a series of cases of fast
in stitched structures and propagation of mode dynamic crack propagation.
II delamination have been presented in Cox
et al. (1996) and Massabò and Cox (1999), 3.09.6.2 A Class of Elastic–Plastic Damage
while bridging laws for the study of fiber- Interface Models for Composite
reinforced cementitious composites have been Materials: Formulation and
studied in Karihaloo et al. (1996) and Kar- Numerical Time Integration
ihaloo and Wang (2000). Bridging laws applied
to short fiber composites can be found in The purpose of this section is to give a
Lindhagen et al. (2000). In Point and Sacco sufficiently general class of interface laws
(1996a, 1996b) the Frémond’s adhesion ap- which can be used for the simulation of
proach (Frémond, 1987) has been used to delamination in composites. Consider a surface
formulate a class of elastic-damage interface G which separates two solids or divides a solid
models suitable for the study of delamination in two parts. The two parts are noted with V þ
processes. Voyiadjis and Park (1997) studied and V
. Fix a point P on the surface G and
metal-matrix composites with interfacial da- introduce a local right handed reference frame
mage analysis. 1, 2, 3 centered on P. Axis 3 coincides with the
Among recent papers in which interface normal on G in P directed toward V þ (Figure
models for the study of delamination in 28). The traction vector in P is denoted with t;
composites have been formulated and used, ½u  uþ
u
is the vector of displacement
the following are mentioned: Wisnom and discontinuities (or displacement jump vector),
Chang (2000), Jansson and Larsson (2001), where uþ and u
are the displacements of point
and Borg et al. (2001). Use of cohesive laws for P thought to belong to V þ or V
, respectively.
the study of debonding in multilayered com- The displacement discontinuity vector allows
posites has been tackled with new mathema- for the simulation of fracture along the inter-
tical concepts in Kaziolas and Baniotopoulos face G; mode I (opening mode) corresponds to
(2001). nonzero displacement jumps in direction 3
Interesting contributions can also be found orthogonal to the interface, modes II and III
in the recent literature concerning the problem (shearing and tearing modes) correspond con-
of fiber–matrix debonding treated by means of ventionally to displacement jumps in directions
cohesive or interface models. Budiansky et al. 2 and 3, respectively.
(1995), studied fiber–matrix debonding at the An interface constitutive law is here defined
microscale to the purpose of defining the as an equation between the traction vector t
parameters of a crack-bridging model. Thion- and the displacement jumps vector ½u: Inter-
net and Renard (1998) adopted a multiscale face models have to possess the following main
approach to identify a local debonding criter- features, when applied to the modeling of
ion for the simulation of fiber–matrix debond- composite delamination: the possibility of
ing. McCartney (1999) proposed an analytical modeling a decrease in resistance (softening)
model for debonded interfaces associated with until a critical level ½uc is reached and the
fiber fractures or matrix cracks. Ghosh et al. layers are separated completely; the anisotropy
(2000) studied interfacial debonding between of the response; the difference in tensile and
fibers and matrix by making use of cohesive compressive behavior for the direction normal
models and the Voronoi cell FE model. to the surface (unilateral effect). Another
Besides interface models which have been important point to consider is the possible
proposed specifically for the study of delami- dependence of the interface behavior on the
 The Particular Role of Interfaces and Interface Mechanics 499

Figure 28 Interface G and local reference frame.

orientation of fibers of adjacent layers. A first Equation (100) expresses the displacement
way to take this into account is to define a local discontinuities as the sum of an elastic (rever-
reference system of principal directions which sible) part ½ue and a plastic (irreversible) one
depends on the orientation of fibers of adjacent ½up : In Equation (101) the free energy per unit
layers; for instance define a reference frame by surface in isothermal conditions is given by the
taking as the principal direction 1 (or 2) the sum of two contributions. The first one is the
bisectrix of the angle formed by the direction of elastic deformation energy per unit surface (the
fibers of adjacent layers. Other features that are first four addends); Di (i ¼ 1; 2; 3) are three
important but not crucial can be introduced in damage variables, different for each direction
the interface model, e.g., the existence of 1, 2, 3 to take into account the anisotropy of
irreversible displacement discontinuities and damage evolution, Ki, i ¼ 1; 2; 3 are interface
time-dependent behavior. stiffnesses with the dimension of a force over a
length cube. The positive and negative parts
of ½u3 e are introduced in Equation (101) in
3.09.6.2.1 Model formulation order to take into account the unilateral
A general class of interface models has been effect. The second contribution gp ðZp Þ is a
proposed in Corigliano (1993) governed by the function of the kinematic internal variable Zp
following equations and governs, together with Equation (102c),
possible modifications of the activation func-
½u ¼ ½ue þ½up ð100Þ tion f. Equation (102) represent equations of
state which define the variables conjugate to
½ue ; D and Z; respectively. Equation (102a)
C ¼ ED þ gp defines the traction vector t as function of
¼ 12ð1
D1 ÞK1 ½u1 e
2
elastic displacement discontinuities and da-
2 2
mage variables. Equation (102b) defines the
þ 12ð1
D2 ÞK2 ½u2 e þ12ð1
D3 ÞK3þ h½u3 e iþ vector Y of variables conjugate to the damage
2
þ 12K3
h½u3 e i
þgp ðZp Þ ð101Þ variables Di (i ¼ 1; 2; 3) collected in vector D;
notice that Yi (i ¼ 1; 2; 3) have the dimension
of an energy per unit area, they are the
@C @C @C @gp ðZp Þ analogous of the ERR in FM. In Equation
t¼ ; Y ¼
; wp ¼ ¼ ð102Þ (102c) the static internal variable wp is defined
@ ½ue @D @Zp @Zp
as conjugate to Zp. In Equations (103) a
  function f which governs the activation of
f ¼ f t; Y; wp ; D r0; f l’ ¼ 0; ’
lZ0 ð103Þ damage and plasticity is introduced, together
with the loading-unloading conditions and the
    plastic-damage multiplier l: ’ Nonassociated
’
½u’ p ¼ l up t; Y; wp ; D l; Z’ p ¼ lZp t; Y; wp ; D l’ ð104Þ plastic flow rules for plastic displacement
discontinuities ½up and for the kinematic
  internal variable Zp are given in Equations
D ’
’ ¼ l d t; Y; wp ; D l; Di A½0; 1 ð105Þ (104). Finally, the evolution of the damage
500 Damage and Fracture Mechanics Techniques for Composite Structures
variables is given in Equation (105), again in a

nonassociated format. @gp Zp 

The above class of interface models has the wpnþ1 ¼   wpnþ1 Zpnþ1 ð110Þ
main characteristics necessary to the modeling @Zp 
nþ1
of composite delamination: anisotropic da-
mage; unilateral effect; and irreversible displa-  
cement discontinuity. It is worth noticing the fnþ1 ¼ f tnþ1 ; Y nþ1 ; wpnþ1 ; Dnþ1 r0
formal analogy between the above interface ð111Þ
fnþ1 Dl ¼ 0; DlZ0
models and the class of elastic–plastic-damage
models for composites presented in Section
 
3.09.5.2 (Equations (35)–(41)). The main dif- D½up ¼ l up tnþ1 ; Y nþ1 ; wpnþ1 ; Dnþ1 Dl
ference consists in the simplification of the flow   ð112Þ
DZp ¼ lZp tnþ1 ; Y nþ1 ; wpnþ1 ; Dnþ1 Dl
rule that in the present IDM context depends
on one activation function only. Some of the
symbols in Equations (35)–(41) and (100)–  
DD ¼ l d tnþ1 ; Y nþ1 ; wpnþ1 ; Dnþ1 Dk; Di A½0; 1
(105) coincide; variables with the same symbols
play in fact an equivalent role in a CDM ð113Þ
description and in the present IDM one. As for
the models of Section 3.09.5.2, it is important The solution of the above nonlinear constitu-
for the various specialization considered in tive law integrated in time which relates the
applications, to check the nonnegativity of the traction vector tnþ1 to the total displacement
dissipation function. In the present IDM discontinuity vector ½unþ1 can be obtained
context, for the models represented by Equa- through an elastic predictor phase, a check of
tions (100)–(105), this reads consistency and a plastic-damage corrector
phase, as illustrated below.
o ¼ tT ½u’ 
C ’
wp Z’ p
’ ¼ tT ½u’ p þY T D ð106Þ
(i) Step 1. Elastic predictor
A purely elastic trial response to ½unþ1 is
first computed by fixing at the values
3.09.6.2.2 Numerical time integration reached at tn the variables D; w; ½up ; Z; l;
An interface model can be practically used in governing the nonlinear, irreversible inter-
nonlinear FE analyses when introduced in face behavior
special interface FEs, as discussed in Section  p
ttrial
inþ1 ¼ ð1
Din ÞKi ½ui nþ1
½ui n ; i ¼ 1; 2
3.09.7.2.2, after a numerical time integration þ
p
along a time interval ½tn ; tnþ1 : The remarks at ttrial
3nþ1 ¼ ð1
D3n ÞK3 ½u3 nþ1
½u3 n þ ð114Þ

the beginning of Section 3.09.5.2.2 still hold in þ K3
½u3 nþ1
½u3 pn

the present IDM framework; a backward
difference integration rule is here applied in
order to numerically integrate in time Equa- 1  2
trial
Yinþ1 ¼ Ki ½ui nþ1
½ui pn ; i ¼ 1; 2
tions (100)–(105). 2 ð115Þ
Given all quantities at the beginning of the 1 2
trial
Y3nþ1 ¼ K3þ ½u3 nþ1
½u3 pn þ
time step tn and given the value of total 2
displacement discontinuities ½unþ1 at time
instant tn þ 1, all other variables are sought at (ii) Step 2. Check of consistency
time instant tn þ 1 by solving the following set The plastic-damage condition is checked on
of algebraic nonlinear equations the basis of the elastic trial response
 
½unþ1 ¼ ½uenþ1 þ½upnþ1 ð107Þ trial
fnþ1 nþ1 ; Y nþ1 ; wpn ; Dn r0 ?
¼ f ttrial trial
ð116Þ

tinþ1 ¼ ð1
Dinþ1 ÞKi ½ui enþ1 ; i ¼ 1; 2

trial
If fnþ1 r0; then the response is purely elastic
 and the solution at tn þ 1 is given by the
t3nþ1 ¼ ð1
D3nþ1 ÞK3þ ½u3 enþ1 þ ð108Þ
 elastic trial response, the process stops. If
þ K3
½u3 enþ1
trial
fnþ1 40 then the response is plastic-damage
and a plastic-damage corrector phase is
necessary, the process continues from Step 3.
1 2 (iii) Step 3. Plastic-damage corrector
Yinþ1 ¼ Ki ½ui enþ1 ; i ¼ 1; 2
2 ð109Þ The process is plastic damage, hence the
1 2 solution will give a positive multiplier
Y3nþ1 ¼ K3þ ½u3 enþ1 þ
2 increment Dl: Equations (107)–(113) must
 The Particular Role of Interfaces and Interface Mechanics 501
be solved with fnþ1 ¼ 0 for given ½unþ1 ; i.e., can be found in Corigliano (1993), Schellekens
  and de Borst (1994), Hashagen et al. (1999),
tinþ1 ¼ ð1
Dinþ1 ÞKi ½ui nþ1
½ui pnþ1 ; i ¼ 1; 2 and Hashagen and de Borst (2000).

t3nþ1 ¼ ð1
D3nþ1 ÞK3þ ½u3 nþ1
½u3 pnþ1 þ

þ K3
½u3 nþ1
½u3 pnþ1
ð117Þ 3.09.6.3 Examples of Interface Models
In this section some examples of interface
1  2 models proposed and used in the literature are
Yinþ1 ¼ Ki ½ui nþ1
½ui pnþ1 ; i ¼ 1; 2
2 ð118Þ presented as particular cases of the more
1 2 general formulation (Equations (100)–(105)).
Y3nþ1 ¼ K3þ ½u3 nþ1
½u3 pnþ1 þ
2

 3.09.6.3.1 Elastic-damage interface model

@gp ðZp Þ
wpnþ1 ¼   wpnþ1 ðZpnþ1 Þ ð119Þ This model was presented in Allix and
@Zp 
nþ1 Corigliano (1996) as a modification suitable
for mixed-mode crack propagation studies of
  previous elastic-damage models proposed in
fnþ1 ¼ f tnþ1 ; Y nþ1 ; wpnþ1 ; Dnþ1 ¼ 0 ð120Þ
Allix and Ladevèze (1992), Corigliano (1993),
 
and Allix et al. (1995). A similar model was
D½up ¼ l up tnþ1 ; Y nþ1 ; wpnþ1 ; Dnþ1 Dl also discussed in Allix et al. (1998) and
  ð121Þ Corigliano and Allix (2000) with reference to
DZp ¼ lZp tnþ1 ; Y nþ1 ; wpnþ1 ; Dnþ1 Dl
parameter identification. The model in point
 
can be obtained from Equations (100)–(105) by
DD ¼ l d tnþ1 ; Y nþ1 ; wpnþ1 ; Dnþ1 Dk dropping all the terms related to the plastic
ð122Þ behavior (with superscript or index p) and by
Di A½0; 1
choosing the following format for the damage
activation function f and for the damage
The above set of Equations (117)–(122) is a evolution vector-valued function ld
nonlinear system which can be usually sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 a  a  affi
solved by means of a Newton–Raphson Y1 Y2 Y3
f ¼ þ þ
l
1 ð124Þ
procedure. In Corigliano (1993) an example Y01 Y02 Y03
of solution has been presented for a subclass
of the interface Equations (100)–(105). ’
D’ i ¼ gi l; Di A½0; 1; i ¼ 1; 2; 3 ð125Þ

3.09.6.2.3 Consistent tangent matrix D3 ¼ 1 ) D1 ¼ D2 ¼ 1 ð126Þ

The remarks in Section 3.09.5.2.3 with where a; Y0i ; gi ; i ¼ 1; 2; 3 are material model
respect to the problem of computing the parameters. Equations (126) imply that when
consistent tangent operator hold without mod- complete damage is reached for mode I (open-
ifications also in the present IDM context. The ing mode, in direction 3, cf. Figure 28) the
consistent tangent operator is the Jacobian of interface is considered completely damaged in
the equation connecting tractions and total modes II and III (shearing and tearing modes,
displacement discontinuities at the time instant in directions 1 and 2), too.
tn þ 1, as defined by the numerical integration The mechanical dissipation (Equation (106))
scheme adopted, i.e., by Equations (107)–(113) becomes, in the present case
dt ¼ Jd½u ð123Þ ’
’ ¼ Y TD
o ¼ tT ½u’ 
C ð127Þ

where J is the consistent tangent 3  3 matrix which is never negative.

which has to be computed from the system of As observed analogously in Section
Equations (107)–(113). As in the case of CDM 3.09.5.2.3 for the example of the 1D-CDM
of Section 3.09.5, the computation of J implies model, the damage flow rule given by Equa-
taking the derivatives of each equation and tions (103), (124), and (125) can be expressed in
distinguishing the cases of elastic unloading an equivalent format
and plastic-damage loading; moreover the "sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 a  a  affi#
unilateral effect also obliges to distinguish Y1 Y2 Y3
cases with positive or negative tractions in Y%  max þ þ
t0 rt Y01 Y02 Y03 ð128Þ
direction 3. Examples of computation of
consistent tangent matrix for interface models Di ¼ gi hY%
1iþ ; i ¼ 1; 2; 3
502 Damage and Fracture Mechanics Techniques for Composite Structures
The pure mode response of the above interface Z 1
model for an assigned history of monotonically a
Gc ¼ Y ðDÞdD ¼ gY0
increasing displacement discontinuities reads 0 ða þ 2Þ
"  #
g þ 1 ðaþ2Þ=a

1 ð130Þ
t ¼ K ½u g
rffiffiffiffiffiffiffiffi
2Y0
if 0r½ur½u0 ¼ When loaded in mixed-mode conditions the
K
  total energy dissipated in the damage process is
K a=2 aþ1 ð129Þ
a combination of the energy involved in the
t ¼ K ð1 þ gÞ½u
Kg ½u
2Y0 single modes; the following holds
 1=a
1þg Z Z Z
if ½u0 r½ur½uc ¼ ½u0
g GcT ¼ Y1 dD þ Y2 dD þ Y3 dD
DP DP DP
 GII þ GIII þ GI ð131Þ
Parameters K; g; Y0 in the above equations
must be considered as particular values of para- where the symbol DP means a decohesion
meters K1 ; K2 ; K3 ; g1 ; g2 ; g3 ; Y01 ; Y02 ; Y03 process (DP), i.e., a history of displacement
depending on the pure mode considered (3 discontinuities which leads to complete deco-
for mode I, 1 for mode II, 2 for mode III). hesion. The energy dissipated in a DP generally
In Figure 29 the above analytical pure-mode depends on the combination of different modes
response is shown at varying value of para- in the loading process.
meters a and g, having fixed the other para- An interesting feature of the above elastic-
meters K and Y0 to the values given in the damage interface model is the fact that in Allix
figure legend. The value of fracture energy Gc is and Corigliano (1996) it has been shown that
equal to 0.4 (N mm
1) for the three cases of when gi ¼ g; 8i ¼ 1; 2; 3 it is possible to deduce
Figure 29. Notice that the above analytical analytically a failure locus which satisfies an
pure mode response coincides with the 1D equation similar to Equation (18), which holds
p ffiffiffiffiffiffiffiffiffiffiffiffiffi of Equation (71) when a ¼ 1; g ¼
response for a FM study of mixed-mode conditions of
Y0 =Yc : fracture propagation. In other words, mixed-
In the introduction to Section 3.09.6 it was mode failure conditions as governed by the
recalled that the area under the traction- interface model in point satisfy the following
displacement discontinuity plot of a cohesive equations
or interface model coincides with the value of  a    
fracture energy Gc. For the interface model in GI GII a GIII a
þ þ ¼1 ð132Þ
point this in turn coincides with the mechanical GIc GIIc GIIIc
dissipation necessary to completely damage the
material (cf. Equation (72)), its analytical Equations (131) and (132) complete the con-
expression is the following nection between FM and IDM given by
Equation (99), extending the energetic equiva-
lence also to mixed-mode conditions.
It is also interesting to compare the initial
delamination criterion of the present interface
model as given in the traction space by the
activation function (Equation (124)) to the
delamination criterion of Hashin given by
Equation (31) discussed in Section 3.09.4.1.
For the special case a ¼ 1; t01 ¼ t02 ; being
t0i ; i ¼ 1; 2 the values of tractions in pure
modes II and III at the end of the elastic range,
the damage process starts when
 2  2 
t3 t þ t22
þ 1 ¼1 ð133Þ
t03 t01

which coincides with Hashin’s delamination

criterion (Equation (31)). The elastic-damage
model presented above, and its version given in
Figure 29 Pure-mode elastic-damage interface re- Corigliano and Allix (2000), can be used for the
sponse (Equation (129)) (K ¼ 1  105 N mm
2, simulation of delamination processes in 3D
Y0 ¼ 0.0125 N mm
1, Gc ¼ 0.4 N mm
1). mixed-mode conditions.
 The Particular Role of Interfaces and Interface Mechanics 503
3.09.6.3.2 Elastic–plastic softening interface puted also
models
t0
t ¼ K ½u if 0r½ur½u0 ¼
Elastic–plastic softening interface laws for K
t0 K
pffiffiffi 
the simulation of decohesion processes which
t ¼  pffiffiffi  b þ h½u ð138Þ
lead to debonding and delamination in com- K b þ Ht0
posites are an alternative to the use of elastic- pffiffiffi
b
damage interface models. Two possible aniso- if ½u0 r½ur½uc ¼

tropic elastic–plastic softening interface models H
are presented. The first model was proposed in The area under the traction-displacement dis-
Corigliano (1993); it can be obtained from continuity plot is in the present case given by
Equations (100)–(105) with the following
pffiffiffi
special choices t0 b
  Gc ¼
ð139Þ
2H
l d t; Y; wp ; D ¼ 0 ð134Þ
Notice that in Equations (138) and (139) a

 negative value for the parameter H must be
gp Zp ¼ 12HZ2p ; assumed.
  The plasticity model with isotropic softening
f ¼ f t; wp
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð135Þ proposed by Schipperen and de Borst (2001) is
 2  2  
t1 t2 ht3 iþ 2 now considered. Its main feature is the plastic
¼ þ þ
wp
1 activation function f which, in the notation
t01 t02 tþ
03
used in this chapter, can be given the following
form
bi ti ’
½u’ i p ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi l; i ¼ 1; 2;  
 2  2
t1 t2
b1 t1 þ b2 t22 þ b3 ht3 i2þ
2
f ¼ f t; wp ¼ þ
ð136Þ t01 t02
b3 ht3 iþ  
½u’ 3 p ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ’
l; Z’ p ¼ l’ t23 1 1
þ þ t3 þ


wp
1 ð140Þ
b1 t1 þ b2 t22 þ b3 ht3 i2þ
2 þ

t03 t03 t03 t03

In the above equations H is a parameter where t%0 is an initial equivalent yield traction,
governing the hardening (when H40) or t01 and t02 are initial yield limits in directions 1
softening (when Ho0) law, and t01 ; t02 are and 2 respectively, and tþ

