1, E
ij
reduces to the innitesimal strain tensor e
ij
given by
e
ij
=
1
2
@u
i
@x
j
@u
j
@x
i
_ _
=
1
2
u
i;j
u
j;i
_ _
: (2:4)
From Eq. (2.4) it is seen that the strain tensor is symmetric. Thus, there are six
independent strain components, which in the innitesimal version are three normal
strains (e
11
, e
22
, and e
33
), and three shear strains (e
12
= e
21
, e
23
= e
32
, and e
13
= e
31
).
To ensure singlevalued displacements u
i
, the strain components e
ij
cannot be
assigned arbitrarily but must satisfy certain integrability or compatibility condi
tions, given by
e
ij;kl
e
kl;ij
e
ik;jl
e
jl;ik
= 0: (2:5)
Of the 81 equations included in Eq. (2.5), only six are independent. The remainder
are either identities or repetitions due to symmetry of e
ij
. For the special case of
plane stress conditions, the only surviving compatibility equation is
e
11;22
e
22;11
2e
12;12
= 0: (2:6)
2.1.2 Conservation of linear and angular momenta
Ingeneral, the forces exertedona continuumbody are body forces andsurface forces.
Body forces, such as gravitational and magnetic forces, act on all particles within the
volume of the body and are described in terms of force intensity per unit mass or per
unit volume, while surface forces are contact forces that act across an internal surface
or an external (bounding) surface. The continuum description of surface forces is
given by the traction vector t acting on a surface element dS with a unit normal n (see
Figure 2.2(a)). Let dP be the total force exerted on dS by the material points on the
side of dS toward which n is pointing. The traction vector t is then dened as
t = lim
dS0
dP
dS
: (2:7)
At an internal point P there are innitely many surface elements, each with a different
unit normal vector. According tothe Cauchy theorema tractionvector onany of these
Initial configuration
Deformed configuration
X
x
u
X
2
X
1
O
P
P
X
3
Figure 2.1. Initial and deformed geometry of a continuum body.
10 Review of mechanics of composite materials
planes can be expressed in terms of the traction vectors on three orthogonal planes
passingthroughthe point P. InaCartesianreference frame, the three planes are chosen
parallel to the coordinate planes and the resultant traction vectors on these planes are
decomposed along the three coordinate axes. These 3 3 = 9 components taken
together form the components of the second rank stress tensor associated with the
considered point P. They are indicated in Figure 2.2(b) where their positive directions
are shown. Inindex notation, they are denoted by s
ij
, where the rst index refers to the
directionof the unit normal onthe surface (the face of the cube inFigure 2.2(b)) andthe
second index stands for the direction of the resolved traction component. The stress
components withtwoequal indices, e.g., s
11
, are callednormal stresses while those with
unequal indices, e.g., s
23
, are termed shear stresses.
The traction vector components are related to the stress tensor components by
the following equation
t
i
= s
ij
n
j
; (2:8)
where n
j
are components of the normal vector associated with the traction vector.
The conservation of linear momentum at a material point inside the continuum
body gives the following relation
s
ji; j
f
i
= r u
i
; (2:9)
where f
i
are components of the body force vector, and r is the mass density. For
quasistatic problems the righthand side of Eq. (2.9) vanishes, and if the body
forces are neglected, the equations of equilibrium reduce to
s
ji; j
= 0: (2:10)
When there are no body moments, the conservation of angular momentum results
in the symmetry of the stress tensor, i.e.,
s
ij
= s
ji
: (2:11)
2.1.3 Constitutive relations
For an elastic material there exists a positivedenite, singlevalued, potential
function of strains e
kl
, dened as
s
22
s
23
s
21
s
13
s
11
s
32
s
31
P
n
t
dS
(a) (b)
s
33
s
12
Figure 2.2. (a) Traction vector; (b) a volume element with components of stress tensor.
11 2.1 Equations of elasticity
U =
_
e
kl
0
s
ij
de
ij
: (2:12)
This function is termed as the strain energy density. U is independent of the
loading path and thus a function of nal strains only. Differentiating Eq. (2.12)
with respect to the strains, the stress tensor can be written as
s
ij
=
@U
@e
ij
: (2:13)
If we consider a linear elastic material, then U can be written as a quadratic
function in e
kl
U e
kl
( ) =
1
2
C
ijkl
e
ij
e
kl
; (2:14)
where C
ijkl
is a fourthorder tensor of material stiffness coefcients known as
the stiffness tensor. Using Eqs. (2.13) and (2.14), one obtains the generalized
Hookes law
s
ij
= C
ijkl
e
kl
: (2:15)
A potential function of stresses known as the complementary energy density is
dened as
U
+
s
ij
_ _
= s
ij
e
ij
U: (2:16)
Differentiation of Eq. (2.16) with respect to stress tensor yields the relation
e
ij
=
@U
+
@s
ij
: (2:17)
Analogous to Eq. (2.14), U
*
can also be represented as a quadratic function as
U
+
s
ij
_ _
=
1
2
S
ijkl
s
ij
s
kl
; (2:18)
where S
ijkl
are components of the compliance tensor. Using Eqs. (2.17) and (2.18),
one obtains the inverse constitutive law
e
ij
= S
ijkl
s
kl
: (2:19)
In all, the stiffness matrix C
ijkl
has 81 coefcients. However, not all of these
coefcients are independent. Note rst that the symmetry of the strain compon
ents (e
kl
= e
lk
) leads to C
ijkl
= C
ijlk
, which reduces the number of coefcients from
81 to 54. Similarly, the symmetry of the stress tensor further reduces the number of
these coefcients to 36. Finally, differentiating Eq. (2.14) twice with respect to
strains, one obtains
C
ijkl
=
@
2
U
@e
ij
@e
kl
: (2:20)
12 Review of mechanics of composite materials
Since the order of differentiation in the above equation is arbitrary, one infers
that
C
ijkl
= C
klij
; (2:21)
which reduces the number of independent material coefcients to 21.
The coefcient matrix C
ijkl
is expressed in compact form by using the Voigt
notation, in which stress and strain tensor components are denoted using a single
subscript, whereas two subscripts are used to denote the stiffness tensor. With this,
the constitutive relation, Eq. (2.15), can be written as s
p
= C
pq
e
q
; p, q = 1,2, . . ., 6,
or in expanded matrix form as
s
1
s
2
s
3
s
4
s
5
s
6
_
_
_
_
=
C
11
C
12
C
13
C
14
C
15
C
16
C
21
C
22
C
23
C
24
C
25
C
26
C
31
C
32
C
33
C
34
C
35
C
36
C
41
C
42
C
43
C
44
C
45
C
46
C
51
C
52
C
53
C
54
C
55
C
56
C
61
C
62
C
63
C
64
C
65
C
66
_
_
_
_
e
1
e
2
e
3
e
4
e
5
e
6
_
_
_
_
; (2:22)
where C
pq
= C
qp
, and
s
1
= s
11
; s
2
= s
22
; s
3
= s
33
; s
4
= s
23
; s
5
= s
31
; s
6
= s
12
e
1
= e
11
; e
2
= e
22
; e
3
= e
33
; e
4
= 2e
23
; e
5
= 2e
31
; e
6
= 2e
12
C
11
= C
1111
; C
22
= C
2222
; : : : ; etc:
(2:23)
These constitutive relationships are for an anisotropic material. If material sym
metry exists, then further reduction occurs in the number of independent coef
cients of the stiffness matrix. It should be noted that the stiffness matrix in the
Voigt notation does not follow the transformation rule for tensors. The fourth
order stiffness tensor C
ijkl
transforms as
C
/
ijkl
=
ip
jq
kr
ls
C
pqrs
; (2:24)
where
ij
is the matrix of direction cosines associated with coordinate trans
formation from one coordinate system (x
1
, x
2
, x
3
) to another (x
/
1
; x
/
2
; x
/
3
).
A material with one plane of symmetry is called monoclinic, and if this plane is
parallel to the x
1
x
2
plane then it can be shown that the constitutive relation is
given by
s
1
s
2
s
3
s
4
s
5
s
6
_
_
_
_
=
C
11
C
12
C
13
0 0 C
16
C
21
C
22
C
23
0 0 C
26
C
31
C
32
C
33
0 0 C
36
0 0 0 C
44
C
45
0
0 0 0 C
54
C
55
0
C
61
C
62
C
63
0 0 C
66
_
_
_
_
e
1
e
2
e
3
e
4
e
5
e
6
_
_
_
_
: (2:25)
Here, the stiffness matrix has 13 independent material coefcients. If a mater
ial has two mutually orthogonal planes of symmetry, then the plane
13 2.1 Equations of elasticity
orthogonal to these planes is also a plane of symmetry. In this case, the
material symmetry is described as orthotropic, and the number of independent
constants in the stiffness matrix reduces to nine. The stressstrain relations
when the symmetry planes are parallel to the three coordinate planes take the
following form
s
1
s
2
s
3
s
4
s
5
s
6
_
_
_
_
=
C
11
C
12
C
13
0 0 0
C
21
C
22
C
23
0 0 0
C
31
C
32
C
33
0 0 0
0 0 0 C
44
0 0
0 0 0 0 C
55
0
0 0 0 0 0 C
66
_
_
_
_
e
1
e
2
e
3
e
4
e
5
e
6
_
_
_
_
: (2:26)
In terms of the engineering elastic constants the inverse strainstress relations for
the orthotropic case become as follows
e
1
e
2
e
3
e
4
e
5
e
6
_
_
_
_
=
1
E
1
n
21
E
2
n
31
E
3
0 0 0
n
12
E
1
1
E
2
n
32
E
3
0 0 0
n
13
E
1
n
23
E
2
1
E
3
0 0 0
0 0 0
1
G
23
0 0
0 0 0 0
1
G
31
0
0 0 0 0 0
1
G
12
_
_
_
_
s
1
s
2
s
3
s
4
s
5
s
6
_
_
_
_
; (2:27)
where E
1
, E
2
, E
3
are Youngs moduli in the three material symmetry directions
(x
1
, x
2
, x
3
) respectively, n
ij
; i ,= j; are the six Poissons ratios dened in the
conventional way, e.g., n
12
= e
2
/ e
1
with s
1
applied, and G
23
, G
31
, and G
12
are
shear moduli in the x
2
x
3
, x
1
x
3
, and x
1
x
2
planes, respectively. The compliance
matrix in Eq. (2.27), being the inverse of a symmetric matrix, is also symmetric.
From this symmetry follows the reciprocal relationship,
n
ij
E
i
=
n
ji
E
j
(no sumon i; j); (2:28)
which can be used to eliminate three of the six Poissons ratios.
If the material is isotropic in a plane, i.e., with same elastic properties in all
directions in the plane, it is called transversely isotropic. Let the x
2
x
3
plane be the
plane of isotropy, i.e., E
3
= E
2
; n
31
= n
12
; G
31
= G
12
; G
23
=
E
2
2 1 n
23
( )
: The com
pliance tensor is then given by
14 Review of mechanics of composite materials
S [ [ =
S
11
S
12
S
12
0 0 0
S
12
S
22
S
23
0 0 0
S
12
S
23
S
22
0 0 0
0 0 0 2 S
22
S
23
( ) 0 0
0 0 0 0 S
66
0
0 0 0 0 0 S
66
_
_
_
_
=
1
E
1
n
12
E
1
n
12
E
1
0 0 0
n
12
E
1
1
E
2
n
23
E
2
0 0 0
n
12
E
1
n
23
E
2
1
E
2
0 0 0
0 0 0
E
2
2 1 n
23
( )
0 0
0 0 0 0
1
G
12
0
0 0 0 0 0
1
G
12
_
_
_
_
:
(2:29)
As seen above, a transversely isotropic material has ve independent stiffness
coefcients, viz. E
1
, E
2
, n
23
, n
12
, and G
12
. For a completely isotropic material there
are only two independent material coefcients, namely the Youngs modulus (E)
and Poissons ratio (n) or, alternatively, the Lame constants (l and m). The
constitutive relations can now be written as
s
ij
= le
kk
d
ij
2me
ij
; (2:30)
where d
ij
is the Kronecker delta. Alternatively,
e
ij
=
1
E
1 n ( )s
ij
ns
kk
d
ij
_
: (2:31)
2.1.4 Equations of motion
The equations governing the motion of a deformable body can be obtained by
combining kinematic relations, Eq. (2.4), equilibrium equations, Eq. (2.10), and
the constitutive relations, Eq. (2.15). For the particular case of linear elastic
isotropic materials, they can be written as
l m ( )u
j; ji
mu
i; jj
f
i
= r u
i
: (2:32)
These equations are known as Naviers equations. The displacement eld obtained
from these equations is unique and results into the determination of strains and
stresses by use of kinematic and constitutive relations.
2.1.5 Energy principles
Energy principles for a continuum body allow formulating the relationships
between stresses, strains or deformations, displacements, material properties,
and external effects in the form of energy or work done by internal and external
forces. They are also useful for obtaining approximate solutions of complex
boundary value problems, e.g., nite element methods. Detailed treatment of these
concepts can be found in [1820].
15 2.1 Equations of elasticity
Principle of virtual work
In the context of an elastic boundary value problem, consider a solid continuum
body (Figure 2.3), occupying a volume V and bounded by surface S = S
t
S
u
, to
be in static equilibrium under prescribed body forces f
i
over volume V, surface
tractions t
i
on S
t
, and displacements u
i
over remaining portion of the boundary S
u
.
For a statically admissible stress eld ~ s
ij
(such that ~ s
ij; j
= 0 in V, and
~
t
i
= ~ s
ij
n
j
on S
t
) and a kinematically admissible displacement eld ^ u
i
(such that
^e
ij
=
1
2
^ u
i; j
^ u
j;i
_ _
), the principle of virtual work states
_
S
~
t
i
^ u
i
dS
_
V
f
i
^ u
i
dV =
_
V
~ s
ij
^e
ij
dV: (2:33)
It should be noted that the displacement eld ^ u
i
and the stress eld ~ s
ij
are
completely independent of each other.
Principle of minimum potential energy
For a kinematically admissible displacement eld ^ u
i
, the potential energy of a
linear elastic continuum body under the action of conservative forces f
i
and
prescribed surface tractions t
i
on S
t
is dened as
^ u
i
( ) =
1
2
_
V
^ s
ij
^e
ij
dV
_
S
t
i
^ u
i
dS
_
V
f
i
^ u
i
dV: (2:34)
The principle of minimum potential energy states that among all the kinematically
admissible displacement elds the actual displacement eld minimizes the poten
tial energy. Thus, if u
i
represents the actual displacement eld, then
^ u
i
( ) _ u
i
( ) : (2:35)
V
t
j
n
i
S
t
S
u
u
i
f
i
Figure 2.3. A continuum body loaded with body forces inside its volume, and traction and
displacement on the boundary.
16 Review of mechanics of composite materials
Principle of minimum complementary energy
For a statically admissible stress eld ^ s
ij
, the complementary potential energy of a
linear elastic body is dened as
+
^ s
ij
_ _
=
1
2
_
V
^ s
ij
^e
ij
dV
_
S
^
t
i
u
i
dS; (2:36)
where
^
t
i
= ^ s
ij
n
j
is the reaction on S
u
. The principle of minimum complementary
energy states that among all the statically admissible stress elds the actual stress
eld minimizes the complementary potential energy. Thus, if u
i
represents the
actual displacement eld, then
+
^ s
ij
_ _
_
+
s
ij
_ _
: (2:37)
For actual stress, strain, anddisplacement elds, additionof Eqs. (2.34) and(2.36) yields
u
i
( )
+
s
ij
_ _
=
_
V
s
ij
e
ij
dV
_
S
t
i
u
i
dS
_
V
f
i
u
i
dV: (2:38)
The righthand side of Eq. (2.38) vanishes by virtue of the principle of virtual
work. Hence,
u
i
( ) =
+
s
ij
_ _
: (2:39)
Using Eqs. (2.35), (2.37), and (2.39), we obtain the lower and upper bounds to the
potential energy of a continuum body
^ u
i
( ) _ u
i
( ) =
+
s
ij
_ _
_
+
^ s
ij
_ _
: (2:40)
For the purpose of illustration, the potential and complementary energies for a
typical loaddisplacement response are shown in Figure 2.4.
2.2 Micromechanics
Micromechanics is a welldeveloped advanced eld that treats the response of a
heterogeneous solid based on the behavior of its constituents and their geometrical
congurations. For a detailed exposition the reader may refer to, e.g., [8]. Here a
brief summary of simple micromechanics estimates of the linear elastic properties
P
* P, D
D
Figure 2.4. A typical loaddisplacement diagram.
17 2.2 Micromechanics
of a unidirectional berreinforced composite is provided. These estimates are
useful in selecting bers and matrix materials and their volume fractions. In many
structural applications a unidirectional composite, fabricated as a thin layer,
called lamina or ply, is used as a basic unit and a laminate is constructed by
stacking these layers as illustrated in Figure 2.5.
2.2.1 Stiffness properties of a unidirectional lamina
Linear elastic properties of a lamina can be referred to a coordinate system (x
1
, x
2
,
x
3
) where the x
1
axis is aligned with bers, x
2
axis is transverse to bers in
the plane of the lamina, and the x
3
axis is normal to the plane of lamina
(see Figure 2.6). Noting that the lamina has orthotropic symmetry, the nine
independent elastic constants, as described in Section 2.1.3 above, in this reference
system are the three Youngs moduli (E
1
, E
2
, E
3
), the three Poissons ratios (n
12
,
n
13
, n
23
), and the three shear moduli (G
12
, G
13
, G
23
). For a subset of these constants
that represents inplane properties, i.e., E
1
, E
2
, n
12
, n
21
, and G
12
, in the x
1
x
2
plane,
the following expressions hold
E
1
= E
f
V
f
E
m
V
m
; (2:41)
n
12
= n
f
V
f
n
m
V
m
; (2:42)
1
E
2
=
V
f
E
f
V
m
E
m
; (2:43)
1
G
12
=
V
f
G
f
V
m
G
m
; (2:44)
where E, n, G, and V stand for the Youngs modulus, Poissons ratio, shear
modulus, and volume fraction, respectively, with the subscripts f and m indicating
bers and matrix, respectively. The minor Poissons ratio n
21
can be estimated
using the reciprocal relationship n
21
= n
12
(E
2
/E
1
).
Equations (2.41) and (2.42) have the form of the familiar rule of mixtures and
Eqs. (2.43) and (2.44) follow that rule for the inverse of the respective properties.
A unidirectional lamina
Lamination
Laminate
(a) (b) (c)
Figure 2.5. Stacking of a number of laminae makes up a laminate.
18 Review of mechanics of composite materials
The rst two expressions predict the experimental properties usually well while the
third and fourth expressions are found to be less accurate. Halpin and Kardos [21]
and Halpin and Tsai [22] proposed semiempirical relationships based on numer
ical computations. These relations can be used in place of Eqs. (2.43) and (2.44),
and are together expressed as
p
p
m
=
1 xV
f
1 V
f
; (2:45)
where
=
p
f
p
m
1
p
f
p
m
x
: (2:46)
Here p represents E
2
or G
12
, and p
f
and p
m
are the corresponding moduli for ber
and matrix, respectively. The tting parameter x needs to be determined by
comparing predictions with experimental data.
More advanced micromechanics approaches, such as HashinShtrikman vari
ational bounds [2329], MoriTanaka model [30], composite sphere and cylinder
assemblage model [31, 32], selfconsistent method [33], method of cells [3436], etc.
have also been developed in the past four decades. Interested readers are referred to
texts on micromechanics, e.g., [8, 37, 38], for detailed treatment of these approaches.
