A continuum damage model for composite laminates:

Part I Constitutive model

P. Maim
a
, P.P. Camanho
b,
*
, J.A. Mayugo
a
, C.G. Davila
c
a
AMADE, Escola Polite` cnica Superior, Universitat de Girona, Girona, Spain
b
DEMEGI, Faculdade de Engenharia, Universidade do Porto, Rua Dr. Roberto Frias, 4200-465 Porto, Portugal
c
NASA Langley Research Center, Hampton, VA 23681, USA
Received 21 March 2006; received in revised form 9 December 2006
Abstract
A continuum damage model for the prediction of the onset and evolution of intralaminar failure mechanisms and the
collapse of structures manufactured in ber-reinforced plastic laminates is proposed. The failure mechanisms occurring in
the longitudinal and transverse directions of a ply are represented by a set of scalar damage variables. Crack closure eects
under load reversal are taken into account by using damage variables that are established as a function of the sign of the
components of the stress tensor. Damage activation functions based on the LaRC04 failure criteria are used to predict the
dierent failure mechanisms occurring at the ply level.
2007 Elsevier Ltd. All rights reserved.
Keywords: Fracture mechanics; Continuum damage mechanics; Composite materials
1. Introduction
The methodology for designing high-perfor-
mance structures of composite materials is still
evolving. The complexity of the response of com-
posite materials and the diculty in predicting
structural modes of failure result in the need for a
well-planned test program. The recommended prac-
tice to mitigate the technological risks associated
with such materials is to substantiate the perfor-
mance and durability of the design in a sequence
of steps known as the building block approach
(BBA) (MIL-HDBK-17-3F, 2002). The BBA
ensures that cost and performance objectives are
met by testing greater numbers of smaller less
expensive specimens, assessing technology risks
early in the program, and building on the knowl-
edge acquired at a given level of structural complex-
ity before progressing to a level of more complexity.
Achieving substantiation of structural perfor-
mance by testing alone can be prohibitively expen-
sive because of the number of specimens and
components required to characterize all loading
conditions. BBA programs can achieve signicant
cost reductions by seeking a synergy between testing
and analysis. The more the development relies on
analysis, the less expensive it becomes.
0167-6636/$ - see front matter 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mechmat.2007.03.005
*
Corresponding author. Tel.: +351 225081753; fax: +351
225081315.
E-mail address: pcamanho@fe.up.pt (P.P. Camanho).
Mechanics of Materials 39 (2007) 897908
www.elsevier.com/locate/mechmat
The use of advanced analytical or numerical
models for the prediction of the mechanical behav-
ior of composite structures can replace some of
the mechanical tests and can signicantly reduce
the cost of designing with composites while provid-
ing to the engineers the information necessary to
achieve an optimized design.
Strength-based failure criteria are commonly
used to predict failure in composite materials. A
large number of continuum-based criteria have been
derived to relate stresses and experimental measures
of material strength to the onset of failure (Soden
et al., 1998; Davila et al., 2005; Pinho et al., 2004).
Failure criteria predict the onset of the dierent
damage mechanisms occurring in composites and,
depending on the material, the geometry and the
loading conditions, may also predict nal structural
collapse.
For composite structures that can accumulate
damage before structural collapse, the use of failure
criteria is not sucient to predict ultimate failure.
Simplied models, such as the ply discount method,
can be used to predict ultimate failure, but they can-
not represent with satisfactory accuracy the quasi-
brittle failure of laminates that results from the
accumulation of several failure mechanisms.
The study of the non-linear response of quasi-
brittle materials due to the accumulation of damage
is important because the rate and direction of dam-
age propagation denes the damage tolerance of a
structure and its eventual collapse. To model the
phenomena of damage propagation, non-linear con-
stitutive models dened in the context of the
mechanics of continuum mediums have been devel-
oped and implemented in nite elements codes in
recent years. The formalism of the thermodynamics
of irreversible processes is a rigorous framework
from which the constitutive models can be
developed.
The simplest way to describe damage is using a
single scalar damage variable, as proposed by
Kachanov (1958). Damage can be interpreted as
the creation of microcavities, and the damage vari-
ables as a measure of the eective surface density
of the microdefects. Such a mechanical interpreta-
tion of damage assumes that the loads are resisted
only by the undamaged ligaments in the material.
The stresses ~ r in the ligaments, referred to as eec-
tive stresses, continue to increase until all ligaments
are severed and the material has failed.
The tensorial representation of damage is a for-
mal and general procedure to represent the direc-
tionality of microcracks, which can take any
direction in a medium depending on the load his-
tory, geometry, boundary conditions, and material
properties. After Kachanovs pioneering work, sev-
eral damage models have been developed that
describe damage as a second order tensor (Chab-
oche, 1995; Carol et al., 2001a,b; Luccioni and
Oller, 2003) or as a fourth order tensor (Ortiz,
1985; Simo and Ju, 1987a,b). Second order tensors
describe an initially isotropic material as an ortho-
tropic one when damage evolves, whereas fourth
order tensor models can remove all material symme-
tries and provide a more general procedure to simu-
late damage (Cauvin and Testa, 1999).
The application of continuum damage models in
orthotropic or transversely isotropic materials, such
as ber-reinforced plastics (FRP), results in addi-
tional diculties. The nature and morphology of
the material induces some preferred directions for
crack growth, i.e., crack orientations are not only
induced by the loads, geometry and boundary con-
ditions, but also by the morphology of the material.
The interface between ber and matrix is weaker
than the surrounding material and interfacial deb-
onding is normally the rst damage mechanism to
occur. Furthermore, residual thermal stresses occur
in the composite plies due to dierent coecients of
thermal expansion of the ber and matrix (micro-
mechanical residual thermal stresses) and due to
the dierent coecients of thermal expansion in
the longitudinal (ber) and transverse (matrix)
directions (macromechanical residual thermal
stresses).
Multiscale models and mesomodeling are two
approaches used to evaluate the elastic and inelastic
response of a material. Using homogenization laws,
multiscale models dene relations between a meso-
scale, normally the scale of the nite elements,
where material is considered homogeneous, and
the microscale, which is the scale of the ber and
matrix constituents. The constitutive models are
dened at the microscale, and the strain and stress
elds at the microscale and the mesoscale are related
via transformation eld tensors (Chaboche et al.,
2001; Fish and Yu, 2001; Fish et al., 1999; Voyiadjis
and Deliktas, 2000; Car et al., 2002), or solved using
nite elements (Car et al., 2002; Oller et al., 2005).
To reduce the amount of computations that need
to be performed, periodicity of the material is
invoked.
Mesomodeling is an alternative way to dene
damage models for composite materials that is more
898 P. Maim et al. / Mechanics of Materials 39 (2007) 897908
appropriate for large scale computations. Meso-
models treat the composite lamina (Ladeve`ze
et al., 2000; Barbero and Lonetti, 2001; Barbero
and Devivo, 2001; Matzenmiller et al., 1995; Oller
et al., 1995) or sub-laminate (Williams et al., 2003)
as a homogeneous material. When diuse damage
localizes in a narrow band and becomes a macro-
crack, the response is dominated by the crack tip
and its ability to dissipate energy. On the other
hand, the material morphology, which is the main
basis of homogenization techniques, loses impor-
tance due to the loss of periodicity. Therefore, in
structures exhibiting stable crack propagation, i.e.,
when the macrocrack length does not increase under
constant load, mesomodeling is more appropriate to
predict the structural collapse than multiscale
models.
The main objective of the present paper is to
develop a continuum damage model able to repre-
sent the quasi-brittle fracture of laminated compos-