03 ; t03 are the tensile
the traction values at the end of the elastic and compressive yield tractions in the direction
phase for pure mode II and pure mode III 3, normal to the interface plane.
loading, respectively; tþ 03 is the elastic limit for
The model is completed by a nonlinear
mode I loading in tension; bi ; i ¼ 1; 2; 3 are softening rule and by an associated flow rule
model parameters governing the nonassociated for the irreversible displacement discontinuities
flow rule for plastic displacement discontinuity (see Schipperen and de Borst, 2001).
rates.
The positive part of traction t3 has been 3.09.6.4 Interface Model Identification
introduced in the activation function f and in
the flow rule for ½u’ 3 p in order to avoid damage As anticipated in Section 3.09.6.1, the issue
(i.e., softening) for compressive loading in of parameter identification is particularly im-
direction normal to the interface. Notice that portant in the case of interface models because
the initial activation function coincides with only indirect experimental information can be
the Hashin’s delamination criterion (Equation obtained. Two alternative identification proce-
(31)) when t01 ¼ t02 : The mechanical dissipa- dures can in general be followed. In the first
tion (106) becomes one, an attempt is done to assign to each model
parameter a precise physical meaning; from
o ¼ tT ½u’ 
C ’ ¼ tT ½u’ p
wp Z’ p
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  this, usually, a specific experimental test is
defined which can give data for the direct
¼ b1 t21 þ b2 t22 þ b3 ht3 i2þ
Hl l’ ð137Þ
parameter determination. The second possible
approach to parameter identification is a
which is never negative in the case of softening completely indirect one; in this case the model
behavior with Ho0: parameters are determined as optimal values
The pure mode response of the above which minimize the discrepancy (measured by
elastic–plastic softening interface model for an ad hoc defined norm) between experimental
an assigned history of monotonically increas- data and equivalent data coming from the
ing displacement discontinuities can be com- numerical FE simulation of the experiment in
504 Damage and Fracture Mechanics Techniques for Composite Structures
which the model to be identified has been identified starting from the knowledge of
introduced. fracture energies in pure mode situations
The first approach to parameter identifica- GIc ; GIIc ; GIIIc ; of the coordinates of a point
tion of interface models was discussed in belonging to the failure locus in mixed-mode
Corigliano (1993), Allix and Corigliano conditions and by exploiting Equation (132).
(1996), and Corigliano and Allix (2000); it is Finally, parameters gi can be evaluated
here exemplified with reference to the elastic- through Equation (130) applied for each
damage model of Section 3.09.6.3.1. pure-mode, separately.
The model in point depends on 11 para- The above identification must be considered
meters: the interface stiffnesses K1 ; K2 ; K3þ ; as a procedure to obtain a first guess for
K3
; parameters Y01 ; Y02 ; Y03 ; and a which interface parameters. In Allix et al. (1998) and
appear in the damage activation function; and Corigliano and Allix (2000) the identification
parameters gi ; i ¼ 1; 2; 3 which govern the of an elastic-damage interface model similar to
evolution of damage variables. A possible the one here discussed has been based on the
way to determine the values of interface hypothesis that in the delamination process
stiffnesses is to make the hypothesis that the initiation and propagation phases must be
interface is equivalent to a fictitious layer of considered as two different phenomena. Due
small thickness e which separates the laminas. to the fact that the interface model tries to
By considering the strain field constant along simulate both initiation and propagation, the
the thickness of the fictitious layer and defining interface parameters can be divided in two
E3 its Young modulus groups: parameters which mainly govern the
initiation phase and parameters which mainly
½u3  govern the propagation phase.
t3 ¼ s33 ¼ E3 e33 ¼ E3 ¼ K3 ½u3  ð141Þ
e The above remark can help in particular for
the identification of interface elastic stiffnesses.
From the above equation it can be deduced The identification of K1 ; K2 ; K3þ ; K3
by
that K3 ¼ E3 =e; similar equations hold also for means of Equations (141) and (142) is in fact
other directions, the values of interface stiff- based on the hypothesis that the properties of
nesses can, therefore, be computed as the interface can be considered equivalent to
2G13 2G23 E3 those of pure resin. This not completely true:
K1 D ; K2 D ; K3þ D ð142Þ the properties of the resin inside the laminate
e e e
are in general different from those of the pure
where G13, G23 are shear moduli of the resin. A possible way to better identify the
fictitious layer. For numerical applications, interface stiffnesses is to make reference to the
G13, G23, and E3 can be assumed equal to the influence of these parameters on the interlami-
analogous values of an homogenized layer of nar stress concentration at the edges of
the composite or to the values attributed to the laminates, i.e., to edge effects. In particular
matrix; e can be assumed equal to a fraction of the length of penetration xc of interlaminar
the layer thickness. Parameter K3
is the stresses is directly influenced by the ratio of the
interface stiffness for compression in a direc- interface stiffness over an equivalent Young
tion normal to the interface; it is introduced in modulus of the layer. By assuming that in the
order to model the unilateral effect and it can initiation process of delamination the distance
be considered as a penalty parameter that xc coincides with the penetration length of
allows for unilateral constraint conditions to damage, interface stiffnesses can in principle be
be satisfied: for this reason it has to be chosen identified starting from experimental data
with a very high value, for simplicity it can be concerning delamination damage initiation.
made equal to K3þ : Parameters Y01 ; Y02 ; Y03 The second procedure which can be adopted
are initiation damage energies in pure mode for interface parameter identification is alter-
situations, from the definition of Yi, they can native to the first one and is based on a
be directly related to the critical level of coupling between numerical simulation and
interlaminar stresses t0i or, equivalently, of experimental data, as usually done in the
displacement discontinuities ½ui 0 at the begin- framework of indirect parameter identification.
ning of the damage process These procedures have been applied for identi-
fication of nonlinear, irreversible constitutive
1 1t2 models of various kind (see, e.g., Mahnken and
Y0i ¼ Ki ½ui 20 ¼ 0i ; i ¼ 1; 2; 3 ð143Þ Stein, 1996; Geers et al., 1999), they are based
2 2Ki
on the comparison of data coming from
Parameter a is strictly connected to the shape experiments and analogous data coming from
of the failure locus in mixed-mode conditions a complete FE numerical simulation of the test
of fracture propagation; it can, therefore, be used in the experiment (see Figure 30). In
 Computational Damage Mechanics for Composites 505
the body of the composite by means of classical
FE and on the interface surfaces by means of
IFE. Once the general formulation for FE
modeling is available, the problem arises in
choosing particular FE to be used for spatial
discretization.
The first possible choice is to discretize fully
the composite under study by means of brick
FE for the single layers, coupled with interface
FE; in this way it is in principle possible to
describe all the details of the structure under
study. As already observed, this solution is
only applicable to special situations or inside
particular global strategies due to the very
large number of degrees of freedom (dof)
which should be managed in order to avoid
Figure 30 General scheme for an indirect identifi-
locking phenomena induced by large length/
cation procedure.
thickness ratios.
The above remarks have motivated research-
Corigliano and Mariani (2001) indirect para- ers to look for different solutions. Three basic
meter identification based on the use of the ways can be followed: the first one tries to
extended Kalman filter has been applied to the formulate special FE which allow for the
identification of time independent and time description of layered solids; the second one
dependent interface models. The experimental exploits particular geometrical and/or loading
data considered were taken from interlaminar conditions in an attempt to simplify the whole
fracture specimens like the DCB test for mode numerical model; the third one tries to
I or the ENF test for mode II; the same tests reformulate the whole problem by decoupling
were completely simulated via FE. The same computations at different levels. An example of
issue has been considered in Bolzon et al. the first approach is given by the 3D brick
(2002a). layered FE (see, e.g., Hellweg et al., 1993)
which allow for the description of more layers
3.09.7 COMPUTATIONAL DAMAGE by means of a brick element. Multilayer
MECHANICS FOR COMPOSITES volume elements for the simulation of damage
in composites have been also used in Nguyen
Sections 3.09.5 and 3.09.6 of this chapter (1998).
have presented CDM and IDM formulations Another noteworthy example of the first
for the description of damage phenomena in approach to composite damage simulation
composite materials and structures. The pur- concerns the formulation of solid-like shell
pose of this section is to highlight possible elements put forward (Parisch, 1995; Hashagen
computational procedures which can be used et al., 1995, 1999), in an attempt to analyze
in order to exploit these formulations. As in both intralaminar and interlaminar stresses by
the whole chapter, the presentation will focus means of a shell element. The formulation
on particular aspects which are relevant to proposed in Hashagen et al. (1995) is based on
damage composite mechanics, with particular a 3D continuum theory with 16 geometrical
reference to laminates. For a comprehensive nodes. Three translational dof are defined for
description of computational procedures, with each node, additionally four internal dof are
particular reference to FE, the reader can refer introduced to account for internal stretching,
to other chapters of this series and to the thus yielding a fully 3D field of membrane and
references therein. CDM and IDM models can bending strains. The proposed solid-like shell
be used in the framework of the so-called elements can be coupled with interface ele-
mesomechanical approach to laminate struc- ments which can be connected at their top and
tures; laminates are described by means of bottom surfaces. The use of special FE for the
single layers and interfaces which separate the simulation of propagating delaminations in
layers (see Figure 25). layered composite structures is also adopted in
The first point to be discussed is the way in Sprenger et al. (2000) and Wagner et al. (2001).
which discontinuous displacements associated Special FE with embedded-displacement-
with the discrete crack description of interfaces discontinuities able to describe discrete crack
can be accommodated in the FE model. A processes have been proposed starting from
general formulation is needed which allows the different formulations (Lotfi and Shing, 1995;
introduction of spatial discretizations inside Bolzon and Corigliano, 1997b, 2000; Möes
506 Damage and Fracture Mechanics Techniques for Composite Structures
et al., 1999; Wells and Sluys, 2001; Mariani and Multiscale approaches belong to the class of
Perego, 2001; Wells et al., 2002). These kind of methods which try to de-couple the computa-
FE, originally conceived for the simulation of tion of the response of an heterogeneous
complex crack processes in homogeneous material or structure in various parts. These
materials like concrete, can be applied also to methods couple the homogenization theory
the study of delamination processes in compo- (see, e.g., Sanchez-Palencia, 1980) with the use
sites, as in Remmers et al. (2001). of numerical FE models of the composite
The generalized plane strain problem which under study. The basic idea of multiscale
can be used for the study of special configura- methods is to compute the response of the
tions of laminates under tension (see, e.g., whole composite by iteratively solving micro-,
Schellekens and de Borst, 1993a; Herakovich, meso-, and macro-FE models which are related
1998; de Borst and Schipperen, 2002) is an to each other by equations coming from some
example of the second approach. In this case homogenization procedure.
the problem is reduced to 2D and the spatial Multiscale and multigrid methods have been
discretisation can be carried out by means of developed originally for the computation of
plane FE for the single layers and linear elastic responses of composites (see, e.g., Fish
interface elements. and Wagiman, 1993; Fish and Belsky, 1995a,
Another typical example of the second 1995b) and subsequently applied to the study
approach is the use of plate/shell laminate of plastic and damage processes in composites
theories (see, e.g., Herakovich, 1998; Hyer, (Devries et al., 1989; Fish et al., 1997, 1999;
2000) which reduce the spatial dimension of the Fish and Yu, 2001a, 2001b; Thionnet and
problem by making use of structural theories. Renard, 1998; Ghosh et al., 1996, 2001; Lee
An abundant literature exists concerning linear et al., 1999; Voyiadjis et al., 2001; Xia and
elastic analyses of laminate structures, a great Gurtin, 2001).
number of plate and shell FE have been The majority of the above multiscale proce-
proposed (see, e.g., Ochoa and Reddy, 1992; dures are based on the introduction of a double
Alfano et al., 2001) but unfortunately these are scale asymptotic expansion of the unknown
usually not suitable for the study of delamina- fields. In contrast to this approach, in Wu et al.
tion processes due to the difficulties in recover- (1989) (see also Karihaloo et al., 2001), an FE
ing the interlaminar stresses. Laminate plate procedure for the simulation of nonlinear
and shell FE can, nevertheless, be used for the behavior of composites with periodic hexagonal
study of damage processes which involve the arrays of elastic fibers and plastic matrix has
layers only; in this case a single layer damage been proposed in which the computation of the
model like that presented in Section 3.09.5.3 constitutive law at the macroscale is based on a
can be integrated in a layered shell element microscale FE model, without making explicit
(see, e.g., Herakovich, 1998). In Shu and use of the asymptotic expansion method.
Soldatos (2001), a generalized 2D plate theory The advantage of multiscale methods is
is applied to the study of delamination for represented by the possibility to decouple the
weakly bonded laminates. response of the heterogeneous solid at different
Examples of the third approach for the levels thus avoiding heavy spatial discretisation
damage mesomodeling of composites are: the for the whole solid under consideration. The
study of edge effects by means of a boundary meshes used in multiscale analysis concern in
layer problem; the study of damage processes fact RVE for the micro- and mesoscale with the
through local reanalysis procedures; and the whole system treated as homogenous.
use of multiscale strategies. Due to the fact that multiscale approaches
The study of boundary or edge effects can be are based on homogenization theory, the
decoupled from that of the whole laminate hypothesis of periodicity (or statistical periodi-
when nonlinear damage effects are concen- city) must be satisfied. This represents a
trated along the interfaces, in this case a 2D limitation, in particular for the analysis of
problem can be formulated which allows for boundary layer effects.
the study of delamination near external edges A systematic use of basic principles of
or around holes; this approach has been homogenization can also be used to formulate
applied, e.g., by Allix et al. (1989) and in a consistent way constitutive models of the
Daudeville and Ladevèze (1993). composite at different scales (e.g., Allen, 2001).
Local reanalysis can be formulated by Caiazzo and Costanzo (2001a, 2001b) pro-
first computing the overall behavior of a posed a computational procedure called dis-
laminate by classical laminate theories and crete damage space homogenization method
subsequently re-analyzing the stress state in (DDSHM) to predict the constitutive response
each Gauss point by means of nonlinear of layered composite materials containing
damage models. growing cracks.
 Computational Damage Mechanics for Composites 507
Worth mentioning among the possible stra- (1995), and Bolzon and Corigliano (1997a).
tegies which can be chosen in order to simulate General conditions for the loss of uniqueness
the nonlinear behavior of composite materials in the incremental response and the possible
and structures, are the applications of indirect instabilities in the response have been proved
parameter identification techniques (Kalman in Bolzon and Corigliano (1997a) with refer-
filtering, least square methods) coupled with ence to a class of interface models which satisfy
numerical homogenization method presented the maximum dissipation principle (see also
in Giampieretti et al. (2000) and Bolzon et al. Nguyen, 1987).
(2002b). These innovative procedures aim at The availability of bifurcation and instability
identifying a homogenized model for the conditions unfortunately does not resolve
behavior of the composite system at the chosen another important point for the correct simu-
scale (meso or macro) by determining a set of lation of a fracture process in the presence of
unspecified parameters through the compar- multiple responses, like those of multiple
ison of the numerical response of a refined delaminations. This concerns the choice of
model at the lower scale (micro or meso) with the bifurcated path among the multiplicity of
that of the homogenized phenomenological possible solutions. Various proposals have
model. been made; the most physically sound seems
Once the spatially discretized FE modeling to be that of Bažant (see Bažant and Cedolin,
have been defined and the global strategy for 1991) based on thermodynamical considera-
the computation chosen, a series of problems tions.
related to the softening character of constitu- Besides the choice of one equilibrated path
tive laws remain to be solved in order to obtain among multiple responses, one additional
an efficient simulation of damage and fracture problem is the way in which the critical point
in composite materials and structures. in the global response is reached and the way in
The first problem is the already mentioned which a global softening response is followed.
mesh dependency of the response. This issue is It has been shown in various works (see, e.g.,
common to all simulations of damage pro- Schellekens and de Borst, 1993a; Corigliano,
cesses, and various solutions have been pro- 1993; Crisfield et al., 1997) that delamination
posed in the literature (see, e.g., de Borst, 2001 processes can lead to severe snap-through and
for a review paper), all of them trying to snap-back global responses. This strongly
introduce a characteristic length in the model. suggests the use of path-following techniques
A simple solution adopted, e.g., in Ladevèze based on an indirect control of the global
et al. (2000b), which can be applied in the case response (Riks, 1972; Crisfield, 1981) neither in
of laminate structures, is to force damage and/ loads nor in displacement. Algorithms belong-
or other softening variables to be constant ing to this family have been proposed for the
along the thickness of each layer (see Section study of delamination in Corigliano (1993),
3.09.5.3). The characteristic length is in this Schellekens and de Borst (1993a), and Hellweg
way quite naturally assumed equal to the layer and Crisfield (1998).
thickness. Another provision to avoid spurious An additional issue which has not been
mesh dependency is to make use of rate considered in this chapter is the necessity to
dependent damage models. As far as the take into consideration geometrical non-linear-
interface modeling is concerned, mesh depen- ities when delamination-buckling induced phe-
dency is not observed provided that the process nomena must be simulated. This is also
zone near the crack-tip is described with necessary for a correct description of the elastic
sufficient accuracy by FE. The reason of this behavior of thin laminates plates and shells. A
is the fact that in the interface models the possible way to formulate the problem in this
amount of fracture energy is introduced case is to make use of a Lagrangian formula-
naturally in the constitutive law and therefore tion for the description of the layer behavior
it does not depend on the mesh size. and to make the hypothesis that displacement
The second important difficulty to be over- discontinuities are small in the zones which are
come strictly related to the softening character not delaminated (see Allix and Corigliano,
of the constitutive models is the possibility to 1999). This allows for a simple formulation in
have bifurcations and instabilities in the global which the interface constitutive behavior can
response. This aspect is in particular important be derived without difficulties from that
in the presence of multiple delamination cracks formulated for linear geometry problems.
and is common to the study of multiple crack Formulations for delamination studies in the
processes in the framework of FM. Bifurcation presence of large displacements have been
and instabilities for a solid containing soft- proposed and used in Schellekens and de Borst
ening interfaces have been discussed in Cor- (1993a), Hashagen et al. (1995), de Borst and
igliano and Bolzon (1995), Costanzo and Allen Schipperen (2002), Remmers and de Borst
508 Damage and Fracture Mechanics Techniques for Composite Structures
(2001), and Qiu et al. (2001). More generally scribed in Section 3.09.6. The constitutive
interface models and interface FE have been behavior of the solid V can be nonlinear, in
used widely in a large displacement and strain particular damageable, and described by a law
context by Ortiz and co-workers (see, e.g., of the kind presented in Section 3.09.5. Each
Pandolfi et al., 1999). Some of the above- interface Gj is characterized by an interface
mentioned issues concerning computational constitutive law of the kind described in
aspects of damage and fracture simulations Section 3.09.6.
for composite materials and structures will be The body is subjected to body forces f in V
discussed below. and to surface loading F on the part SF of its
In Section 3.09.7.1 a general FE formulation external surface; kinematic constraints u ¼ u0
for the study of damage processes in the are imposed on Su. Inertia effects are neglected,
framework of the mesomodeling approach is small strain and displacement theory is con-
presented and subsequently specialized to the sidered. Define G as the union of all interfaces
generalized plane strain problem and to the Gj and V 0  V
G0 : The class of displacement
boundary layer solution. In Section 3.09.7.2 a vectors u which are zero on Su and regular in
layered brick FE element and an interface V 0 (virtual displacements) is named U; notice
element for plane problems are presented. that u (and the real displacements) can be
Path-following techniques are discussed in discontinuous on G0 .
Section 3.09.7.3 and two particular procedures The boundary value problem (bvp) consists
are described in detail. Section 3.09.7 closes of finding the fields of displacements u and
with a series of examples, presented in Section stresses r which satisfy the following equations
3.09.7.4, concerning PFA of a sandwich panel;
the formulation of failure loci for composite (i) Compatibility
materials by means of FE analyses; mesoscale
modeling of laminate structures with particular u is regular in V 0 and u ¼ u0 on Su ð144Þ
reference to delamination; parameter identifi-
cation issues.
(ii) Equilibrium
For any u AU:
3.09.7.1 FE Formulation for Mesomodeling of Z Z
Composite Materials and Structures rT eðu Þ dV þ tT ½u  dG
V0 G0
FE formulations suitable for mesomodeling Z Z
of composite materials and structures are ¼ f T u dV þ F T u dS ð145Þ
V0 Su
presented in this section. A general 3D
approach for damageable bodies with damage-
able interfaces is first presented in Section (iii) Constitutive law
3.09.7.1.1. The approach in Section 3.09.7.1.1
is then specialized to the case of an elastic body r ¼ rðeðuÞ; historyÞ in V 0 ;
with nonlinear interfaces in Section 3.09.7.1.2, ð146Þ
t ¼ tð½u; historyÞ on G0
to generalized plane strain in Section
3.09.7.1.3, and to the boundary layer solution
case in Section 3.09.7.1.4. In the above equations r and e are stress and
strain vectors defined as in Section
3.09.7.1.1 General 3D formulation 3.09.5.2.1.