2.2.2 Thermal properties of a unidirectional lamina
Simple micromechanics estimates for the linear coefcient of thermal expansion of
a lamina can be obtained in the same way as the linear elastic properties. The
expressions obtained are as follows
a
1
=
1
E
1
a
f
E
f
V
f
a
m
E
m
V
m
( );
a
2
= 1 n
f
( )a
f
V
f
1 n
m
( )a
m
V
m
a
1
n
12
;
(2:47)
x
2
x
1
x
y
x
3
, z
Figure 2.6. Coordinate systems for a unidirectional ply. The material system is denoted by
x
1
, x
2
, x
3
; while the laminate system is denoted by x, y, z.
19 2.2 Micromechanics
where a
1
and a
2
are the thermal expansion coefcients in the ber and transverse
directions, respectively, and E
1
and n
12
are given by Eqs. (2.41) and (2.42).
2.2.3 Constitutive equations for a lamina
A lamina is thin compared to other dimensions of the entire laminate. Therefore,
the lamina can be assumed to be in a state of generalized plane stress. Conse
quently, all the throughthickness stress components are zero, i.e., s
4
= s
5
=
s
6
= 0. In such a case, the constitutive relation for an individual lamina referred
to the three axes of symmetry can be written in Voigt notation as
s
1
s
2
s
6
_
_
_
_
_
_
=
Q
11
Q
12
0
Q
12
Q
22
0
0 0 Q
66
_
_
_
_
e
1
e
2
e
6
_
_
_
_
_
_
; (2:48)
with
Q
11
=
E
1
1n
12
n
21
; Q
22
=
E
2
1n
12
n
21
; Q
12
=
n
12
E
2
1n
12
n
21
=
n
21
E
1
1n
12
n
21
; Q
66
=G
12
: (2:49)
The inverse constitutive relation for the lamina is given by
e
ij
= S
ijkl
s
kl
=
e
1
e
2
e
6
_
_
_
_
_
_
=
1
E
1
n
21
E
2
0
n
12
E
1
1
E
2
0
0 0
1
G
12
_
_
_
_
s
1
s
2
s
6
_
_
_
_
_
_
: (2:50)
The above constitutive relations are written in the lamina coordinate system (i.e.,
with x
1
along the ber direction, x
2
normal to the ber direction, and x
3
along the
lamina thickness). The constitutive relation for the lamina in another coordinate
system (xyz), which, for instance, could be aligned with the coordinate system
chosen for the laminate, is
s
xx
s
yy
s
xy
_
_
_
_
_
_
=
Q
11
Q
12
Q
16
Q
12
Q
22
Q
26
Q
16
Q
26
Q
66
_
_
_
_
e
xx
e
yy
2e
xy
_
_
_
_
_
_
; (2:51)
where
Q
ij
are known as reduced stiffness coefcients. These are related to Q
ij
,
dened by Eq. (2.49), by the transformation rules for stresses and strains. Thus,
s
1
s
2
s
6
_
_
_
_
_
_
=
m
2
n
2
2mn
n
2
m
2
2mn
mn mn m
2
n
2
_
_
_
_
s
xx
s
yy
s
xy
_
_
_
_
_
_
= T [ [
s
xx
s
yy
s
xy
_
_
_
_
_
_
(2:52)
e
1
e
2
e
6
_
_
_
_
_
_
=
m
2
n
2
mn
n
2
m
2
mn
2mn 2mn m
2
n
2
_
_
_
_
e
xx
e
yy
e
xy
_
_
_
_
_
_
= T [ [
1
_ _
T
e
xx
e
yy
e
xy
_
_
_
_
_
_
(2:53)
20 Review of mechanics of composite materials
where m = cos y, n = sin y, where y is the angle between the x and x
1
axes (Figure
2.6). Then by inverting Eqs. (2.52) and (2.53), substituting these in Eq. (2.51), and
on using Eq. (2.48) one obtains
Q
_
= T [ [
1
Q [ [ T [ [
T
_ _
1
: (2:54)
[T]
1
is simply given by changing y to y, i.e., [T(y)]
1
= [T(y)]. Expanding the
above relation, we have
Q
11
= Q
11
m
4
2 Q
12
2Q
66
( )m
2
n
2
Q
22
n
4
;
Q
22
= Q
11
n
4
2 Q
12
2Q
66
( )m
2
n
2
Q
22
m
4
;
Q
12
= Q
11
Q
22
4Q
66
( )m
2
n
2
Q
12
m
4
n
4
_ _
;
Q
16
= Q
11
Q
12
2Q
66
( )m
3
n Q
12
Q
22
2Q
66
( )mn
3
;
Q
26
= Q
11
Q
12
2Q
66
( )mn
3
Q
12
Q
22
2Q
66
( )m
3
n;
Q
66
= Q
11
Q
22
2Q
12
2Q
66
( )m
2
n
2
Q
66
m
4
n
4
_ _
:
(2:55)
The transformation rules described above enable us to express engineering moduli
for the lamina referred to arbitrary inplane axes (xy) in terms of moduli in the
principal (x
1
x
2
) directions as
1
E
x
=
m
4
E
1
n
4
E
2
1
G
12
2n
12
E
1
_ _
m
2
n
2
;
1
E
y
=
n
4
E
1
m
4
E
2
1
G
12
2n
12
E
1
_ _
m
2
n
2
;
n
xy
E
x
=
n
12
E
1
1 2n
12
E
1
1
E
2
1
G
12
_ _
m
2
n
2
;
1
G
xy
=
1
G
12
4m
2
n
2
1 n
12
E
1
1 n
21
E
2
1
G
12
_ _
:
(2:56)
To account for thermal stresses, we need to modify strains in Eq. (2.51) to include
thermal strains, as
e
xx
e
yy
2e
xy
_
_
_
_
_
_
=
e
0
xx
e
0
yy
2e
0
xy
_
_
_
e
th
xx
e
th
yy
2e
th
xy
_
_
_
_
; (2:57)
where the superscripts 0 and th denote mechanical and thermal strains, respectively, with
e
th
xx
e
th
yy
2e
th
xy
_
_
_
_
= T [ [
e
e
th
1
e
th
2
2e
th
12
_
_
_
_
_
_
= T [ [
e
a
1
a
2
0
_
_
_
_
_
_
T : (2:58)
2.2.4 Strength of a unidirectional lamina
Phenomenological failure (strength) criteria that use experimental data to deter
mine material constants have been proposed for composite materials along the
21 2.2 Micromechanics
lines of those used for metals such as the von Mises yield criterion. Failure
mechanisms in composite materials are, however, signicantly more complex,
resulting in a large number of criteria. Here, some common criteria will be stated
for reference; the interested reader is encouraged to consult [39] for more indepth
treatment.
For a unidirectional berreinforced lamina, the ve basic strength parameters
under inplane loading are as follows:
X = ultimate tensile strength in the ber direction
X
/
= ultimate compressive strength in the ber direction
Y = ultimate tensile strength transverse to bers
Y
/
= ultimate compressive strength transverse to bers
S = ultimate shear strength in the lamina plane.
These parameters are obtained by experimental testing. See, e.g., [39, 40] for
further details.
Maximum stress theory
According to this theory, a lamina fails if
s
1
=
X s
1
> 0;
X
/
s
1
< 0;
_
s
2
=
Y s
2
> 0;
Y
/
s
2
< 0;
_
s
6
[ [ = S:
(2:59)
For combined loading, theoretical predictions of the theory are inaccurate because
the maximum stress criterion does not account for stress interactions. For an off
axis normal loading, i.e., loading axis inclined to bers, this theory can be applied
by transforming the stresses to the principal material directions and then using the
criteria in Eq. (2.59).
Maximum strain theory
This theory states that failure occurs when
e
1
=
X
e
e
1
> 0;
X
/
e
e
1
< 0;
_
e
2
=
Y
e
e
2
> 0;
Y
/
e
e
2
< 0;
_
e
6
[ [ = S
e
;
(2:60)
where X
e
= X=E
1
; X
/
e
= X
/
=E
1
; Y
e
= Y=E
2
; Y
/
e
= Y
/
=E
2
; and S
e
= S=G
12
are the
ultimate failure strains analogous to the stressbased parameters mentioned above.
22 Review of mechanics of composite materials
Distortional energy (TsaiHill) criterion
This criterion is based on the distortional energy failure (yield) theory of
von Mises. Hill [41] further developed this yield criterion for anisotropic
materials and Azzi and Tsai [42] modied it to describe failure of a composite
lamina as follows
s
2
1
X
2
s
1
s
2
X
2
s
2
2
Y
2
s
2
6
S
2
= 1; (2:61)
where s
1
and s
2
are the tensile normal stresses along bers and normal to
bers, respectively, and s
6
is the inplane shear stress. When the normal
stresses are compressive, the compressive strength values in Eq. (2.61) are to
be used.
TsaiWu criterion
A polynomial function of stress components can be formulated with the multiply
ing terms of the polynomial expressing strength properties. Restricted to quadratic
terms of inplane stress components, such a function is known as the TsaiWu
criterion [43] and can be expressed as
F
1
s
1
F
2
s
2
F
11
s
2
1
F
22
s
2
2
F
66
s
2
6
2F
12
s
1
s
2
= 1: (2:62)
The product terms s
1
s
6
and s
2
s
6
are not present in Eq. (2.62) because the
multiplying coefcients to these terms can be shown to vanish. Also, the linear
term in s
6
is absent because of the shear strength being independent of the sign
of shear stress, which renders the coefcient of this term to be zero.
The six material constants in the TsaiWu criterion require two tests (tension
and compression) in the ber direction, two similar tests normal to bers, an in
plane shear test, and a biaxial normal load test.
Hashins criterion
Hashin [44] formulated threedimensional failure criteria for unidirectional ber
composites in terms of quadratic stress polynomials. The terms used in the
polynomials were functions of the stress invariants for transversely isotropic
symmetry. Thus the crosssectional plane of a unidirectional ber composite was
assumed as an isotropic plane. For relatively thick layers this may be a good
assumption.
A unidirectional ber composite was assumed to fail in one of four possible
separate modes: tensile ber mode (s
1
> 0), compressive ber mode (s
1
< 0),
tensile matrix mode (s
2
+ s
3
> 0), and compressive matrix mode (s
2
+ s
3
< 0)
For a thin unidirectional ber composite layer (lamina), the four failure criteria
are given by
23 2.2 Micromechanics
s
2
X
_ _
2
s
6
S
_ _
2
= 1; s
1
> 0;
s
1
= X
/
; s
1
< 0;
s
2
Y
_ _
2
s
6
S
_ _
2
= 1; s
2
> 0;
s
2
2S
/
_ _
2
Y
/
2S
/
_ _
2
1
_ _
s
2
Y
/
s
6
S
_ _
2
= 1; s
2
< 0;
(2:63)
where S
/
is the strength in transverse shear, while S here is the same in axial shear.
The difference between the two shear strengths is not fully unambiguous.
Over the years, a wide variety of failure criteria have been proposed. There is no
single failure theory that seems to capture all the complexities of composite failure.
A worldwide failure exercise was conducted to evaluate applicability of most
theories by comparing their predictions with test data [45].
2.3 Analysis of laminates
Laminates used in most engineering applications are fabricated by stacking plies in
different orientations. An example of a laminate with layup [0/90/45]
s
is shown in
Figure 2.7. A commonly used method of determining stresses and strains for such
laminates is based on the classical laminate plate theory (CLPT). More advanced
theories are treated in [9, 46]. The geometrical conditions needed for the application
of the CLPT are: (a) the individual plies are of uniform thickness, (b) they are
perfectly bonded to their neighboring plies, and (c) the total thickness of the
laminate follows the socalled thin plate assumption, which states that the thickness
dimension is much smaller than other structural dimensions (width and length).
The kinematic assumptions of the CLPT derive from the Kirchhoff assumptions,
which state that (a) a line element normal to the midplane in the undeformed state
0
90
45
45
90
0
y
z
x
z
0
z
1
z
2
z
n
q
Figure 2.7. Stacking of unidirectional plies in different orientations to make a
multidirectional [0/90/45]
s
laminate. The subscript s denotes that the laminate is symmetric
about the midplane.
24 Review of mechanics of composite materials
of the plate remains straight and perpendicular to the midplane after deformation,
and (b) such a line element does not change its length when the plate deforms.
2.3.1 Straindisplacement relations
The Kirchoff assumptions stated above lead to the x, y, and zdisplacements u, v,
and w, respectively, in the coordinate system shown in Figure 2.7 as follows
u(x; y; z) = u
0
(x; y) z
@w
0
(x; y)
@x
;
v(x; y; z) = v
0
(x; y) z
@w
0
(x; y)
@y
;
w(x; y; z) = w
0
(x; y);
(2:64)
where (u
0
, v
0
, w
0
) are the displacements of the laminate midplane. The corres
ponding straindisplacement relations are given by
e
xx
=
@u
@x
=
@u
0
@x
z
@
2
w
0
@x
2
;
e
yy
=
@v
@y
=
@v
0
@y
z
@
2
w
0
@y
2
;
e
zz
=
@w
@z
= 0;
e
xy
=
1
2
@u
@y
@v
@x
_ _
=
1
2
@u
0
@y
@v
0
@x
_ _
z
@
2
w
0
@x@y
;
e
xz
=
1
2
@u
@z
@w
@x
_ _
= 0;
e
yz
=
1
2
@v
@z
@w
@y
_ _
= 0:
(2:65)
The nonzero equations can be written in the following form
e
xx
e
yy
g
xy
_
_
_
_
_
_
=
e
0
xx
e
0
yy
g
0
xy
_
_
_
_
z
k
xx
k
yy
k
xy
_
_
_
_
_
_
; (2:66)
where e
0
xx
; e
0
yy
; g
0
xy
_ _
are the midplane strains and k
xx
; k
yy
; k
xy
_ _
are the laminate
curvatures, given by
e
0
xx
e
0
yy
g
0
xy
_
_
_
_
=
@u
0
@x
@v
0
@y
@u
0
@y
@v
0
@x
_
_
_
_
and
k
xx
k
yy
k
xy
_
_
_
_
_
_
=
@
2
w
0
@x
2
@
2
w
0
@y
2
2
@
2
w
0
@x@y
_
_
_
_
: (2:67)
25 2.3 Analysis of laminates
2.3.2 Constitutive relationships for the laminate
Using the lamina constitutive relations described earlier, the constitutive equation
for the kth (k = 1, 2, . . .) layer of the laminate can be written as
s
(k)
=
Q
_
(k)
e
(k)
: (2:68)
In the above equation, the square bracket represents a 33 matrix and the curly
bracket is for a 31 vector. The strains in the kth ply are given by
e
(k)
= e
0
_ _
z k : (2:69)
The thermal strains can be added to these strains, such that
e
(k)
= e
0
_ _
z k a
k
T: (2:70)
The kth ply stresses on using Eq. (2.68) can now be written as
s
(k)
=
Q
_
(k)
e
0
_ _
a
k
T
_ _
z
Q
_
(k)
k : (2:71)
At the laminate level the force and moment resultants are dened as
N [ [ =
N
xx
N
yy
N
xy
_
_
_
_
=
_
h=2
h=2
s
xx
s
yy
s
xy
_
_
_
_
dz;
M [ [ =
M
xx
M
yy
M
xy
_
_
_
_
=
_
h=2
h=2
s
xx
s
yy
s
xy
_
_
_
_
z dz:
(2:72)
In terms of ply stresses that generally vary from ply to ply, we have
N
xx
N
yy
N
xy
_
_
_
_
_
_
=
N
k=1
_
z
k1
z
k
s
xx
s
yy
s
xy
_
_
_
_
_
_
dz; (2:73)
which gives us
N
xx
N
yy
N
xy
_
_
_
N
th
xx
N
th
yy
N
th
xy
_
_
_
_
=
N
k=1
_
z
k1
z
k
Q
11
Q
12
Q
16
Q
12
Q
22
Q
26
Q
16
Q
26
Q
66
_
_
_
_
e
0
xx
e
0
yy
g
0
xy
_
_
_
_
dz
N
k=1
_
z
k1
z
k
Q
11
Q
12
Q
16
Q
12
Q
22
Q
26
Q
16
Q
26
Q
66
_
_
_
_
k
xx
k
yy
k
xy
_
_
_
_
z dz;
(2:74)
where the force resultants due to thermal stresses are given by
26 Review of mechanics of composite materials
N
th
xx
N
th
yy
N
th
xy
_
_
_
_
=
N
k=1
_
z
k1
z
k
Q
11
Q
12
Q
16
Q
12
Q
22
Q
26
Q
16
Q
26
Q
66
_
_
_
_
a
x
T
a
y
T
0
_
_
_
_
_
_
dz: (2:75)
The relations in Eq. (2.74) can be rewritten in more compact form by using
matrices [A] and [B] as follows
N
xx
N
yy
N
xy
_
_
_
_
_
_
N
th
xx
N
th
yy
N
th
xy
_
_
_
_
=
A
11
A
12
A
16
A
12
A
22
A
26
A
16
A
26
A
66
_
_
_
_
e
0
xx
e
0
yy
g
0
xy
_
_
_
B
11
B
12
B
16
B
12
B
22
B
26
B
16
B
26
B
66
_
_
_
_
k
xx
k
yy
k
xy
_
_
_
_
_
_
:
(2:76)
Similarly,
M
xx
M
yy
M
xy
_
_
_
_
_
_
=
N
k=1
_
z
k1
z
k
s
xx
s
yy
s
xy
_
_
_
_
_
_
zdz; (2:77)
i.e.,
M
xx
M
yy
M
xy
_
_
_
M
th
xx
M
th
yy
M
th
xy
_
_
_
_
=
N
k=1
_
z
k1
z
k
Q
11
Q
12
Q
16
Q
12
Q
22
Q
26
Q
16
Q
26
Q
66
_
_
_
_
e
0
xx
e
0
yy
g
0
xy
_
_
_
_
zdz
N
k=1
_
z
k1
z
k
Q
11
Q
12
Q
16
Q
12
Q
22
Q
26
Q
16
Q
26
Q
66
_
_
_
_
k
xx
k
yy
k
xy
_
_
_
_
z
2
dz;
(2:78)
where
M
th
xx
M
th
yy
M
th
xy
_
_
_
_
=
N
k=1
_
z
k1
z
k
Q
11
Q
12
Q
16
Q
12
Q
22
Q
26
Q
16
Q
26
Q
66
_
_
_
_
a
x
T
a
y
T
0
_
_
_
_
_
_
zdz: (2:79)
Introducing a new matrix [D], Eq. (2.78) can be rewritten as
M
xx
M
yy
M
xy
_
_
_
_
_
_
M
th
xx
M
th
yy
M
th
xy
_
_
_
_
=
B
11
B
12
B
16
B
12
B
22
B
26
B
16
B
26
B
66
_
_
_
_
e
0
xx
e
0
yy
g
0
xy
_
_
_
D
11
D
12
D
16
D
12
D
22
D
26
D
16
D
26
D
66
_
_
_
_
k
xx
k
yy
k
xy
_
_
_
_
_
_
:
(2:80)
The material coefcients (A
ij
, B
ij
, D
ij
) are known as the extensional stiffness, the
extensionbending coupling stiffness, and the bending stiffness coefcients, respect
ively. These are given by
A
ij
; B
ij
; D
ij
_ _
=
_ h
2
h
2
Q
ij
1; z; z
2
_ _
dz; (2:81)
27 2.3 Analysis of laminates
or
A
ij
=
N
k=1
Q
ij
z
k1
z
k
( ) ;
B
ij
=
1
2
N
k=1
Q
ij
z
2
k1
z
2
k
_ _
;
D
ij
=
1
3
N
k=1
Q
ij
z
3
k1
z
3
k
_ _
:
(2:82)
The laminate constitutive relations can now be written in compact form as
N
M
_ _
=
A [ [ B [ [
B [ [ D [ [
_ _
e
0
_ _
k
_ _
(2:83)
where {N}, {M} include thermal resultants.