ite structures, from the onset of non-critical failure
mechanisms up to nal structural collapse.
The majority of the material properties required
by the present model can be measured using stan-
dard test methods. Most of these properties can be
obtained from ply-based test methods. The use of
ply properties, rather than laminate properties, is
an advantage because it avoids the need to test lam-
inates every time the lay-up or stacking sequence is
modied.
The proposed constitutive model accounts for
crack closure eects under load reversal, an impor-
tant phenomenon in cases where a composite struc-
ture is subjected to multiaxial loading.
One important issue regarding the numerical
modeling of damage is that the convergence of the
solution through successive mesh renement must
be ensured. The objectivity of the numerical model
is ensured by adjusting the energy dissipated by each
failure mechanism using a characteristic element
length (Bazant and Oh, 1983). The constitutive
model proposed herein can be integrated explicitly,
making it computationally ecient and, therefore,
suitable to be used in large scale computations.
This paper is organized as follows: a brief
description of the failure mechanisms occurring in
laminated composites is presented. Based on the
mechanisms identied, a new constitutive model
which relates the failure mechanisms with a set of
internal variables is proposed. An accompanying
paper (Maim et al., in press) presents the details
of the computational implementation of the model
and the comparison between the model predictions
and experimental data.
2. Mechanisms of damage and fracture in laminated
composites
The model proposed in this work predicts
intralaminar failure mechanisms in laminated
composites: matrix cracking and ber fracture.
Intralaminar failure mechanisms trigger structural
collapse of unidirectional laminates almost immedi-
ately. In matrix-dominated failure modes, laminate
collapse occurs as soon as a matrix crack is created.
Failure of unidirectional laminates loaded in the
longitudinal (ber) direction results from the accu-
mulation of ber fractures (Rosen, 1964).
However, multidirectional laminates can sustain
increasing amounts of intralaminar failure mecha-
nisms before structural collapse occurs. Considering
the multidirectional laminate as a representative
volume element, the intralaminar failure mecha-
nisms can be regarded as damage mechanisms, i.e.
distributed microcracks in a laminate whose tangent
stiness tensor is positive denite. When the tangent
stiness tensor of the laminate ceases to be positive
denite, a macrocrack is formed and structural col-
lapse ensues.
The main characteristics of the intralaminar fail-
ure mechanisms occurring in laminated composites
are briey described in the following sections. These
characteristics are the basis for the denition of the
failure criteria and damage variables used in this
work.
2.1. Longitudinal failure
In ber-reinforced composites, the largest por-
tion of the loads is resisted by the bers. When these
fail under either tension or compression, the internal
loads must redistribute to other areas of the struc-
ture, and may cause a structural collapse.
In composites with high ber volume fraction
and those where the strain to failure of the resin
matrix is higher than the one of the reinforcing ber,
such as carbonepoxy composites, longitudinal fail-
ures start by isolated ber fractures in weak zones.
The localized fractures increase the normal and
interfacial shear stresses in the adjoining bers,
and the local stress concentrations promote matrix
cracking, ber matrix debonding and, for ductile
matrices, conical shear failures (Daniel and Ishai,
P. Maim et al. / Mechanics of Materials 39 (2007) 897908 899
1994). When increasing the load further, additional
ber fractures occur, leading to nal collapse.
Failure under longitudinal tensile loading occurs
in both constituents, and fracture occurs along a
plane whose normal is parallel to the ber direction.
A simple non-interacting failure criterion based on
maximum stress or maximum strain along the longi-
tudinal direction can usually provide an accurate
measure of longitudinal tensile failure (Davila
et al., 2005).
Compressive failure of aligned ber composites
occurs from the collapse of the bers as a result of
shear kinking and damage of the supporting matrix
(Fleck and Liu, 2001; Schultheisz and Waas, 1996).
The analysis of the kinking phenomena is usually
based on the assumption of a local initial ber mis-
alignment. Fiber misalignment causes shear stresses
between bers that rotate the bers, which in turn
increases the shear stress and leads to instability.
2.2. Transverse failure
Failure in the transverse direction encompasses
both matrix cracking and bermatrix debonding.
Under the presence of transverse tensile stresses
and in-plane shear stresses, the combined eect of
small defects present in a ply, such as small ber
resin debonds, resin-rich regions, and resin voids,
trigger a transverse crack that extends through-
the-thickness of the ply. The transverse cracks are
formed without disturbing the bers: they occur at
the berresin interface and in the resin.
When an unidirectional laminate is loaded
in shear, a non-linear shear stress-shear strain
response is observed before the laminate fails by
through-the-thickness transverse matrix cracking.
This non-linearity results from the visco-plastic
behavior of the matrix, and from the nucleation
and growth of microcracks. The non-linear shear
response which occurs before transverse matrix
cracking is not taken into account in the model
proposed here, and will be the subject of future
research. According to Chang et al. (1984), shear
non-linear eects should be taken into account in
the strength prediction of cross-ply and angle-ply
laminates.
A fundamental issue that needs to be considered
in the analysis models is the eect of ply thickness
on the ply strength, usually called the in situ eect.
As shown in Parvizi et al. (1978) and Chang and
Chens (1987) experiments, the constraints imposed
by the neighboring plies of dierent ber orienta-
tions cause an apparent increase in the tensile and
shear strengths of a ply compared to those of an
unconstrained ply. The in situ eect is a determinis-
tic size eect that can be represented using fracture
mechanics models of plies containing defects (Dav-
ila et al., 2005; Camanho et al., 2006; Dvorak and
Laws, 1987).
Experimental results have shown that moderate
values of transverse compression have a benecial
eect on the strength of a ply (Soden et al., 1998):
when the in-plane shear stress is large compared
to the transverse compressive stress, the fracture
plane is perpendicular to the mid-plane of the ply.
However, increasing the compressive transverse
stress causes a change in the angle of the fracture
plane. Normally, for carbonepoxy and glassepoxy
composites loaded in pure transverse compression,
the fracture plane is at an angle (fracture angle,
a
0
) of 53 3 with respect to the thickness direc-
tion (Puck and Schu rmann, 1998). Therefore,
matrix cracking does not occur in the plane of the
maximum transverse shear stress (45).
1
3
1
2 2
3
3 3
a) Longitudinal tensile fracture
b) Longitudinal compressive fracture
c) Transverse fracture with =0 d) Transverse fracture with =53
Fig. 1. Ply fracture planes considered in the model.
900 P. Maim et al. / Mechanics of Materials 39 (2007) 897908
Following the observations outlined above, four
possible fracture planes are considered in the model
proposed here, as schematically represented in
Fig. 1.
The purpose of present work is to propose a
model for the calculation of the initiation and prop-
agation of intralaminar failure mechanisms. Conse-
quently, delamination is not considered here.
3. Constitutive damage model
The thermodynamics of irreversible processes is a
general framework that can be used to formulate
constitutive equations. It is a logical framework
for incorporating observations and experimental
results and a set of rules for avoiding incompatibil-
ities. In this section, we present a constitutive dam-
age model for laminated composites that has its
foundation in irreversible thermodynamics, and that
uses the LaRC04 criteria as damage activation
functions.
3.1. Complementary free energy and damage
operator
To establish a constitutive law, it is possible to
dene a scalar function corresponding to the com-
plementary free energy density in the material. This
function must be positive denite, and it must be
zero at the origin with respect to the free variables
(the stresses) (Malvern, 1969). The proposed deni-
tion for the ply complementary free energy density
is
G
r
2
11
21 d
1
E
1