Consider a composite material or structure In Equation (145), e(u) is related to the

occupying the domain V; suppose that the displacement field u through the linear differ-
composite contains a number n of internal ential equations of small strain theory. The
interfaces Gj (Figure 31) which can undergo vector t in Equation (146a) is the vector of
damage-decohesion processes like those de- tractions on surfaces Gj while [u] is the vector of
displacement discontinuities.
Equations (146(a) and (b)) represent gener-
ally nonlinear, irreversible constitutive laws for
the volume V 0 and for the interfaces Gj,
respectively. Due to the history-dependent
character of the laws (Equation (146(b)), initial
conditions on the value of history variables
must be supplemented to completely solve the
structural problem in point, which should be
Figure 31 A composite structure with interfaces G. defined more precisely as an initial bvp. The
 Computational Damage Mechanics for Composites 509
definition as a bvp is maintained due to the fact In the above equation U þ and U
are the
that dynamic effects are not considered. nodal displacement vectors for nodes which
The bvp (Equations (144)–(146)) is nonlinear belong to V þ and V
respectively and are on
due to the constitutive laws, it can be surface G. By means of the above description it
discretized in space by using the compatible is possible to define interface elements k ¼
FE method, as commented below. 1; y; nei which discretize the whole set G0 : A
The body V 0 is divided in nev FE and a complete formulation for an interface FE is
displacement model is introduced by means of presented in Section 3.09.7.2.2.
standard interpolation functions By introducing the interpolations given by
Equations (147)–(149) given by the weak
ui ¼ N i U i ; i ¼ 1; y; nev ð147Þ condition for equilibrium (Equation (145)),
the following space-discretized equilibrium
where N i are matrices of shape functions and equation is obtained
U i are vectors of nodal dof (nodal displace- Z Z
ments). The strain model is recovered from the X
nev X
nei
H vT
i BvT
i ri dV þ H intT
k BintT
k t k dG
above displacement model by applying to i¼1 Vii k¼1 Gki
Equation (147) the linear compatibility opera-
¼ mP ð150Þ
tor, thus obtaining Z Z
X
nev X
nes
P H vT
k N Ti F i dV þ H sT
l N Tl f l dS
ei ¼ B i U i ; i ¼ 1; y; nev ð148Þ k¼1 Vii l¼1 Suli

where matrices B i contain derivatives of shape

functions introduced in matrices N i : Vectors u In Equation (150) H vi ; H int s
k and H l are Boolean
of virtual displacements and e of virtual matrices which extract the dof of volume
strains are interpolated as u and e in Equations element i, interface element k, element l which
(147) and (148). has a side on the surface Su, respectively, from
As far as the interface variables are con- the global dof vector U
cerned, the interpolation of the displacement
discontinuity vector [u] on a generic surface Gj U vi ¼ H vi U; i ¼ 1; y; nev;
is derived directly from the interpolation used U int int ð151Þ
k ¼ H k U; k ¼ 1; y; nei;
for the displacement field u in the adjacent FEs
(Figure 32) U l ¼ H sl U; l ¼ 1; y; nes
s

½uk ¼ uþ
þ þ

int int P in Equation (150) is the vector of nodal loads

k
uk ¼ N k U k
N k U k  B k U k ð149Þ
equivalent to body and surface forces; m is a
nondimensional load parameter.
The bvp discretized in space is governed by
the equilibrium Equation (150) and the con-
stitutive laws (Equation (146)). The problem
can be solved after time discretization, i.e.,
subdivision of the global time interval in time
steps; each time step can in turn be solved by
means of a suitable iterative procedure, as
discussed in Section 3.09.7.3.

3.09.7.1.2 Elastic body with nonlinear

interfaces
An interesting specialization of the bvp
presented in Section 3.09.7.1.1 is obtained by
making the hypothesis that the volume V is
linear elastic; the constitutive law (Equation
(146a)) must in this case be substituted by the
following linear one
r ¼ EeðuÞ in V 0 ð152Þ

where E is the elastic stiffness matrix of the

body V. By substituting Equation (152) into
Figure 32 FEs neighbouring a discontinuity inter- Equation (150), taking into account the strain
face. model (Equation (148)) and the assembling
510 Damage and Fracture Mechanics Techniques for Composite Structures
operator (Equation (151a)), the following scopic effects are treated exactly in the same
special form of the space-discretized equili- manner.
brium equation is obtained By introducing the hygro-thermo-elastic
Z constitutive law in Equation (153) the follow-
X
nei
ing equilibrium equation can be obtained
K EU þ H intT
k BintT
k t k dV ¼ mP
k¼1 Vi
X
nei Z
Z  ð153Þ
X
nev
K EU þ H intT
k BintT
k t k dV ¼ mP þ P DT þ P DM
KE  H vT
i BvT
i E B
i i
v
dV H vT
i k¼1 Vi
i¼1 Vi
X
nev Z 
PDT  H vT
i BvT
i E i DTa dV ; ð155Þ
where the assembled elastic stiffness matrix KE i¼1 Vi
for the domain V has been defined. Xnev Z 
The formulation of an elastic body with PDM  H vT
i BvT
i E i DMb dV
i¼1 Vi
material nonlinearities concentrated on the
interface G0 only, has a particular relevance in
This is suitable for direct specialisation to the
the case of fracture propagation studies for
case of generalized plane strain, as discussed in
concrete-like materials (see, e.g., Bažant and
Section 3.09.7.1.3.
Oh, 1983; Carpinteri, 1989; Maier et al., 1993).
The same formulation can be applied advanta-
geously in the study of delamination problems 3.09.7.1.3 Generalized plane strain
where the single fiber-reinforced layers can be
conceived as purely elastic (the domain V 0 ) and Generalized plane problems usually are
connected by nonlinear interfaces (the set G0 ); defined as those problems in which stresses
this approach has been used, e.g., in Corigliano and strains generally have all the components
(1993), Schellekens and de Borst (1993a), and nonzero, but do not vary along a prescribed
Allix et al. (1995). direction. A meaningful example of generalized
One noteworthy computational advantage plane problems is the generalized plane strain
of Equation (153) is the fact that the dof applied to the study of delamination testing
relevant to the elastic domain V only can be specimens subjected to a uniaxial tensile or
condensed statically as shown in Bolzon and compressive load. If the length of the laminate
Corigliano (1997b); the nonlinear analysis can, is large when compared to its width and
therefore, be reduced to the dimension of thickness and if all loadings are constant along
interface dof only with a drastic reduction of the laminate length, it is possible to assume
computational effort. The same advantage can that, at a certain distance from the ends of the
be obtained by means of the boundary element specimen, the in-plane displacements in the
method (BEM) as shown in Maier et al. (1993) x2
x3 plane are independent of the coordinate
and Maier and Frangi (1998). x1 (Figure 33). The following displacement
Remaining in the framework of Equation field proposed by Pipes and Pagano (1970),
(153), it is easy to introduce hygro-thermal
effects in the computation. Both thermal and
hygroscopic effects can be important in the
analysis of composite behavior; they can in fact
induce nonuniform straining which in turn can
be the reason for the development of additional
stresses. As discussed, e.g., in Herakovich
(1998), a hygro-thermo-elastic constitutive
law for the body V can be formulated in
following form

e ¼ ee þ et þ eh ; ee ¼ E
1 r; et ¼ DTa;
ð154Þ
eh ¼ DMb

where ee ; et ; eh are the elastic, thermal and

hygroscopic strains, respectively; DT is the
temperature change with respect to a reference
value; a is the vector of coefficients of thermal
expansion; DM is the percent of moisture (by
weight) absorbed; b is the vector of coefficients
of hygroscopic (or moisture) expansion. As Figure 33 Geometry and loading for a uniaxial
shown by Equation (154), thermal and hygro- laminate specimen.
 Computational Damage Mechanics for Composites 511
can, therefore, be introduced By taking into account the hygro-thermo-
elastic constitutive law (Equation (154)), the
u1 ðx1 ; x2 ; x3 Þ ¼ m%e11 x1 þ u1 ðx2 ; x3 Þ virtual displacement field (Equation (158)) and
u2 ðx1 ; x2 ; x3 Þ ¼ u2 ðx2 ; x3 Þ ð156Þ the strain fields (Equation (160)), and recalling
u3 ðx1 ; x2 ; x3 Þ ¼ u3 ðx2 ; x3 Þ the fact that all unknowns depend on the
coordinate x2 and x3 only, the weak form of
where e%11 is a (normalized) strain prescribed in equilibrium can be rewritten in the following
the direction x1, and m is a load factor. format
From the displacement field (Equation
(156)), the strain field follows applying the For any u AU :
linear compatibility equations Z Z
eTr Eer ðu Þ dO þ tT ½u  dG
@u2 ðx2 ; x3 Þ @u3 ðx2 ; x3 Þ O0
Z Z G0 -O
e11 ¼ m%e11 ; e22 ¼ ; e33 ¼
@x2 @x3 ¼ f T u dO þ F T u ds
1 @u1 ðx2 ; x3 Þ O0
Z
@Ou
Z
e12 ¼ ;
2 @x2 ð157Þ þ DTaT Eer ðu Þ dO þ DMbT Eer ðu Þ dO
  O0 O0
1 @u2 ðx2 ; x3 Þ @u3 ðx2 ; x3 Þ Z
e23 ¼ þ ;
2 @x3 @x2
m eTm Eer ðu Þ dO ð161Þ
O0
1 @u1 ðx2 ; x3 Þ
e13 ¼
2 @x3
where the volume integrals have been substi-
Notice that the strain field does not depend on tuted by integrals on the plane x2
x3 of area O
the coordinate x1. and surface integrals by line integrals on the
A mesomodel approach for the solution of line s (see Figure 33); G-O means that the
laminates under tension loading can be for- integration for the interface internal force must
mulated by exploiting the above kinematic for be done along the linear interfaces which
each single layer, by introducing interfaces belong to the area O. The volume, surface,
between adjacent layers and solving the bvp thermal and hygroscopic loadings have been
(Equations (144)–(146)) reduced to 2D. This is assumed independent of the coordinate x1. The
here done in the hypothesis of linear elastic generalized plane strain formulation (Equation
behavior of the body V, in the presence of (161)) can be discretized spatially by means of
hygro-thermal effects. FE for the modeling of functions u1 ðx2 ; x3 Þ;
To this purpose first define the following u2 ðx2 ; x3 Þ; u3 ðx2 ; x3 Þ:
virtual fields of displacements and strains This approach, in the context of finite
displacements and strains, has been applied
u1 ¼ u1 ðx2 ; x3 Þ by Schellekens and de Borst (1993a, 1994) for
u2 ¼ u2 ðx2 ; x3 Þ ð158Þ the study of free edge delamination in speci-
u3 ¼ u2 ðx2 ; x3 Þ mens subject to tensile or compressive strain
loading.

@u2 ðx2 ; x3 Þ
e11 ¼ 0; e22 ¼ ;
@x2 3.09.7.1.4 Boundary layer solution
@u ðx2 ; x3 Þ
e33 ¼ 3 The purpose of boundary layer solutions is
@x3
the study of edge effects which can induce
1 @u1 ðx2 ; x3 Þ
e12 ¼ ; ð159Þ interlaminar stresses responsible of free edge
2 @x delamination. Being based on a superposition
  2 
1 @u2 ðx2 ; x3 Þ @u3 ðx2 ; x3 Þ principle, the use of boundary layer solutions is
e23 ¼ þ ;
2 @x3 @x2 usually made in the case of elastic composites.
1 @u1 ðx2 ; x3 Þ In Daudeville and Ladevèze (1993) it was
e13 ¼
2 @x3 observed that a boundary layer solution can
still be used in the particular case of an elastic
Then write the strain field (Equation (157)) and solid containing damageable interfaces (see
the virtual strain field (Equation (159)) in the Section 3.09.7.1.2).
following compact form Consider first an elastic laminate plate or
 T shell, of thickness 2 h, occupying the domain
e ¼ mem þ er ; em  e%11 0 0 0 0 0 ; V ¼ V1 ,V2 (Figure 34). When the thickness
 T
er  0 e22 e33 2e12 2e23 2e13 ð160Þ of the plate is small compared to the other
 T dimensions, the 3D solution of the elastic bvp
e ¼ er  0 e22 e33 2e12 2e23 2e13 in V can be seen as the superposition of two
512 Damage and Fracture Mechanics Techniques for Composite Structures
An interesting further specialization con-
cerns situations in which the curvature of the
plate edge boundary is small compared to
the thickness. In this case the variation of the
boundary solution along the direction tangent
to the external edge can be considered negli-
gible, hence a 2D bvp can be formulated in the
plane domain O drawn in Figure 34.
Taking into account the above considera-
tions, the bvp for the determination of edge
effects can be formulated. The class of dis-
placement vectors u (virtual displacements)
which are zero on S12 (the boundary between
the layer zone and the internal domain (see
Figure 34) and regular in O) is named U; hence
the edge bvp reads

(i) Compatibility
uboun is regular in O and uboun ¼ 0 on S12 ð164Þ
Figure 34 Boundary layer problem.
(ii) Equilibrium
solutions
For any u AU
u ¼ udom þ uboun ; r ¼ rdom þ rboun ð162Þ Z Z
rTboun eðu Þ dO ¼ F TR u ds ð165Þ
O S2 -O
The solution of the plate problem in the
internal domain has been noted with the
symbol dom, while the solution of the bound- (iii) Constitutive law
ary problem, localized near the edge of the
rboun ¼ Eeðuboun Þ in O ð166Þ
structure, has been marked with boun. The
meaning of the boundary solution is equivalent The above boundary layer problem can be
to the 3D effects which exist at the end of a extended to the presence of nonlinear softening
beam under flexure and which rapidly vanish interfaces between the layers of the laminate.
inside the beam by De Saint Venant’s principle. This is possible on the hypothesis that the
The boundary solution can be computed as a nonlinear interface behavior is activated in the
correction term to the domain solution by boundary zone only. In this case the interior
considering the fact that the latter generally solution is not affected by the nonlinear
does not satisfy the local static boundary behavior of the interfaces. This has been put
conditions on the external edge S2 (see Figure forward in Daudeville and Ladevèze (1993),
34); this means that the following residual term where a nonlinear, time-dependent bvp has
is generally nonzero been formulated with softening interfaces for
the study of free-edge delamination. In this
F R  F ðxÞ
rdom ðxÞna0 xAS2 ð163Þ
case the equilibrium Equation (165) must be
where F is the assigned surface load and n is the substituted with the following one (see also the
bvp Equations (144)–(146))
normal to the surface S2. The boundary
condition (Equation (163)) is satisfied only in For any u AU
an average sense along the thickness by the Z Z Z
domain solution. rTboun eðu Þ dO þ tT ½u  dG ¼ F TR u ds
O G0 -O S2 -O
The edge solution can, therefore, be com-
puted by applying the loading term FR on the ð167Þ
external boundary S2 and imposing zero value
for the displacements on the internal boundary 3.09.7.2 Special FEs for the Simulation of
S12, thus satisfying the De Saint Venant’s Damage in Composites
principle (Figure 34). The dimension of the
boundary layer zone can be estimated of the As discussed in the introduction to Section
order of the plate/shell thickness 2h. This 3.09.7, special FE have been formulated in
computation can be carried out in a postpro- order to build efficient computational tools for
cessing phase, after the plate solution of the the analysis of composite materials and struc-
laminate has been found. tures. The problem of computational efficiency
 Computational Damage Mechanics for Composites 513
is extremely important in the case of damage By defining an eight-component vector N p
and fracture simulations like those carried out which collects the above eight-plane functions
in the framework of the mesomodeling ap-
proach. Among the large number of proposals N Tp  ½N1 ðx; ZÞ N2 ðx; ZÞ N3 ðx; ZÞ N4 ðx; ZÞ
in the literature (enhanced laminates, layered N5 ðx; ZÞ N6 ðx; ZÞ N7 ðx; ZÞ N8 ðx; ZÞ ð169Þ
elements, solid-like elements, special inter-
faces), two examples of FE modeling for the complete set of shape functions of the brick
composites are shown in this section. These element, introduced in the 16-component
concern a layered FE and an interface FE, vector N Tbr ; can be built by interpolation along
discussed, respectively, in Sections 3.09.7.2.1 the thickness coordinate z
and 3.09.7.2.2. The elements described are
shown for their simplicity with the aim of 1h i
N Tbr ¼ð1
zÞN Tp ð1 þ zÞN Tp
clarifying the basic aspects which can be met 2
in FE modeling for damage simulation in 1h i zh i
¼ N Tp N Tp þ
N Tp N Tp  FT þ zCT ð170Þ
composites. 2 2