2.3.3 Stresses and strains in a lamina within a laminate
The laminate constitutive relations in Eq. (2.83) can be reverted to yield the mid
plane strains and curvatures in terms of the stress and moment resultants.
A partial inversion is rst done by inverting Eq. (2.76) and substituting into
Eq. (2.80) to obtain
e
0
M
_ _
=
A
+
B
+
C
+
D
+
_ _
N
k
_ _
; (2:84)
where
A
+
= A
1
; B
+
= A
1
B; C
+
= BA
1
= B
+
( )
T
; D
+
= D BA
1
B; (2:85)
and the brackets for matrix/vector representation have been dropped for conveni
ence. Solving Eq. (2.84) for k and its substitution back gives
e
0
k
_ _
=
A
/
B
/
B
/
D
/
_ _
N
M
_ _
; (2:86)
where
A
/
= A
+
B
+
D
+
( )
1
B
+
( )
T
; B
/
= B
+
D
+
( )
1
; D
/
= D
+
( )
1
: (2:87)
Once midplane strains and curvatures are known, the strains and stresses in each
lamina can be determined using Eqs. (2.70) and (2.68), respectively.
2.3.4 Effect of layup conguration
The sequence of ply layup has a signicant impact on the stiffness properties of the
designed laminate. Some interesting ply congurations are described below.
28 Review of mechanics of composite materials
v Balanced laminate: If for each +yply, we have another identical ply of
the same thickness, but y orientation, we have A
16
= A
26
= 0. Such
laminates are known as balanced laminates. If additionally these plies are
at the same distance about the midplane (one above and another below
the midplane), then D
16
= D
26
= 0. An example of a balanced
laminate is [0/+45/45/90
2
/0]
T
, where the subscript T denotes total laminate
sequence.
v Symmetric laminate: If a laminate has plies stacked in such a way that through
its thickness the plies are symmetrical about the midplane, then B
ij
= 0. Thus,
such laminates will not exhibit any extensionbending coupling, e.g., [0/30/
45
2
/90
2
/45
2
/30/0]
T
= [0/30/45
2
/90]
s
, where the subscript s represents sym
metry about midplane.
v Crossply laminate: If the plies are stacked in two orthogonal directions, e.g., in
longitudinal (0
2pr
_ cos
y
2
_ _
1 sin
y
2
_ _
sin
3y
2
_ _ _ _
;
s
yy
=
K
I
2pr
_ cos
y
2
_ _
1 sin
y
2
_ _
sin
3y
2
_ _ _ _
;
t
xy
=
K
I
2pr
_ cos
y
2
_ _
sin
y
2
_ _
cos
3y
2
_ _
;
(2:89)
where r and y are as shown in the gure and K
I,
known as the stress intensity
factor, is given by
K
I
= s
pa
_
: (2:90)
where the subscript I denotes the opening mode (mode I). It can be noted that the
stress eld is singular at the crack tip with a r
1/2
singularity.
The condition of failure, i.e., unstable crack growth, is assumed when
K
I
_ K
IC
: (2:91)
K
IC
, known as the critical stress intensity factor, or fracture toughness, is a
parameter representing the material resistance to fracture, and can be obtained
experimentally.
The energy criterion
In the energybased approach, one considers a cracked body and examines the
changes brought about by an incremental crack growth in the potential energy of
30 Review of mechanics of composite materials
applied forces the stored elastic strain energy and the crack surface energy. The
condition for unstable crack growth is then expressed as
G _ G
C
; (2:92)
where G is the energy available for crack growth per unit of crack surface area,
called the energy release rate, and G
C
is its critical value, which depends on the
material in which the crack is advancing. G
C
is viewed as the resistance to crack
growth induced by the material. For a linear elastic material undergoing small
scale yielding at the crack front, the energy release rate is found to be related to the
stress intensity factor, described above, as
G =
K
2
I
E
/
; (2:93)
where E
/
= E for plane stress condition, and E
/
=
E
1 n
2
for plane strain condition.
2.4.2 Crack separation modes
A crack is activated, i.e., it produces stresses at its front, when the two crack
surfaces separate. The separation can take place in combination of three
a
x
y
r
s
yy
t
xy
q
s
xx
s
s
Figure 2.8. Edge crack in a plate in tension.
31 2.4 Linear elastic fracture mechanics
independent modes, denoted as modes I, II, and III, illustrated in Figure 2.9. In
mode I, also called the crack opening mode, the two crack surfaces separate
symmetrically about the crack plane. Mode II is a sliding mode, in which the two
crack surfaces remain in contact and slide past each other in the crack plane.
Finally, mode III, described as the tearing mode, is driven by outofplane shear,
resulting in displacement of the two crack surfaces in the x
3
direction.
Any displacement of the crack surfaces for a general loading can be viewed as
a superposition of these three modes. Denoting the stress intensity factors in individ
ual modes as K
I
, K
II
, and K
III
, the energy release rate for mixedmode is given by
G =
K
2
I
E
/
K
2
II
E
/
1 n
E
K
2
III
; (2:94)
where
K
I
= s
11
pa
_
; K
II
= s
12
pa
_
; K
III
= s
13
pa
_
; (2:95)
where the stresses s
11
, etc. refer to the axes shown in Figure 2.9.
2.4.3 Crack surface displacements
The displacement jump across the two crack surfaces, expressed as
u
i
= u
i
u
i
; (2:96)
where u
i
and u
i
represent the displacements of the upper and lower crack
surfaces, respectively, is a quantity of interest in fracture analysis. For the opening
mode of crack separation, mode I, illustrated in Figure 2.10, i = 2, this quantity is
described as crack opening displacement (COD). For an innite isotropic homo
geneous medium the COD value is given by
u
2
= k
1
x
1
a
_ _
2
_
; (2:97)
which describes an elliptical crack opening prole.
(a) (b) (c)
x
3
x
2
x
1
Figure 2.9. Crack separation modes: (a) opening; (b) sliding; and (c) tearing.
32 Review of mechanics of composite materials
2.4.4 Relevance of fracture mechanics for damage analysis
Fracture mechanics developed and matured well before damage mechanics
emerged. For both elds, the impetus came from the need to analyze failure of
metals. Fracture mechanics initially addressed brittle failure from sharp defects
based on idealized stress analysis of cracks. In contrast, damage mechanics was
concerned with the effect of distributed voids and cracks on the average response of
a solid. For composite materials, the complexity of failure processes involving
a multitude of cracks gave rise to further development of damage mechanics. Today,
damage mechanics of composite materials stands on its own as a mature eld solidly
founded in thermodynamics and having a variety of analytical and computational
methodologies associated with it. Fracture mechanics has aided the development of
damage mechanics of composite materials in providing energybased concepts for
addressing evolution of failure states. However, the stress analysis of cracks, char
acterized by stress intensity factors, is less relevant to composite damage analysis.
Other than a few cases where single crack growth is a dominant failure mechanism,
such as delamination emanating from free edges in laminates, crack front singular
ities are of little interest. Indeed, individual cracks constituting damage modes are
usually arrested at interfaces. Therefore, their growth is of little interest. Instead,
energy dissipation occurs due to crack multiplication. Therefore, an appropriate
energybased analysis, needed to treat this type of situation, does not resort to stress
intensity factors, as is the case in brittle fracture of metals.
x
1
x
2
u
2
+
u
2
s
s
+
Figure 2.10. Crack opening displacement for a crack of size 2a.
33 2.4 Linear elastic fracture mechanics
References
1. Y. C. Fung, A First Course in Continuum Mechanics. (Englewood Cliffs, NJ: Prentice
Hall, Inc., 1977).
2. L. E. Malvern, Introduction to the Mechanics of a Continuous Medium. (Englewood
Cliffs, NJ: PrenticeHall, Inc., 1969).
3. I. S. Sokolnikoff, Mathematical Theory of Elasticity, 2ndedn. (NewYork: McGrawHill, 1956).
4. A. E. H. Love, A Treatise on the Mathematical Theory of Elasticity. (New York: Dover
Publications, 1944).
5. S. P. Timoshenko and J. N. Goodier, Theory of Elasticity. (NewYork: McGrawHill, 1951).
6. R. M. Jones, Mechanics of Composite Materials, 2nd edn. (Philadelphia, PA: Taylor &
Francis, 1999).
7. R. M. Christensen, Mechanics of Composite Materials. (Dover Publications, 2005).
8. J. Qu and M. Cherkaoui, Fundamentals of Micromechanics of Solids. (Hoboken, NJ:
Wiley, 2006).
9. J. N. Reddy, Mechanics of Laminated Composite Plates and Shells: Theory and Analysis,
2nd edn. (Boca Raton, FL: CRC Press, 2004).
10. I. M. Daniel and O. Ishai, Engineering Mechanics of Composite Materials. (Oxford:
Oxford University Press, 1994).
11. A. K. Kaw, Mechanics of Composite Materials. (Boca Raton, FL: CRC Press, 1997).
12. L. P. Kollar and G. S. Springer, Mechanics of Composite Structures. (Cambridge:
Cambridge University Press, 2003).
13. D. Broek, Elementary Engineering Fracture Mechanics, 4th revised edn. (Dordrecht,
Netherlands: Kluwer Academic Publishers, 1991).
14. T. L. Anderson, Fracture Mechanics: Fundamentals and Applications, 2nd edn. (Boca
Raton, FL: CRC Press, 1994).
15. E. E. Gdoutos, Fracture Mechanics: An Introduction, 2nd edn. Solid Mechanics and its
Applications, Vol. 123. (Dordrecht: Norwell, MA: Springer, 2005).
16. K. Friedrich, Application of Fracture Mechanics to Composite Materials. (Amsterdam:
Elsevier, 1989).
17. R. W. Hertzberg, Deformation and Fracture Mechanics of Engineering Materials, 3rd
edn. (New York: Wiley, 1989).
18. K. Washizu, Variational Methods in Elasticity and Plasticity, 2nd edn. (Oxford: Perga
mon Press, 1974).
19. C. L. Dym and I. H. Shames, Solid Mechanics: A Variational Approach. (New York:
McGrawHill, 1973).
20. J. N. Reddy, Energy Principles and Variational Methods in Applied Mechanics, 2nd edn.
(Hoboken, NJ: Wiley, 2002).
21. J. C. Halpin and J. L. Kardos, HalpinTsai equations review. Polym Eng Sci, 16:5
(1976), 34452.
22. J. C. Halpin and S. W. Tsai, Environmental Factors Estimation in Composite Materials
Design. US Airforce Technical Report, AFML TR (1967). pp. 67423.
23. Z. Hashin, On elastic behaviour of bre reinforced materials of arbitrary transverse
phase geometry. J Mech Phys Solids, 13:3 (1965), 11934.
24. Z. Hashin and S. Shtrikman, A variational approach to the theory of the
effective magnetic permeability of multiphase materials. J Appl Phys, 33:10 (1962),
312531.
34 Review of mechanics of composite materials
25. Z. Hashin and S. Shtrikman, A variational approach to the theory of the elastic
behaviour of multiphase materials. J Mech Phys Solids, 11:2 (1963), 12740.
26. Z. Hashin and S. Shtrikman, On some variational principles in anisotropic and non
homogeneous elasticity. J Mech Phys Solids, 10:4 (1962), 33542.
27. Z. Hashin and S. Shtrikman, A variational approach to the theory of the elastic
behaviour of polycrystals. J Mech Phys Solids, 10:4 (1962), 34352.
28. Z. Hashin, Variational principles of elasticity in terms of the polarization tensor. Int
J Eng Sci, 5:2 (1967), 21323.
29. Z. Hashin, Analysis of composite materials a survey. J Appl Mech, Trans ASME, 50:3
(1983), 481505.
30. T. Mori and K. Tanaka, Average stress in matrix and average elastic energy of
materials with mistting inclusions. Acta Metall, 21:5 (1973), 5714.
31. Z. Hashin, The elastic moduli of heterogeneous materials. J Appl Mech, TASME,
29 (1962), 14350.
32. R. M. Christensen and K. H. Lo, Solutions for effective shear properties in 3phase
sphere and cylinder models. J Mech Phys Solids, 27:4 (1979), 31530.
33. R. Hill, A selfconsistent mechanics of composite materials. J Mech Phys Solids,
13:4 (1965), 21322.
34. J. Aboudi, Micromechanical analysis of composites by the method of cells. Appl Mech
Rev, 42:7 (1989), 193221.
35. J. Aboudi, Micromechanical analysis of composites by the method of cells update.
Appl Mech Rev, 49:10 Part 2 (1996), S8391.
36. J. Aboudi, Micromechanical prediction of the effective coefcients of thermo
piezoelectric multiphase composites. J Intell Mater Syst Struct, 9:9 (1999), 71322.
37. S. NematNasser and M. Hori, Micromechanics: Overall Properties of Heterogeneous
Materials, 2nd edn. (Amsterdam: North Holland, 1999).
38. T. Mura, Micromechanics of Defects in Solids, 2nd edn. (New York: Springer, 1987).
39. C. T. Sun, Strength analysis of unidirectional composites and laminates. In Fiber
Reinforcements and General Theory of Composites, ed. T. W. Chou. (Amsterdam:
Elsevier, 2000), pp. 64166.
40. L. A. Carlsson, D. F. Adams, and R. B. Pipes, Experimental Characterization of
Advanced Composite Materials. (Boca Raton, FL: CRC Press, 2003).
41. R. Hill, The Mathematical Theory of Plasticity. (New York: Oxford University Press,
1998).
42. V. D. Azzi and S. W. Tsai, Anisotropic strength of composites, Exp Mech, 5 (1965),
28388.
43. S. W. Tsai and E. M. Wu, General theory of strength for anisotropic materials.
J Compos Mater, 5:January (1971), 5880.
44. Z. Hashin, Failure criteria for unidirectional ber composites. J Applied Mech, Trans
ASME, 47 (1980), 32934.
45. P. D. Soden, A. S. Kaddour, and M. J. Hinton, Recommendations for designers and
researchers resulting from the worldwide failure exercise. Compos Sci Technol, 64:34
(2004), 589604.
46. J. N. Reddy, Mechanics of Laminated Composite Plates: Theory and Analysis. (Boca
Raton, FL: CRC Press, 1997).
47. T. L. Anderson, Fracture Mechanics: Fundamentals and Applications, 3rd edn. (Boca
Raton, FL: Taylor & Francis, 2005).
35 References
3 Damage in composite materials
All structures are designed for a purpose. If the purpose is to carry loads, then a
designer must assure that the structure has sufcient loadbearing capacity. If the
structure is to function over a period of time, then it must be designed to meet its
functionality over that period without losing its integrity.
These are generic structural design issues irrespective of the material used. There
are, however, signicant differences in design procedures depending on whether the
material used is a socalled monolithic material, e.g., a metal or a ceramic, or
whether it is a composite material with distinctly different constituents. The hetero
geneity of microstructure as well as the anisotropy of properties provide signi
cantly different characteristics to composite materials in how they deform and fail
when compared to metals or ceramics. This chapter will review those characteris
tics. However, before proceeding we need to introduce certain denitions.
Fracture: Conventionally, fracture is understood to be breakage of material,
or at a more fundamental level, breakage of atomic bonds, manifesting itself in
formation of internal surfaces. Examples of fracture in composites are ber
breakage, cracks in matrix, ber/matrix debonds, and separation of bonded
plies (delamination). The eld known as fracture mechanics deals with condi
tions for formation and enlargement of the surfaces of material separation.
Damage: Damage, on the other hand, refers to a collection of all the irreversible
changes brought about in a material by a set of energy dissipating physical or
chemical processes, resulting from the application of thermomechanical load
ings. Damage may inherently be manifested by atomic bond breakage. Unless
specied differently, damage is understood to refer to distributed changes.
Examples of damage in composites are multiple berbridged matrix cracking
in a unidirectional composite, multiple intralaminar cracking in a laminate,
local delamination distributed in an interlaminar plane, and ber/matrix inter
facial slip associated with multiple matrix cracking. These damage mechanisms
will be explained in some detail later in this chapter. The eld of damage
mechanics deals with conditions for initiation and progression of distributed
changes as well as with consequences of those changes on the response of a
material (and by implication, a structure) to external loading.
Failure: The inability of a given material system (and consequently, a structure
made from it) to perform its design function. Fracture is one example of a
possible failure; but, generally, a material could fracture (locally) and still
perform its design function. Upon suffering damage, e.g., in the form of
multiple cracking, a composite material may still continue to carry loads and,
thereby, meet its loadbearing requirement but fail to deform in a manner
needed for its other design requirements, such as vibration characteristics and
deection limits. It is a common practice for engineers to predict composite
failure based on any of the multitude of lamina failure criteria described in the
previous chapter. These criteria only predict the nal event of failure, and
generally cannot characterize the damage mechanisms leading to the nal
failure. In reality, the failure event in a composite structure is preceded and
inuenced by the progressive occurrence and interaction of various damage
mechanisms.
Structural integrity: The ability of a loadbearing structure to remain intact and
functional upon the application of loads. In contrast to metals, remaining intact
(not breaking up in pieces) for composites is not necessarily the same as
remaining functional. For instance, composites can lose their functionality by
suffering degradation in their stiffness properties while still carrying signicant
loads.
Durability: A term very close in meaning to structural integrity. Specically,
durability is dened as the ability of a structure to retain adequate properties
(strength, stiffness, and environmental resistance) throughout its life to the
extent that any deterioration can be controlled and repaired [1]. The longterm
durability of a composite structure is an important design requirement in civil,
infrastructure, and aircraft industries.
3.1 Mechanisms of damage
The heterogeneous microstructure of composites, the large differences between
constituent properties, the presence of interfaces as well as directionality of
reinforcement that induces anisotropy in overall properties, are reasons for the
complexities observed in geometrical features of microlevel failure (microcracks)
in composites. Additionally, when interfaces are present, such as between bers
and matrix and between plies in a laminate, the stress transfer via interfaces
provides conditions for multiple cracking (to be discussed later). The wealth of
observations reported in the literature on various cracking processes, collectively
referred to as damage mechanisms, are summarized below for the purpose of
treatments in later chapters related to deformation and failure of composite
materials at a macro level.
3.1.1 Interfacial debonding
The performance of a berreinforced composite is markedly inuenced by the
properties of the interface between the ber and the matrix resin. The
37 3.1 Mechanisms of damage
adhesion bond at the interfacial surface affects the macroscopic mechanical
properties of the composite. The interface plays a signicant role in stress
transfer between ber and matrix. For instance, if the bers are weakly held
by the matrix, the composite starts to form a matrix crack at a relatively low
stress. On the other hand, if the bers are strongly bonded to the matrix, the
matrix cracking is delayed and the composite fails catastrophically because of
ber fracture as the matrix cracks. The constraint between the ber and the
matrix also inuences other damage mechanisms such as interfacial slipping,
and ber pullout. Controlling interfacial properties can thus provide a way to
control the performance of a composite structure. In unidirectional compos
ites, debonding occurs at the interface between ber and matrix when the
interface is weak. Figure 3.1 shows debond surfaces observed in a ber
reinforced composite [2].
The longitudinal interfacial debonding behavior of singleber composites has
been studied in detail by the use of the pullout [37] and fragmentation [811]
tests. The mechanics of ber/matrix interfacial debonding in a unidirectional
berreinforced composite is depicted in Figure 3.2. When fracture strain of the
ber is greater than that of the matrix, a crack originating at a point of stress
concentration, e.g., voids, air bubbles, or inclusions, in the matrix is either halted
by the ber, if the stress is not high enough (Figure 3.2(b)), or it may pass
around the ber without destroying the interfacial bond. As the applied load
increases, the ber and matrix deform differentially, resulting in a buildup of
large local stresses in the ber. This causes local Poisson contraction and even
tually when the shear stress developed at the interface exceeds the interfacial
shear strength, debonding extending over a distance along the ber results
(Figure 3.2(c)). Shear lag and cohesive zone models are commonly used
approaches to predict initiation of debonding and stress transfer at the interface
[3, 5, 1217].