r
2
22
21 d
2
E
2

m
12
E
1
r
11
r
22

r
2
12
21 d
6
G
12
a
11
r
11
a
22
r
22
DT
b
11
r
11
b
22
r
22
DM 1
where E
1
, E
2
, m
12
and G
12
are the in-plane elastic
orthotropic properties of a unidirectional lamina.
The subscript 1 denotes the longitudinal (ber)
direction, and 2 denotes the transverse (matrix)
direction. a
11
and a
22
are the coecients of thermal
expansion in the longitudinal and transverse direc-
tions, respectively. b
11
and b
22
are the coecients
of hygroscopic expansion in the longitudinal and
transverse directions, respectively. DT and DM are
the dierences of temperature and moisture content
with respect to the corresponding reference values.
The stress tensor r corresponds to the average stress
tensor over a representative volume that is assumed
to be much larger than the diameter of a ber.
The set of scalar damage variables d
1
,d
2
, and d
6
ensure that the composite ply maintains the original
planes of material symmetry, regardless of the dam-
age state of the material. The damage variable d
1
is
associated with longitudinal (ber) failure, whereas
d
2
is the damage variable associated with transverse
matrix cracking and d
6
is a damage variable inu-
enced by longitudinal and transverse cracks.
To ensure the thermodynamically irreversibility
of the damage process, the rate of change of the
complementary free energy
_
G minus the externally
supplied work to the solid at constant strains,
_ r : e, must not be negative:
_
G _ r : e P0 2
This inequality corresponds to the positiveness of
the dissipated energy and has to be fullled by any
constitutive model (Malvern, 1969). Expanding the
inequality in terms of the stress tensor and damage
variables gives:
oG
or
e
_ _
: _ r
oG
od
1
_
d
1

oG
od
2
_
d
2

oG
od
6
_
d
6
P0 3
Since the stresses are variables that can vary freely,
the expression in the parenthesis must be equal to
zero to ensure positive dissipation of mechanical en-
ergy. Therefore, the strain tensor is equal to the
derivative of the complementary free energy density
with respect to the stress tensor:
e
oG
or
H : r aDT bDM 4
The lamina compliance tensor can be represented in
Voigt notation as:
H
o
2
G
or
2