By means of the above vector N Tbr of shape

3.09.7.2.1 Layered brick element functions it is possible to build the complete
displacement model for the brick element
The element discussed here has been pre-
2 3 2 T 32 3
sented in Hellweg et al. (1993); the basic idea of ux N br 0T 0T Ux
the formulation is to introduce economies in 6 7 6 76
u ¼ 4 uy 5 ¼ 4 0T N Tbr 0T 54 U y 5
7
the computation of a layered FE by separating
the through-thickness from the in-plane inte- uz 0T 0T N Tbr Uz
2 3
gration. This approach can avoid the full T
U þ zW T
0T
0T
computation of contributions to the strain 6 7
¼64 0T UT þ zWT 0T 7
5
displacement matrix at every integration point.
T T T T
A 16-noded isoparametric brick is consid- 0 0 U þ zW
ered in its natural reference frame x; Z; z 2 3
Ux
(Figure 35). The shape functions of the top 6 7
 4 Uy 5 ð171Þ
and bottom surfaces of the brick element in
Figure 35 coincide with those of an eight- Uz
noded plane isoparametric FE; they can be
written in following way where the vector of nodal dof Ux have been
defined in the following way:
Ni ¼ 14ð1 þ xi xÞð1 þ Zi ZÞðxi x þ Zi Z
1Þ h i
U Tx  U Txbott U Txtop ;
for corner nodes
  ð168Þ " #
Ni ¼ 12 1
Z2i x2
x2i Z2 ðxi x þ Zi Z þ 1Þ Ux1 Ux2 Ux3 Ux4 ð172Þ
T
U xbott 
for mid-side nodes Ux5 Ux6 Ux7 Ux8

where bott and top mean bottom and top

surfaces of the brick; analogous expressions
define vectors Uy and Uz.
The equations between natural coordinates
x; Z; z and cartesian coordinates x; y; z are
governed by a regular transformation; this
is eight-noded isoparametric for the transfor-
mation of coordinates x and y, and a simple
linear equation for the transformation of
coordinate z
x ¼ xðx; ZÞ
x ¼ xðx; ZÞ ð173Þ
z ¼ zðzÞ ¼ zm þ hz

where 2h is the constant thickness of the brick

element in the cartesian space and zm is the
coordinate of the average plane along the
Figure 35 Topology of a 3D layered isoparametric thickness. This transformation rule implies
brick FE. that the brick element in the cartesian space
514 Damage and Fracture Mechanics Techniques for Composite Structures
can assume a curved shape on planes parallel
to the cartesian plane x–y, while the sides B ¼ B1 þ zB2
2 3
which are transformed of the sides parallel to z @UT T T
axis in the natural space must remain recti- 6 @x 0 0 7
6 7
linear. This kind of rule is, therefore, suitable 6 T 7
6 T @U T 7
for the description of layered composites with 6 0 0 7
6 @y 7
constant layer thicknesses. 6 7
6 1 7
As a consequence of Equation (173), the 6 0T 0 T T7
W 7
6 h
transformation of elementary volumes from 6 7
6 7
6 @UT @UT T 7
the natural space to the cartesian space assume 6 0 7
6 @y @x 7
the following format: 6 7
6 1 T @U 7 T7
6 T
6 0 W 7
  6 h @y 7
 x;x x;Z x;z  6 7
  41 @UT 5
  WT 0T
dV ¼  y;x y;Z y;z  dx dZ dz h2 @x
  3
 z;x z;Z z;z 
  @WT T T
 x; x;  6 0 0 7
 x Z 6 @x 7
¼ z;z   dx dZ dz 6 T 7
 y;x y;Z  6 T @W 7
6 0 0T 7
  6 @y 7
 x; x;  6 7
 x Z 6 T T T 7
¼ h  dx dZ dz ¼ hjJ j dx dZ dz ð174Þ 6 0 0 0 7
 y;x y;Z  6 7
þ z6
6 @W
T
@WT T 7
7 ð176Þ
6 0 7
6 @y @x 7
6 7
By means of the displacement model (172)) and 6 T7
@W 7
6 0T 0T
of the coordinate transformation (Equation 6 7
6 @y 7
(173)) it is possible to compute the strain field 6 7
4 @W 5
T
T T
model, governed by matrix B, after applying 0 0
the linear compatibility equation @x

It is now possible to compute the element

e ¼ BU stiffness matrix k; for a layered brick with n
2 3 elastic layers this reads
exx
6
6 eyy 7
7 Z Z Z ( "X nlay Z
#
6 7 k¼ T
B EB dV ¼ B T1 E n dz B1
6 e 7
6 zz 7 V x Z n¼1 zlayn
¼6 7 " #
6 2exy 7 nlay Z
X
6 7
6 2e 7 þ B T1 zE n dz B2
4 yz 5 zlayn
n¼1
2exz " #
nlay Z
X
2 3
@UT @WT þ B T2 zE n dz B1
6 @x þ z @x 0T 0T 7 n¼1 zlayn
6 7 " # )
6 7 Xnlay Z
6 @UT @WT 7
6 0T þz 0T 7 þBT2 2
z E n dz B2 hjJ j dZ dx ð177Þ
6 @y @y 7
6 7 n¼1 zlayn
6 1 T 7
6 0T 0T W 7
6 7
6 h 7 The advantage and efficiency of the layered
¼6 7
6 @UT @WT @UT @WT T 7 brick element can be appreciated from Equa-
6 þz þz 0 7
6 @y @y @x @x 7 tion (177); the dependence on the thickness
6 7
6 1 T @U T T7
@W 7 variables is condensed in the integration of the
6
6 0T W þz 7 material matrices and the strain displacement
6 h @y @y 7
6 7 matrices have to be computed as functions of
4 1 T @UT @WT 5
W 0T þz in plane coordinates x and Z only.
2 h3 @x @x
Ux
6 7 3.09.7.2.2 Interface element
 4 Uy 5 ð175Þ
Uz The formulation of an interface element for
plane problems which can be coupled with
eight-noded isoparametric elements with recti-
The matrix B in this equation can be split into linear sides is presented. In the case of plane
a linear combination of two matrices problems the geometry of an interface element
 Computational Damage Mechanics for Composites 515
is represented by a line which separates two The contribution to the equilibrium Equation
solid elements (Figure 36). The behavior of the (145) of the interface element can now be
interface element in its local reference frame computed as function of nodal dof in the local
(xl, yl) (Figure 36) is first derived. In order to reference frame
do this let us first define the displacement
Z Z 1
models for the upper ( þ ) and lower (
) nodes
tT ½u  dG ¼ tT Bint
l l dxU l ð181Þ
" # G
1
u

xl ðxÞ
u

l ðxÞ ¼
u

yl ðxÞ where 2l is the length of the interface element.
2 3T 2 3 The nodal force vector qintl and the stiffness
N1 ðxÞ 0 Ux1
6 7 6 7 matrix kint
l for an elastic interface can therefore
6 0 N1 ðxÞ 7 6 Uy1 7 be computed in the local reference frame as
6 7 6 7
6 N ðxÞ 0 7 6U 7
6 2 7 6 x2 7 Z
¼6 7 6 7  NU

l 1
6 0 N2 ðxÞ 7 6 Uy2 7 qint  BintT tl dx;
6 7 6 7 l l
6 N ðxÞ 0 7 6U 7
1
4 3 5 4 x3 5 Z 1 ð182Þ
0 N3 ðxÞ Uy3 l kint BintT E int Bint
l  l l l dx
" #
1
þ
uþ
xl ðxÞ
ul ðxÞ ¼ þ
uyl ðxÞ where matrix Eint collects elastic interface
2 3T 2 3 stiffnesses. An expression analogous to Equa-
N1 ðxÞ 0 Ux4
6 7 6 7 tion (182) can be obtained for the tangent
6 0 N1 ðxÞ 7 6 Uy4 7 matrix of the interface element by substituting
6 7 6 7
6 N ðxÞ 0 7 6U 7 to Eint the interface tangent matrix.
6 2 7 6 x5 7
¼6 7 6 7  NU þ
l ð178Þ In order to introduce the above formulation
6 0 N2 ðxÞ 7 6 Uy5 7
6 7 6 7 in a global FE analysis a transformation of
6 N ðxÞ 0 7 6U 7
4 3 5 4 x6 5 coordinates is needed between the local refer-
0 N3 ðxÞ Uy6 l ence frame and the global one. By defining y
the angle between the interface and the global x
where the parabolic shape functions Ni ðxÞ; i ¼ axis, the nodal dof vector Ul in the local frame
1; 2; 3 are defined as function of the nondimen- can be expressed as function of the dof vector
sional coordinate x: Ug in the global one as follows
N1 ðxÞ ¼ 12xðx
1Þ; N2 ðxÞ ¼ ð1
xÞð1 þ xÞ; " #
ð179Þ cos y sin y
N3 ðxÞ ¼ 12xðx þ 1Þ; U l ¼ diag ½rU g  RU g ; r
1212
sin y cos y
Given the model for the upper and lower ð183Þ
displacements, it is possible to obtain the
model for displacement discontinuities in the By means of the above rotation matrix R,
local reference frame (cf. Equation (149)) the nodal force vector qint
g and the stiffness
matrix kint
g in the global reference frame result
½uðxÞl ¼ uþ ðxÞ
u
ðxÞ ¼ NU þ
NU

l
" # l
 U
l qint T int
g  R ql ; kint T int
g  R kl R ð184Þ
¼
N N  B int
l Ul ð180Þ
Uþl
Interface FE suitable for 3D analyses can be
formulated in a manner analogous to the above
one. In the case of nonlinear geometric effects
the reader can refer to Allix and Corigliano
(1999), Pandolfi et al. (1999), and de Borst and
Schipperen (2002).
A noteworthy issue concerns the behavior
of interface elements for different numerical
space integrations necessary to compute the
element stiffness matrix and force vector. This
problem has been studied extensively in
Schellekens and de Borst (1993b) where it has
been highlighted that, usually, a Newton–
Cotes numerical integration can help in avoid-
Figure 36 Quadratic interface element for plane ing spurious oscillations of the interfacial
problems. stresses.
516 Damage and Fracture Mechanics Techniques for Composite Structures
3.09.7.3 Path Following Procedures additional scalar equation which is the so-
called constraint condition.
The particular difficulties which can be In the analysis of composites damage and
found in the numerical solution of bvp in the delamination, algorithms of the above kind
presence of softening had already been en- have been used. Worth mentioning is the new
visaged in the literature at the end of the 1960s arc-length method proposed in Hellweg and
(Maier, 1968); in particular snap-through and Crisfield (1998), where the authors have
snap-back behaviors can be found. This kind improved the original arc-length method by
of global responses can be tackled by making defining a procedure for the correct choice of
use of path following algorithms, originally the root in the cylindrical arc-length method.
proposed for instabilities caused by geometri- The method by Hellweg and Crisfield makes
cal nonlinearities, among them in primis those use of a global constraint; this is at variance
of Riks (1972) and Crisfield (1981) (see also with the local constraint introduced in Cor-
Crisfield and Shi, 1991). igliano (1993) who originated the path-follow-
The common idea characterizing the major- ing algorithm presented in Section 3.09.7.3.1.
ity of these algorithms is that to drive the Another kind of local control is that used by de
analysis through an additional condition im- Borst and co-workers (Schellekens and de
plying the control of a quantity which in Borst, 1993a; de Borst and Schipperen, 2002),
general is neither a loading parameter nor a discussed in Section 3.09.7.3.2.
displacement. The choice of an alternative
driving quantity can be achieved in general
by treating the loading parameter as a variable 3.09.7.3.1 Local control with total
in the equilibrium equation and by adding an displacement discontinuity
equation often called the constraint condition.
In the algorithm proposed in Corigliano
The constraint can be of a global kind, as in the
(1993) the constraint condition is derived from
arc-length method (Crisfield, 1981) or of a local
the local control of the displacement disconti-
one as proposed by de Borst (1987). Chen and
nuity in a suitably chosen point R of an
Schreyer (1990) proposed to locally control the
interface Gj. The chosen point R for the time
analysis for softening continua through a total
step Dt ¼ tnþ1
tn is that with the greater
strain norm in a suitably chosen point.
value of the plastic-damage multiplier l of the
In order to show the meaning of the
interface constitutive law at instant tn among
constraint condition, let us reconsider the bvp
the Gauss points of interface elements which
(Equation (150)): its solution must be obtained
are not already completely damaged; this
in a time interval [0; tend ]. By choosing
choice is motivated by the fact that l is a
convenient time instants 0¼
measure of the degree of degradation for
t0 ; t1 ; y; tn ; y; tN ¼ tend the time interval is
the class of models of Section 3.09.6.2.1. The
subdivided in time steps Dt ¼ tnþ1
tn : When
variable controlled in the time step is a norm of
the response is known at a time instant tn,
the displacement discontinuity increment
Equation (150) is solved at time instant tnþ1 :
D½uR ¼ ½uR R
nþ1
½un in the point R:
X
nev Z
qðU nþ1 Þ
mnþ1 P  H vT BvT cT D½uR ¼ g ð186Þ
i i rinþ1 dV
i¼1 Vii

X
nei Z where g is a fixed parameter, c is a three-
þ H intT
k BintT
k t knþ1 dG
mnþ1 P ¼ 0 ð185Þ component vector of weighting coefficients.
k¼1 Gki Vector c governs the kind of  imposed fractur-
ing process, e.g., for cT ¼ 0 0 1 a pure
where qðU nþ1 Þ can be considered as a known mode-I opening situation will be imposed. The
nonlinear vector-valued function of variables choice of c can be made automatically in the
Un þ 1. The dependence of q on Un þ 1 is given algorithm by considering the results obtained
by Equation (185), by the FE interpolation of at the end of the last step:
strains (148) and interface displacement dis- qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
continuities (149) and by the numerically ½uR
c ¼  nR ; jj½ujj  ½u21 þ½u22 þ½u23 ð187Þ
integrated constitutive laws (see Sections ½u 
n
3.09.5.2.2 and 3.09.6.2.2) for the volume V
and for the interface G. The constraint (Equation (186)) can be ex-
In contrast to classical procedures in which pressed in terms of the vector U nþ1 by making
the loading is assumed as a known quantity, in use of the interpolation of displacement
Equation (185) Un þ 1 and mn þ 1 are both discontinuities Equation (149) and the defini-
considered as variables. In order to solve tion of the Boolean matrix Hint given in
Equation (185) it is necessary to add an Equation (151); the problem to be solved at
 Computational Damage Mechanics for Composites 517
each time step of the analysis consists, there- 3.09.7.3.2 Local control with crack opening
fore, of the following equations in the un- displacement
knowns U nþ1 and mnþ1 :
The algorithm proposed in Schellekens and
qðU nþ1 Þ
mnþ1 P ¼ 0 de Borst (1993a) can be described by restarting
cT D½uR
g ¼ cT B int int
ð188Þ
from the linearized equilibrium Equation
R H R ðU nþ1
U n Þ
g
(189a)
 cT B ðUnþ1
Un Þ
g ¼ 0


q U jnþ1
mjþ1 tgj jþ1 j
nþ1 P þ K nþ1 U nþ1
U nþ1 ¼ 0 ð192Þ
where B int
denotes the matrix which governs
R
the displacement discontinuity model com- This equation can be solved to give the
puted in the integration point R, while H int
R is increment dU jþ1
nþ1 of the global dof vector from
the Boolean matrix which extracts from the an iteration j to the following iteration j þ 1 in
global displacement vector the dof vector of the following format:
the interface element to which point R belongs.

The nonlinear system (Equation (188)) is jþ1
dU nþ1 jþ1
 U nþ1
U jnþ1
solved iteratively by applying a method of the

1


Newton–Raphson kind. Notice that the con- ¼ K tgj mn P
q U jnþ1
nþ1
straint Equation (188b) is linear in the un-

1
known Un þ 1; hence, only a linearization of jþ1
þ Dmnþ1 K tgj P
nþ1
Equation (188a) around the value U jnþ1 ob-
I
II
tained at iteration j is necessary: jþ1
 dU nþ1 þDmnþ1jþ1
dU jþ1
nþ1 ð193Þ



jþ1
q U nþ1
mjþ1 j jþ1
nþ1 PDq U nþ1
mnþ1 P The knowledge of dU jþ1

 nþ1 allows for the
þ K tgj jþ1 j determination of the updated dof vector in
nþ1 U nþ1
U nþ1 ¼ 0

 ð189Þ the step as follows:
cT B U jþ1
nþ1
U n
g ¼ 0 jþ1
U nþ1 ¼ U n þ DU jnþ1 þ dU nþ1
jþ1
ð194Þ
K tgj
nþ1  ð@q=@U ÞjT
nþ1
In order to solve Equation (193) the constraint
where the consistent tangent matrix of the condition must be added.
whole system K tg has been introduced. The In Schellekens and de Borst (1993a) the
solution for each iteration can be obtained in choice has been made that the crack opening
the following way: displacement (COD) of the interface between


1
 two layers where delamination occurs have the
j
cT B K tgj
nþ1 q U nþ1
same value for each iteration. This means
mjþ1
nþ1 ¼

1 ð j þ 1Þ41; choosing a nodal point R on a particular
T  tgj
c B K nþ1 P interface and writing
 
1 ð190Þ
þ


g þ cT B K tgn qðU n Þ dðCODÞ ¼ 0 ) dUnþ1 jþ1

dUnþ1jþ1
m1nþ1 ¼  
1 R R
cT B K tgn P 
Iþ
IIþ 


1

 ¼ dUnþ1jþ1 jþ1
þDmnþ1 dUnþ1 jþ1
U jþ1
nþ1
j
¼ U nþ1 þ K nþ1tgj
mjþ1 j
nþ1 P
q U nþ1
R R

I

II

jþ1

dUnþ1 þDmjþ1
nþ1 dUnþ1
jþ1
¼0 ð195Þ
To derive the above Equations (190) use has R R
been made of the fact that the constraint
Equation (189b) also holds at iteration j40 where the symbols þ and
denote the nodes
and that U j¼0
nþ1 ¼ U n :
on the two sides of the interface where COD is
The iteration procedure is stopped when the maintained constant. From this equation the
following conditions are satisfied increment of the load factor in the step can be
    recovered as
 jþ1 
U nþ1
U nþ1 rTolu U 1nþ1
U n1 ;
j
Iþ
I

jþ1 jþ1

    ð191Þ dUnþ1
dUnþ1
 jþ1  Dmjþ1 R R
mnþ1 P
q U nþ1 rTolm m1nþ1 P
qðU n Þ
j ¼
ð196Þ
nþ1 IIþ
II

jþ1 jþ1
dUnþ1
dUnþ1
R R
where Tolu and Tolm are two tolerance para-
meters. It has to be observed that the first step 3.09.7.4 Examples of Applications
of the analysis is to be done by following
another strategy, for instance by imposing a This section shows some examples of da-
value of the load parameter m which is below mage, fracture, and plastic computations re-
the initial elastic limit. levant to composite materials and structures.
518 Damage and Fracture Mechanics Techniques for Composite Structures
The examples are taken from the author’s (and been simulated numerically in Corigliano et al.
the author’s co-workers) experience on the (2000) on the basis of some simplifying
subject. The reason for this choice is simply assumptions. Such choice has been made in
because of the possibility to describe some order to check the possibility to simulate the
examples correctly and completely. It does not main rupture mechanisms observed in the tests
imply a disregard of the works existing in the by making use of a commercial code, with the
scientific literature. The reader is invited to addition of few, ad hoc, procedures. All the
study the numerous meaningful contributions numerical simulations have been performed
given in the reference list that show the with the commercial FE code ABAQUS
potentialities of fracture and damage computa- (Hibbit and S
rensen 1997) available at the
tions in the study of composite materials. time the research was carried out. The purpose
In Section 3.09.7.4.1 an example of PFA for of the numerical simulations was to correctly
sandwich composites is shown. Section capture single collapse mechanisms (and the
3.09.7.4.2 concerns a microscale computational corresponding failure loads), considered as
study of a long-fiber-reinforced composite. In independent and occurring separately in the
Section 3.09.7.4.3 some examples of delamina- specimen, as those shown in Figure 38. Among
tion simulated by means of interface laws and them the formation of macroscopic cracks in
interface FE elements are discussed. The the core (Figures 38(a) and (b)) or at the
possible use of IDM for the simulation of interface core/skins (delamination, Figure
reinforcement with GFRP composites is shown 38(d)) and the skin collapse in tension (Figure
in Section 3.09.7.4.3(i). Finally, in Section 38(c)). The different materials in the sandwich
3.09.7.4.3(ii), examples of the combined use thickness were reproduced by the superposition
of identification and homogenization for the of three strips of elements with different
determination of composite models at the mechanical properties: two external strips
meso- and macroscales are presented. representing the skins and the extra skins
(3 mm thick) and a central layer for the core
3.09.7.4.1 PFA of a sandwich beam (9 mm thick). In order to simulate core
collapse, skin collapse, or delamination, re-
Three point bending (TPB) (Figure 37(a)) spectively, the numerical simulations were
and Four point bending (FPB) tests (Figure done by activating separately a simplified
37(b)) of a syntactic-foam (for the core)/glass procedure for the simulation of the progressive
fiber (for the skins) composite sandwich have damage in the core, in the skins or in the line of

Figure 37 Bending test on a fiber-glass-syntactic foam sandwich composite beam: (a) TPB and (b) FPB.
 Computational Damage Mechanics for Composites 519

Figure 38 Recorded rupture mechanisms in TPB and FPB fiber-glass-syntactic foam sandwich composite
beams: (a) fractures in the core of a specimen tested in FPB conditions; (b) fractures in the core of a specimen
tested in TPB conditions; (c) tensile rupture in the skin of a specimen tested in FPB conditions; and (d)
delamination between skin and core of a specimen tested in TPB conditions.