20m
Figure 3.1. Debonds in a berreinforced composite. Reprinted, with kind permission, from
Compos Sci Technol, Vol. 59, E. K. Gamstedt and B. A. Sjo gren, Micromechanisms in
tensioncompression fatigue of composite laminates containing transverse plies, pp. 16778,
copyright Elsevier (1999).
38 Damage in composite materials
3.1.2 Matrix microcracking/intralaminar (ply) cracking
Fiberreinforced composites offer high strength and stiffness properties in the
longitudinal direction. Their properties, however, in the transverse directions are
generally low. As a result, they readily develop cracks along bers. These cracks
are usually the rst observed form of damage in berreinforced composites [19].
In laminates with plies in different ber orientations, these cracks can form from
defects in a given ply and grow traversing the thickness of the ply and running
parallel to the bers in that ply. The terms matrix microcracks, transverse cracks,
intralaminar cracks, and ply cracks are invariably used to refer to these very same
cracks. Such cracks are found to be caused by tensile loading, fatigue loading, as
well as by changes in temperature or by thermal cycling. They can originate from
ber/matrix debonds or manufacturinginduced defects such as voids and inclu
sions [20] (see Figure 3.3). Matrix cracks can also form in ceramic matrix compos
ites (CMC), and in short ber composites (SFC). The eld of damage mechanics
deals with prediction of formation, growth, and effects of matrix cracks on overall
material behavior. Analysis, design, and behavior of composites subjected to
intralaminar cracking will be dealt with in detail in the subsequent chapters.
Figure 3.4 illustrates matrix cracks observed on the free edges of continuous
ber and woven fabric polymer composite laminates induced due to fatigue
loading [21, 22]. Although matrix cracking does not cause structural failure by
itself, it can result in signicant degradation in material stiffness and can also
induce more severe forms of damage, such as delamination and ber breakage,
and give pathways for entry of uids.
3.1.3 Interfacial sliding
Interfacial sliding between constituents in a composite can take place by differential
displacement of the constituents. One example of this is when bers and matrix in a
composite are not bonded together adhesively but by a shrinkt mechanism
due to difference in thermal expansion properties of the constituents. On
(a) (b) (c)
Figure 3.2. Mechanics of interfacial debonding in a simple composite: (a) perfect laminate;
(b) differential deformation of ber and matrix crack causes high stresses at ber/matrix
interface; (c) shear stress exceeds the interfacial shear strength nucleating a debond.
Reprinted from [18], with kind permission from Maney Publishing.
39 3.1 Mechanisms of damage
thermomechanical loading, the shrinkt (residual) stresses can be removed,
leading to a relative displacement (sliding) at the interface. The relief of interfacial
normal stress can also occur when a matrix crack tip approaches or hits the
interface.
Debond
Matrix
Void
Fiber
5 m 10 m
(a)
(b)
Figure 3.3. Matrix crack initiation from: (a) ber debonds; (b) void results. Reprinted, with
kind permission, from Compos Sci Technol, Vol. 57, C. A. Wood and W. L. Bradley,
Determination of the effect of seawater on the interfacial strength of an interlayer Eglass/
graphite/epoxy composite by in situ observation of transverse cracking in an environmental
SEM, pp. 103343, copyright Elsevier (1997).
(a)
(b)
Figure 3.4. Examples of matrix cracks observed in (a) continuous ber and (b) woven fabric
polymer composite laminates [22]. Part (a) reprinted, with kind permission, from Compos Sci
Technol, Vol. 68, D. T. G. Katerelos, J. Varna, and C. Galiotis, Energy criterion for modeling
damage evolution in crossply composite laminates, pp. 231824, copyright Elsevier (2008).
40 Damage in composite materials
When the two constituents are bonded together adhesively, interfacial sliding
can occur subsequent to debonding if a compressive normal stress on the interface
is present. The debonding can be induced by a matrix crack, or it can result from
growth of interfacial defects. Thus, interfacial sliding that follows debonding can
be a separate damage mode or it can be a damage mode coupled with matrix
damage.
Interfacial sliding in ceramic matrix composites (CMCs) can be signicant if the
temperature change imposed is high and the thermal expansion mismatch between
the bers and matrix is also large. When the matrix in a CMC cracks, the resulting
interfacial debonding affects interfacial sliding, causing interactive effect on the
composite deformation [23].
3.1.4 Delamination/interlaminar cracking
Interlaminar cracking, i.e., cracking in the interfacial plane between two adjoining
plies in a laminate, causes separation of the plies (laminae) and is referred to as
delamination. In composite laminates, delamination can occur at cut (free) edges,
such as at holes, or at an exposed surface through the thickness. When loaded in
the plane, the laminate develops throughthickness normal and shear stresses at
the tractionfree surface extending a short distance into the laminate plane. These
stresses can result in local cracking in the interlaminar planes. Delaminations can
also form as a result of lowvelocity impact [2426]. In contrast to metals, in
polymer composite laminates delamination can occur below the surface of a
structure under a relatively light impact, such as that from a dropped tool, while
the surface appears undamaged to visual inspection [25, 27, 28]. The growth of
delamination cracks under the subsequent application of external loads leads to a
rapid deterioration of the mechanical properties and may cause catastrophic
failure of the composite structure [29, 30]. Another source of delamination is the
local interlaminar cracking induced by ply cracks. This delamination can grow
and separate the region between two adjacent ply matrix cracks as illustrated in
Figure 3.5.
Delamination can be a substantial problem in designing composite structures
as it can diminish the role of strong bers and make the weaker matrix
properties govern the structural strength [31]. In initiating delamination the
critical material property is the interlaminar strength, which is determined by
the matrix [26, 31]. Once the interlaminar cracks are formed, their growth is
determined by the interlaminar fracture toughness, which is also governed by
the matrix. If delamination is viewed as decohesion of the cohesive zone
between the separating plies, then both the matrix strength and the fracture
toughness act as material parameters [32]. As a design approach, delamination
can be reduced either by improving the interlaminar strength and fracture
toughness or by modifying the ber architecture to reduce the driving forces
for delamination [33, 34].
41 3.1 Mechanisms of damage
3.1.5 Fiber breakage
The failure (separation) of a berreinforced composite ultimately comes from
breakage of bers. In a unidirectional composite loaded in tension along bers the
individual bers fail at their weak points and stress redistribution between bers
and matrix occurs, affecting other bers in the local vicinity of the broken bers
and possibly breaking some. The ber/matrix interface transfers the stress from
the broken ber back to the ber at a certain distance, making another ber break
possible if the strength is exceeded by the stress. The ber breakage process is of
a statistical nature because of the nonuniformity of ber strength along the ber
length and the stress redistribution. When plies of unidirectional bers are stacked
in a laminate, the stress on bers is enhanced in the vicinity of ply cracks in the
adjacent plies, causing a narrow distribution of ber failure sites [35]. A greater
number of ber breaks per unit volume is found closer to the interface where the
ply crack terminates than away from the interface where the local stress concen
tration falls off [35].
The ultimate tensile strength of a ply within a general laminate is difcult to
predict from the tensile strength of bers due to the statistical nature of ber
failure and the progression of ber failures [36, 37]. Fracture (crack growth)
properties such as the fracture toughness of a composite depend not only on the
failure properties of the constituents but signicantly also on the efciency of
bonding across the interface [38].
3.1.6 Fiber microbuckling
When a unidirectional composite is loaded in compression, the failure is governed
by a mechanism known as microbuckling of bers. There are two idealized
basic modes of microbuckling deformation, denoted extensional and shear
modes [39], illustrated in Figure 3.6, depending upon whether the bers deform
out of phase or in phase with one another. The corresponding compressive
strength for the onset of instability is given as
Figure 3.5. Interlaminar delamination crack formed due to joining of two adjacent matrix
cracks in a berreinforced composite laminate.
42 Damage in composite materials
s
c
2V
f
V
f
E
f
E
m
31 V
f
s
; 3:1
for the extension mode, and
s
c
G
m
1V
f
; 3:2
for the shear mode, where E and G denote Youngs modulus and shear modulus,
respectively, and the subscripts f and m designate ber and matrix, respectively.
These expressions for idealized deformation modes do not generally agree with
experimental data for compression strength. It has been argued that in practical
composites the manufacturing process tends to cause misalignment of bers,
which can induce localized kinking of bers. The kinking process is driven by
local shear, which depends on the initial misalignment angle f
0
[40]. The critical
compressive stress corresponding to instability then is given by
s
c
t
y
f
0
; 3:3
where t
y
represents the inplane shear strength (yielding). Budiansky [41] con
sidered the kink band geometry (see Figure 3.7) and derived the following estimate
for the kink band angle b in terms of the transverse modulus E
T
and shear
modulus G of a composite layer:
2
p
1
2
G s
c
E
T
<tan
2
b<
G s
c
E
T
: 3:4
To account for shear deformation effects, Niu and Talreja [42] modeled the ber
as a generalized Timoshenko beam with the matrix as an elastic foundation. It was
observed that not only an initial ber misalignment but also any misalignment in
the loading system can affect the critical stress for kinking.
Extensional Mode Shear Mode
Figure 3.6. Extensional and shear modes of ber microbuckling.
43 3.1 Mechanisms of damage
3.1.7 Particle cleavage
If brittle particles (e.g., ceramics) are placed in a ductile but strong and tough
matrix, particle cleavage is the main mode of damage in initial stages of deform
ation. This mode of damage is found in particulate metal matrix composites.
Cleavage refers to the breakage of the reinforcing particle. The cleavage crack
typically forms perpendicular to the global maximum principal stress. The damage
analysis has been performed assuming viscoplastic material behavior [43]. Failure
of many practically relevant particulate twophase composites can typically be
attributed to cleavage fracture of the brittle particles followed by ductile crack
growth in the matrix [44]. To account for particle geometry and distribution,
statistical methods are employed to predict inclusion fracture. To fully character
ize brittle fracture of a particle embedded in a ductile metallic matrix, careful
computational modeling (FEA) sometimes becomes necessary (see, e.g., [4547]).
3.1.8 Void growth
A composite structure may contain an appreciable amount of manufacturing
induced defects. For polymer matrix composites, the defects induced during
manufacturing can be in the ber architecture, e.g., ber misalignment, irregular
ber distribution in the cross section, and broken bers; in the matrix, e.g., voids;
or at the ber/matrix interface, e.g., disbonds and delaminations. Voids are one of
the primary defects found virtually in all types of composite materials. The
formation of voids is controlled by manufacturing parameters, such as vacuum
pressure, cure temperature, cure pressure, and resin viscosity.
The presence of voids, even at low volume fractions, is found to have a
signicant detrimental effect on the overall material behavior. The exural, trans
verse, and shear properties are affected the most. Their shape, size, and distribu
tion also play role in material degradation. Micromechanics homogenization
methods, such as MoriTanaka [48], are commonly used to estimate the average
composite property assuming voids as inclusions with zero properties. More
sophisticated methods have also been developed to analyze the effect of voids on
overall composite elastic and failure properties [49].
Voids can also lead to appreciable inelastic deformations in the material locally,
which can act as precursors to initiation of damage processes, such as crazing,
b
W
j
0
s
s
Figure 3.7. Kink band geometry assumed in Budiansky [41]. Reprinted, with kind
permission, from Computers & Structures, Vol. 16, B. Budiansky, Micromechanics,
pp. 312, copyright Elsevier (1983).
44 Damage in composite materials
shear yielding, brillation, and local fracture. These damage processes in the nal
stage may have signicant inuence on the deformation response and failure
properties of the composite material.
In composites with metallic and polymer matrices, the matrix phase undergoes
ductile fracture due to nucleation, growth and coalescence of voids and cavities.
These voids grow and expand due to high local inelastic strains and high stress
triaxiality in the matrix. Ductile fracture models, such as RiceTracy [50] can be
used to model the initiation and growth of voids in ductile matrices [51]. These
voids can sometimes coalesce to form matrix cracks, and may also cause ber
matrix debonds.
3.1.9 Damage modes
The damage mechanisms described above have different characteristics depending
on a variety of geometric and material parameters. Each mechanism has different
governing length scales and evolves differently when the applied load is increased.
Interactions between individual mechanisms further complicate the damage pic
ture. As the loading increases, stress transfer takes place from a region of high
damage to that of low damage, and the composite failure results from the critical
ity of the last loadbearing element or region. For clarity of treatment, the full
range of damage can be separated into damage modes, treating them individually
followed by examining their interactions.
Which damage mechanisms become active in a given life period of a composite
structure depends mainly on the properties of the base material (e.g., matrix),
architecture, orientation, distribution, and volume fraction of the reinforcing agent
(ber), the properties of the interface, and loading and environmental conditions.
Intralaminar and interlaminar cracking, ber fracture, and microbuckling are the
dominant damage mechanisms in long ber composites. Short ber composites
show three basic mechanisms of interfacial failure [52], as depicted in Figure 3.8:
Mode a: Localized matrix yielding at the interface due to the stress concen
tration at the ber end (see Figure 3.8(a)). Typically, this occurs in combination
with debonding of the ber end and the formation of a pennyshaped crack.
Mode b: If the interface is relatively weak, an interface crack propagates from
the debonded ber end (Figure 3.8(b)). This is different than the ber end
pennyshaped crack and remains closed upon increase in tensile loading on
the composite, and the load transfer occurs by frictional stress transfer.
Mode g: If the interface is relatively strong, a conical matrix crack propagates
from the debonded ber end at an angle y
c
to the ber axis (Figure 3.8(c)). This
matrix crack opens with increasing applied load and suppresses load transfer
across the crack faces.
For particulate composites, the major damage mechanisms are dewetting
(debonding) of the particle and cavity nucleation [53] (see Figure 3.9). At a critical
tensile load, the particles separate from the matrix causing dewetting. Dewetting
45 3.1 Mechanisms of damage
of the particle eventually leads to cavity formation which grows on subsequent
loading. Dewetting introduces volume dilatation and results in nonlinearity in the
stressstrain behavior. For wellbonded particles, cavities and cracks may form
entirely within the matrix [54].
Damage modes in continuous ber laminates are thus rich in complexity. These
will be described below in the context of their evolution with loading.
3.2 Development of damage in composite laminates
A schematic description of damage development in composite laminates in tension
is depicted in Figure 3.10, where the ve identiable damage mechanisms are
indicated in the order of their occurrence. Although the gure is developed on
the basis of fatigue experiments [5560], it provides the basic details for quasi
static loading as well.
In the early stage of damage accumulation, multiple matrix cracking dominates
in the layers which have bers aligned transverse to the applied load direction.
(a)
(b)
(c)
q
c
q
c
Figure 3.8. Failure mechanisms of interfacial failure in short ber/epoxy composites:
(a) mode a; (b) mode b; (c) mode g. Reprinted, with kind permission, from Compos Sci
Technol, Vol. 60, S. Sirivedin, D. N. Fenner, R. B. Nath, and C. Galiotis, Matrix crack
propagation criteria for model shortcarbon bre/epoxy composites, pp. 283547,
copyright Elsevier (2000).
46 Damage in composite materials
Static tensile tests on crossply laminates have shown that the transverse matrix
cracks can initiate as early as at 0.4% applied strain depending upon the laminate
conguration. They initiate from the locations of defects such as voids, or areas of
high ber volume fraction or resin rich areas. Ply cracks grow unstably through
the width direction and quickly span the specimen width. As the applied load is
increased (or the specimen is loaded cyclically), more and more cracks appear. The
accumulation of ply cracks in a cracked ply is depicted in Figure 3.11. Initially
these cracks are irregularly spaced and isolated from each other, i.e., have no
interaction among themselves. However, as cracks become closer they start inter
acting, i.e., the inbetween tensile stresses diminish and can no longer build up to
earlier levels. Thus further increase in load is required to produce new cracks. This
is well illustrated in Figure 3.12 by plots of diminishing crack spacing versus load
or number of cycles. The conguration to which crack density saturates, often
reached only under fatigue loading, has been termed the characteristic damage
state (CDS) [5759]. This state seems to mark the termination of the intralaminar
cracking. The uniqueness of the CDS for a given laminate irrespective of the
loading path has, however, not been found to hold in all cases [61].
Subsequent loading causes initiation of cracks transverse to the primary (intra
laminar) cracks lying in plies adjacent to the ones with those primary cracks (see
Figure 3.10). These cracks, known as secondary cracks, are small in size and they
can cause interfacial debonding, thereby initiating interlaminar cracks. The inter
laminar cracks are also initially small, isolated and distributed in the interlaminar
planes. Subsequently, some interlaminar cracks merge into striplike zones leading
to large scale delaminations. This results into loss of the integrity of the laminate
s s s
s s s
Matrix
Particle
Debond
Cavity
Figure 3.9. Damage mechanisms in particulate composites. Reprinted, with kind permission,
from Int J Solids Struct, Vol. 32, G. Ravichandran and C. T. Liu, Modeling constitutive
behavior of particulate composites undergoing damage, pp. 97990, copyright Elsevier (1995).
47 3.2 Development of damage in composite laminates
in those regions. Further development of damage is highly localized, increasing
unstably, and involving extensive ber breakage. The nal failure event is mani
fested by the formation of a failure path through the locally failed regions and is
therefore highly stochastic.
The damage prior to localization is sometimes referred to as subcritical
damage. The intralaminar (ply) cracking in this stage causes loss of stiffness
properties in the laminate and can by itself lead to loss of functionality (failure)
of the composite structure. The eld of damage mechanics addresses the initi
ation and progression of the subcritical damage. Later chapters will be devoted to
this subject. The next section discusses the phenomenon of multiple cracking and
its effects on overall (average) laminate response.
100
PERCENT OF LIFE
5. Fracture
D
A
M
A
G
E
CDS
0
0
1. Matrix Cracking
0
0
0
0
0
0
3. Delamination
2. Crack coupling
Interfacial debonding
4. Fiber Breakage
Figure 3.10. Development of damage in composite laminates [62].
Figure 3.11. Accumulation of intralaminar cracks in an offaxis ply of a composite laminate.
Based on Xray radiographs reported in [64].
48 Damage in composite materials
3.3 Intralaminar ply cracking in laminates
One of the earliest observations of ply cracking in laminates was reported by
Broutman and Sahu [65]. However, the rst major explanation of multiple matrix
cracking was proposed by Aveston, Cooper, and Kelly [66, 67]. They argued that
multiple fracture occurs in a brous composite when one of the constituents (ber or
matrix) fractures at a much lower elongation than the other and when the unbroken
constituent is able to take the additional load; otherwise single fracture results. Later,
a group of experimentalists, Garret, Bailey, Parvizi, and colleagues [6874] carried
out signicant tests to analyze the ply cracking behavior in crossply laminates. These
experiments showed the initiation of microcracking in glassber reinforced polyster
and epoxy crossply laminates. It was observed that for thick 90
plies, transverse
cracks initiated at the edge of the specimen and propagated instantly through the
width of entire cross section. As the 90
plies
0
0
0
0.2
4.0
6.0
8.0
10.0
C
r
a
c
k
s
p
a
c
i
n
g
(
m
m
)
2.0
0.4 0.6 0.8 1.0
100 200 300 400 500 600 700
Applied stress (MPa)
Cycles ( 10
6
)
Fatigue data
Quasistatic data
Figure 3.12. Spacing of cracks in 45
crossply laminates
of a glass brereinforced polyester. J Mater Sci, 12:1 (1977), 15768.