1
1 d
1
E
1

m
21
E
2
0

m
12
E
1
1
1 d
2
E
2
0
0 0
1
1 d
6
G
12
_

_
_

_
5
The closure of transverse cracks under load reversal,
also known as the unilateral eect, is taken into ac-
count by dening four damage variables associated
with longitudinal and transverse damage. To deter-
mine the active damage variables, it is necessary to
dene the longitudinal and transverse damage
modes as follows:
P. Maim et al. / Mechanics of Materials 39 (2007) 897908 901
d
1
d
1
hr
11
i
jr
11
j
d
1
hr
11
i
jr
11
j
d
2
d
2
hr
22
i
jr
22
j
d
2
hr
22
i
jr
22
j
6
where hxi is the McCauley operator dened as
hxi := (x + jxj)/2.
The present model tracks damage caused by ten-
sion loads (d
+
) separately from damage caused by
compression loads (d

). Depending on the sign of

the corresponding normal stress, a damage mode
can be either active or passive.
The model also assumes that the shear damage
variable, d
6
, is not aected by the closure eect.
Shear damage is caused mainly by transverse cracks
and these do not close under shear stresses (r
12
).
Transverse cracks are inuenced by transverse stres-
ses (r
22
) producing the closure of cracks and a fric-
tion retention whereas longitudinal cracks produce
the same eect under longitudinal stresses (r
11
)
(Chaboche and Maire, 2002). The eect of friction
is neglected in the present model.
3.2. Damage activation functions
The determination of the domain of elastic
response under complex stress states is an essential
component of an accurate damage model. Based on
the previously described mechanisms of crack gen-
eration in advanced composites, a strain space is
considered where the material is linear elastic. In
the present model, it is assumed that the elastic
domain is enclosed by four surfaces, each of them
accounting for one failure mechanism: longitudinal
and transverse fracture under tension and compres-
sion. Those surfaces are formulated by the damage
activation functions based on the LaRC03 and
LaRC04 failure criteria. The LaRC03-04 failure
criteria have been shown to represent accurately
the physical process of failure onset in laminated
composites. The LaRC04 criteria represent an evo-
lution of the LaRC03 criteria: some criteria such as
the one for ber kinking, are more accurate in
LaRC04. However, the improvement in accuracy
is associated with a signicant increase in the com-
putational eort. The present damage model uses a
combination of both sets of criteria to achieve a
compromise between accuracy and computational
eciency. The full details of the derivation and val-
idation of the LaRC04 failure criteria are presented
in references (Davila et al., 2005; Pinho et al.,
2004).
The four damage activation functions, F
N
, asso-
ciated with failure mechanisms in the longitudinal
(N = 1+, 1) and transverse (N = 2+, 2) direc-
tions, shown in Fig. 1, are dened as
F
1
/
1
r
1
6 0; F
1
/
1
r
1
6 0
F
2
/
2
r
2
6 0; F
2
/
2
r
2
6 0
7
where the loading functions /
N
(N = 1+, 1, 2+,
2) depend on the strain tensor and material con-
stants (elastic and strength properties). The elastic
domain thresholds r
N
(N = 1+, 1, 2+, 2) take
an initial value of 1 when the material is undam-
aged, and they increase with damage. The elastic do-
main thresholds are internal variables of the
constitutive model, and are related to the damage
variables d
M
(M = 1+, 1, 2+, 2, 6) by the dam-
age evolution laws. The elastic domain threshold de-
nes the level of elastic strains that can be attained
before the accumulation of additional damage.
3.2.1. Longitudinal tensile fracture
The LaRC04 criterion for ber tensile failure,
Fig. 1a, is a non-interacting maximum allowable
strain criterion dened as
/
1

E
1
X
T
e
11

~ r
11
m
12
~ r
22
X
T
8
where the eective stress tensor ~ r is computed as
~ r H
1
0
: e. H
0
is the undamaged compliance tensor
obtained from Eq. (5) using d
1
= d
2
= d
6
= 0.
3.2.2. Longitudinal compressive fracture
The LaRC03 failure criterion for longitudinal
compressive fracture, Fig. 1b, postulates that a kink
band is triggered by the onset of damage in the sup-
porting matrix. Under this circumstance, the bers
lose lateral support and fail under the eect of lon-
gitudinal compressive stresses. The initial ber mis-
alignment and the rotation of the bers as a
function of the applied stress state are the parame-
ters used in the damage activation function.
The damage activation function used to predict
damage under longitudinal compression ~ r
11
< 0
and in-plane shear (ber kinking) is established as
a function of the components of the stress tensor
~ r
m
in a coordinate system (m) representing the ber
misalignment:
/
1

hj ~ r
m
12
j g
L
~ r
m
22
i
S
L
9
where the longitudinal friction coecient can be
approximated as (Davila et al., 2005).
902 P. Maim et al. / Mechanics of Materials 39 (2007) 897908
g
L
%
S
L
cos2a
0

Y
C
cos
2
a
0
10
The components of the eective stress tensor in the
coordinate system associated with the rotation of
the bers are calculated as
~ r
m
22
~ r
11
sin
2
u
C
~ r
22
cos
2
u
C
2j~ r
12
j sin u
C
cos u
C
~ r
m
12
~ r
22
~ r
11
sinu
C
cos u
C
j~ r
12
jcos
2
u
C
sin
2
u
C

11
where the absolute value of the shear stress is taken
because the misalignment angle can be positive or
negative.
The misalignment angle (u
C
) is determined using
standard shear and longitudinal compression
strengths, S
L
and X
C
(Davila et al., 2005):
u
C
arctan
1