elements near the interface between the extra

skin and the core.
The simplified procedure belongs to the class
of PFA described in Section 3.09.4 and consists
in a local stiffness release at the Gauss point
level, implemented through a user subroutine.
When a threshold value of a scalar failure
index is reached in a single Gauss point, the
tensile elastic modulus Et is set to zero locally;
the contribution of that Gauss point to the
element stiffness matrix is then brought to
zero. In the numerical calculations, a Rankine
criterion of maximum principal stress was
assumed for the simulation of damage in the
core and the skins, while a control on the
maximum shear stress was adopted for the Figure 39 Schematic representation of the elastic/
strip of elements at the boundary core/lower brittle behavior assumed for the numerical simula-
skin for the simulation of skin debonding. This tion of syntactic-foam-glass fiber sandwich beams.
procedure implies that the mechanical behavior
of the single constituents was assumed to be The numerical analyses were conducted
elastic-perfectly brittle as depicted schemati- under the assumption of plane strain, since
cally in Figure 39. Moreover, in the simula- the specimen width to span ratio is equal to 0.5
tions, the core and the skins were considered as (span 60 mm, height 15 mm, width 30 mm).
homogeneous and isotropic. Although the specimen geometry and loading
The critical value of shear stress for the configurations were symmetrical, since the
simulation of skin debonding was assumed behavior at rupture was nonsymmetrical in
equal to the interlaminar shear stress (ILSS) for some cases (Figure 38(b)); the whole cross-
the glass-epoxy resin fabric: tmax ¼ 16.4 (MPa). section was modeled. To simulate single,
The skin and core model parameters used nonsymmetric, crack propagation, half of the
for the numerical simulations are collected in section was considered indefinitely elastic,
Table 3; they were obtained from the experi- while in the other half the local stiffness release
ments described in Corigliano et al. (2000). procedure was applied.
520 Damage and Fracture Mechanics Techniques for Composite Structures
The meshes adopted in the simulations are mental ones in Figure 38; from the comparison
shown in Figures 40(a) and (b) for a symmetric it can be observed that the crack pattern is
TPB case and an unsymmetric FPB one, reproduced correctly, at least qualitatively. As
respectively. The meshes are composed of in the experiments, during the numerical simu-
four-node plane strain elements. The loading lations the first elements which fail are near the
and support rollers are simulated as rigid edge of the loading cylinders and the crack
bodies. proceeds from top to bottom and is inclined
In Figures 40(a) and (b) the numerically towards the lower cylindrical support.
computed crack patterns at the end of the A numerical load–displacement plot ob-
analyses in the numerically simulated TPB and tained for the TPB test by activating the
FPB are shown. In Figure 40 the elements which rupture criterion in the core only is compared
were concerned in the stiffness reduction proce- in Figure 41(a) with two experimental plots
dure are marked in black. Crack patterns in concerning TPB tests which registered non-
Figure 40 can be compared with the experi- symmetric failure in the core. The elastic
stiffness and the fracture load are captured
adequately, considering also the great simpli-
Table 3 Mechanical data adopted for the numer- city of the adopted model. In Figure 41(b) two
ical simulations of TPB and FPB sandwich tests. experimental load–displacement plots concern-
Young’s Tensile ing specimen failure by skin delamination are
modulus Poisson’s strength compared with a numerically simulated re-
(MPa) ratio (MPa) sponse. The numerical analysis was carried out
by activating the PFA simulation in the strip of
Skin 14,000 0.2 225 elements at the boundary core/lower skin. In
Core 1,100 0.34 15
this case the agreement between the numerical

Figure 40 FE meshes adopted for numerical simulation of TPB and FPB sandwich tests. Marked elements
represent the numerically simulated crack pattern.
 Computational Damage Mechanics for Composites 521
and the experimental failure loads is particu- convergence. The example discussed here
larly good. shows that PFA could be used to have simple
As shown by the results displayed in Figures estimation of the overall composite resistance,
40–41, the PFA used in the present examples also making use in the numerical simulation of
leads to results which are in good overall commercial codes with a minimum amount of
qualitative agreement with the experimental additional effort for the user.
tests. As a matter of fact, it can be noticed that
the stiffness release procedure was already
attempted in Rizzi et al. (2000) with reference 3.09.7.4.2 Assessment of material resistance at
to the simulation of the plain syntactic foam the micromechanical level
(the material for the sandwich core) behavior The second example concerns a microme-
in notched TPB specimens; however, in that chanical study of the transversal resistance
case, the numerical results were not completely properties of a long-fiber composite. This
satisfactory due to the considerable brittleness unpublished numerical study (carried out in
of the numerical responses which did not take the framework of a Masters thesis project co-
advantage of the extra structural resources ordinated by Carvelli and the author) follows
available here from the sandwich geometry. In the approach put forward in Taliercio and
the simulations of the plain foam behavior an Sagramoso (1995) and Carvelli and Taliercio
alternative, more refined procedure, based on (1999) for the determination of strength of
the discrete crack approach of IDM was also composites by means of numerical homogeni-
adopted, leading to a considerable improve- zation. A composite with an hexagonal peri-
ment of the numerical results. odic distribution of elastic fibers in a perfectly
As already observed in Section 3.09.4, PFA plastic matrix governed by the von Mises yield
suffer from mesh dependency and difficulties of criterion was considered. The mechanical
properties of fibers and matrices are collected
in Table 4.
Periodic boundary conditions were imposed
on the single hexagonal RVE of Figure 42,
modeled by FE in plane stress or plane strain
conditions, as suggested by the homogeniza-
tion theory (see Sanchez-Palencia, 1980; Ta-
liercio and Sagramoso, 1995; Carvelli and
Taliercio, 1999). In order to assess the influence
of the fiber–matrix interface on the composite
resistance, two interface conditions were con-
sidered in the FE analyses: (i) perfectly bonded
interface; (ii) completely damaged interface in
tension, with unilateral contact conditions to
avoid inter-penetration.
A series of FE simulations (with the com-
mercial code ABAQUS (Hibbit and S
rensen
1997) were made for various imposed history
of macroscopic average strains. For each
simulation: the complete elastic–plastic history
was computed until plastic collapse; the
macroscopic stresses were computed as volume
averages of microscopic stresses in the RVE;
the value of macroscopic stresses at plastic
collapse were drawn in the plane of macro-
scopic stresses. By means of the above

Table 4 Periodic composite with hexagonal RVE:

mechanical properties of fibers and matrix.
Young’s Tensile
Figure 41 Comparison between experimental and modulus Poisson’s strength
numerical load/displacement plots for sandwich (MPa) ratio (MPa)
beam under bending: (a) TPB test with core rupture Fiber 71,600 0.22 N
and (b) FPB test with skin delamination. Marked Matrix 3,280 0.31 99.5
lines represent experimental data.
522 Damage and Fracture Mechanics Techniques for Composite Structures

Figure 42 FE mesh of the hexagonal RVE of a

long-fiber composite material.

Figure 44 Computed limit domains for an hex-

agonal RVE: (a) plane strain analysis and (b) plane
stress analysis.

small volume fraction the decrease in resistance

of the RVE due to the imperfect interface is
noteworthy; this means that in order to
correctly evaluate the real macroscopic resis-
tance properties of a composite the nature of
the fiber–matrix interface must be considered
carefully. This is an example of how micro-
mechanical analyses could result in useful
Figure 43 Deformed meshes at plastic collapse for outputs for the assessment of composite
a hexagonal RVE: (a) perfectly bonded fiber–matrix resistance at the macroscale.
interface and (b) completely damaged fiber–matrix
interface.
3.09.7.4.3 Delamination analysis of
briefly-described procedure, the failure locus in interlaminar fracture specimens
the macroscopic stress space was computed for
plane strain, plane stress situations at varying In this section some examples of numerical
value of fiber volume fraction, and for the two simulations concerning interlaminar fracture
interface conditions examined. specimens (Figure 17) are presented, taken
In Figure 43 the deformed meshes at plastic from Corigliano and Allix (2000) and Cor-
collapse obtained for plane strain conditions igliano and Ricci (2001). The formulation
for a perfectly bonded and a perfectly damaged described in Section 3.09.7.1.2 is first adopted,
fiber–matrix interface are compared. The i.e., damageable interface FE are introduced
difference in the mechanical behavior for the along the line of delamination, the remaining
two considered situations can be appreciated part of the interlaminar specimen is considered
from the figure. In Figure 44 the limit domains as elastic. Hence, all the nonlinearities are
computed for a fiber volume fraction Vf ¼ 0:16 concentrated along the damageable interfaces
are shown for plane strain and plane stress G. In the last example of Section 3.09.7.4.3
situations. It can be observed that also for a damage is considered also in the layers.
 Computational Damage Mechanics for Composites 523
Numerical simulations of interlaminar frac- governed by a simple linear softening law with
ture tests are useful for two main reasons: the a critical energy in pure mode I GIc ¼
first one is the necessity to test the numerical 0:4ðN mm
1 Þ and in pure mode II GIIc ¼
model in situations in which experimental 0:8ðN mm
1 Þ; this law can be considered as a
results are easily available; the second one is particular case of the elastic–plastic softening
connected with the necessity to build indirect interface law presented in Section 3.09.6.3.2. In
identification procedures in which the numer- the case of the ENF test, it is interesting to
ical model is compared to experimental results compare the responses in terms of interlaminar
to the purpose of finding the correct para- normal stresses, shown in Figure 46(b), which
meters of the interface model (see Section can be modeled thanks to the unilateral
3.09.6.4). The spatial discretization of inter- behavior (in the normal direction) of the
laminar fracture tests can be done by means of interface model. From these simple examples
beam elements, 2D meshes under a plane-strain the use of a beam model for the simulation of
hypothesis or even 3D meshes if the shape of interlaminar fracture tests appears justified,
the crack front along the width of the specimen even for small aspect ratios compared to those
is thought to be nonrectilinear or to vary usually encountered in laminated composites;
during the test. other examples of the use of beam FE for the
simulation of interlaminar fracture tests can be
(i) Comparison of 2D and beam FE simulations found (Allix et al., 1995; Allix and Corigliano,
for DCB and ENF tests 1996).
As a first example, a DCB specimen having
the following dimensions (Figure 17(a)) (ii) Responses at varying mesh size
l ¼ 20ðmmÞ; h ¼ 1ðmmÞ; B ¼ 1ðmmÞ; As already observed, FE analyses with
ð197Þ interface models require sufficiently refined
a0 ¼ 5ðmmÞ meshes in order to correctly describe the
process zone at the crack-tip. A simple example
has been analyzed (Corigliano and Allix, 2000) of DCB simulation is described to underline
and two load–displacement responses are
shown in Figure 45. One of the responses has
been obtained by means of 2D four-noded 80
isoparametric elements, the other by means of
beam elements with Timoshenko kinematic
and cubic interpolation for the transversal 60
displacement. In Figure 46(a) the same com-
parison is done for an ENF specimen with the
P [N]

40
same dimensions as the previous case but with
a ratio a0/L ¼ 0.35. The two arms of the
specimens have been considered isotropic 20 2D elements
elastic with the following parameters: Young’s beam elements
modulus E ¼ 1:35  105 ðMPaÞ; Poisson’s ratio
n ¼ 0:3: The interface has been assumed to be 0
0.1 0.2 0.3 0.4 0.5
(a) s [mm]

10

-10
[MPa]
t3

-20
2D elements
-30 beam elements

-40
0 4 8 12 16 20
(b) x [mm]

Figure 46 Response of an ENF specimen: (a) load–

Figure 45 Load–displacement plots for a DCB displacement response and (b) interlaminar normal
specimen. stresses.
524 Damage and Fracture Mechanics Techniques for Composite Structures
choice of parameters:
 
K ¼ 1  105 N mm
3 ;
 
t0 ¼ 80 ðMPaÞ; H ¼
120 1 mm
1 ; ð200Þ
(a)  
g ¼ 180 N mm
1 s
1 ; N ¼ 10

In Figures 48(a) and (b) the response of the

DCB obtained with the four meshes is shown
(b)
in terms of the load-displacement plots and of
the interlaminar traction t3 along the interface.
From a combined examination of the two
figures, it can be concluded that the strong
over-estimation of the response obtained with
the coarser mesh is due to the fact that the
(c) interlaminar stress distribution is not repro-
duced correctly, in particular the high stress
concentration at the crack tip. The necessity to
use small elements when IDM is used is well
known and can be considered one of the main
drawbacks related to its application. As shown
(d) in the above example, mesh dependence is
shown in the response of the specimen until the
Figure 47 Viscoplastic interface model. Response
element size is small enough to reproduce the
of a DCB test at varying mesh size. The four meshes
used: (a) le =l ¼ 0:05; (b) le =l ¼ 0:025; (c) le =l ¼ interlaminar stresses correctly. For element size
0:0125; and (d) le =l ¼ 0:0625: smaller than the above one, no pathological
mesh dependence is shown. The important
this aspect. The DCB considered has the issue of pathological mesh dependence remains
dimensions of the previous example, given by an open question for 3D fracture phenomena
Equation (197). The response of the specimen described through interface elements.
at imposed opening displacement velocity v ¼
1 (mm s
1) is computed at varying mesh size. (iii) Time dependence in DCB test
The four meshes of Figure 47 are considered.
The two arms of the specimen are assumed to The response of the DCB specimen of the
have an elastic, transversely isotropic behavior previous example is computed at varying
governed by the following parameters: imposed opening displacement velocity at the
free end, with the finer mesh among those in
E11 ¼ 84; 766 ðMPaÞ; E33 ¼ 0:1E11 Figure 47. The results in term of the displace-
G13 ¼ 1; 000 ðMPaÞ; v13 ¼ 0:035; v32 ¼ 0:35 ment-load plots are shown in Figure 49. The
plots in Figure 49 increase with the imposed
ð198Þ
velocity, this is a consequence of the increase of
fracture energy as given by the visco-plastic
where 1 is the beam axis and 3 is the direction
interface model adopted.
orthogonal to it in the plane of the analysis.
In Figure 50 load-displacement plots at
The four meshes used have elements measuring
varying imposed opening displacement velocity
in the crack propagation direction 1, 0.5, 0.25,
at the free end of a DCB test simulated with a
0.125 (mm), respectively, with the element-
time-dependent elastic-damage interface law
length/beam-length ratios le =l marked in the
are presented. The geometry of the specimen
figure caption.
is as before, the behavior of the arms is this
The interface model used is the viscoplastic
time elastic isotropic, governed by the follow-
one proposed in Corigliano and Ricci (2001),
ing parameters: Young’s modulus E ¼ 84,750
which can be obtained from the elastic–plastic
(MPa); Poisson’s ratio n ¼ 0:35: The time-
softening of Section 3.09.6.3.2 by substituting
dependent interface model, proposed in Cor-
the evolution law of plastic displacement
igliano and Allix (2000) can be obtained from
discontinuities (Equation (136)) with the fol-
that discussed in Section 3.09.6.3.1 by sub-
lowing viscoplastic one:
stituting the time independent evolution law
Z t qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi for damage governed by Equation (128), with
@f T
½u’ vp ¼ gh f iN
þ ; l¼ ð½u’ vp Þ ð½u’ vp Þ dt0 ð199Þ the time-dependent one:
@t 0

This model has been used with the following D’ i ¼ gi h f iN

þ; i ¼ 1; 2; 3 ð201Þ
 Computational Damage Mechanics for Composites 525
25
mesh Le/L = .05

mesh Le/L = .025

20
mesh Le/L = .0125

mesh Le/L =. 00625

15

load P [N]

10

0
0.0 0.2 0.4 0.6 0.8 1.0
(a) displacement U [mm]

1.2

mesh le = 1. [mm]
0.8 mesh le= .5 [mm]

mesh le =. 25 [mm]
traction t [MPa]

mesh le = .125 [mm]

0.4

0.0

-0.4
5 6 7 8 9 10
(b) x [mm]

Figure 48 Viscoplastic interface model. Response of a DCB test at varying mesh size. Opening displacement
velocity v ¼ 1 mm s
1: (a) load displacement plots and (b) traction distribution along the interface.

The model parameters used for the mode I sions of the DCB and ENF specimens analyzed
simulation are as follows: in the previous examples. The behavior of the
  two arms of the specimen and of the interface is
K ¼ 105 N mm
3 ; a ¼ 1; assumed as elastic transversely isotropic and
    viscoplastic, respectively, as in the previous
Y0 ¼ 0:0333 mm N
1 ; g ¼ 180 1 s
1 ; ð202Þ
example.
N ¼ 16
The response of the specimen at varying
velocity of imposed displacement in the loaded
From Figure 50 the strong rate dependence point is shown in Figure 51(a); as in the DCB
shown by the global response of the specimen simulations, the increase with velocity of the
can be appreciated. energy dissipated during crack propagation
can be appreciated, due to the increase of
(iv) Time dependence in an ENF test fracture energy as modeled by the interface
law. It is also worth noticing that the response
The ENF specimen (Figure 17(b)) consid- of the specimen can be divided in three main
ered for the simulation has the same dimen- parts: an elastic part with increasing load, a
526 Damage and Fracture Mechanics Techniques for Composite Structures
16 v = 100 [mm/s] 80

v = 10 [mm/s]

v = 1 [mm/s]
12 60
v = 0.1 [mm/s]

v = 0.01 [mm/s]

Load P [N]
Load P [N]

8 40
v = 100 [mm/s]

v = 10 [mm/s]

v = 1 [mm/s]
20
4
v = 0.1 [mm/s]

v = 0.01 [mm/s]

0
0
0.0 0.2 0.4 0.6 0.8 1.0
0.0 0.4 0.8 1.2 (a) Displacement [mm]
Displacement [mm]
80
Figure 49 Viscoplastic interface model. Load–
displacement plots of a DCB test at varying imposed
velocity of displacements.
60

10 v = 100 [mm/s]
a0/l = 0.2
Load P [N]

v = 10 [mm/s]
40 a0/l = 0.25
8 v = 1 [mm/s]
a0/l = 0.3
v = 0.1 [mm/s]
a0/l = 0.35

20 a0/l = 0.4
6
Load P [N]

a0/l = 0.45

a0/l = 0.5
4 0
0.0 0.2 0.4 0.6 0.8 1.0
(b) Displacement [mm]
2
Figure 51 Viscoplastic interface model: (a) load–
displacement plots of a ENF test at varying imposed
velocity of displacements and (b) load–displacement
0
plots of a ENF test at varying initial crack length for
0.0 0.4 0.8 1.2 imposed displacement velocity v ¼ 1 (mm s
1).
Displacement [mm]