70. A. Parvizi, K. W. Garrett, and J. E. Bailey, Constrained cracking in glass bre
reinforced epoxy crossply laminates. J Mater Sci, 13:1 (1978), 195201.
71. M. G. Bader, J. E. Bailey, P. T. Curtis, and A. Parvizi, eds. The mechanisms of
initiation and development of damage in multiaxial brereinforced plastic laminates.
Proc Third Int Conf Mech Behav Mater (ICM3), Vol. 3. (Cambridge, 1979), pp. 22739.
72. J. E. Bailey, P. T. Curtis, and A. Parvizi, On the transverse cracking and longitudinal
splitting behaviour of glass and carbon bre reinforced epoxy crossply laminates and
the effect of Poisson and thermally generated strain. Proc R Soc London A, 366:1727
(1979), 599623.
73. J. E. Bailey and A. Parvizi, On ber debonding effects and the mechanism of
transverseply failure in crossply laminates of glass berthermoset composites.
J Mater Sci, 16:3 (1981), 64959.
74. A. Parvizi and J. E. Bailey, Multiple transverse cracking in glassber epoxy crossply
laminates. J Mater Sci, 13:10 (1978), 21316.
75. R. Talreja, Transverse cracking and stiffness reduction in composite laminates.
J Compos Mater, 19:4 (1985), 35575.
76. P. W. Manders, T. W. Chou, F. R. Jones, and J. W. Rock, Statistical analysis of multiple
fracture in [0/90/0] glass ber/epoxy resin laminates. J Mater Sci, 19 (1983), 287689.
77. H. Fukunaga, T. W. Chou, P. W. M. Peters, and K. Schulte, Probabilistic failure strength
analysis of graphite epoxy crossply laminates. J Compos Mater, 18:4 (1984), 33956.
78. H. Fukunaga, T. W. Chou, K. Schulte, and P. W. M. Peters, Probabilistic initial failure
strength of hybrid and nonhybrid laminates. J Mater Sci, 19:11 (1984), 354653.
79. P. S. Steif, Parabolic shear lag analysis of a [0/90]
s
laminate. Transverse ply crack growth
and associated stiffness reduction during the fatigue of a simple crossply laminate. In
S. L. Ogin, P. A. Smith, and P. W. R. Beaumont (eds.), Report CUED/C/MATS/TR
105, Cambridge University, Engineering Department, UK (September 1984).
80. R. J. Nuismer and S. C. Tan, Constitutive relations of a cracked composite lamina.
J Compos Mater, 22:4 (1988), 30621.
81. S. C. Tan and R. J. Nuismer, A theory for progressive matrix cracking in composite
laminates. J Compos Mater, 23:10 (1989), 102947.
55 References
82. Z. Hashin, Analysis of cracked laminates: a variational approach. Mech Mater, 4:2
(1985), 12136.
83. J. A. Nairn, The strain energy release rate of composite microcracking: a variational
approach. J Compos Mater, 23:11 (1989), 110629.
84. J. Varna and L. A. Berglund, Multiple transverse cracking and stiffness reduction in
crossply laminates. J Compos Tech Res, 13:2 (1991), 97106.
85. L. N. McCartney, Theory of stress transfer in a 0degrees90degrees0degrees cross
ply laminate containing a parallel array of transverse cracks. J Mech Phys Solids, 40:1
(1992), 2768.
86. P. Gudmundson and W. L. Zang, An analytic model for thermoelastic properties of
composite laminates containing transverse matrix cracks. Int J Solids Struct, 30:23
(1993), 321131.
87. E. Adolfsson and P. Gudmundson, Thermoelastic properties in combined bending and
extension of thin composite laminates with transverse matrix cracks. Int J Solids Struct,
34:16 (1997), 203560.
88. P. Lundmark and J. Varna, Constitutive relationships for laminates with ply cracks in
inplane loading. Int J Damage Mech, 14:3 (2005), 23559.
56 Damage in composite materials
4 Microdamage mechanics
4.1 Introduction
As explained in the previous chapter, damage affects the overall stressstrain
response of the solid continuum body. Damage mechanics pertains to the study
of this effect. Two widely different subelds have emerged over the years in this
eld. One concerns study of damage directly at the scale of formation of cracks,
i.e., the microstructural scale, and hence can be called microdamage mechanics
(MIDM). The other approach, on the contrary, looks at the overall response at
the macro or structural scale by using some internal variables to characterize
damage, and thus can be termed as macrodamage mechanics (MADM). These
terms were originally coined by Hashin [1]. MADM is the same as continuum
damage mechanics (CDM), which is still the commonly used terminology.
MIDM for composite materials is derived from an older and more mature eld
called micromechanics that deals with overall properties of heterogeneous mater
ials (see, e.g., [2]). In micromechanics one views heterogeneities such as inclusions
and voids as microstructure and estimates overall properties by various
methods, e.g., averaging schemes such as selfconsistent and differential schemes,
or variational methods to obtain bounds to average properties. Microcracks are
treated as limiting geometry of microvoids, such as ellipsoidal voids with one
dimension much smaller than the other two. As illustrated in the previous chapter,
damage in composite materials has signicant complexities concerning the
geometry as well as evolution characteristics such as multiplication of cracks
within a xed volume. For these reasons a simple extension of micromechanics
to damage in composites is generally not possible. A separate eld identied as
MIDM has therefore emerged. This chapter will treat the features of MIDM that
have been developed to specically treat certain cases of damage in composite
materials. Since determining local (microlevel) stress or displacement elds is a
necessary feature of micromechanics, it is expected that not all cases within the
wide range of damage in composites can be handled by MIDM. However, this
limitation can be alleviated by incorporating computational solutions of the local
stress or displacement elds, thereby broadening classical micromechanics to
include socalled computational micromechanics. In the most recent versions of
MIDM this strategy has been used. More on this will be discussed toward the end
of this chapter.
In the following we will rst treat the aspect of damage in composite mater
ials that is due to the presence of continuous interfaces between dissimilar
materials, such as bers and matrix or plies oriented differently in a laminate.
In fact this aspect is fundamental to understanding damage in composite
materials. Historically, it was rst analyzed in a classical work by Aveston,
Cooper, and Kelly [3] who explained conditions that lead to failure of a
composite from a single crack versus when multiple cracks precede the nal
failure. Their work has become known as the ACK theory. Although the case
treated by them is of simple geometry and loading, namely a unidirectional
berreinforced brittle matrix composite loaded in tension along bers, it
explains the basic mechanism underlying multiple cracking in a wide range of
cases. The simplied stress analysis and the associated energy balance consider
ations in the ACK paper have later been extended to include more accurate
solutions, but little further insight into the multiple cracking mechanism has
resulted by these efforts.
4.2 Phenomena of single and multiple fracture: ACK theory
The mode of failure (separation in two or more pieces) in homogeneous mater
ials such as metals and ceramics may be described as single fracture in the
sense that the failure is attributable to a single source a crack. Heterogeneous
materials, on the other hand, can fail in the mode of single fracture or sustain
multiple fractures of one of the phases before ultimately separating in two or
more pieces. The latter phenomenon is known as multiple fracture and can
commonly occur in brous composites with brittle matrices, such as cement
plaster, glass, etc.
Damage in composite materials usually initiates with matrix cracking.
Figure 4.1 depicts the isotropic stressstrain response of an unreinforced glass
and of a berreinforced glass in tension loading along bers. Other than the
enhancement of the stressstrain response in the ber direction, a striking aspect is
the nonlinearity shown by the reinforced specimen. This nonlinearity is a type of
ductility which occurs due to multiple cracking of the matrix (glass) [4].
Although the phenomenon of multiple fracture was observed earlier, e.g., by
Cooper and Sillwood [5] it was systematically investigated in a landmark paper by
Aveston, Cooper, and Kelly [3]. Their work, the ACK theory, is the basis of the
treatment presented below.
Consider a unidirectional brous composite loaded in tension along bers as
shown in Figure 4.2 and assume the following:
1. Fibers are of the same diameter and uniformly distributed in the matrix.
2. All the bers are aligned parallel to one another.
3. There are no preexisting aws in the matrix such as voids and cracks.
4. Both the matrix and the bers are linearly elastic.
58 Microdamage mechanics
Assuming that bers and matrix have different failure strain in tension, when one
of the constituents fails, the other will either fail simultaneously or continue
deforming by carrying the additional load. In the latter case, the constituent that
failed rst will fail again at a different site. Thus, there are two necessary condi
tions for multiple fracture to occur in a composite:
1. One of the constituents has a lower failure strain than the other.
2. When the weaker constituent fails, i.e., when it no longer carries any load,
the stronger constituent must be able to carry the additional load thrown
upon it.
If P
c
is the total tensile load on the composite and P
f
and P
m
represent the load
taken up by the bers and matrix, respectively, then by force balance we have
P
c
= P
f
P
m
: (4:1)
s
m
s
f
s
c
Figure 4.2. A unidirectional brous composite loaded in tension.
0
200
S
t
r
e
s
s
/
M
P
a
400
600
800
1000
0.4 0.8 1.2
Strain (%)
Unreinforced glass
Fibre
reinforced
glass
Figure 4.1. The stressstrain curves for borosilicate glass alone (dotted line) and
reinforced with aligned carbon bers. Reprinted from [4] with kind permission
from Royal Society Publishing, London.
59 4.2 Phenomena of single and multiple fracture: ACK theory
Dividing by the composites crosssectional area A, we get
P
c
A
=
P
f
A
P
m
A
; (4:2)
or
P
c
A
=
P
f
A
f
A
f
A
P
m
A
m
A
m
A
: (4:3)
Assuming unit composite length, we obtain
s
c
= s
f
V
f
s
m
V
m
; (4:4)
where V
f
and V
m
are the volume fractions of bers and matrix, respectively.
Depending upon which phase fails rst, two cases as shown in Figure 4.3 arise. Case
1 is whenbers have a lower breaking strainthan the matrix (e
fu
<e
mu
), while Case 2 is
for the opposite (e
mu
< e
fu
). In the rst case, bers will undergo multiple fracture if the
matrix is able to carry the additional load thrown upon it due to ber failure, i.e., if
P
mu
> P
c
[
e
fu
;
i:e:; if s
mu
V
m
> s
fu
V
f
s
/
m
V
m
(4:5)
where P
mu
is the maximum load that can be carried by the matrix, s
fu
and s
mu
are the tensile strength values for bers and matrix, respectively, and
s
/
m
= E
m
e
fu
= (s
mu
=e
mu
)e
fu
is the stress inthe matrix requiredtoproduce a strainequal
to the breaking strain of the bers. In this case, the bers will be successively fractured
intoshorter lengths until the matrix attains its failure strainandat that instant the whole
compositefails. Ontheother hand, for Case2, thematrixwill undergomultiplefractureif
P
fu
> P
c
[
e
mu
;
i:e:; if s
fu
V
f
> s
mu
V
m
s
/
f
V
f
;
(4:6)
A
Multiple
fracture
of fibers
s s
s
f
Single
fracture
V
f
s
m
f
s
m
(a)
A
Multiple
fracture
of matrix
Single
fracture
V
f
(b)
s
m
Figure 4.3. Single and multiple fractures in a unidirectional composite. Fracture stress is
plotted against the ber volume fraction: (a) case 1: e
mu
> e
fu
; (b) case 2: e
mu
< e
fu
.
60 Microdamage mechanics
where P
mu
is the maximum load that can be carried by the matrix, s
fu
and s
mu
are the tensile strength values for bers and matrix, respectively, and
s
/
f
= E
f
e
mu
= (s
fu
=e
fu
)e
mu
is the stress in the bers required to produce a strain
equal to the breaking strain of the matrix.
4.2.1 Multiple matrix cracking
From this point on we will focus on the case of multiple matrix cracking, assuming
e
mu
< e
fu
. In addition to the assumptions described in the previous section, the
following analysis will assume:
1. Fibers remain intact throughout entire loading history.
2. Matrix cracks extend in the entire cross section.
3. Fibers debond completely between adjacent matrix cracks.
If we concentrate on the matrix region between two bers, the force P
m
shed by the
matrix, subsequent to its failure, is carried by bers in the cracked cross section,
and is transferred back to the matrix over a distance x
/
. This load transfer takes
place through shear at the ber/matrix interface with constant shear stress t. The
mechanism of interfacial load transfer is illustrated in Figure 4.4. The load balance
between the total matrix load P
m
and the total shear force at the bermatrix
interfaces yields
P
m
= s
mu
A
m
= t 2pr x
/
N; (4:7)
where r is the ber radius and N is the number of bers in the composite cross
section of area A.
P
m
x
s
mu
E
f
V
m
V
f
= e
f
s
mu
E
m
E
m
E
f
V
m
V
f
= e
mu
1 a ( ); (4:13)
with
a =
E
m
E
f
V
m
V
f
: (4:14)
The mean strain over crack spacing 2x
/
is equal to
e
mean;2x
/ =
1
2
e
mu
e
mu
1 a ( ) [ [ = e
mu
1
a
2
_ _
: (4:15)
When the crack spacing reduces to x
/
, the mean strain increases to
e
mean;x
/ = e
mu
1
3a
4
_ _
: (4:16)
62 Microdamage mechanics
Energy considerations in multiple cracking
Consider the unidirectional composite of Figure 4.2 at xed applied load P
c
=
s
c
A. Let its initial conguration be denoted as state 1 and let state 2 refer to its
conguration with multiple matrix cracking. The energy changes in going from
state 1 to state 2 are described below.
Supply of energy
1. DW: the work done per unit crosssectional area A by an external (xed) load
through specimen extension caused by cracking is given by
W =
1
A
P
c
2x
/
= s
c
2 e
mu
a
2
x
/
_ _
= E
c
e
mu
e
mu
ax
/
= E
c
e
2
mu
ax
/
:
(4:17)
2. DU
m
: the reduction in elastic strain energy of matrix over distance 2x
/
is given by
U
m
= 2
_
x
/
0
1
2
E
m
V
m
e
2
mu
1
2
E
m
V
m
e
mu
x
x
/
_ _
2
_ _
dx
=
2
3
E
m
V
m
e
2
mu
x
/
=
E
m
V
m
3t
e
3
mu
ar:
(4:18)
Consumption of energy
1. Energy spent in formation of matrix cracks. If g
m
is the surface energy per unit
area of crack surface, then the energy spent in formation of a matrix crack per
unit crosssectional area A is
2g
m
A
m
A
= 2g
m
V
m
: (4:19)
2. Energy spent in ber/matrix interfacial debonding. If we take G
II
to be the
energy released per unit area of the debond surface, then the debond energy g
db
per unit crosssectional area A can be expressed as
g
db
A = G
II
2pr 2x
/
N; (4:20)
i.e.,
g
db
= G
II
2prN
A
2x
/
= 2G
II
A
f
Ar
V
m
V
f
s
mu
t
r
= 2G
II
V
m
s
mu
t
:
(4:21)
3. U
s
: the energy spent in sliding of the matrix onto the ber surface over a
distance 2x
/
, per unit crosssectional area A is given by
U
s
=
1
A
N 2
_
x
/
0
v t 2pr dx (4:22)
63 4.2 Phenomena of single and multiple fracture: ACK theory
where Dn is the sliding displacement at x. This sliding displacement is equal to the
difference in displacements of the ber and the matrix. It can be found by
integrating the strain in matrix and ber:
_
x
/
0
tv dx = t
_
x
/
0
e
mu
1 a ( )x
a
2
x
2
x
/
x
/
1
a
2
_ _
x
2
x
/
x
/
2
_ _
dx
=
te
mu
x
/
2
6
1 a ( ):
(4:23)
Thus,
U
s
=
E
f
E
m
V
m
6t
e
3
mu
ra 1 a ( ): (4:24)
4. DU
f
: the increase in the elastic energy of bers due to additional extension
caused by additional ber stress, per crosssectional area A, is given by
U
f
= U
(2)
f
U
(1)
f
= 2
_
x
/
0
1
2
E
f
V
f
e
mu
a 1
x
x
/
_ _
e
mu
_ _
2
1
2
E
f
V
f
e
2
mu
_ _
dx
= E
f
V
f
e
2
mu
x
/
a 1
a
3
_ _
=
E
f
E
m
V
m
2t
e
3
mu
ra 1
a
3
_ _
:
(4:25)
Conditions for multiple matrix cracking
1. Stress in the matrix is greater than or equal to the matrix failure stress, i.e.,
s
m
_ s
mu
or e
m
_ e
mu
: (4:26)
2. The supply of energy in going from state 1 to state 2 is greater than or equal
to the consumption of energy, i.e.,
2g
m
V
m
g
db
U
s
U
f
_ W U
m
: (4:27)
Substituting Eqs. (4.17)(4.25) derived above into Eq. (4.27), one obtains
2V
m
g
m
G
II
s
mu
t
_ _
_
E
c
E
f
e
3
mu
a
2
r
6t
: (4:28)
It can be argued that the energy term G
II
is much smaller than the other energy
contributions. Assuming G
II
= 0 then gives
2V
m
g
m
_
E
c
E
f
e
3
mu
a
2
r
6t
: (4:29)
Thus, the strain required to cause multiple matrix cracking is given by the
following expression
e
muc
=
12tg
m
E
f
V
2
f
E
c
E
m
V
m
r
_ _
1=3
: (4:30)
64 Microdamage mechanics
Stressstrain response
When the composite is loaded to an applied strain level equal to the failure strain of
the matrix, multiple cracking in the matrix starts occurring. If the matrix has a well
dened singlevalued breaking strain, the cracking will continue at a constant applied
stress E
c
e
mu
until the matrix is broken down into a set of blocks of length between x
/
and 2x
/
. The composite stressstrain behavior subsequent to multiple fracture of the
matrix is shown in Figure 4.5. During multiple matrix cracking (Point A to Point B),
the mean strain varies between e
mu
1
a
2
_ _
and e
mu
1
3a
4
_ _
while going from a crack
spacing of 2x
/
to x
/
. The total strain at the limit of multiple cracking e
mc
is therefore
e
mu
1
a
2
_ _
<e
mc
<e
mu
1
3a
4
_ _
: (4:31)
When the applied load is increased, bers are stretched further and start slipping
through the blocks of matrix (point Bto Point C). Since the matrix can no longer take
any load the Youngs modulus of the specimen is reduced to E
f
V
f
. The composite will
eventually fail at a stress s
fu
V
f
. The failure strain of the composite e
cu
is given by
e
fu
ae
mu
2
_ _
<e
cu
< e
fu
ae
mu
4
_ _
: (4:32)
4.2.2 Perfectly bonded ber/matrix interface: a modied shear lag analysis
The previous analysis was based on the assumption that the bers debond com
pletely from the matrix during the fracture process. In reality the displacements of
matrix and bers are interrelated. The complete debonding scenario can be viewed
as one extreme, the other extreme being the fully bonded case. For the latter, a
modied shear lag model [6] is applicable.
1+
A
C
fu f
V s
s
D
O
B
1+
4
3a
4 2
2
E
c
E
f
V
f
ae
mu
ae
mu
mu
e
mu
e
mu
e
fu
e
fu
e
e
fu
e
a
Figure 4.5. The stressstrain response subsequent to multiple fracture for a unidirectional
composite according to the ACK theory.
65 4.2 Phenomena of single and multiple fracture: ACK theory
For both cases the fundamental equation governing load transfer between bers
and matrix is obtained from a simple force balance and is, for discrete bers of
radius r in a continuous matrix, given by
dF
dy
=
2V
f
t
i
r
; (4:33)
where dF is the load per crosssectional area A transferred over the distance dy and
t
i
is the shear stress acting at the interface.