1 4
S
L
X
C
g
L
_ _
S
L
X
C
_
2
S
L
X
C
g
L
_ _
_
_
_
_
_
_
_
_
12
It should be noted that the LaRC03 failure crite-
rion, derived to predict the onset of damage in lam-
inated composites, calculates the misalignment
angle as a function of the applied stress. The
LaRC03 failure criterion is modied here assuming
a constant misalignment angle, corresponding to the
rotation of the bers at failure under pure longitudi-
nal compression. This modication of LaRC03 fail-
ure criterion ensures that /
1
is a monotonic
increasing function under any state of proportional
loading.
It should be pointed out that two criteria are used
in LaRC03 for ber kinking: Eq. (9) for ~ r
m
22
6 0 and
a second equation for ~ r
m
22
P0. However, the omis-
sion of the second equation results in a minor loss
of accuracy because the equation is the same as
Eq. (13.a) with the stresses transformed into the
misaligned coordinate frame.
3.2.3. Transverse fracture perpendicular to the
laminate mid-plane (a
0
= 0)
Transverse matrix cracks perpendicular to the
mid-plane of the ply, i.e. with a
0
= 0 (Fig. 1c),
are created by a combination of in-plane shear stres-
ses and transverse tensile stresses, or in-plane shear
stresses and small transverse compressive stresses.
These conditions are represented by the following
LaRC04 failure criterion:
/
2

1 g
~ r
22
Y
T
g
~ r
22
Y
T
_ _
2

~ r
12
S
L
_ _
2
_
if ~ r
22
P0
1
S
L
hj ~ r
12
j g
L
~ r
22
i if ~ r
22
< 0
_

_
13
where g is the fracture toughness ratio dened as:
g
G
Ic
G
IIc
.
3.2.4. Transverse fracture with a
0
= 53
The LaRC04 matrix failure criterion for high
transverse compressive stresses, Fig. 1d, consists of
a quadratic interaction between the eective shear
stresses acting on the fracture plane:
/
2

~s
T
eff
S
T
_ _
2

~s
L
eff
S
L
_ _
2

if ~ r
22
< 0 14
where the eective stresses ~s
T
eff
and ~s
L
eff
are computed
as (Pinho et al., 2004)
~s
T
eff
h~ r
22
cosa
0
sina
0
g
T
cosa
0
coshi
~s
L
eff
hcosa
0
j~ r
12
j g
L
~ r
22
cosa
0
sinhi
15
The sliding angle h is calculated as (Pinho et al.,
2004)
h arctan
j~ r
12
j
~ r
22
sina
0

_ _
16
The transverse shear strength and transverse friction
coecient can be approximated as
S
T
Y
C
cosa
0
sina
0

cosa
0

tan2a
0

_ _
g
T

1
tan2a
0

17
The fracture angle a
0
is approximately 53 in uniax-
ial compression. With increasing amounts of in-
plane shear, the fracture angle diminishes up to
about 40 and then abruptly switches to 0. To nd
the correct angle of fracture, a maximization of the
LaRC03-04 failure criteria as a function of a should
be performed. However, in order to improve the
computational eciency of the present model, it is
assumed that the fracture angle can only take one
of two discrete values: 0 or 53.
The elastic domain in the ~ r
11
, ~ r
22
, ~ r
12
space rep-
resented by the LaRC04 failure criteria is shown in
Fig. 2.
3.3. Dissipation
The rate of energy dissipation per unit volume
resulting from the evolution of damage is given by
P. Maim et al. / Mechanics of Materials 39 (2007) 897908 903
N
oG
od
1
_
d
1

oG
od
2
_
d
2

oG
od
6
_
d
6
Y
1
_
d
1
Y
2
_
d
2
Y
6
_
d
6
P0 18
The form of the complementary free energy dened
in Eq. (1) ensures that the thermodynamic forces
(Y
M
) conjugated to their respective damage vari-
ables (d
M
) are always positive:
Y
1

oG
od
1

r
2
11
21 d
1

2
E
1
P0
Y
2

oG
od
2

r
2
22
21 d
2

2
E
2
P0
Y
6

oG
od
6

r
2
12
21 d
6

2
G
12
P0
19
Therefore, the condition of positive evolution of
damage variables
_
d
M
P0 is a sucient condition
for the fulllment of the second law of
thermodynamics.
It is important to note that the proposed model
does not generate spurious energy dissipation, i.e.,
the loss or gain of mechanical energy, under crack
closure or opening. At load reversal, the time deriv-
ative of the damage variable is non-zero
_
d
M
6 0.
Considering Eq. (18), the thermodynamical forces,
Y
M
, associated with the damage variables, d
M
, must
be zero to avoid spurious energy dissipation at load
reversal (Carol and Willam, 1996). This condition is
trivially satised in the present model (Eq. (19)).
Damage evolution without energy dissipation is
physically inadmissible. Therefore, it is necessary
to avoid damage evolution when the corresponding
conjugated thermodynamic force is zero. Consider
the load history represented in Fig. 3: the material
is loaded in transverse tension and shear to t
1
and
then loaded to t
2
. At time t
2
, the damage variable
d
2
evolves because the corresponding damage acti-
vation function is activated. However, the corre-
sponding thermodynamic force is zero (r
22
= 0,
Y
2
= 0).
This non-physical response is avoided by modify-
ing the longitudinal and transverse damage activa-
tion functions. The transverse damage activation
function is modied using the following equation:
/
2
min

~s
T
eff
S
T
_ _
2

~s
L
eff
S
L
_ _
2

;
~ r
22
X
_
_
_
_
_
_
20
where the constant X is equal to ~ r
22
when
/
2
= /
2+
(Fig. 3).
The longitudinal damage activation function is
modied by taking into account that under shear
dominated loads, matrix cracking is the rst form
Fig. 2. Elastic domain in the ~ r
11
, ~ r
22
, ~ r
12
space.
Fig. 3. Evolution of elastic domain in the ~ r
22
~ r
12
space.
904 P. Maim et al. / Mechanics of Materials 39 (2007) 897908
of damage to occur. After matrix cracking, the
transverse and shear stresses are zero and the ber
misalignment angle u tends to p/4. Under these cir-
cumstances, the kink band criteria, Eq. (9), reads
/
1
min
hj ~ r
m
12
j g
L
~ r
m
22
i
S
L
;
g
L
1
2S
L
~ r
11
_ _
21
3.4. Damage evolution
The evolution of the damage threshold values r
N
is expressed by the KuhnTucker conditions:
_ r
N
P0; F
N
6 0; _ r
N
F
N
0 22
Neglecting viscous eects, the damage activation
functions, Eqs. (7), always have to be non-positive.
While the damage activation function F
N
is nega-
tive, the material response is elastic. When the strain
state activates a criterion, F
N
= 0, it is necessary to
evaluate the gradient
_
/
N
. If the gradient is not posi-
tive, the state is one of unloading or neutral loading.
If the gradient
_
/
N
is positive, there is damage evolu-
tion, and the consistency condition has to be
satised:
_
F
N