Figure 50 Time dependent elastic–damage inter- The ENF specimen is then analyzed at
face model. Load–displacement plots of a DCB test varying initial crack length. The response of
at varying imposed velocity of displacements. the specimen can be unstable also when the
displacement is imposed (snap-back response)
softening branch corresponding to crack pro- if initial crack length is sufficiently small. In
pagation, and a part with increasing load Figure 51(b), the load-displacement plots,
corresponding to the response after the dela- obtained for a velocity of imposed displace-
mination crack has passed the beam half. A ment v ¼ 1(mm s
1), show that when
simple explanation of this behavior can be a0 =Lo0:35 the response tends to have snap-
given on the basis of a LEFM model of the back; this is in agreement with the critical
ENF test. This has been done in Allix et al. theoretical value a0 =Lo0:347 found in Allix
(1995) by making use of the CM described in et al. (1995).
Section 3.09.3.1.2. It can be said that the
response of the specimen has as a limit the
linear elastic response of two superposed (v) Numerical and experimental time-dependent
beams (the two arms of the specimen) without responses for a DCB test
interlaminar connections; this limit is reached
when complete delamination in mode II has As a further example, we discuss the simula-
occurred. tion of time-dependent interlaminar fracture in
 Computational Damage Mechanics for Composites 527
a DCB carbon fiber-PEI composite specimen been used with the following set of parameters:
(Corigliano et al., 1997, 1998; Corigliano and  
Ricci, 2001). K ¼ 200;000 N mm
3 ;
Frassine et al. (1993, 1996) and Frassine and  
t0 ¼ 60 ðMPaÞ; H ¼
70 mm
1 ; ð203Þ
Pavan (1995) obtained results concerning the  
behavior of DCB specimens at varying velocity g ¼ 7 N mm
1 s
1 ; N ¼ 16
and temperature. They considered 16-ply uni-
directional laminates 0.3 mm thick with poly These parameters have been identified in
ether-imide (PEI) resin and carbon fibers and Corigliano et al. (1997) (see also Corigliano
carried out interlaminar fracture tests on DCB and Ricci, 2001), following a direct approach
specimens 20 mm wide and 170 mm long, similar to that discussed in Section 3.09.6.4.
having an initial crack length of 60 mm, In Figures 53(a) and (b) the numerical and
following the ESIS protocol. The thickness experimental responses are compared in terms
varies between 3.6 mm and 4.2 mm, depending of the load-displacement plot and of the
on the moulding conditions. The range of fracture length vs. time plot, for two opening
crack propagation velocities considered was displacement velocities. In Figure 54 an
such that dynamic crack effects were negligible. example of deformed meshes during crack
By using for composites the time-tempera-
ture equivalence postulate valid for most
polymers (Ferry, 1980), the plots were obtained
of the fracture toughness Gc of the composite
and of pure resin shown in Figure 52 (see also
Corigliano et al., 1997, 1998). From Figure 52
an increase of the composite fracture toughness
with crack propagation velocity can be ob-
served, while no clear dependence of pure resin
fracture toughness with the crack velocity can
be observed. These experimental results have
motivated the application of time-dependent
interface models for the simulation of the DCB
specimen at varying velocity. A simulation of
the test conducted by the above referenced
authors is presented here.
The elastic arms of the DCB specimen are
modeled with the parameters given in Equation
(198). The viscoplastic interface model has

1.0

matrix
0.8
composite

0.6
Log(Gc) [N/mm]

0.4

0.2

0.0

-0.2

-0.4
-16 -12 -8 -4 0
Log(crackspeed [m/s])
Figure 52 Fracture toughness of the PEI matrix
and of the composite vs. crack speed master curve at
T0 ¼ 231: The line represents power-law least-square Figure 53 DCB specimen. Experimental vs. numer-
fitting (experimental results from Frassine and ical responses: (a) load displacement and (b) fracture
Pavan, 1995). length vs. time.
528 Damage and Fracture Mechanics Techniques for Composite Structures

Figure 54 DCB specimen. Example of deformed meshes during crack propagation.

propagation is shown. As it can be seen from Figure 14 shows the evolution of the X-ray
the above results, the comparison between damage map near the hole for an increasing
simulation and experiments is satisfactory applied load. The first damage, appearing at
and confirms the potentiality of the IDM also 55% of rupture, is transverse cracking in 901-
in the case of time-dependent phenomena. plies near the hole, and matrix cracking in the
01-plies tangent at the hole and in the fiber
(vi) Delamination in a holed plate in tension direction called splitting. Delamination only
begins at B80% of rupture (Figure 14(b)). Just
A plate with a central circular hole, loaded in before the rupture (Figure 14(c)), the delami-
tension (see Figure 14), is a delamination test nated area is always found to be located
which can be considered as an alternative to between the splittings and in the 01-direction
the more classical interlaminar fracture speci- with about two hole diameters in length. The
mens like those discussed previously. In this damage is well developed in several ways:
test the delamination crack initiation is repro- splittings, transverse cracking not only in the
ducible and the growth is often stable within a 901-plies but also in the 7451-plies, and
certain range. Also of interest in this test is the multiple delamination at the 01/ þ 451, 7451
valuable information it provides in terms of the (the most damaged), and
451/901 interfaces.
shape and size of the delaminated area revealed From the computation, the splitting can be
by means of X-ray photography. In spite of seen as a shear damage in the 01-layer (see
these positive features, the exploitation of the Figure 55(a)). In fact, when the first 01-fibers
test presents some difficulties: the delamination near the hole break, the local load is trans-
appears as the result of complex damage ferred by shear in the matrix to the adjacent
interactions (see Figure 14); an inverse proce- fibers. The delaminated area (corresponding to
dure (like that mentioned briefly in Section the interface damage variable D3 ¼ 1) com-
3.09.6.4) would require the use 3D nonlinear puted in the 7451 interface is shown in Figure
numerical simulations with localisation of 55(b) as an example. In the same manner, the
damage both in the plies and in the interfaces; other interfaces, except for the midplane, are
the way to compare computations and tests found to be less delaminated.
should take into account different information
in nature (measurement of strains, crack 3.09.7.4.4 Simulation of GFRP reinforcement
density, opening displacementy).
Therefore, the practical use of tests based on GFRP laminates are used more and more in
holed plates still remains an open issue. The civil engineering applications, especially for the
comparison made in Corigliano and Allix reinforcement of existing structures (see, e.g.,
(2000) between experimental observations in a Karbhari and Zhao, 2000). One of the main
M55J/M18 [03/7452/90]s laminate loaded in problems to be solved in order to optimise the
tension and computation, is presented. The behavior of these kind of structures is repre-
comparison is made, qualitatively, by making sented by the possible delamination of the
use of a specialized software delamination whole laminate from the structure surface. An
simulation by damage mechanics (DSDM) unpublished example of 3D interface numerical
which was developed in Allix (1992). The simulation for a simply supported beam loaded
numerical model follows the mesoscale ap- in TPB and reinforced on the tensile side with a
proach described in Sections 3.09.5 and 3.09.6 GFRP laminate is presented here.
and includes damage behavior for the single The beam is shown in Figure 56; its length is
layers (model of Section 3.09.5.3) and an 3,000 mm; its heigth 150 mm; and its width
elastic-damage model for the interfaces similar 120 mm. A 1,000 mm long and 10 mm thick
to that described in Section 3.09.6.3.1. laminate is applied symmetrically at the center
 Computational Damage Mechanics for Composites 529

Figure 55 Damage maps computed in a [03/7452/90]s holed specimen at the rupture load: (a) d indicator in
the 01 layers and (b) d3 indicator at the 7451 interface.

ing to the softening part of the global response;

after complete debonding, the elastic beam
shows a reduced value of the global stiffness,
corresponding to that of the beam without
reinforcement.

3.09.7.4.5 Homogenization and parameter

identification applied to the
constitutive modeling of composites
With reference to laminated composites, a
Figure 56 Simply supported beam with GFRP procedure has been proposed in Giampieretti
reinforcement on the tensile side. et al. (2000) for the generation of constitutive
models, at the level of the single lamina as well
as at the level of the whole laminate, by means
of the beam. The 3D FE mesh used in the of parameter identification (calibration) based
analysis is composed of 5,058 4-noded tetra- on pseudo-experimental data generated nu-
hedral FE for the beam and the laminate and merically through homogenization by FEs.
by 202 6-noded triangular interface elements The procedure is outlined briefly and two
for the interface between the beam and the examples of applications taken from Giampier-
laminate. In Figure 57 the mesh relevant to etti et al. (2000) are presented. The purpose of
interface elements only is shown; it can be seen the presentation is to show an example of
that it is quite a coarse mesh. Both the beam decoupled computations at micro-, meso-, and
and the laminate have been considered as macroscales.
linear elastic and isotropic with the elastic The proposed procedure consists of the
moduli given in Table 5. The interface between following steps:
the beam and the laminate has been modeled (i) The macroscopic behavior of the compo-
with a linear softening law with a critical site system is described by a constitutive model
energy in the three modes equal to Gc ¼ (in generalized macroscopic variables) selected
0:5ðN mm
1 Þ; this law can be considered as a according to the peculiar computational re-
particular case of the elastic–plastic softening quirements of the relevant scale. Specifically,
interface law presented in Section 3.09.6.3.2. an elastic-viscoplastic generalized standard
In Figures 58(a)–(c) the distributions of material (GSM) model has been assumed: it
displacement discontinuities in the axial, width, is defined by a (macroscopic) thermodynamic
and vertical directions are shown. The figures potential and a dissipation potential, both
reveal the 3D character of the interlaminar assigned a priori and dependent on a set of
response. In Figure 59 the load-displacement material parameters to be identified. The
plot is shown. During the simulation, brittle inelastic behavior is governed by a set of
mode II delamination is observed correspond- internal state variables contained in the
530 Damage and Fracture Mechanics Techniques for Composite Structures

Figure 57 Simply supported beam with GFRP reinforcement: FE interface mesh.

thermodynamic potential. In the examples a

Table 5 Elastic moduli for the simulation of a Norton–Hoff type dissipation potential (Le-
GFRP reinforced beam. maitre and Chaboche, 1990), based on the
Hill’s yielding function has been assumed.
Young’s modulus
(MPa) Poisson’s ratio (ii) An RVE of the heterogeneous texture
(Figure 60) (at the microstructure) is modeled
Beam 3  104 0.3 by FEs, assigning appropriately accurate con-
Laminate 8  104 0.25 stitutive models to the various constituents.
The macroscopic (average) strain variables are

Figure 58 Simply supported beam with GFRP reinforcement: distributions of displacement discontinuities
on the beam/laminate interface: (a) axial direction; (b) width direction; and (c) vertical direction.
 Computational Damage Mechanics for Composites 531
applied to the RVE and the consequent This procedure, briefly called the mixed
macroscopic stresses are computed by aver- method, is outlined schematically in Figure
aging the microscopic stress field resulting from 61, with reference to the problem of a single
the numerical solution. lamina. In Figure 61 the symbol E denotes
(iii) The assumed time-histories of the macroscopic strains, S macroscopic stresses, y
macroscopic strains and of the computed parameters to be identified.
stresses are fed into a parameter-identification The first example (see Figure 62) concerns an
tool designed for the specific form of the epoxy-resin lamina reinforced with unidirec-
chosen macroscopic constitutive model, fol- tional glass fibers. The local material behavior
lowing a scheme similar to that briefly de- at the microscale is assumed to be elastic–
scribed in Section 3.09.6.4 (see Figure 30); the plastic with strain hardening. A similar beha-
extended Kalman filter has been adopted. The vior is then expected also at the macroscale, it
identification process yields optimal estimates was then modeled with a Norton–Hoff dissipa-
of the unknown constitutive parameters. tion potential with the creep exponent n set to a
sufficiently high value (say 30) to eliminate
time-dependence. In fact, the Norton–Hoff
viscoplastic model can be shown to reduce to
a classical elasto-plastic model as the creep
exponent tends to infinity. A differentiable
dissipation potential such as the Norton–Hoff’s
allows an easy development of the identification
procedure, which requires differentiation to
compute the sensitivity of the system. In this
example the total number of unknown para-
meters identified with the mixed method
amounts to 10. Figure 62 reports the progres-
sive estimates yielded by the identification
procedure for two elastic parameters and two
yield stresses. In each diagram, the attached
sketch indicates the type of loading test that
was modeled with FEs to build the pseudo-
experimental data employed for the identifica-
tion process. Values obtained by traditional
Figure 59 Simply supported beam with GFRP analytical methods (Hashin and Rosen, rule of
reinforcement: load–displacement plot. mixtures, see for example, Herakovich, 1998)
are indicated for comparison.
Similar considerations hold for the example
in Figure 63, which concerns a laminate made
up of five homogeneous orthotropic laminas
oriented at 7301. The single laminas are
assumed to be an homogeneous orthotropic
elasto-plastic material that follows Hill’s yield
criterion. The laminate behavior in terms of
plate generalized variables is assumed to be
described by a pair of potentials which are
formally identical to those of the previous case,
except for the fact that the macroscopic stresses
and strains R and E are to be replaced by
generalized plate stresses Q and strains q,
respectively. Again the creep exponent n is
Figure 60 RVE of the lamina microstructure. set to a high value (say 50) to obtain an

Figure 61 The mixed method for constitutive modeling of a composite lamina.

532 Damage and Fracture Mechanics Techniques for Composite Structures

Figure 62 Homogenization of a lamina: (a) the RVE; (b) local microscopic material models; (c) and (d)
results of the identification process for two macroscopic elastic parameters; and (e) results of the identification
process for two yield stresses.

elasto-plastic behavior. The total number of of damage and fracture phenomena in compo-
unknown parameters identified with the mixed sites. The subject treated is extremely wide; to
method amounts to 31. It is important to have a comprehensive description it is neces-
remark that the choices of Norton–Hoff sary to have knowledge of solid mechanics,
potential and Hill’s yield criterion have been linear and nonlinear FM, CDM, linear and
made in Giampieretti et al. (2000) with the aim nonlinear computational mechanics, and
of verifying the proposed procedure; they are homogenization theories. Due to the extensive
not the optimal ones for a good description of nature of the subject, the author has been
the composite nonlinear behavior. forced to make choices and select a series of
subjects to be reviewed.
A first consequence of this is the fact that
3.09.8 DISCUSSION
some of the issues which are relevant to the
This chapter focused on computational mechanics of composite materials have been
methods that can be applied to the simulation only mentioned and not discussed fully.
 Discussion 533

Figure 63 Homogenization of a laminate: (a) the RVE; (b) the constitutive behavior of the single laminas; (c)
results of the identification process for an elastic parameter; and (d) results of the identification process for a
yield stress.