For the unbonded case, the load transfer from ber to matrix is given by
Eq. (4.11), whereas for the bonded case the bers in the plane of the rst crack
will be subjected to an additional stress, Ds
0
, given by
s
0
=
s
a
V
f
E
f
e
mu
; (4:34)
where s
a
is the applied stress. This additional stress has its maximum value at the
plane of the matrix crack and decays with distance from the crack surface. Using a
modied shear lag analysis, Aveston and Kelly [6] showed that the variation of
this additional stress along the ber is given by
s y ( ) = s
0
e
w
_
y
; (4:35)
with
w =
2G
m
E
c
E
f
E
m
V
m
_ _
1
r
2
ln
R
r
_ _ ; (4:36)
where G
m
is the shear modulus of the matrix.
Carrying out a force balance over an element of ber of length dy,
ds pr
2
t
i
2pr dy = 0; (4:37)
from which the rate of change of stress in the ber is given by
ds
dy
=
2
r
t
i
: (4:38)
Differentiating Eq. (4.35) w.r.t. y and substituting into Eq. (4.38), the shear stress
at the interface between the bers and matrix is given by
t
i
=
r
2
s
0
we
w
_
y
: (4:39)
Substituting Eq. (4.39) into Eq. (4.33), and integrating, the load F per cross
sectional area A transferred to the matrix over any distance l from the crack
surface can be found as
F = V
f
s
0
1 e
w
_
l
_ _
: (4:40)
66 Microdamage mechanics
If Ds
0
_ s
mu
(V
m
/V
f
), the matrix will undergo further cracking into blocks of
length between l and 2l, where l is obtained by setting F = s
mu
V
m
into Eq. (4.40),
l =
1
w
_ ln 1
s
mu
s
0
V
m
V
f
_ _
: (4:41)
The energetic considerations in this case result into the crack initiation strain for
the matrix, as given by
e
mu
=
2g
m
V
m
w
_
aE
c
_ _
1=2
: (4:42)
The effective Youngs modulus for the matrix can be determined by averaging the
stress distribution in the matrix, i.e.,
E
m
=
1
e
mu
1
s
_
s
0
s
m
y ( )dy; (4:43)
where 2s is the mutual crack spacing, and the stress distribution in the matrix is
given by
s
m
y ( ) =
E
m
e
mu
s
V
f
V
m
: (4:44)
Using Eqs. (4.34), (4.43), and (4.44),
E
m
can be derived as [7]
E
m
= E
m
E
m
s
w
_ e
s
w
_
1
_
: (4:45)
The effective modulus for the unidirectional composite can then be determined
using the rule of mixture
E
c
= E
f
V
f
E
m
V
m
: (4:46)
4.2.3 Frictional ber/matrix interface
As suggested by Wang and ParviziMajidi [7], another important case is to assume
that the ber/matrix interface is frictional. For this case, the load transfer between
ber and matrix is assumed constant. After the development of matrix cracking,
the additional stress transferred to the ber at the crack surface should be
transferred back to the matrix by a constant interfacial shear stress over the
distance of the limiting matrix crack spacing, x. Accordingly, the stress in the
matrix away from the crack surface will vary linearly, from a zero value at the crack
to a maximum of s
m
at a distance x from the crack. This stress will stay at its
maximum level for distances longer than x. The average stress and the stiffness
values in the matrix can therefore be represented by the following equations
s
m
= s
m
2s x
x
;
E
m
= E
m
2s x
x
:
(4:47)
The overall stiffness for the composite can then be calculated from Eq. (4.46).
67 4.2 Phenomena of single and multiple fracture: ACK theory
The onedimensional analyses of stress transfer used in the ACK theory and its
later version are inadequate to give accurate prediction, e.g., of the strain to onset
of multiple cracking. The main value of these works was in explaining the phe
nomenon of multiple cracking. In subsequent sections we will address the problem
of multiple cracking more generally and describe a wide range of approaches
taken since the appearance of the ACK theory.
4.3 Stress analysis (boundary value problem) for cracked laminates
4.3.1 Complexity and issues
Damage analysis of cracked laminates of a general layup is a highly complex task. The
major issues in analyzing damage in a multidirectional laminate are discussed below:
1. Anisotropy and heterogeneity: The commonly used laminate analysis is based
on the assumption that the plies are homogeneous and anisotropic, with the
effective properties of a unidirectional composite. This is a valid assumption
for undamaged laminates for the usual cases of membrane force and moment
loading when the variation of the primary, inplane, laminate stresses through
the ply thickness is constant or linear. But the presence of intralaminar cracks
may cause local conditions such as high stress gradients to invalidate assump
tions needed for homogenization of plies [1]. Furthermore, an inplane stress
analysis will be inadequate to account for interactions between cracks in
different plies.
2. Stress singularity: An assumption of ideal cracks within plies will produce
stress singularity. In reality, however, the crack tips will be blunted due to the
presence of nitely sized bers near the ply interfaces and by local ow of the
matrix at crack tips. Hence, the actual stress eld in the vicinity of the crack
tip may not be accurately given by the usual stress analysis of cracks in
homogeneous elastic medium. In such instances, numerical approaches such
as the nite element method may become necessary.
3. Interaction between cracks: At sufcient crack densities, the stress elds
around two adjacent cracks in a ply start interacting, thereby relaxing the
region between those two cracks. This crack interaction affects the crack
surface displacements as well as the overall stress elds. Accurate modeling
of crack interaction is a complex task.
4. Threedimensionality of the boundary value problem: Following all the
previous points, and realizing that in a real scenario the ply cracks in a
general laminate may be curved, irregularly spaced, or not fully grown
through the laminate width, a complex 3D boundary value problem
(BVP) arises. In the simpler case of cracked crossply laminate, assuming
that cracks are periodic, straight, and fully grown through the ply thickness
and width, the resulting BVP can be reduced to a generalized plane strain
problem. For the offaxis ply cracking, however, the stress analysis problem
68 Microdamage mechanics
is still a truly 3D BVP, and any reduction to a 2D problem will not produce
accurate predictions.
5. Dening RVE size: In homogenizing a heterogeneous solid such as a composite
ply the RVE size must be large enough to contain sufciently many bers to
provide average properties. With bers of typically 0.01 mm in diameter, and a
ply thickness of typically 0.125 mm, the RVE extending across a ply thickness
may or may not sufce, depending on the ber volume fraction and ber
distribution irregularity, but it is implicitly assumed to sufce in the classical
laminate theory. However, when cracks appear within a ply, the local stress
gradients increase sharply, leading to a breakdown of the homogenized ply
properties. Away from the cracks, nevertheless, the properties hold. In
obtaining the overall (average) composite properties with multiple cracks, the
RVE size must be large enough to contain a representation of the cracks. To
satisfy this requirement the RVE must extend in the laminate length direction
while it is limited in the thickness direction by the laminate thickness.
6. Multiscale considerations: Connected with the RVE issue is the consideration
of the multiscale nature of the stress and failure analysis. The RVE scale is the
socalled meso (intermediate) scale, while the scale of discrete entities within
the RVE is the micro scale. The scale at which structural analysis is conducted
is the macro scale. Most of the MIDM is concerned with determining changes
of mesoscale properties using stress elds at the microscale. In fact the
analyses described in the following mostly assume uniform distribution of
cracks, giving repeating unit cells for BVP solutions.
7. Constraint effects: In a cracked laminate, stress perturbations are caused by
the surface displacements of the ply cracks in response to the applied loading.
These surface displacements do not occur freely, as they would if the cracks
were to lie in a homogeneous ply of innite thickness, but are affected by
constraint from the neighboring plies. Understanding these constraint effects
is the key in determining the effective properties of cracked laminate. They
will be discussed in further detail in a subsequent chapter.
8. Complexity of offaxis ply cracking: Unlike in crossply laminates, intralami
nar cracking in offaxis plies of orientations other than 90
can be complex.
Microscopic observations suggest that depending on the offaxis angle, ply
thickness, and orientation of neighboring plies these cracks may be partially
grown and erratic in shape and size, and of nonuniform distribution [8, 9].
Raman Spectroscopy experiments on [0/45]
s
laminates show that a crack
developing in the 45

ply of a crossply laminate [10], which seemed to suggest that the initiation and
propagation strains for a 45
ply, the cracks usually initiate in that ply under axial tension, while cracks
in other offaxis plies initiate at higher loading. The observations on multidir
ectional laminates indicate that the angle between the two adjacent plies may
have signicant effects on damage initiation and progression. When this angle
is small, partially formed cracks are observed before propagation in the ber
69 4.3 Stress analysis (boundary value problem) for cracked laminates
direction. However, fully developed cracks mainly form in the cases of large
intersecting angles [11, 12]. The damage development in the 60
ply of a [0/60
2
/
90]
s
laminate is shown in Figure 4.6. Moreover, shearextension coupling may
introduce some additional complexity in analysis of offaxis laminates [13, 14].
9. Randomness in cracking process: In general, damage models assume a uni
form distribution of transverse cracks, i.e., they are assumed to be periodic
and selfsimilar. In reality the variation in crack spacing can be considerable,
particularly at large crack spacing. Recently, there have been some develop
ments to account for the inuence of the spatial nonuniformity of matrix
cracking on the stress transfer and the effective mechanical properties of
cracked crossply laminates [1517].
10mm
90
60
(a)
x
= 315[MPa]
x
= 0.71%
x
= 347[MPa]
x
= 0.80%
x
= 388[MPa]
x
= 0.89%
x
= 428[MPa]
x
= 1.00%
(b)
(c)
(d)
Figure 4.6. Consecutive matrix cracking behavior in contiguous plies in a [0/60
2
/90]
s
laminate. Parts (a)(d) show damage state at different levels of applied strain. Reprinted,
with kind permission, from Int J Solids Struct, Vol. 42, T. Yokozeki, T. Aoki, and
T. Ishikawa, Consecutive matrix cracking in contiguous plies of composite laminates,
pp. 2785802, copyright Elsevier (2005).
70 Microdamage mechanics
10. Multiple damage mechanisms: In general, composite laminates can display
a multitude of damage modes. Here, our focus is on ply cracking, which
usually occurs much before other damage mechanisms such as delamina
tions and ber fracture. The inuence of manufacturinginduced defects
such as voids and ber clusters may further complicate the analysis. These
interactions may become important for failure analysis and have been
considered in some studies [1824].
The above list of complex issues concerning cracked laminates, even for simple
congurations, makes the task of stress and failure analysis daunting. The
accuracy of stress analysis by itself may not always be the answer to the
engineering problem at hand. Often approximations and simplications need to
be made in a judicious manner guided by the application. In the following the
efforts made at analyzing damage and its effects are treated in increasing order of
complexity.
4.3.2 Assumptions
Consider a crossply laminate loaded in tension (Figure 4.7(a)). At a certain value
of the load, it develops transverse cracks in the 90
ply
2t
90
= Thickness of 90
o
ply
2h = 2 t
0
t
90
( ) = Total laminate thickness
2l = Average spacing between two adjacent cracks
(4:49)
t
0
t
90
h
x
z
2l
90
0
0
c
(a)
(b)
N
xx
N
xx
x
z
y
O
Figure 4.7. Construction of a unit cell for stress analysis of a cracked crossply laminate:
(a) cracked laminate in tension; (b) equivalent 2D unit cell.
72 Microdamage mechanics
v Loading:
N
xx
=Applied tensile load per unit width in xdirection
(inplane stress resultant)
(4:50)
v Stresses and strains:
s
c
= Total applied stress for laminate along xdirection
s
0
xx
= Total xdirection stress in 0
ply
s
90
xx
= Total xdirection stress in 90
ply
s
0
xx0
= Initial (virginlaminate) xdirection stress in 0
ply
s
90
xx0
= Initial (virginlaminate) xdirection stress in 90
ply
(4:51)
Other notations will be dened as and when needed. The boundary value problem
can be stated as:
For a cracked crossply laminate loaded in tension determine the displacement and stress
elds which satisfy equilibrium and boundary conditions, and further determine its effective
stiffness properties for a xed state of damage (given crack spacing).
That is, determine s
ij
which satisfy the following conditions:
1. Force balance:
N
xx
= s
c
A: (4:52)
2. Equilibrium conditions:
s
ij; j
= 0: (4:53)
3. Boundary and continuity conditions:
Laminate midplane symmetry: s
90
xz
x; 0 ( ) = 0
Traction continuity across interface: s
90
xz
x; t
90
( ) = s
0
xz
x; t
90
( )
s
90
zz
x; t
90
( ) = s
0
zz
x; t
90
( )
Tractionfree boundary: s
0
xz
x; h ( ) = 0
s
0
zz
x; h ( ) = 0
Tractionfree crack surfaces: s
90
xz
l; z ( ) = 0; t
90
_ z _ t
90
s
90
xx
l; z ( ) = 0; t
90
_ z _ t
90
:
(4:54)
4.4 Onedimensional models: shear lag analysis
A class of onedimensional models that has been useful in the analysis of
multiple cracking is the socalled shear lag models. Although the stress analysis
is inherently inaccurate in these models, their ability to capture the stress
73 4.4 Onedimensional models: shear lag analysis
transfer at the interface by shear stress makes them useful. Historically, the
shear lag analysis was rst used by Cox [25] to describe the stress transfer
between a ber and a matrix for discontinuous ber composites. He considered
an axisymmetrical model of a single ber embedded in the matrix. Later,
Aveston and Kelly [6] modied the model for predicting strain to initiate
multiple matrix cracking in a unidirectional ber composite with brittle matrix.
A recent work considering axisymmetrical model has shown that Coxs original
result can be derived by a series of approximations to the elasticity theory [26].
Garrett, Bailey, and Parvizi [27, 28] and Manders et al. [29] adopted Coxs
approach for the particular case of transverse cracking. Most of the early
studies on transverse cracking used shear lag analysis for the unit cell shown
in Figure 4.7(b). There have been many modications and extensions of the
same analysis. For a more detailed discussion on shear lag analysis, the reader
is referred to [3058].
All shear lag analyses are based on the following basic concept:
In the plane of a transverse crack the transverse ply does not carry the axial load,
while away from the crack a part of this load is transferred back to the transverse ply
by axial shear at the interface between the cracked transverse ply and the adjacent
uncracked ply.
The shear lag analysis is essentially a 1D analysis and it uses the following basic
assumptions:
1. The axial shear stress, t
xy
a dn/dx, where v is the displacement in the
axial, ydirection. This violates the relationship of linear elasticity,
t
xy
= G
xy
dn
dx
du
dy
_ _
. Hence, this is equivalent to assuming that du/dy = 0 or
du/dy dn/dx.
2. The axial normal stress remains constant over the thickness of the transverse
ply after cracking. In other words, concentration of local stress near cracks is
neglected.
3. Cracks remain sufciently far apart so that their mutual interactions can be
neglected.
The shear lag analyses will be presented here with consistent notations and
expressions that may differ in form from those given in the original articles.
4.4.1 Initial shear lag analysis
Here we describe the analysis presented by Garrett, Bailey, and Parvizi [27, 43]
and modied by Manders et al. [29] to account for the presence of neighboring
cracks. For convenience, we follow a treatment similar to the one described in a
review paper by Berthelot [41].
The objective is to determine the variation of the axial (xdirection) stress
in the transverse plies on crack formation in these plies. To begin, one
74 Microdamage mechanics
assumes that the xdisplacement in the 0
ply, u
90
is the xdisplacement at midplane,
and G
90
xz
is the transverse shear modulus of the 90
ply
across the thickness (zdirection). Substituting Eq. (4.56) into Eq. (4.55), one
obtains
ds
90
xx
dx
= G
90
xz
u
90
u
0
t
2
90
_ _
: (4:57)
The axial stresses in the 0
 and 90
 and 90
plies are:
s
0
xx
= E
0
x0
e
0
xx
; with e
0
xx
=
du
0
dx
;
s
90
xx
= E
90
x0
e
90
xx
; with e
90
xx
=
du
90
dx
;
(4:60)
90
0
0
90
xx
x
dx
90
xx
+
90
xx
d
Figure 4.8. Stresses acting on an element of 90
ply.
75 4.4 Onedimensional models: shear lag analysis
where E
0
x0
= E
1
and E
90
x0
= E
2
are the initial axial Youngs moduli of the 0
 and
90
1
lE
0
x0
_ _
; (4:62)
where G
90
xz0
is the initial inplane shear modulus of the 90
ply and E
x0
= E
c
is the axial modulus of the undamaged laminate given by the rule of
mixtures as
E
x0
=
lE
0
x0
E
90
x0
1 l
: (4:63)
This value can be estimated accurately by using the classical laminates plate
theory. Since the crack surfaces are traction free, the axial stress in the transverse
ply vanishes on the crack planes (x = l). Hence, the solution of Eq. (4.61) is
given by
s
90
xx
= s
c
E
90
x0
E
x0
1
cosh b
x
t
90
cosh b
l
t
90
_
_
_
_
_
_
_
_
: (4:64)
Thus b appears as a load transfer parameter, and is sometimes known as the shear
lag parameter.
A quite similar shear lag analysis was conducted by Dvorak et al. [44], except
that they included residual thermal stresses and in place of Eq. (4.55) assumed that
the shear stress is given by
t = K u
90
u
0
( ); (4:65)
where K is a shear parameter, which is to be determined from experimental data.
Dvorak et al. [44] suggested determining b from the experimentally meas
ured stress at rst ply failure. The differential equation for the axial stress in this
case is
d
2
s
90
xx
dx
2
b
2
t
2
90
s
90
xx
=
b
2
t
2
90
s
90
xxR
E
90
x0
E
x0
s
c
_ _
; (4:66)
where the shear lag parameter is now given by
b
2
=
Kt
90
t
0
E
0
x0
t
90
E
90
x0
_ _
t
0
E
0
x0
E
90
x0
= Kt
90
1
E
90
x0
1
lE
0
x0
_ _
; (4:67)
76 Microdamage mechanics
and the solution obtained for axial stress in the transverse ply is
s
90
xx
= s
90
xxR
s
c
E
90
x0
E
x0
_ _
1
cosh b
x
t
90
cosh b
l
t
90
_
_
_
_
_
_
_
_
; (4:68)
where s
90
xxR
is the axial residual thermal stress in the 90
ply.
4.4.2 Interlaminar shear lag analysis
Based on extensive experimental observations of cracks, Highsmith and Reifsnider
[30] found that shear deformations in a given ply were restricted to a thin region
near interfaces between adjacent plies. This region is resin rich and thus is less stiff in
its response to shear stress than the central portion of the ply. Transverse cracks
were observed to extend up to the region, but not into it. Based on these observa
tions, interlaminar shear lag models were developed [30, 45], in which shear stresses
were assumed to develop only within this thin region, whose thickness and shear
modulus are unknown. The unit cell for such models is shown in Figure 4.9.
We present here the interlaminar shear lag analysis reported by Fukunaga et al.
[45], who incorporated the thermal residual stresses and Poissons effect into the
analysis. The displacements in the x and ydirections of each ply are still assumed
to be constant across thickness, as above, but are expressed as
u
0
= e
c
x U
0
x ( ); v
0
=
A
12
A
22
e
c
y V
0
x ( );
u
90
= e
c
x U
90
x ( ); v
90
=
A
12
A
22
e
c
y V
90
x ( );
(4:69)
where the coefcients A
ij
are the inplane stiffness components of half the crossply
laminate. These are the same as the components of the [A] matrix in classical
laminates plate theory, and can be expressed in terms of the reduced stiffness
components Q
ij
of the 0
 and 90
plies as
A
11
= Q
11
t
0
Q
22
t
90
; A
12
= Q
12
h; A
22
= Q
22
t
0
Q
11
t
90
: (4:70)
t
0
t
90
h
x
z
2l
90
c
t
s
Figure 4.9. Unit cell for interlaminar shear lag analysis.