_
/
N
_ r
N
0 23
Two important characteristics of the model pro-
posed here are that the damage threshold values
are a function of the damage variables, and that
the loading functions depend on the strain tensor.
Under these conditions, it is possible to integrate
the constitutive model explicitly (Simo and Ju,
1987a,b).
In the denition of the constitutive model, it is
necessary to represent the relation between active
and inactive elastic domains. The evolution of an
active elastic domain is dened by the consistency
condition, i.e., it is dened with respect to the corre-
sponding damage activation function. However, it is
also necessary to specify how the inactive elastic
domain evolves if other damage modes are active.
It is assumed that the longitudinal and transverse
domains are not coupled. On the other hand, com-
pression damage is coupled with tension damage, as
explained in the next section.
3.4.1. Transverse loading
As described previously, transverse damage in
the form of matrix cracks can have dierent orienta-
tions as a result of tension, shear, or compression-
dominated loads. Under load reversal, transverse
cracks, which are perpendicular to the ply mid-
plane, do not aect the compression response: the
elastic domain and the compressive damage variable
(d
2
) are unaected by r
2+
.
On the other hand, matrix cracks at a fracture
angle a
0
= 53 caused by high compressive transverse
stresses have the same eect as cracks perpendicular
to the mid-plane (a = 0) when the load is reversed
from compression to tension. Therefore, the evolu-
tion of the transverse tensile elastic domain threshold
(r
2+
) is governed by both damage mechanisms.
Based on the above considerations, the evolution
of the elastic domain in the transverse direction can
be represented by the following equations:
Tension loading: _ r
2

_
/
2
and _ r
2
0
Compression loading: _ r
2

_
/
2
and
_ r
2

_
/
2
if r
2
6 r
2
0 if r
2
> r
2
_
The integration of the previous expressions results in
r
2
max 1; max
s0;t
f/
s
2
g; max
s0;t
f/
s
2
g
_ _
r
2
max 1; max
s0;t
f/
s
2
g
_ _ 24
3.4.2. Longitudinal loading
Under longitudinal tensile stresses, the fracture
plane is perpendicular to the ber direction. When
reversing the load, the cracks close and can still
transfer load. However, the broken and misaligned
bers do not carry any additional load. Therefore,
the compressive stiness is inuenced by longitudi-
nal damage. However, the elastic domain is
assumed to remain unchanged.
Under longitudinal compression, damaged mate-
rial consisting of broken bers and matrix cracks
forms a kink band, and there is not a unique orien-
tation for the damage planes. When the loads are
reversed, the cracks generated in compression open
and the elastic domain threshold increases.
Therefore, the evolution of damage thresholds
for longitudinal damage are dened as
Tension loading: _ r
1

_
/
1
and _ r
1
0
Compression loading: _ r
1

_
/
1
and
_ r
1

_
/
1
if r
1
6 r
1
0 if r
1
> r
1
_
25
P. Maim et al. / Mechanics of Materials 39 (2007) 897908 905
The integration of the previous expressions results in
r
1
max 1; max
s0;t
f/
s
1
g; max
s0;t
f/
s
1
g
_ _
r
1
max 1; max
s0;t
f/
s
1
g
_ _ 26
3.5. Damage evolution laws
The internal variables r
N
dene the threshold of
the elastic domains, and are related to the damage
state of each lamina, i.e., the damage variables
depend on the values of the internal variables. In
order to fully dene the constitutive model, it is nec-
essary to dene the relation between the internal
variables and the damage variables.
When the material is undamaged the internal
variables r
N
take the initial value of 1, and
d
N
(r
N
= 1) = 0. Eqs. (24) and (26) dene the evolu-
tion of the internal variables ensuring that _ r
N
P0.
As shown in Eqs. (18) and (19), the condition for
positive dissipation is satised if
_
d
N
P0. The condi-
tion for positive dissipation is automatically fullled
if the damage evolution law satises the condition
od
N
/or
N
P0. When the material is completely dam-
aged, a fracture plane is created, the strains are
localized in a plane in which r
N
! 1 and the related
components of the stiness tensor are zero,
d
N
(r
N
! 1) = 1.
Matrix cracks are related to the internal variables
r
2+
and r
2
. The internal variable r
2
accounts for
compressive damage only, whereas r
2+
accounts
for both, compressive and tensile, failure mecha-
nisms. Therefore, for tensile transverse stresses dam-
age is a function of d
2+
(r
2+
) because both types of
cracks (a = 0 and a = 53) are open. Under trans-
verse compressive loads, the damage is only inu-
enced by the inclined cracks, d
2
(r
2
).
Kink bands are related to the internal variable
r
1
. The internal variable r
1+
accounts for kink
bands and ber tensile fracture. For tensile longitu-
dinal stresses, the material loses stiness as a result
of both damage modes because the cracks open.
Therefore, the damage variable can be expressed
as d
1+
(r
1+
).
When a lamina which is fully damaged in tension
(d
1+
= 1) is subjected to load reversal and the crack
closes, some of the original stiness is recovered
because the tractions can be transmitted through
the crack faces. However, the broken bers lose
their alignment. Assuming that the bers do not
carry any load, which can be considered as a limit
case, the compressive stiness can be approximated
by the rule of mixtures applied for components in
parallel as: (1 d
1
)E
1
= V
m
E
m
. The generalization
of the above arguments for an intermediate damage
state can be expressed as d
1
A