Among these: experimental research; micro- to explain how and in which cases the different
mechanical, bounding, and homogenization computational tools can be advantageously
theories; transformation field methods; multi- applied. Section 3.09.4 focused on the so-called
scale approaches; and stochastic approaches. PFA, which are still used for the simulation of
A second important consequence of the progressive degradation in composites.
choices made is the fact that attention has The more recent subject of CDM applied to
been mainly paid throughout the chapter to the composites has been discussed in Section
study of laminates with long fibers and 3.09.5, where specific material models and
polymeric matrices. Metal and ceramic matrix techniques for their numerical treatement have
composites, short fiber composites, textile, and been presented. The description of discrete
3D and 4D composites have not been dis- fracture phenomena like delamination has been
cussed. extensively treated in Section 3.09.6 in the
After the selection of the subject, the chapter framework of what has been defined in this
has been structured so as to discuss methodol- chapter as IDM. It is in fact the author’s
ogies for the description of damage and opinion that the wide range of applications
fracture from their basic motivations to the that interface models and correlated numerical
final formulation. To this end, after a first methods have found in the recent literature,
introductory section, the Section 3.09.2 was justify the introduction of a specific terminol-
devoted to the description of the basic failure ogy.
mechanisms which can occur in composite A description of numerical procedures for
materials or structures. Among them, rupture the use of damage and interface models for
of fibers, fiber/matrix debonding, matrix composites in the framework of FE methods
microcracking, and delamination have been has been given in Section 3.09.7, together with
enucleated as those responsible of the majority some examples of numerical simulations con-
of failures and, therefore, as those to take into cerning the non-linear behavior of composites.
account when formulating numerical simula- Many possibilities are offered to the engineers
tion tools. who want to simulate realistically the nonlinear
In Section 3.09.3, an overview of computa- behavior of composites. Naturally, each numer-
tional methods of FM have been given, trying ical procedure has its own limitations. For some
534 Damage and Fracture Mechanics Techniques for Composite Structures
of them the use is limited by the computational 3.09.9 REFERENCES
cost, for others it is limited by the lack of
experimental data available for a correct defini- M. M. Aghdam, D. J. Smith and M. J. Pavier, 2000, Finite
element micromechanical modeling of yield and collapse
tion of constitutive models, in yet other cases behavior of metal matrix composites. J. Mech. Phys.
the limitation comes from the intrinsic character Solids, 48, 499–528.
of the procedure and its impossibility to capture G. Alfano and M. A. Crisfield, 2001, Finite element
important phenomena. Besides the limitations interface models for the delamination analysis of
typical of a specific approach for composites, it laminated composites: mechanical and computational
issues. Int. J. Num. Meth. Eng., 50(7), 1701–1736.
should be remembered that the simulation of G. Alfano, F. Auricchio, L. Rosati and E. Sacco, 2001,
fracture and damage remains a difficult task MITC finite elements for laminated composite plates.
also for homogeneous materials. Bifurcations Int. J. Num. Meth. Eng., 50, 707–738.
and instabilities, mesh dependency, difficulties D. H. Allen, 2001, Homogenization principles and their
application to continuum damage mechanics. Compo-
related to the description of discrete crack sites Sci. Technol., 61, 2223–2230.
processes are all items which must still, at least D. H. Allen, C. E. Harris and S. E. Groves, 1987, A
partially, be solved. thermomechanical constitutive theory for elastic compo-
The above mentioned problems of computa- sites with distributed damage-I. Theoretical develop-
tional cost, lack of complete experimental data, ment. Int. J. Solids Struct., 23, 1301–1318.
D. H. Allen, R. H. Jones and J. G. Boyd, 1994,
and difficulties related to the description of Micromechanical analysis of a continuous fiber metal
discrete crack processes are all issues which matrix composite including the effects of matrix visco-
have stimulated and continue to stimulate the plasticity and evolving damage. J. Mech. Phys. Solids,
research in the field. New approaches seem to 42, 505–529.
O. Allix, 1992, Damage analysis of delamination around a
be well suited for the reduction of computa- hole. In: ‘‘New Advances in Computational Structural
tional cost, among them the nonlinear multi- Mechanics,’’ eds. P. Ladevèze and O. C. Zienkiewicz,
scale methods and the use of domain Elsevier, Amsterdam, pp. 411–421.
decomposition methods which can be coupled O. Allix, 2001, A composite damage meso-model for
impact problems. Composites Sci. Technol., 61, 2193–
naturally with the use of parallel computing.
2205.
The discrepancies between the complexity of O. Allix and A. Corigliano, 1996, Modeling and simulation
constitutive models for composites and experi- of crack propagation in mixed-modes interlaminar
mental data available for an efficient parameter fracture specimens. Int. J. Fract., 77, 111–140.
identification have stimulated research in the O. Allix and A. Corigliano, 1999, Geometrical and
interfacial non-linearities in the analysis of delamination
field of indirect parameter identification tools. in composites. Int. J. Solids Struct., 36, 2189–2216.
New approaches are available which allow for O. Allix, M. Dommanget, M. Gratton and P. L. Héreil,
a complete experiment design and for the 1997, Impact on 3D carbon/carbon composites: A meso-
formulation of models on the basis of a scale approach. J. Phys. IV France, 7, 675–680.
O. Allix and P. Ladevèze, 1992, Interlaminar interface
comparison between numerical simulation modeling for the prediction of laminates delamination.
and experiments. Composite Struct., 22, 235–242.
The description of discrete crack processes O. Allix, P. Ladevèze and A. Corigliano, 1995, Damage
require the use of techniques based on the analysis of interlaminar fracture specimens. Composite
introduction in finite element analyses of Struct., 31, 61–74.
O. Allix, P. Ladevèze, D. Gilletta and R. Ohayon, 1989, A
displacement discontinuities. IDM has there- damage prediction method for composite structures. Int.
fore become a very useful tool for numerical J. Num. Meth. Eng., 27, 271–283.
simulations. New procedures for the introduc- O. Allix, P. Ladevèze and E. Vittecoq, 1994, Modeling and
tion of discrete cracks have been developed, identification of the mechanical behavior of composite
like the element free Galerkin methods, the laminates in compression. Composite Sci. Technol., 51,
35–42.
embedded crack finite elements, and the parti- O. Allix, D. Leveque and L. Perret, 1998, Identification and
tion of unity method. The problem of transi- forecast of delamination in composite laminates by an
tion from diffuse damage to the formation of interlaminar interface model. Composite Sci. Technol.,
discrete cracks is, nevertheless, still an open 58, 671–678.
G. S. Amrutharaj, K. Y. Lam and B. Cotterell, 1996,
issue. Delaminations at the free edge of a composite laminate.
In spite of the great amount of research Composites: Part B, 27B, 475–483.
which has been done on the subject of this A. S. Argon, 2000, Fracture: strength and toughness
chapter, no universal method exists which can mechanisms. In ‘‘Comprehensive Composite Materials,’’
solve every problem, and probably it will never eds. A. Kelly and C. Zweben, Elsevier, Oxford, UK, vol.
1, chap. 24, pp. 763–802.
exist. As always, when the reality is modeled X. Aubard, 2001, New advances in damage mechanics and
and simulated by means of computer codes, the computational methods for composites: from research to
judgment of the engineer is an ingredient of industry for spatial applications. Composites Sci. Tech-
paramount importance for the correct choice nol., 61, 2337–2344.
F. Auricchio and E. Sacco, 1998, Delamination analysis of
and application of one specific approach; it can stitched layered plates. In: ‘‘Computational Mechanics:
only be built by means of intense study and a New Trends and Applications,’’ eds. S. Idelshon, E.
large amount of experience on the specific field. Oñate and A. Dvorkin, Cimne, Barcelona, Spain.
 References 535
Y. A. Bahei-El-Din and G. J. Dvorak, 2000, Micromecha- J. L. Chaboche and J. F. Maire, 2001, New progress in
nics of inelastic composite materials. In: ‘‘Comprehensive micromechanics-based CDM models and their applica-
Composite Materials,’’ eds. A. Kelly and C. Zweben, tion to CMCs. Composites Sci. Technol., 61, 2239–2246.
Elsevier, Oxford, UK, vol. 1, chap. 14, pp. 403–430. J. Chen, M. Crisfield, A. J. Kinloch, E. Busso, F. L.
J. Backlund, 1981, Fracture analysis of notched compo- Matthews and Y. Qiu, 1999, Predicting progressive
sites. Comp. Struct., 13, 145–1154. delamination of composite material specimens via inter-
G. Bao and Y. Song, 1993, Crack bridging models for fiber face elements. J. Mech. Composite Mater. Struct., 6, 301–
composites with slip-dependent interfaces. J. Mech. 317.
Phys. Solids, 41, 1425–1444. Z. Chen and H. L. Schreyer, 1990, A numerical solution
G. I. Barrenblatt, 1959, The formation of equilibrium scheme for softening problems involving total strain
cracks during brittle fracture-general ideas and hypoth- control. Comp. Struct., 37, 1043–1050.
esis, axially symmetric cracks. Prikl. Math Mekh., 23, C. Comi, A. Corigliano and G. Maier, 1991, Extremum
434–444. properties of finite-step solutions in elastoplasticity with
Z. L. Bažant and L. Cedolin, 1991, ‘‘Stability of nonlinear mixed hardening. Int. J. Solids Struct., 27,
structures,’’ Oxford University Press, Oxford, UK. 965–981.
Z. P. Bažant and B. H. Oh, 1983, Crack band theory for A. Corigliano, 1993, Formulation, identification and use of
fracture of concrete. Materiaux et Constructions, 16, interface models in the numerical analysis of composite
155–177. delamination. Int. J. Solids Struct., 30, 2779–2811.
V. V. Bolotin, 1996, Delaminations in composite struc- A. Corigliano and O. Allix, 2000, Some aspects of
tures: its origin, buckling, growth and stability. Compo- interlaminar degradation in composite. Comp. Meth.
sites: Part B, 27B, 129–145. Appl. Mech. Eng., 185, 203–224.
G. Bolzon and A. Corigliano, 1997a, A discrete formula- A. Corigliano and G. Bolzon, 1995, Numerical simulation
tion for elastic solids with damaging interfaces. Comp. of debonding phenomena in composites materials. In:
Meth. Appl. Mech. Eng., 140, 329–359. ‘‘Computational Plasticity: Fundamentals and Applica-
G. Bolzon and A. Corigliano, 1997b, An embedded-crack tions,’’ eds. D. R. J. Owen, E. Oñate and E. Hinton,
finite element approach to quasi-brittle fracture. In: Cimne, Barcelona, Spain, pp. 1179–1190.
‘‘Advances in Fracture Research,’’ eds. B. L. Karihaloo, A. Corigliano, R. Frassine and M. Ricci, 1997, Rate-
Y. W. Mai, M. I. Ripley and R. O. Ritchie, Pergamon, dependent interface models for the analysis of delamina-
Oxford, UK, pp. 2127–2134. tion in polymer-matrix composites. ed. H. P. Rossma-
G. Bolzon and A. Corigliano, 2000, Finite elements with nith, Balkema, Rotterdam, The Netherlands.
embedded displacement discontinuity: a generalised A. Corigliano, R. Frassine and M. Ricci, 1998, Rate-
variable formulation. Int. J. Num. Meth. Eng., 49, dependent fracture properties in the delamination of
1227–1266. polymer-matrix composites. In: ‘‘Computational Me-
G. Bolzon, R. Fedele and G. Maier, 2002aa, Identification chanics: New Trends and Applications,’’ eds. S. Idel-
of cohesive crack models by Kalman filter. Comp. Meth. shon, E. Oñate and A. Dvorkin, Cimne, Barcelona,
Appl. Mech. Eng., 191, 2847–2871. Spain.
G. Bolzon, D. Ghilotti and G. Maier, 2002bb, Strength of A. Corigliano and S. Mariani, 2001, Simulation of damage
periodic elastic-brittle composites evaluated through in composites by means of interface models: Parameter
homogenization and parameter identification. Euro. J. identification. Composites Sci. Technol., 61, 2299–2315.
Mech. A. Solids, 21, 355–378. A. Corigliano and S. Mariani, 2002, Identification of
R. Borg, L. Nilsson and K. Simonsson, 2001, Simulation of constitutive models for the simulation of time-dependent
delamination in fiber composites with a discrete cohesive interlaminar debonding processes in composites. Comp.
failure model. Composites Sci. Technol., 61, 667–677. Meth. Appl. Mech. Eng., 191, 1891–1894.
J. C. Brewer and P. A. Lagace, 1988, Quadratic stress A. Corigliano and M. Ricci, 2001, Rate-dependent inter-
criterion for initiation of delamination. J. Composites face models: Formulation and numerical applications.
Mater., 22, 1141–1155. Int. J. Solids Struct., 38/4, 547–576.
D. Bruno and A. Grimaldi, 1990, Delamination failure of A. Corigliano, M. Ricci and R. Contro, 1997, Rate
layered composite plates loaded in compression. Int. J. dependent delamination in polymer-matrix composites.
Solids Struct., 26, 313–330. In: ‘‘Computational Plasticity: Fundamentals and Ap-
B. Budiansky, A. G. Evans and J. W. Hutchinson, 1995, plications,’’ eds. D. R. J. Owen, E. Oñate and E. Hinton,
Fiber–matrix debonding effects on cracking in aligned Cimne, Barcelona, Spain, pp. 1168–1175.
fiber ceramic composites. Int. J. Solids Struct., 32, 315– A. Corigliano, E. Rizzi and E. Papa, 2000, Experimental
328. characterization and numerical simulations of a syntac-
A. A. Caiazzo and F. Costanzo, 2001a, Effective 3D tic-foam/glass-fiber composite sandwich. Composites Sci.
constitutive equations of composites with evolving Technol., 60(11), 2169–2180.
damage. J. Eng. Mech., 127, 661–666. F. Costanzo and D. H. Allen, 1995, A continuum
A. A. Caiazzo and F. Costanzo, 2001b, Modeling the thermodynamic analysis of cohesive zone models. Int.
constitutive behavior of layered composites with evol- J. Eng. Sci., 33, 2197–2219.
ving cracks. Int. J. Solids Struct., 38, 3469–3485. B. Cox, R. Massabò and K. T. Kedward, 1996, The
G. T. Camacho and M. Ortiz, 1996, Computational suppression of delaminations in curved structures by
modeling of impact damage in brittle materials. Int. J. stitching. Composites, A, 27A, 1133–1138.
Solids Struct., 33, 2899–2938. M. A. Crisfield, 1981, A fast incremental/iterative solution
W. J. Cantwell, R. Scudamore, J. Ratcliffe and P. Davies, procedure that handles snap-through. Comp. Struct., 13,
1999, Interfacial fracture in sandwich laminates. Com- 55–62.
posites Sci. Technol., 59, 2079–2085. M. A. Crisfield, Y. Mi, G. A. O. Davies and H.-B. Hellweg,
A. Carpinteri, 1989, Decrease of apparent tensile and 1997. Finite element methods and the progressive failure
bending strength with specimen size: two different modeling of composite structures. In: ‘‘Computational
explanations based on fracture mechanics. Int. J. Solids Plasticity: Fundamentals and Applications,’’ eds. D. R. J.
Struct., 25, 407–429. Owen, E. Oñate and E. Hinton, Cimne, Barcelona,
V. Carvelli and A. Taliercio, 1999, A micromechanical Spain, Part 1, pp. 239–254.
model for the analysis of unidirectional elastoplastic M. A. Crisfield and J. Shi, 1991, A review of solution
composites subjected to 3D stresses. Mech. Res. Comm., procedures and path-following techniques in equation to
26, 544–553. the nonlinear finite element analysis of structures. In:
536 Damage and Fracture Mechanics Techniques for Composite Structures
‘‘Nonlinear Computational Mechanics: State of the R. Frassine, M. Rink and A. Pavan, 1996, Viscoelastic
Art,’’ eds. P. Wriggers and W. Wagner, Springer, Berlin, effects on the interlaminar fracture behavior of thermo-
pp. 47–68. plastic matrix composites: II. Rate and temperature
I. M. Daniel and O. Ishai, 1994, ‘‘Engineering mechanics of dependence in unidirectional PEEK/carbon-fiber lami-
composite materials,’’ Oxford University Press, Oxford, nates. Composites Sci. Technol., 56, 1253–1260.
UK. M. Frémond, 1987, Adhérence des solides. J. Mécanique
L. Daudeville, O. Allix and P. Ladevèze, 1995, Delamina- Théorique et Appliquée, 6, 383–407.
tion analysis by damage mechanics: some applications. A. C. Garg, 1988, Delamination—a damage mode in
Composite Eng., 5(1), 17–24. composite structures. Eng. Fract. Mech., 29, 557–584.
L. Daudeville and P. Ladevèze, 1993, A damage mechanics M. G. D. Geers, R. de Borst and T. Peijs, 1999, Mixed
tool for laminate delamination. J. Composite Struct., 25, numerical-experimental identification of non-local char-
547–555. acteristics of random-fiber-reinforced composites. Com-
R. de Borst, 1987, Computation of post bifurcation and posite Sci. Technol., 59, 1569–1578.
post-failure behavior of strain-softening solids. Compu- S. Ghosh, K. Lee and S. Moorthy, 1996, Two scale
ter Struct., 25, 211–224. analysis of heterogeneous elastic–plastic materials with
R. de Borst, 2001, Some recent issues in computational asymptotic homogenization and Voronoi cell finite
failure mechanics. Int. J. Num. Meth. Eng., 52, 63–95. element model. Comp. Meth. Appl. Mech. Eng., 132,
R. de Borst and J. H. A. Schipperen, 2002, Computational 63–116.
methods for delamination and fracture in composites. In: S. Ghosh, K. Lee and P. Raghavan, 2001, A multi-level
‘‘Continuum Damage Mechanics of Materials and computational model for multiscale damage analysis in
Structures,’’ eds. O. Allix and F. Hild, Elsevier, Oxford, composite and porous materials. Int. J. Solids Struct., 38,
UK. 2335–2385.
K. Derrien, J. Fitoussi, G. Guo and D. Baptiste, 2000, S. Ghosh, Y. Ling, B. Majumdar and R. Kim, 2000,
Prediction of the effective damage properties and failure Interfacial debonding analysis in multiple fiber rein-
properties of nonlinear anisotropic discontinuous re- forced composites. Mech. Mater., 32, 561–591.
inforced composites. Comp. Meth. Appl. Mech. Eng., R. Giampieretti, A. Corigliano and G. Maier, 2000,
185, 93–107. Constitutive modeling of composites and laminates via
F. Devries, H. Dumontet, G. Duvaut and F. Lene, 1989, homogenization and parameter identification. In: ‘‘Pro-
Homogenization and damage for composite structures. ceedings: FRC 2000, Newcastle-upon-Tyne, 13–15 Sep-
Int. J. Num. Meth. Eng., 27, 285–298. tember,’’ ed. A. G. Gibson, Woodhead, Cambridge, UK,
D. S. Dugdale, 1960, Yielding of steel sheets containing pp. 449–457.
slits. J. Mech. Phys. Solids, 8, 100–104. J. W. Gillespie, L. A. Carlsson and R. B. Pipes, 1986, Finite
G. J. Dvorak, 2000, Composite materials: Inelastic element analysis of the end notched flexure specimen for
behavior, damage, fatigue and fracture. Int. J. Solids measuring mode II fracture toughness. Composites Sci.
Struct., 37, 155–170. Technol., 27, 177–197.
G. J. Dvorak and J. Zhang, 2001, Transformation field L. J. Hart-Smith, 2000, Strength of unnotched laminates
analysis of damage evolution in composite materials. J. under multiaxial stress. In: ‘‘Comprehensive Composite
Mech. Phys. Solids, 49, 2517–2541. Materials,’’ eds. A. Kelly and C. Zweben, Elsevier,
J. D. Ferry, 1980, ‘‘Viscoelastic Properties of Polymers,’’ Oxford, UK, vol. 1, chap. 21, pp. 667–681.
3rd edn., Wiley, New York. F. Hashagen and R. de Borst, 2000, Numerical assessment
J. Fish and V. Belsky, 1995a, Multigrid method for of delamination in fiber metal laminates. Comp. Meth.
periodic heterogeneous media. Part 1: convergence Appl. Mech. Eng., 185, 141–159.
studies for one-dimensional case. Comp. Meth. Appl. F. Hashagen, R. de Borst and T. de Vries, 1999,
Mech. Eng., 126, 1–16. Delamination behavior of spliced fiber metal laminates.
J. Fish and V. Belsky, 1995b, Multi-grid method for Part 2: Numerical investigation. Composite Struct., 46,
periodic heterogeneous media. Part 2: multiscale model- 147–162.
ing and quality control in multidimensional case. Comp. F. Hashagen, J. C. J. Schellekens and R. de Borst, 1995,
Meth. Appl. Mech. Eng., 126, 17–38. Finite element procedure for modeling fiber metal
J. Fish, K. Shek, M. Pandheeradi and M. S. Shepard, 1997, laminates. Composite Struct., 32, 255–264.
Computational plasticity for composite structures based Z. Hashin, 1980, Failure criteria for unidirectional fiber
on mathematical homogenization: theory and practice. composites. J. Appl. Mech., 47, 329–334.
Comp. Meth. Appl. Mech. Eng., 148, 53–73. Z. Hashin, 1985, Analysis of cracked laminates: a varia-
J. Fish and A. Wagiman, 1993, Multiscale finite element tional approach. Mech. Mater., 4, 121–136.
method for locally nonperiodic hetereogeneous medium. Z. Hashin, 1986, Analysis of stiffness reduction of cracked
Comp. Mech., 12, 164–180. cross-ply laminates. Eng. Fract. Mech., 25, 771–778.
J. Fish and Q. Yu, 2001a, Two-scale damage modeling of Z. Hashin, 1996, Finite thermoelastic fracture criterion
brittle composites. Composites Sci. Technol., 61, 2215– with application to laminate cracking analysis. J. Mech.
2222. Phys. Sol., 44, 1129–1145.
J. Fish and Q. Yu, 2001b, Multiscale damage modeling for H. B. Hellweg and M. A. Crisfield, 1998, A new arc-length
composite materials: theory and computational frame- method for handling sharp snap-backs. Comp. Struct.,
work. Int. J. Num. Meth. Eng., 52, 161–191. 66(5), 705–709.
J. Fish, Q. Yu and K. Shek, 1999, Computational damage H. B. Hellweg, M. A. Crisfield, G. A. Duffett and D. J.