77 4.4 Onedimensional models: shear lag analysis
These relations suppose that the 0
 and 90
ply is obtained as
s
90
xx
= s
90
xxR
s
c
Q
22
E
0
x
_ _
1
Q
12
A
12
Q
22
A
22
_ _
1
cosh b
x
t
90
cosh b
l
t
90
_
_
_
_
_
_
_
_
; (4:75)
where the shear lag parameter b is given by
b
2
= G
t
90
t
s
1
Q
22
1
lQ
11
_ _
: (4:76)
The interlaminar shear lag models by Highsmith and Reifsnider [30] differ from
Fukunagas analysis [45] in their denition of the shearlag parameter. For the
HighsmithReifsnider model,
b
2
= G
t
90
t
s
1
E
90
x0
1
lE
0
x0
_ _
: (4:77)
It can be seen that the expressions for the axial stress in the 90
ply is given by
s
90
xx
= s
90
xx0
1 a
1
e
bx=t
90
a
2
e
bx=t
90
_ _ _ _
; (4:78)
where
a
1
=
1 e
2bl=t
90
e
2bl=t
90
e
2bl=t
90
; a
2
=
e
2bl=t
90
1
e
2bl=t
90
e
2bl=t
90
; (4:79)
and the shear lag parameter b is given by Eq. (4.76). The solution in Eq. (4.78) is
different than in Eq. (4.75) due to differences in the assumed boundary conditions.
It is noted that s
90
xx0
in Eq. (4.78) is the total stress in the transverse ply before
cracking, and thus includes thermal residual stress, if any.
To account for the progressive shear in the 90
layer
thickness but varied parabolically through the 90
1
lE
0
x0
_ _
; (4:80)
which is three times the value of b
2
as given by Eq. (4.62). Later, the same
approach was applied by Ogin et al. [47, 48] to study the stiffness reduction of
glassber crossply laminates subjected to quasistatic or fatigue loading. The
transverse shear effects in both 0
 and 90
and 90
1
lE
0
x0
_ _
: (4:81)
It can be seen that Eq. (4.81) reduces to Eq. (4.80) when G
0
xz0
G
90
xz0
.
79 4.4 Onedimensional models: shear lag analysis
4.4.4 2D shear lag models
Flaggs [32], and Nuismer and Tan [33, 49] developed twodimensional shear lag
models. Contrary to the authors claims, these models are essentially equivalent to
1D models with a minor modication to correct for the contraction due to the
Poissons effect [42].
The analysis by Flaggs [32] accounts for both applied normal and shear load
ings and results into a system of two coupled secondorder differential equations.
For only normal applied loading, it reduces to a single ODE identical to general
1D analysis equations, except that the shear lag parameter is given by
b
2
=
2
1
lQ
0
xx0
Q
90
yy0
Q
90
xy0
Q
0
xy0
=Q
0
xx0
_ _
Q
90
xx0
Q
90
yy0
Q
90
xy0
_ _
2
_
_
_
_
_
_
1
k
2
1
2
_ _
1
G
90
xz0
l
2G
0
xz0
; (4:82)
where k is a transverse shear correction factor and the Qs are stiffness coefcients
for the virgin laminate. The Flaggs analysis therefore uses a minor correction for
the Poissons contraction introduced by approximate inclusion of the transverse
coordinate (y).
Another 2D elasticity analysis was performed by Nuismer and Tan [33, 49].
This analysis uses the shear lag parameter dened as
b
2
=
1
Q
90
xx0
1
lQ
0
xx0
1
3G
90
xz0
l
3G
0
xz0
: (4:83)
The analysis also yields a nonzero o (see Eq. (4.85) described in the next subsec
tion), given by
o =
1
l
t
90
t
/
i
b
2
s
90
x0
_
; (4:84)
where t
/
i
is the slope of the interfacial shear stress at the crack location. This
analysis is also equivalent to other 1D models with a minor correction for the
Poissons ratio.
More recently, there have been efforts to modify and extend the 1D and 2D
shear lag models to enable predictions for laminates of other than crossply
layups, such as the equivalent constraint models [13, 5153] and 2D displacement
analysis [54, 55]. These models also suffer from deciencies that are common in the
shear lag models described above.
4.4.5 Summary of shear lag models
It can be shown [42] that all onedimensional stress analyses discussed above can
be reduced to a generalized form of Garrett and Baileys equation [27],
80 Microdamage mechanics
d
2
s
dx
2
b
2
s = o P ( ); (4:85)
where Ds represents the total stress transferred from the 90
layer to the 0
layer,
and is given by
s = s
0
xx
s
0
xx0
; (4:86)
and x = x/t
90
is a dimensionless xcoordinate, b is the shear lag parameter, and
o(P) is a function which may depend on the laminate structure, crack spacing, and
the applied load (P). The corresponding boundary conditions, equivalent to
tractionfree crack surfaces, are
s x = r ( ) =
s
90
xx0
l
; (4:87)
where r = l/t
90
denotes the crack spacing normalized by the cracked ply thickness.
Other than in the onedimensional model by Nuismer and Tan [33, 49], the
function o is chosen to be zero. It can be clearly observed that all the one
dimensional shear lag models are essentially the same; the difference being in the
choice of the shear lag parameter. The different shear lag models are summarized
in Table 4.1.
The solution of the differential equation (4.85) provides
s =
o
b
2
s
90
xx0
l
o
b
2
_ _
cosh bx
cosh br
: (4:88)
With this, the stress in the 0
layer is given by
s
0
xx
= s
0
xx0
o
b
2
s
90
xx0
l
o
b
2
_ _
cosh bx
cosh br
; (4:89)
and in the 90
layer, it is given by
s
90
xx
= s
90
xx0
lo
b
2
_ _
1
cosh bx
cosh br
_ _
: (4:90)
Now, the rate of load transfer from the 90
layer to the 0
1
lE
0
x0
_ _
0
Laws and
Dvorak [36]
Use rst ply
failure data to
determine b
Kt
90
1
E
90
x0
1
lE
0
x0
_ _
0
Steif [46],
Ogin et al. [47, 48]
Parabolic
displacement
prole
3G
90
xz0
1
E
90
x0
1
lE
0
x0
_ _
0
Highsmith and
Reifsnider [30]
Use an effective
shear transfer
layer
G
t
90
t
s
1
E
90
x0
1
lE
0
x0
_ _
0
Fukunaga et al. [45] Use an effective
shear transfer
layer
G
t
90
t
s
1
Q
22
1
lQ
11
_ _
0
Limand Hong [38] Use an effective
shear transfer
layer, account
for crack
interaction
G
t
90
t
s
1
Q
22
1
lQ
11
_ _
0
Nuismer and Tan
[33, 49]
Account for
Poissons effect
1
Q
90
xx0
1
lQ
0
xx0
1
3G
90
xz0
l
3G
0
xz0
1
l
t
90
t
/
i
b
2
s
90
x0
_
Flaggs [32] 2D shear lag
analysis to
account for
both transverse
and shear
loading
2
1
lQ
0
xx0
Q
90
yy0
Q
90
xy0
Q
0
xy0
Q
0
xx0
Q
90
xx0
Q
90
yy0
Q
90
xy0
_ _
2
_
_
_
_
_
_
_
_
_
_
1
k
2
1
2
_ _
1
G
90
xz0
l
2G
0
xz0
0
Lee and Daniel [31] Parabolic
displacement
variation,
account for
Poissons effect
3G
90
xz0
1 l
G
90
xz0
G
0
xz0
1
E
90
x0
1
lE
0
x0
_ _
0
82 Microdamage mechanics
from the crack. The nonzero shear stress on the crack surfaces is a clear violation of
the boundary condition. This makes all 1D analyses inherently inaccurate.
Let us now turn our attention to the stiffness degradation. The treatment
provided here follows the approach covered in a book chapter by Nairn and
Hu [42]. Before cracking, both the 0
 and 90
layers at the crack planes do not carry any load. The total
applied load at these planes is carried by the 0
plies take up load again through the shear transfer mechanism. To determine
the modulus of the cracked laminate, let us rst determine the total xdisplacement
of the load bearing layer in the region between two cracks in the transverse ply.
It can be found by integrating the axial strain as
u P ( ) =
_
l
l
e
0
xx
dx = t
90
_
r
r
s
0
xx
E
0
x
n
0
xz
s
0
zz
E
0
x
n
0
xy
s
0
yy
E
0
x
_ _
dx; (4:93)
where P is the applied load. It is noted that the thermal strain is ignored because for
linear thermoelastic materials the modulus is independent of the residual stresses.
The compliance of a unit cell of the laminate with cracks is dened as
C =
u P ( ) u 0 ( )
P
: (4:94)
The effective modulus is then given by
E =
2l
hWC
: (4:95)
60
60
4 3 2 1 0
1 2
Manders et al.
Laws & Dvorak
Fukunaga et al.
Lee & Daniel
Steif
3 4
40
20
0
(
M
P
a
)
9
0
9
0
x
x
x
z
20
40
Figure 4.10. Variation of the axial normal and interfacial stresses between two cracks in a
[0/90
2
]
s
laminate from shear lag analysis assuming normalized crack spacing r = 4 and
applied stress of 50 MPa.
83 4.4 Onedimensional models: shear lag analysis
Assuming a plane stress state,
s
0
yy
= 0: (4:96)
Moreover there is no load applied in the thickness (z) direction, i.e.,
_
r
r
s
0
zz
dx = 0: (4:97)
Thus, for both onedimensional analysis and twodimensional plane stress analy
sis the total displacement reduces to
u P ( ) = t
90
_
r
r
s
0
xx
E
0
x
dx: (4:98)
Substituting stress from Eq. (4.89) for 1D shear lag analyses, we obtain
1
E
x
=
1
E
0
c
1
E
90
x0
lE
0
x0
tanh br
br
_ _
o P ( ) o 0 ( )
E
0
x0
b
2
s
0
1
tanh br
br
_ _
: (4:99)
Assuming o(P)o(0) = 0, as is the case for most onedimensional analyses, the
effective axial modulus is then
1
E
x
=
1
E
0
c
1
E
90
x0
lE
0
x0
tanh br
br
_ _
: (4:100)
Thus, the axial modulus for the cracked laminate normalized with its virgin value
is given by
E
x
E
0
c
= 1
E
90
x0
lE
0
x0
tanh br
br
_ _
1
: (4:101)
The predictions for stiffness reductionusing various shear lag models as a function of
crack density for [0/90
3
]
s
glass/epoxy (Scotch Ply 1003) laminate are shown in Figure
4.11. It should be noted, however, that these predictions may vary depending upon
the parameters used in the models. It is important to note that the original shear lag
model by Garrett, Bailey, and Parvizi [27, 43], or equivalently by Manders et al. [29],
is reasonably accurate. More complicated shear lag models not only involve more
complicated analysis, and adjustable parameters, but also do not yield more accurate
predictions. The basic deciency of shear lag models lies in their onedimensionality
of stress eld, and no signicant improvement is essentially possible.
4.5 Selfconsistent scheme
The selfconsistent method is widely employed for estimating properties of elastic
solids containing entities such as inclusions, voids, and cracks; see for example
NematNasser and Hori [2]. Using this method generally requires assuming
84 Microdamage mechanics
an innite solid. For composite laminates containing ply cracks, Laws and
Dvorak [59] estimated elastic properties using this method. They rst assumed that
a unidirectional ply (within a laminate) with cracks along its bers can be modeled
as an innite solid with the same ber volume fraction and crack density. Taking
then the estimated average properties of the innite solid as those of the cracked ply,
they replaced the cracked ply in a composite laminate by a homogeneous ply of the
degraded properties of the innite solid. Figure 4.12 illustrates the assumed scheme.
Elastic properties of the cracked composite laminate were then calculated by the
classical laminate plate theory. This procedure in principle allows calculating
stiffness degradation of any composite laminate with given crack density in any
ply, i.e., for any orientation of cracks. It is doubtful, however, that the assumption
of an innite solid will give accurate elastic property estimates for plies that are
typically of a fraction of a millimeter in thickness. The predicted properties of
various laminates with cracks in different plies and with different crack density
values were reported in Laws and Dvorak [59] but these properties were not
compared with experimental data or independent numerical computations.
Readers interested in details of the selfconsistent method for composite mater
ials with cracks are referred to Laws et al. [60]. We provide here a brief account
of the concepts involved. Essentially, Hill [61] and Budiansky [62] developed the
selfconsistent method for composite materials, i.e., homogeneous solids with
inclusions. The idea in this method is that a single inclusion is embedded in a
homogeneous elastic solid which has the yetunknown overall properties of the
heterogeneous solid and the local elds thus estimated are then used to obtain the
overall properties. Laws et al. [60] use a key result obtained for ellipsoidal
inclusions by Eshelby [63, 64] that the elastic eld within the inclusion is uniform.
This result was used for two models: a threephase system and a twophase system.
1.0
0.9
0.8
0.7
R
e
l
a
t
i
v
e
S
t
i
f
f
n
e
s
s
0.6
0.5
0.0 0.1 0.2 0.3
Hashin
Garrett & Bailey
Reifsnider
Nuismer & Tan
Glass/Epoxy
Flaggs
[0/90
3
]s
Ogin et al.
0.4 0.5
Microcrack Density (1/mm)
0.6 0.7 0.8 0.9 1.0
Figure 4.11. Variation of normalized Youngs modulus as a function of crack density for a
[0/90
3
]
s
glass/epoxy (Scotch Ply 1003) laminate. The experimental data are from [30].
Reprinted, with kind permission, from Damage Mechanics of Composite Materials,
J. A. Nairn and S. Hu, Micromechanics of damage: a case study of matrix microcracking,
pp. 187243, copyright Elsevier (1994).
85 4.5 Selfconsistent scheme
In the former the bers and cracks are considered as two phases (special cases of
inclusions) embedded in an elastic matrix, while in the latter the bers and matrix are
smeared into a homogeneous solid in which cracks are embedded. Cracks in both
models are introduced as slits that are regarded as limiting cases of elliptical voids.
When the ber diameter is much smaller than the crack size, the twophase
model is more appropriate. Thus if Q is the estimated stiffness matrix of the
cracked ply, and Q
0
is its value without the cracks present, then the two matrices
are related by [60]
Q = Q
0
1
4
bQ
0
Q; (4:102)
where b is a crack density parameter and L is a function of the aspect ratio of the
elliptical crack and the stiffness coefcient matrix Q.
It is seen from this equation that the estimation procedure for stiffness of a
cracked ply is iterative and it requires quantities L that are estimated by assuming
the homogenized composite ply to be of innite extent.
Variations of the selfconsistent method described above have also appeared in
the literature. Noteworthy work is that of Hoiseth and Qu [65, 66]. They followed
the differential selfconsistent method and derived an incremental differential
equation describing the effective axial modulus for the cracked laminate by
representing the change in strain energy due to an increase in terms of the number
of cracks in a lamina. If r = t
90
/l denotes the normalized crack density, then the
axial modulus is given by
(a) (b)
Figure 4.12. Selfconsistent scheme: (a) cracked laminate containing bers of small
diameter and periodic cracks (b) homogenized cracked lamina.
86 Microdamage mechanics
d
E
c
dr
=
d
4t
90
2
E
c
r ( )
E
1
1 n
12
n
21
_ _
; (4:103)
subject to the initial condition
E
c
0 ( ) =
E
c0
; (4:104)
where d is the average crack opening displacement (COD), dened as
d =
1
t
90
_
t
90
0
d z ( ) dz; (4:105)
with d(z) being the COD variation across the zdirection (thickness). The authors
obtained it numerically using nite element calculations on a unit cell. For cross
ply laminates, their predictions compared well with the independent FE simula
tions that they carried out themselves. For multidirectional laminates, these
approaches become complex and do not yield accurate estimates.
4.6 2D stress analysis: variational methods
4.6.1 Hashins variational analysis
An improved 2D stress analysis can be obtained using the principle of minimum
complementary energy applied on a cracked laminate volume. Hashin [67] con
sidered crossply laminates and used this principle to solve the boundary value
problem. He constructed an admissible stress eld assuming that the normal
stresses in the loading direction are constant over the ply thickness. The admissible
stress eld satises equilibrium and boundary and interface conditions. The
boundary volume problem thus solved provides the stresses and the reduced
stiffness coefcients for the cracked crossply laminate.
We followthe general notations for geometry, material, and stress components as
given in Section 4.3. Consider a symmetric crossply laminate containing a parallel
array of transverse cracks and subjected to a uniform axial load (Figure 4.13)
N
xx
per unit laminate width. Let us rst investigate the stress states before and after
cracking. Before cracking, all shear stresses and the transverse normal stress in the
entire laminate are zero, i.e.,
s
0
xx0
; s
90
xx0
,= 0; s
0
yy0
; s
90
yy0
,= 0;
s
0
zz0
= s
90
zz0
= 0; s
0
yz0
= s
90
yz0
= 0;
s
0
xz0
= s
90
xz0
= 0; s
0
xy0
= s
90
xy0
= 0;
(4:106)
where superscripts denote the layer orientations. The nonzero axial stresses in the
0
 and 90
ply and 90
ply,
respectively) of the cracked laminate can be expressed as
s
m
ij
= s
m
ij0
s
m
ij
(4:110)
where the subscript 0 indicates the uncracked laminate. The stress perturbations
can be expressed as
s
90
xx
= s
90
xx0
f
90
x ( );
s
0
xx
= s
0
xx0
f
0
x ( );
(4:111)
where f
90
and f
0
are unknown perturbation functions.
z
x
y
O
t
0
t
0
t
90
t
90
h h
N
xx
Figure 4.13. A crossply laminate loaded in axial tension.
88 Microdamage mechanics
The equilibrium equations for the plies are given by
@s
m
xx
@x
@s
m
xy
@y
@s
m
xz
@z
= 0;
@s
m
yx
@x
@s
m
yy
@y
@s
m
yz
@z
= 0;
@s
m
zx
@x
@s
m
zy
@y
@s
m
zz
@z
= 0:
(4:112)
Using Eqs. (4.108)(4.110), one obtains
@s
m
xx
@x
@s
m
xz
@z
= 0;
@s
m
zx
@x
@s
m
zz
@z
= 0:
(4:113)
Substituting Eq. (4.111) into Eq. (4.113) and integrating, one gets
s
90
xz
= s
90
xx0
f
/
90
x ( )z f
90
x ( ) [ [;
s
90
zz
= s
90
xx0
1
2
f
//
90
x ( )z
2
f
/
90
x ( )z g
90
x ( )
_ _
;
(4:114)
for m = 90 (90
ply), and
s
0
xz
= s
0
xx0
f
/
0
x ( )z f
0
x ( ) [ [;
s
0
zz
= s
0
xx0
1
2
f
//
0
x ( )z
2
f
/
0
x ( )z g
0
x ( )[;
_
(4:115)
for m = 0 (0
ply), where f
0
(x), f
90
(x), g
0
(x), and g
90
(x) are unknown functions
and primes denote derivatives with respect to x.
To obtain the relation between f
90
and f
0
, consider equilibrium in the
xdirection of the undamaged laminate
N
xx
=
_
h
h
s
xx0
dz = 2 s
90
xx0
t
90
s
0
xx0
t
0
_ _
: (4:116)
If the same membrane force is applied to the cracked laminate, the equilibrium in
the xdirection will give
N
xx
= 2 s
90
xx0
t
90
s
0
xx0
t
0
_ _
2 s
90
xx0
t
90
f
90
x ( ) s
0
xx0
t
0
f
0
x ( )
_
: (4:117)
From Eqs. (4.116) and (4.117), we get
s
90
xx0
t
90
f
90
x ( ) s
0
xx0
t
0
f
0
x ( ) = 0; (4:118)
i.e.,
f
0
x ( ) =
s
90
xx0
s
0
xx0
1
l
f
90
x ( ); (4:119)
where l = t
0
/t
90
is the ply thickness ratio (Eq. (4.59)).