1
d
1
, with
A

1
% b
V
f
E
f
V
m
E
m
V
f
E
f
% b
E
1
E
2
E
1
27
where E
f
and E
m
are the ber and matrix Young
modulus, V
f
and V
m
the corresponding volume frac-
tions, and b is an adjustment parameter between 0
and 1. If b = 1, the stiness recovery is due only
to matrix closure, and if b = 0, the stiness recovery
is total and it is assumed that broken bers do not
lose alignment under compressive loads, and the ini-
tial stiness is recovered.
The damage law for longitudinal compression
can be expressed as a combination of the failure
mechanisms caused by tension loads, d
1+
(r
1+
), and
the failure mechanisms generated under compres-
sion, d
1
(r
1
):
d
1
1 1 d

1
r
1
1 A

1
d
1
r
1
28
Longitudinal failure (d
1
) is not inuenced by
matrix cracking (r
2
) as shown in experimental
results carried by Carlsson and Pipes (1987) and in
micromechanical models (Dvorak et al., 1985; Laws
et al., 1983; Nuismer and Tan, 1988; Tan and
Nuismer, 1989) of cracked composites. Therefore,
the longitudinal stiness is not function of trans-
verse matrix cracks.
The shear stiness is reduced as a result of longi-
tudinal and transverse cracks. Under these circum-
stances, the damage variable d
6
is given by
d
6
1 1 d