mechanics for composite materials based on mathema- Irving, 1993, Efficient finite element analysis of 3-D
tical homogenization. Int. J. Num. Meth. Eng., 45, 1657– composite structures. In: ‘‘Proceedings of the Interna-
1679. tional Conference on the Deformation and Fracture of
R. Frassine and A. Pavan, 1995, Viscoelastic effects on the Composites, Manchester, UK,’’ pp. 15/1–15/10.
interlaminar fracture behavior of thermoplastic matrix H. B. Hellweg, M. A. Crisfield and G. A. O. Gao, 1994,
composites: I. Rate and temperature dependence in Interface elements vs. virtual crack closure. In: ‘‘POLY-
unidirectional PEI/carbon-fiber laminates. J. Comp. Sci. MAT ’94,’’ Institute of Materials, pp. 669–672.
Tech., 54, 193–200. C. T. Herakovich, 1989, Edge effects and delamination
R. Frassine, M. Rink and A. Pavan, 1993, Viscoelastic failures. J. Strain Anal., 24, 245–252.
effects on intralaminar fracture toughness of epoxy/ C. T. Herakovich, 1998, ‘‘Mechanics of Fibrous Compo-
carbon fiber. Int. J. Comp. Mater., 27, 921–933. sites,’’ Wiley, New York.
 References 537
K. Hibbit and B. F. S
rensen, 1997, ABAQUS, standard P. Ladevèze, O. Allix, J. F. Deü and D. Leveque, 2000a, A
version 5.7, manuals. mesomodel for localisation and damage computation in
R. Hill, 1948, A theory of the yielding and plastic flow of laminates. Comp. Meth. Appl. Mech. Eng., 183, 105–122.
anisotropic materials. Proc. Roy. Soc., 193, 281. P. Ladevèze and E. le Dantec, 1992, Damage modeling of
A. Hillerborg, M. Modéer and P. E. Petersson, 1976, the elementary ply for laminated composites. Composite
Analysis of crack formation and crack growth in Sci. Technol., 43, 257–267.
concrete by means of fracture mechanics and finite P. Ladevèze, L. Guitard, L. Champaney and X. Aubard,
elements. Cement Concrete Res., 6, 773–782. 2000b, Debond modeling for multidirectional compo-
O. Hoffman, 1967, The brittle strength of orthotropic sites. Comp. Meth. Appl. Mech. Eng., 185, 109–122.
materials. J. Composite Mater., 1, 200. P. Ladevèze and G. Lubineau, 2001, On a damage
D. Hull and T. W. Clyne, 1996, ‘‘An Introduction to mesomodel for laminates: micro–meso equationships,
Composite Materials,’’ Cambridge University Press, possibilities and limits. Composite Sci. Technol., 61,
Cambridge, UK. 2149–2158.
J. W. Hutchinson and Z. Suo, 1992, Mixed mode cracking J. Lamon, 2001, A micromechanics-based approach to the
in layered materials. Adv. Appl. Mech., 28, 63–191. mechanical behavior of brittle-matrix composites. Com-
C. G. Hwang, P. A. Wawrzynek, A. K. Tayebi and A. R. posites Sci. Technol., 61, 2259–2272.
Ingraffea, 1998, On the virtual crack extension method C. M. Landis, I. J. Beyerlein and R. M. McMeeking, 2000,
for calculation of the rates of energy release rate. Eng. Micromechanical simulation of the failure of fiber
Fract. Mech., 59, 521–542. reinforced composites. J. Mech. Phys. Solids, 48, 621–648.
M. W. Hyer, 2000, Laminated plate and shell theory. In: K. Lee, S. Moorthy and S. Ghosh, 1999, Multiple-scale
‘‘Comprehensive Composite Materials,’’ eds. A. Kelly computational model for damage in composite materi-
and C. Zweben, Elsevier, Oxford, UK, vol. 1, chap. 17, als. Comp. Meth. Appl. Mech. Eng., 172, 175–201.
pp. 479–510. J. Lemaitre, 1992, Formulation de l’endommagement des
M. W. Hyer and A. M. Waas, 2000, Micromechanics of interfaces. C. R. Acad. Sci. Paris, 315, 1047–1050.
linear elastic continuous fiber composites. In: ‘‘Compre- J. Lemaitre and J. L. Chaboche, 1990, ‘‘Mechanics of Solid
hensive Composite Materials,’’ eds. A. Kelly and C. Materials,’’ Cambridge University Press, Cambridge, UK.
Zweben, Elsevier, Oxford, UK, vol. 1, chap. 12, pp. 345– F. Lene, 1986, Damage constitutive Equations for compo-
375. site materials. Eng. Fract. Mech., 25, 713–728.
N. E. Jansson and R. Larsson, 2001, A damage model for J. E. Lindhagen, E. K. Gamstedt and L. A. Berglund, 2000,
simulation of mixed-mode delamination growth. Com- Application of bridging-law concepts to short fiber
posite Struct., 53, 409–417. composites: Part 3. Bridging law derivation from
L. M. Kachanov, 1958, Time of the rupture process under experimental crack profiles. Composites Sci. Technol.,
creep conditions. Izv. Akad. Nauk. SSR Otd. Tech. Nauk, 60, 2883–2894.
8, 26–31. H. R. Lotfi and B. Shing, 1995, Embedded representation
M. F. Kanninen and C. H. Popelar, 1985, ‘‘Advanced of fracture in concrete with mixed finite elements. Int. J.
Fracture Mechanics,’’ Oxford University Press, Oxford, Num. Meth. Eng., 38, 1307–1325.
UK. R. E. Lotter and M. H. Santare, 1991, Analysis of mode I
V. M. Karbhari and L. Zhao, 2000, Use of composites for and mode II interlaminar fracture specimens by a
21st century civil infrastructure. Comp. Meth. Appl. comparative finite element method. Composites Sci.
Mech. Eng., 185, 433–454. Technol., 40, 87–107.
B. L. Karihaloo and J. Wang, 2000, Mechanics of fiber G. Maier, 1968, On softening flexural behavior in elastic-
reinforced cementitious composites. Comp. Struct., 76, plastic beams (in Italian). Rend. Ist. Lombardo, Classe
19–34. Sci., A102, 648–677 (English translation in Studi e
B. L. Karihaloo, J. Wang and M. Grzybowski, 1996, Ricerche, 1986, 8, 85–117).
Doubly periodic arrays of bridged cracks and short fibre G. Maier and A. Frangi, 1998, Symmetric boundary
ereinforced cementitious composites. J. Mech. Phys. element method for discrete crack modeling of fracture
Solids, 44, 1565–1586. processes. Comp. Ass. Mech. Eng. Sci., 5, 201–226.
B. L. Karihaloo, Q. Z. Xiao and C. C. Wu, 2001, G. Maier, G. Novati and Z. Z. Cen, 1993, Symmetric
Homogenisation-based multivariable element method Galerkin boundary element method for quasi-brittle
for composite materials. Comp. Struct., 79, 1645–1660. fracture and frictional contact problems. Comp. Mech.,
D. N. Kaziolas and C. C. Baniotopoulos, 2001, On the 13, 74–89.
debonding in multilayered composite structures. A new J. F. Maire and P. M. Lesne, 1998, An explicit damage
numerical approach. Comp. Mech., 27, 160–169. model for the design of composites structures. Compo-
A. Kelly and C. Zweben (eds.), 2000, ‘‘Comprehensive sites Sci. Technol., 58, 773–778.
Composite Materials,’’ Elsevier, Oxford, UK, vol. 1–6. R. Mahnken and E. Stein, 1996, Parameter identification
Z. Kutlu and F. K. Chang, 1995a, Composite panels for viscoplastic models based on analytical derivatives of
containing multiple through the-width delaminations a least-squares functional and stability investigations.
and subjected to compression: Part I. Analysis. Compo- Int. J. Plasticity, 12, 451–479.
sites Struct., 31, 273–296. S. Mariani and U. Perego, 2001, A PU-FE approach to
Z. Kutlu and F. K. Chang, 1995b, Composite panels quasi-brittle fracture. In: ‘‘Proceedings of the XV
containing multiple through the-width delaminations National Congress AIMETA, September 26–29, 2001,’’
and subjected to compression: Part II. Experiments and (on CD), ISSN 1592-8950, eds. G. Augusti, P. M.
verification. Composites Struct., 31, 297–314. Mariano, V. Sepe and M. Lacagina, Taormina, Italy,
P. Ladevèze, 1983, ‘‘Sur une theorie de l’endommagement pp. xx.
anisotrope,’’ Report 34, Laboratoire de Mécanique et R. Massabò and B. N. Cox, 1999, Concepts for bridged
Technologie, Cachan, France. mode II delamination cracks. J. Mech. Phys. Solids, 47,
P. Ladevèze, 1986, Sur la mécanique de l’endommagement 1265–1300.
des composites. In: ‘‘Proceedings of JNC 5, Paris, 1986,’’ A. Matzenmiller, J. Lubliner and R. L. Taylor, 1995, A
eds. C. Bathias and D. Menkès, Pluralis, Paris, pp. 667– constitutive model for anisotropic damage in fiber
683. composites. Mech. Mater., 20, 125–152.
P. Ladevèze, 1995, A damage computational approach for L. N. McCartney, 1993, The prediction of cracking in
composites: basic aspects and micromechanical Equa- biaxially loaded laminates having brittle matrices.
tions. Comp. Mech., 17, 142–150. Composites, 24, 84–92.
538 Damage and Fracture Mechanics Techniques for Composite Structures
L. N. McCartney, 1998, Predicting transverse crack N. Point and E. Sacco, 1996a, A delamination model for
formation in cross-ply laminates resulting from micro- laminated composites. Int. J. Solids Struct., 33, 483–509.
cracking. Composites Sci. Technol., 58, 1069–1081. N. Point and E. Sacco, 1996b, Delamination of beams: an
L. N. McCartney, 1999, Analytical model for debonded application to DCB specimen. Int. J. Fract., 79, 225–247.
interfaces associated with fiber fractures or matrix Y. Qiu, M. A. Crisfield and G. Alfano, 2001, An interface-
cracks. In: ‘‘Proceedings ICCM 12, Paris,’’ ISBN 2-951 element formulation for the simulation of delamination
4526-2-4, TCA-ICCM 12, Gradignan, France. with buckling. Eng. Fract. Mech., 68, 1755–1776.
L. N. McCartney, 2000, Model to predict effects of triaxial Y. N. Rabotnov, 1968, Creep rupture. In: ‘‘Proceedings of
loading on ply cracking in general symmetric laminates. the 12th International Congress on Applied Mechanics,’’
Composites Sci. Technol., 60, 2255–2279. ed. Stanford, CA.
L. N. McCartney, G. A. Schoeppner and W. Becker, 2000, J. J. C. Remmers and R. de Borst, 2001, Delamination
Comparison of models for transverse ply cracks in buckling of fiber-metal laminates. Composites Sci.
composite laminates. Composites Sci. Technol., 60, 2347– Technol., 61, 2207–2213.
2359. J. J. C. Remmers, G. N. Wells and R. de Borst, 2001,
Y. Mi, M. A. Crisfield, H.-B. Hellweg and G. A. O. Davies, Analysis of delamination growth with discontinuous finite
1998, Progressive delamination using interface elements. elements. In: ‘‘Proceedings ECCM-2001, Cracow,’’ ed.
J. Comp. Materials, 32(14), 1246–1272. J. R. Rice, 1988, Elastic fracture mechanics concepts for
N. Möes, J. Dolbow and T. Belytschko, 1999, A finite interfacial cracks. ASME J. A. Mech., 55, 98–103.
element method for crack growth without remeshing. J. R. Rice and G. C. Sih, 1965, Plane problems of cracks in
Int. J. Num. Meth. Eng., 46, 131–150. dissimilar media. ASME J. A. Mech., 32, 418–423.
J. A. Nairn, 2000, Matrix microcracking in composites. In: E. Riks, 1972, The application of Newton’s method to the
‘‘Comprehensive Composite Materials,’’ eds. A. Kelly problem of elastic stability. J. Appl. Mech., 39, 1060–1066.
and C. Zweben, Elsevier, Oxford, UK, vol. 2, chap. 12, E. Rizzi, E. Papa and A. Corigliano, 2000, Mechanical
pp. 403–432. behavior of a syntactic foam: Experiments and modeling.
J. A. Nairn, 2001, Fracture mechanics of composites with Int. J. Solids Struct., 37/40, 5773–5794.
residual stresses, imperfect interfaces, and traction- A. J. Russell and K. N. Street, 1985, In: ‘‘ASTM STP 976,’’
loaded cracks. Composites Sci. Technol., 61, 2159–2167. American Society for Testing and Materials, Philadel-
B. N. Nguyen, 1998, Damage modeling of laminated phia, PA, pp. 349–370.
composites by the use of multilayer volume elements. E. F. Rybicki and M. F. Kanninen, 1977, A finite element
Composites Sci. Technol., 58, 891–905. calculation of stress intensity factors by a modified crack
Q. S. Nguyen, 1987, Bifurcation and post-bifurcation closure integral. Eng. Fract. Mech., 9, 931–938.
analysis in plasticity and brittle fracture. J. Mech. Phys. A. F. Saleeb and T. E. Wilt, 1993, Analysis of the
Solids, 35, 303–324. anisotropic viscoplastic-damage response of composite
K. F. Nilsson, L. E. Asp, J. E. Alpman and L. Nystedt, laminates-continuum basis and computational algo-
2001, Delamination buckling and growth for delamina- rithms. Int. J. Num. Meth. Eng., 36, 1629–1660.
tions at different depths in a slender composite panel. E. Sanchez-Palencia, 1980, ‘‘2 non-homogeneous Media
Int. J. Solids Struct., 38, 3039–3071. and Vibration Theory: Lectures Notes in Physics,’’
K. F. Nilsson, J. C. Thesken, P. Sindelar, A. Giannako- Springer, Berlin.
poulos and B. Storåkers, 1993, A theoretical and J. C. J. Schellekens and R. de Borst, 1993aa, A non-linear
experimental investigation of buckling induced delami- finite element approach for the analysis of mode-I free
nation growth. J. Mech. Phys. Solids, 41, 749–782. edge delamination in composites. Int. J. Solids Struct.,
O. O. Ochoa and J. N. Reddy, 1992, ‘‘Finite Element 30, 1239–1253.
Analysis of Composite Laminates,’’ Kluwer, Dordrecht, J. C. J. Schellekens and R. de Borst, 1993bb, On the
The Netherlands. numerical integration of interface elements. Int. J. Num.
M. H. J. W. Paas and J. Van Den Eikhoff, 1992, Numerical Meth. Eng., 36, 43–66.
analysis of degradation processes in laminated composite J. C. J. Schellekens and R. de Borst, 1994, Free edge
materials. Heron, 37, 17–32. delamination in carbon-epoxy laminates: a novel numer-
N. J. Pagano (ed.), 1989, ‘‘Interlaminar Response of ical/experimental approach. Composite Struct., 28, 357–
Composite Materials,’’ Elsevier, New York. 373.
N. J. Pagano and R. B. Pipes, 1973, Some observations on J. H. A. Schipperen, 2001, An anisotropic damage model
the interlaminar strength of composite laminates. Int. J. for the description of transverse matrix cracking in a
Mech. Sci., 15, 679–688. graphite-epoxy laminate. Composite Struct., 53, 295–299.
N. J. Pagano and G. A. Schoeppner, 2000, Delamination of J. H. A. Schipperen and R. de Borst, 2001, A numerical
polymer matrix composites: problems and assessment. analysis of mixed-mode delamination in carbon-epoxy
In: ‘‘Comprehensive Composite Materials,’’ eds. A. prepregs. Composites Struct., 54, 445–451.
Kelly and C. Zweben, Elsevier, Oxford, UK, vol. 2, J. Schön, T. Nyman, A. Blom and H. Ansell, 2000, A
chap. 13, pp. 433–528. numerical and experimental investigation of delamina-
A. Pandolfi, P. Krysl and M. Ortiz, 1999, Finite element tion behavior in the DCB specimen. Composites Sci.
simulation of ring expansion and fragmentation: the Technol., 60, 173–184.
capturing of length and time scales through cohesive G. T. Sha, 1984, On the virtual crack extension technique
models of fracture. Int. J. Fract., 95, 1–18. for stress intensity factors and energy release rate
H. Parisch, 1995, A continuum-based shell theory for non- calculations for mixed fracture mode. Int. J. Fract., 25,
linear applications. Int. J. Num. Meth. Eng., 38, 1855– 33–42.
1883. F. Shen, K. H. Lee and T. E. Tay, 2001, Modeling
U. Perego, 1988, Explicit backward difference operators delamination growth in laminated composites. Compo-
and consistent predictors for linear hardening elastic- sites Sci. Technol., 61, 1239–1251.
plastic constitutive laws. Solid Mech. Arch., 13, 65–102. X. Shu and K. P. Soldatos, 2001, An accurate delamination
E. A. Phillips, C. T. Herakovich and L. L. Graham, 2001, model for weakly bonded laminates subjected to
Damage development in composites with large stress different sets of edge boundary conditions. Int. J. Mech.
gradients. Composites Sci. Technol., 61, 2169–2182. Sci., 43, 935–959.
R. B. Pipes and N. J. Pagano, 1970, Interlaminar stresses in R. L. Sierakowski and G. M. Newaz, 1995, ‘‘Damage
composite laminates under axial extension. J. Comp. Tolerance in Advanced Composites,’’ Technomic, Lan-
Mater., 4, 538–548. caster, UK.
 References 539
G. C. Sih, P. C. Paris and G. R. Irwin, 1965, On cracks in G. Z. Voyiadjis, B. Deliktas and E. C. Aifantis, 2001,
rectilinearly anisotropic bodies. Int. J. Fract. Mech., 1, Multiscale analysis of multiple damage mechanisms
189–203. coupled with inelastic behavior of composite materials.
J. C. Simo and R. L. Taylor, 1985, Consistent tangent J. Eng. Mech., 127, 636–645.
operators for rate-independent elastoplasticity. Comp. G. Z. Voyiadjis and T. Park, 1997, Local and interfacial
Meth. Appl. Mech. Eng., 48, 101–118. damage analysis of metal matrix composites using
B. F. S
rensen and T. K. Jacobsen, 1998, Large-scale the finite element method. Eng. Fract. Mech., 56,
bridging in composites: R-curves and bridging laws. 483–511.
Composites: Part A, 29, 1443–1451. W. Wagner, F. Gruttmann and W. Sprenger, 2001, A finite
B. F. S
rensen and T. K. Jacobsen, 2000, Crack growth in element formulation for the simulation of propagating
composites: applicability of R-curves and bridging laws. delaminations in layered composite structures. Int. J.
Plastics, Rubber, Composites, 29, 119–133. Num. Meth. Eng., 51, 1337–1359.
S. M. Spottswood and A. N. Palazotto, 2001, Progressive J. Wang and B. L. Karihaloo, 1994, Cracked composite
failure analysis of a composite shell. Composites Struct., laminates least prone to delamination. Proc. Roy. Soc.
53, 117–131. London, A444, 17–35.
W. Sprenger, F. Gruttmann and W. Wagner, 2000, G. N. Wells and L. J. Sluys, 2001, A new method for
Delamination growth analysis in laminated structures modeling cohesive cracks using finite elements. Int. J.
with continuum-based 3D shell elements and a visco- Num. Meth. Eng., 50, 2667–2682.
plastic softening model. Comp. Meth. Appl. Mech. Eng., G. N. Wells, L. J. Sluys and R. de Borst, 2002, Simulating
185, 123–139. the propagation of displacement discontinuities in a
B. Storåkers and B. Andersson, 1988, Nonlinear plate regularized strain-softening medium. Int. J. Num. Meth.
theory applied to delamination in composites. J. Mech. Eng., 53, 1235–1256.
Phys. Solids, 36, 689–718. J. G. Williams, 1988, On the calculation of energy release
N. Sukumar and M. Kumosa, 1993, Finite element analysis rates for cracked laminates. Int. J. Fract., 36, 101–119.
of axial splits in composite Iosipescu specimens. Int. J. J. G. Williams, 1989, Fracture mechanics of delamination
Fract., 62, 55–85. tests. J. Strain Anal., 24, 207–214.
C. T. Sun, 2000, Strength analysis of unidirectional K. V. Williams, A. M. Floyd, R. Vaziri and A. Poursartip,
composites and laminates. In: ‘‘Comprehensive Compo- 1999, Numerical simulation of in-plane damage progres-
site Materials,’’ eds. A. Kelly and C. Zweben, Elsevier, sion in laminated composite plates. In: ‘‘Proceedings
Oxford, UK, vol. 1, chap. 20, pp. 641–666. ICCM-12, Paris,’’ ISBN 2-9514526-2-4, TCA-ICCM 12,
Z. Suo and J. W. Hutchinson, 1990, Interface crack Gradignan, France.
between two elastic layers. Int. J. Fract., 43, 1–18. K. V. Williams, R. Vaziri, A. M. Floyd and A. Poursartip,
A. P. Suvorov and G. J. Dvorak, 2001, Optimized fiber 1998, Simulation of damage progression in laminated
prestress for reduction of free edge stresses in composite composite plates. In: ‘‘Proceedings 5th International LS-
laminates. Int. J. Solids Struct., 38, 6751–6786. DYNA Conference, Southfield, Michigan.’’
A. Taliercio and P. Sagramoso, 1995, Uniaxial strength of M. L. Williams, 1959, The stresses around a fault or
polymeric-matrix fibrous composites predicted through a crack in dissimilar media. Bull. Seismol. Soc. Am., 49,
homogenization approach. Int. J. Solids Struct., 32, 199–204.
2095–2123. M. R. Wisnom and F. K. Chang, 2000, Modeling of
R. Talreja, 1989, Damage development in composites: splitting and delamination in notched cross-ply lami-
mechanisms and modeling. J. Strain Anal., 24, 215–222. nates. Composites Sci. Technol., 60, 2849–2856.
A. Thionnet and J. Renard, 1993, Meso–macro approach J. F. Wu, M. S. Shepard, G. J. Dvorak and Y. A. Bahei-El-
to transverse cracking in laminated composites using Din, 1989, A material model for the finite element
Talreja’s model. Composites Eng., 3, 851–871. analysis of metal matrix composites. Composites Sci.
A. Thionnet and J. Renard, 1998, Multiscale analysis to Technol., 35, 1–20.
determine fiber/matrix debonding criteria in SiC/tita- Z. H. Xia and W. A. Curtin, 2001, Multiscale modeling of
nium composites with and without consideration of the damage and failure in aluminum-matrix composites.
manufacturing residual stresses. Composites Sci. Tech- Composites Sci. Technol., 61, 2247–2257.
nol., 58, 945–955. X. P. Xu and A. Needleman, 1995, Numerical simulations
M. Todo and P. Y. B. Jar, 1998, Study of mode-I of dynamic interfacial crack growth allowing for crack
interlaminar crack growth in DCB specimens of fiber- growth away from the bond line. Int. J. Fract., 74, 253–
reinforced composites. Composites Sci. Technol., 58, 275.
105–118. A. M. Yan, E. Marechal and H. Nguyen-Dang, 2001, A
S. W. Tsai, 1965, ‘‘Strength Characteristics of Composite finite element model of mixed-mode delamination in
Materials,’’ NASA CR-224. laminated composites with an R-curve effect. Composites
S. W. Tsai and E. M. Wu, 1971, A general theory of Sci. Technol., 61, 1413–1427.
strength for anisotropic materials. J. Composite Mater., W. Yang and J. P. Boehler, 1992, Micromechanics
5, 58–80. modeling of anisotropic damage in cross-ply laminates.
R. Vaziri, M. D. Olson and D. L. Anderson, 1992, Finite Int. J. Solids Struct., 10, 1303–1328.
element analysis of fibrous composite structures: a Z. J. Yang, J. F. Chen and G. D. Holt, 2001, Efficient
plasticity approach. Comp. Struct., 44, 103–116. evaluation of stress intensity factors using virtual crack
G. Z. Voyiadjis and B. Deliktas, 2000, A coupled extension technique. Comp. Struct., 79, 2705–2715.
anisotropic damage model for the inelastic response of R. Zinno and F. Greco, 2001, Damage evolution in
composite materials. Comp. Meth. Appl. Mech. Eng., bimodular laminated composites under cyclic loading.
183, 159–199. Composite Struct., 53, 381–402.

Copyright 2003, Elsevier Ltd. All Rights Reserved. Comprehensive Structural Integrity
No part of this publication may be reproduced, stored in any retrieval system or ISBN (set): 0-08-043749-4
transmitted in any form or by any means: electronic, electrostatic, magnetic tape,
Volume 3; (ISBN: 0-08-044158-0); pp. 459–539
mechanical, photocopying, recording or otherwise, without permission in writing
from the publishers.