6
r
2
1 d
1
29
The damage evolution laws used force strain-
softening as soon as one damage activation criterion
is satised. Softening constitutive equations may
result in physically inadmissable responses: the
damage is localized in a plane and fracture occurs
without energy dissipation. The numerical imple-
mentation of softening constitutive equations using
the nite element method may result in mesh-depen-
dent results because the energy dissipated is a func-
tion of the element size.
The solution normally used to ensure the correct
computation of the energy dissipated regardless of
the renement of the mesh is to adjust the damage
evolution laws using a characteristic dimension of
the nite element. The denition of the damage evo-
lution laws is therefore related to the computational
906 P. Maim et al. / Mechanics of Materials 39 (2007) 897908
model, and it is described in the accompanying
paper (Maim et al., in press).
4. Conclusions
A new constitutive model for the prediction of
the onset and growth of intralaminar failure mech-
anisms in composite laminates under plane stress
was proposed. The model is based on four possible
ply fracture planes related to ber tensile fracture,
ber kinking, and matrix cracking with fracture
planes oriented at 0 and 53 with respect to the
ply thickness direction.
The onset of the intralaminar failure mechanisms
related to the four fracture planes is predicted using
a simplication of the LaRC04 failure criteria. The
eect of ply thickness on the onset of transverse
matrix cracking is taken into account by replacing
the unidirectional strengths by the in situ strengths
in the failure criteria.
The unilaterality of damage is taken into account
by the ability of the model to represent complex
load histories, including tensioncompression load
reversals. The computational implementation of
the constitutive model and its validation are
described in part two of this work (Maim et al.,
in press).
Acknowledgements
The research visits of the rst author at the Uni-
versity of Porto, Portugal, were funded by the
University of Girona (BE-UdG-2004) and by the
Spanish government under Acciones Integradas
Hispano-Portuguesas (HIP2004-0031). The rst
and third authors also acknowledge funding from
the Spanish government through DGICYT under
the contract MAT 2003-09768-C03-01.
The nancial support of the Portuguese Founda-
tion for Science and Technology (FCT) under the
project PDCTE/50354/EME/2003 is acknowledged
by the second author.
References
Barbero, E.J., Devivo, L., 2001. A constitutive model for elastic
damage in ber-reinforced PMC laminae. J. Damage Mech.
10 (1), 7393.
Barbero, E.J., Lonetti, P., 2001. A damage model for composites
dened in terms of available data. J. Mech. Compos. Mater.
Struct. 8 (4), 299316.
Bazant, Z.P., Oh, B.H., 1983. Crack band theory for fracture of
concrete. Mater. Struct. 16, 155177.
Camanho, P.P., Davila, C.G., Pinho, S.T., Iannucci, L., Robin-
son, P., 2006. Prediction of in situ strengths and matrix
cracking in composites under transverse tension and in-plane
shear. Composites Part A 37, 164176.
Car, E., Zalamea, F., Oller, S., Miquel, J., On ate, E., 2002.
Numerical simulation of ber reinforced composite materials
two procedures. Int. J. Solids Struct. 39, 19671986.
Carlsson, L.A., Pipes, R.B., 1987. Experimental Characterization
of Composite Materials. Prentice-Hall.
Carol, I., Willam, K., 1996. Spurious dissipation/generation in
stiness recovery models for elastic degradation and damage.
Int. J. Solids Struct. 33, 29392957.
Carol, I., Rizzi, E., Willam, K., 2001a. On the formulation of
anisotropic elastic degradation. I. Theory based on a pseudo-
logarithmic damage tensor rate. Int. J. Solids Struct. 38, 491
518.
Carol, I., Rizzi, E., Willam, K., 2001b. On the formulation of
anisotropic elastic degradation. II. Generalized pseudo-Ran-
kine model for tensile damage. Int. J. Solids Struct. 38, 519
546.
Cauvin, A., Testa, R.B., 1999. Damage mechanics: basic
variables in continuum theories. Int. J. Solids Struct. 36,
747761.
Chaboche, J.-L., 1995. A continuum damage theory with
anisotropic and unilateral damage. Recher. Aerospat. 2,
139147.
Chaboche, J.-L., Maire, J.F., 2002. A new micromechanics based
CDM model and its application to CMCs. Aerospace Sci.
Technol. 6, 131145.
Chaboche, J.-L., Kruch, S., Maire, J.F., Pottier, T., 2001.
Towards a micromechanics based inelastic and damage
modeling of composites. Int. J. Plast. 17, 411439.
Chang, F.K., Chen, M.H., 1987. The in situ ply shear strength
distributions in graphite/epoxy laminated composites. J.
Compos. Mater. 21, 708733.
Chang, F.K., Scott, R.A., Springer, G.S., 1984. Failure strength
of nonlinearly elastic composite laminates containing a pin
loaded hole. J. Compos. Mater. 18, 464477.
Daniel, I.M., Ishai, O., 1994. Engineering Mechanics of Com-
posite Materials. Oxford University Press.
Davila, C.G., Camanho, P.P., Rose, C.A., 2005. Failure criteria
for FRP laminates. J. Compos. Mater. 39, 323345.
Dvorak, G.J., Laws, N., 1987. Analysis of progressive matrix
cracking in composite laminates II. First ply failure. J.
Compos. Mater. 21, 309329.
Dvorak, G.J., Laws, N., Hejazi, M., 1985. Analysis of progressive
matrix cracking in composite laminates I. Thermoelastic
properties of a ply with cracks. J. Compos. Mater. 19, 216
234.
Fish, J., Yu, Q., 2001. Two-scale damage modeling of brittle
composites. Compos. Sci. Technol. 61, 22152222.
Fish, J., Yu, Q., Shek, K., 1999. Computational damage
mechanics for composite materials based on mathematical
homogenization. Int. J. Numer. Methods Eng. 45,
16571679.
Fleck, N.A., Liu, D., 2001. Microbuckle initiation from a patch
of large amplitude bre waviness in a composite under
compression and bending. Eur. J. Mech. A/Solids 20 (1),
2337.
Kachanov, L.M., 1958. Time of the rupture process under
creep conditions. Isv. Nauk SSSR, Otd. Tehk. Nauk 8,
2631.
P. Maim et al. / Mechanics of Materials 39 (2007) 897908 907
Ladeve`ze, P., Allix, O., Deu , J.-F., Leveque, D., 2000. A
mesomodel for localisation and damage computation in
laminates. Comput. Methods Appl. Mech. Eng. 183, 105122.
Laws, N., Dvorak, G.J., Hejazi, M., 1983. Stiness changes in
unidirectional composites caused by crack systems. Mech.
Mater. 2.
Luccioni, B., Oller, S., 2003. A directional damage model.
Comput. Methods Appl. Mech. Eng. 192, 11191145.
Maim , P., Camanho, P.P., Mayugo, J.A., Davila, C.G., in press.
A continuum damage model for composite laminates: part
IIcomputational implementation and validation. Mech.
Mater. doi:10.1016/j.mechmat.2007.03.006.
Malvern, L.E., 1969. Introduction to the Mechanics of a
Continuous Medium. Prentice-Hall, Englewood Clis.
Matzenmiller, A., Lubliner, J., Taylor, R.L., 1995. A constitutive
model for anisotropic damage in ber-composites. Mech.
Mater. 20, 125152.
MIL-HDBK-17-3F, 2002. Military Handbook, Polymer Matrix
Composites. US Department of Defense.
Nuismer, R.J., Tan, S.C., 1988. Constitutive relations of a
cracked composite lamina. J. Compos. Mater. 22, 306321.
Oller, S., Botello, S., Miquel, J., On ate, E., 1995. An anisotropic
elastoplastic model based on an isotropic formulation. Eng.
Comput. 12, 245262.
Oller, S., Miquel Canet, J., Zalamea, F., 2005. Composite
material behavior using a homogenization double scale
method. J. Eng. Mech. 31 (1), 6579.
Ortiz, M., 1985. A constitutive theory for inelastic behaviour of
concrete. Mech. Mater. 4, 6793.
Parvizi, A., Garrett, K., Bailey, J., 1978. Constrained cracking in
glass bre-reinforced epoxy cross-ply laminates. J. Mater. Sci.
13, 195201.
Pinho, S.T., Davila, C.G., Camanho, P.P., Ianucci, L., Robinson,
P., 2004. Failure models and criteria for FRP under in-plane
or three-dimensional stress states including shear non-linear-
ity. NASA/TM-2003-213530.
Puck, A., Schu rmann, H., 1998. Failure analysis of FRP
laminates by means of physically based phenomenological
models. Compos. Sci. Technol. 58, 10451067.
Rosen, B.W., 1964. Tensile failure of brous composites. AIAA
J. 2, 19851991.
Schultheisz, C.R., Waas, A.M., 1996. Compressive failure of
composites, part 1: testing and micromechanical theories.
Progr. Aerospace Sci. 32, 142.
Simo, J.C., Ju, J.W., 1987a. Strain and stress-based continuum
damage models-I. Formulat. Int. J. Solids Struct. 23 (23),
821840.
Simo, J.C., Ju, J.W., 1987b. Strain and stress-based continuum
damage models-II. Comput. Aspects Int. J. Solids Struct. 23
(23), 841869.
Soden, P.D., Hinton, M.J., Kaddour, A.S., 1998. A comparison
of the predictive capabilities of current failure theories for
composite laminates. Compos. Sci. Technol. 58 (7), 1225
1254.
Tan, S.C., Nuismer, R.J., 1989. A theory for progressive matrix
cracking in composite laminates. J. Compos. Mater. 23, 1029
1047.
Voyiadjis, G.Z., Deliktas, B., 2000. A coupled anisotropic
damage model for the inelastic response of composite
materials. Comput. Methods Appl. Mech. Eng. 183, 159199.
Williams, K.V., Vaziri, R., Poursartip, A., 2003. A physically
based continuum damage mechanics model for thin laminated
composite structures. Int. J. Solids Struct. 40, 22672300.
908 P. Maim et al. / Mechanics of Materials 39 (2007) 897908