89 4.6 2D stress analysis: variational methods
The boundary value problem for solving the three stress components s
xx
, s
xz
,
and s
zz
is dened on the unit cell shown in Figure 4.14, in which the origin of the
coordinate system is placed at the midplane of the laminate, and midway between
two cracks.
Clearly z = 0 is the plane of symmetry. Hence, the shear stress s
xz
at all points
along this plane is zero. At the interface between the cracked 90
ply (z =t
90
), the shear stress s
xz
and normal stress s
zz
must be continu
ous. Moreover, the surface at z = h is free from any external loading, i.e., the shear
stress s
xz
and normal stress s
zz
are zero on that surface. Finally, the crack surfaces
x = l are traction free, i.e., s
90
xz
= s
0
xz
= 0 and s
90
xx
= s
90
xx0
s
90
xx
= 0 at x = l.
Thus, all the applicable boundary conditions for the total stresses are written as
Symmetry : s
90
xz
x; 0 ( ) = 0; (a)
Interface : s
90
xz
x; t
90
( ) = s
0
xz
x; t
90
( ); (b)
s
90
zz
x; t
90
( ) = s
0
zz
x; t
90
( ); (c)
Free boundary : s
0
xz
x; h ( ) = 0; (d)
s
0
zz
x; h ( ) = 0; (e)
Traction free : s
90
xz
l; z ( ) = 0; t
90
_ z _ t
90
; (f)
s
90
xx
l; z ( ) = 0; t
90
_ z _ t
90
; (g)
(4:120)
where 2l is the distance between any two adjacent cracks. These conditions are the
same as those given earlier without derivation in Eq. (4.54)
Denoting f
90
(x) by f(x), putting boundary conditions, Eq. (4.120), into Eqs.
(4.114)(4.115), and using Eq. (4.119), after some mathematical manipulations,
the resulting stress eld in the cracked laminate is given by
x
z
l
l
h
t
0
t
90
Figure 4.14. A repeating unit cell for a cracked crossply laminate.
90 Microdamage mechanics
s
90
xx
= s
90
xx0
1 f x ( ) [ [;
s
90
xz
= s
90
xx0
f
/
x ( )z;
s
90
zz
= s
90
xx0
f
//
x ( )
1
2
1 l ( )t
2
90
z
2
_
;
(4:121)
in the 90
ply, and
s
0
xx
= s
0
xx0
s
90
xx0
1
l
f x ( );
s
0
xz
= s
90
xx0
f
/
x ( )
1
l
1 l ( )t
90
z [ [;
s
0
zz
= s
90
xx0
f
//
x ( )
1
2l
1 l ( )t
90
z [ [
2
;
(4:122)
in the 0
ply. Using the crack surface boundary conditions, Eq. (4.120fg), we get
additional conditions for the unknown perturbation function f(x)
At x = l : s
90
xx
= 0 = s
90
xx0
1 f x ( ) [ [;
s
0
xz
= 0 = s
90
xx0
f
/
x ( )z :
(4:123)
Thus,
f l ( ) = l; f
/
l ( ) = 0; t
90
_ z _ t
90
: (4:124)
This completes the description of the resulting boundary value problem. The object
ive is to estimate f(x) in order to determine the stress eld in the cracked laminate.
The stress eld in Eqs. (4.121)(4.122) represents an admissible stress eld as it
satises all equilibrium and boundary conditions. Thus, the principle of minimum
complementary energy (see Section 2.1.5) can be applied to obtain the unknown
function f(x). The procedure for doing this, given in Hashin [67], is described next.
For the problem at hand, we can dene an admissible stress eld s
ij
within the
volume V of the cracked body with associated tractions
T
i
which satisfy only
equilibrium and the traction boundary conditions. Since the tractions are not
altered during the cracking process,
~
T
i
= T
0
i
on S = S
T
; (4:125)
where T
i
0
are the tractions applied on the uncracked body. Also, the crack surfaces
S
c
are traction free, i.e.,
~
T
i
= 0 on S = S
c
: (4:126)
If we write
~ s
ij
= s
0
ij
s
/
ij
;
~
T
i
= T
0
i
T
/
i
; (4:127)
where s
0
ij
are stresses in the uncracked laminate, s
/
ij
are the perturbation stresses
due to cracking, and T
/
is the additional traction due to cracking. Then from the
traction boundary conditions in Eqs. (4.125) and (4.126), we obtain
91 4.6 2D stress analysis: variational methods
T
/
i
= 0 on S = S
T
;
T
/
i
= T
0
i
on S = S
c
:
(4:128)
The complementary energy functional is given by (see Eq. (2.36), Section 2.1.5)
+
0
=
1
2
_
V
S
ijkl
s
0
ij
s
0
kl
dV
_
S
u
T
0
i
^ u
i
dS (4:129)
for the uncracked body, where S
ijkl
is the compliance tensor, and
~
+
=
1
2
_
V
S
ijkl
~ s
ij
~ s
kl
dV
_
S
u
~
T
i
^ u
i
dS (4:130)
for the crackedbody. Since only tractionboundary conditions are applied, the second
term is zero (S
u
= 0; S
T
= S
t
). However, for the sake of completeness, we will follow
the proof given in Appendix 1 of [67], which applies to mixed boundary conditions.
Substituting the stress eld given in Eq. (4.127) into Eq. (4.130) and using Eq. (4.129),
the complementary energy for the cracked solid can be expressed as
~
+
=
+
0
_
V
S
ijkl
s
0
ij
s
/
kl
dV
1
2
_
V
S
ijkl
s
/
ij
s
/
kl
dV
_
S
u
T
/
i
^ u
i
dS : (4:131)
Consider the rst integral in Eq. (4.131). The stresses s
/
ij
satisfy equilibrium since
s
0
ij
and s
ij
do. Also, the strain eld in the uncracked body is given by
e
0
ij
= S
ijkl
s
0
kl
: (4:132)
Therefore, by virtual work (see Eq. (2.33), Section 2.1.5)
J =
_
V
S
ijkl
s
0
kl
s
/
ij
dV =
_
S
u
T
/
i
u
0
i
dS
_
S
c
T
/
i
u
0
i
dS
/
; (4:133)
where the tractions in the second surface integral are dened with respect to the
inward normal to S
c
. Using the boundary condition in Eq. (4.128), we have
J =
_
S
u
T
/
i
^ u
i
dS
_
S
c
T
0
i
u
0
i
dS : (4:134)
The second integral is taken over the two adjacent surfaces of each crack. Since the
normal on one surface is the negative of the normal on the other surface and since
T
0
i
and u
0
i
are continuous across the crack surface, the integrals on the two crack
surfaces cancel one another and therefore the total integral vanishes. Introducing
the remainder of Eq. (4.134) into Eq. (4.131) we obtain the complementary energy
for a cracked solid as
~
+
=
+
0
1
2
_
V
S
ijkl
s
/
ij
s
/
kl
dV =
+
0
/
(4:135)
92 Microdamage mechanics
where
/
0
is the complementary energy due to perturbation stresses. The effective
elastic compliances of the undamaged and damaged laminates can be expressed in
terms of the complementary energy functionals. To obtain the relation, consider
homogeneous boundary conditions on the cracked body
T
i
= s
ij
n
j
; s
ij
= constant; (4:136)
and let
s
ij
=
1
V
_
V
s
ij
dV: (4:137)
Then, from Eq. (4.130) we have
+
=
1
2
S
+
ijkl
s
ij
s
kl
V; (4:138)
where S
+
ijkl
is the effective elastic compliance tensor for the homogenized medium[68].
The laminate under consideration is loaded by a membrane force N
xx
on two
horizontal edges. Thus, volume and the average stress for this laminate are given by
V = 2Ah s
xx
=
N
xx
2h
= s
c
; (4:139)
where A is the area of the plane normal to the xaxis. The complementary energy
for cracked and uncracked laminates takes the form
+
0
=
1
2
s
2
c
E
x0
2Ah
+
=
1
2
s
2
c
E
x
2Ah; (4:140)
where E
x0
, E
x
are the longitudinal moduli of undamaged and damaged laminates,
respectively. Now, from the principle of minimum complementary energy, it
follows that if the admissible stress system, Eqs. (4.121)(4.122), with boundary
conditions, Eq. (4.124), is introduced into Eq. (4.135) then
~
+
_
+
: (4:141)
Using Eqs. (4.135) and (4.140) into Eq. (4.141), we obtain
1
2
s
2
c
E
x0
2Ah
/
_
1
2
s
2
c
E
x
2Ah; (4:142)
where
/
represents the perturbationincomplementary energy due tocracking. Hence,
1
E
x
_
1
E
x0
/
s
2
c
Ah
: (4:143)
Therefore, the variational boundary value problem results in the following calcu
lus of variations problem:
Find f(x) that minimizes
/
.
93 4.6 2D stress analysis: variational methods
The complementary energy change
/
is the sum of energies brought about by
perturbation stresses in the individual laminae, i.e.,
/
= 2
_
l
l
_
t
90
0
W
90
dz dx 2
_
l
l
_
h
t
90
W
0
dz dx; (4:144)
where W
90
and W
0
are stress energy densities due to perturbation stresses in the
90
 and 0
s
2
22
s
2
33
_ _
E
2
2n
12
s
11
s
22
s
33
( )
E
1
2n
23
s
22
s
33
E
2
s
2
23
G
23
s
2
12
s
2
13
_ _
G
12
_ _
;
(4:145)
where 1 is in the ber direction and 2, 3 are transverse directions, and the elastic
moduli are dened in Eq. (4.48). To determine the perturbation stress energy, we
can express stress energy densities in the 90
 and 0
ply, and
s
11
= s
0
xx
; s
13
= s
0
xz
; s
33
= s
0
zz
; s
12
= s
22
= s
23
= 0 (4:147)
for the 0
ply. Thus, the energy densities in the two sets of laminae are
W
90
=
1
2
s
90
xx
_ _
2
E
2
s
90
zz
_ _
2
E
2
2n
23
s
90
xx
s
90
zz
E
2
s
90
xz
_ _
2
G
23
_ _
;
W
0
=
1
2
s
0
xx
_ _
2
E
1
s
0
zz
_ _
2
E
2
2n
12
s
0
xx
s
0
zz
E
1
s
0
xz
_ _
2
G
12
_ _
:
(4:148)
Consider again the laminate region bounded by two transverse planes through
adjacent cracks, Figure 4.14. Due to symmetry about z = 0, it is sufcient to take
one half of the laminate l _ x _ l, 0 _ z _ h with unit width in y direction.
Further, the symmetry across x results into
f x ( ) = f x ( ); (4:149)
Substituting Eqs. (4.121), (4.122), and (4.148) into Eq. (4.144), and carrying out
the integration along z,
/
is given by
/
= s
90
xx0
_ _
2
_
l
l
t
90
C
00
f
2
t
3
90
C
02
ff
//
t
3
90
C
11
f
/
2
t
5
90
C
22
f
//
2
_ _
dx; (4:150)
94 Microdamage mechanics
where
C
00
=
1
E
2
1
lE
1
; C
02
= l
2
3
_ _
u
23
E
2
l
3
u
12
E
1
;
C
11
=
1
3
1
G
23
l
G
12
_ _
; C
22
= l 1 ( ) 3l
2
12l 8
_ _
1
60E
2
:
(4:151)
Introducing a nondimensional geometry parameter x = x/t
90
, Eq. (4.150) can be
rewritten as
/
= s
90
xx0
_ _
2
t
2
90
_
r
r
C
00
f
2
C
02
f
d
2
f
dx
2
C
11
df
dx
_ _
2
C
22
d
2
f
dx
2
_ _2
_ _
dx; (4:152)
where r = l/t
90
denotes the crack spacing normalized with the cracked ply thickness.
The EulerLagrange equation for f(x) is
d
4
f
dx
4
p
d
2
f
dx
2
qf = 0; (4:153)
where the coefcients p and q are given by
p =
C
02
C
11
C
22
; q =
C
00
C
22
: (4:154)
This is a fourthorder ODE whose characteristic equation is
r
4
pr
2
q = 0 : (4:155)
The roots of the characteristic equation are
r = a
1
ia
2
( ); i =
1
_
;
a
1
= q
1=4
cos
y
2
; a
2
= q
1=4
sin
y
2
;
y = tan
1
4q
p
2
1
;
(4:156)
provided that 4q/p
2
> 1. The general solution to Eq. (4.153) will involve terms with
e
a
1
x
cos a
2
x and e
a
1
x
sin a
2
x. However, it is more convenient to use hyperbolic
functions instead of exponentials, because of symmetry condition, Eq. (4.149).
Since only even product functions are admissible solutions, the solution of the
fourthorder ODE, Eq. (4.153), leads to the following expression for f(x)
f = A
1
cosh a
1
x cos a
2
x A
2
sinh a
1
x sin a
2
x; (4:157)
where A
1
and A
2
are constants determined from the boundary conditions as
A
1
=
2 a
1
cosh a
1
r sin a
2
r a
2
sinh a
1
rcos a
2
r ( )
a
1
sin 2a
1
r a
2
sinh 2a
2
r
;
A
2
=
2 a
2
cosh a
1
r sin a
2
r a
1
sinh a
1
rcos a
2
r ( )
a
1
sin 2a
1
r a
2
sinh 2a
2
r
;
(4:158)
95 4.6 2D stress analysis: variational methods
when 4q/p
2
< 1, f(x) is given by [68]
f =
a
2
/
cosh a
1
/
x
sinh a
1
/
r a
2
/
coth a
1
/
r a
1
/
coth a
2
/
r ( )
a
1
/
cosh a
2
/
x
sinh a
2
/
r a
1
/
coth a
2
/
r a
2
/
coth a
1
/
r ( )
;
(4:159)
where a
/
1
; a
/
2
are dened as
a
1
/
; a
2
/
=
p
2
p
2
4
q
_
; (4:160)
assuming that p < 0, which usually holds good for typical glass/epoxy and
graphite epoxy materials [69]. Once f is known, the stress eld can be calculated
from Eqs. (4.121)(4.122). The effective stiffness coefcients for the cracked
laminate are then determined by calculating average stresses and strains.
It is important to note here that variational analysis gives lower bounds of stiffness
properties for the cracked laminates, as canbe seenfromthe inequality inEq. (4.143).
4.6.2 Effect of residual stresses
As a laminate is cooled from the cure temperature, thermal residual stresses are
generated due to differential thermal expansion of the 0
and 90
plies. These
residual stresses can be easily incorporated by laminate theory for uncracked
laminate. The effect of thermal stresses on crack initiation was investigated in
detail by [70]. The variational stress formulation for cracked crossply laminate
including thermal residual stresses is covered in [69, 71]. The complementary
energy analogous to Eq. (4.152) is then given by [69]
/
=
/
0
t
2
90
_
r
r
C
00
c
2
C
02
c
d
2
c
dx
2
C
11
dc
dx
_ _
2
C
22
d
2
c
dx
2
_ _2
2a Tc C
2T
d
2
c
dx
2
_ _
dx;
(4:161)
where
c = s
90
xx0
a T
C
00
_ _
f
a T
C
00
; (a)
a = a
22
a
11
; T = T
0
T
ref
; (b)
C
2T
= a
22
T
n
23
s
c
E
c
_ _
2
3
l
_ _
a
22
T
n
12
s
c
E
c
_ _
l
2
3
; (c)
(4:162)
where a
11
and a
22
are the coefcients of thermal expansion in the longitudinal and
transverse directions, respectively, T
0
is the service temperature, and T
ref
is the
stressfree (reference/curing) temperature of the laminate. The corresponding
EulerLagrange equation for this case is
96 Microdamage mechanics
d
4
c
dx
4
p
d
2
c
dx
2
qc =
a T
C
22
: (4:163)
The solution for f remains the same as in Eqs. (4.158)(4.160), from which c can be
obtainedusingEq. (4.162a). The stresses canthenbe calculatedusingEqs. (4.121)(4.122).
4.6.3 [0
m
/90
n
]
s
vs. [90
n
/0
m
]
s
laminates
The stiffness changes due to cracking in [0
m
/90
n
]
s
laminates are different than in [90
n
/
0
m
]
s
laminates because in the former case we have cracks in an internal 90
layer,
whereas the cracks in the latter are exposed to the free boundary. For the same area
of crack surfaces, the tractions on the crack surfaces (see Eq. (4.128)) do more work
in closing (or opening) the crack surfaces in the latter case, and, therefore, the
perturbation complementary energy has a higher value. This results in a lower
effective axial stiffness as compared to laminates with cracks in internal plies.
Nairn [69] carried out a variation of the solution for transverse cracking in
[90
n
/0
m
]
s
laminates. The procedure is the same as for [0
m
/90
n
]
s
laminates. The
obtained solution for f is also the same, except that some of the constants
appearing in Eqs. (4.150) and (4.162) are different. The newconstants are as follows
C
02
= 1
2
3
l
_ _
n
12
E
1
n
23
3E
2
;
C
22
= l 1 ( ) 3 12l 8l
2
_ _
1
60E
2
;
C
2T
=
1
3
n
23
s
c
E
c
a
22
T
_ _
n
12
s
c
E
c
a
22
T
_ _
l
2
3
l
2
_ _
:
(4:164)
A comparison of the stress proles in [0/90
2
]
s
and [90
2
/0]
s
laminates is illustrated in
Figure 4.15(b)(c). A discussion follows later.
4.6.4 Improved variational analysis
Some major improvements toHashins variational analysis were suggestedbyVarnaand
Berglund [7173] with the most rened model described in [73]. In Hashins approach,
the axial stress variation across the thickness is assumed constant, whereas the Varna
Berglund approach determines these by further application of the principle of minimum
complementary energy. The stress functions in the 90
 and 0
90
= s
90
xx0
z
2
2
c x ( )
z
2
2
A
+
_ _ _ _
c
1
x ( )
2
z ( )
_ _
t
2
90
;
0
= s
0
xx0
z
2
2
s
90
xx0
1
z ( )c x ( )
_ _
s
90
xx0
c
1
x ( )
3
z ( )
_ _
t
2
90
;
(4:165)
where s
0
xx0
ands
90
xx0
are the axial stresses in the 0
 and 90
plies of uncracked
laminate, respectively, x = x=t
90
and z = z=t
90
are nondimensional coordinates,
97 4.6 2D stress analysis: variational methods
and A* is a constant. In the above expressions, the arbitrary function
2
(z) repre
sents nonuniformity in the xaxis stress distribution in the 90
3
(z) along the x and zdirections, respectively. If we neglect
2
(z);
3
(z); c
1
(x) and
set A
+
= h=2t
90
and
1
z ( ) = (h z)
2
=2t
0
t
90
we revert back to Hashins model.
For the current model, the stress components in layer m are given by
10
20
30
40
10
20
30
40
(interfacial)
(MPa)
xz
s
90
xx
s
90
x
Crack
Crack
4
3
(a)
2 1
4
3 2 1
50
60
90
xz
90
xx
(interfacial)
(interfacial)
90
zz
60 (MPa)
40
30
20
10
40
30
20
10
3
4
Crack
Crack
2 1 2 3
4
x
50
1
(b)
70 (MPa)
(interfacial)
60
90
xx
s
90
zz
s
(interfacial)
90
xz
s
50
Crack
Crack
1
4 4
x
40
30
20
10
30
20
10
2 3 3 2 1
(c)
Figure 4.15. Prole of stresses between two adjacent ply cracks in a carbon/epoxy crossply
laminate: (a) using 1D shear lag analysis for [0/90
2
]
s
; (b) using 2D variational analysis for
[0/90
2
]
s
; and (c) using 2D variational analysis for a [90
2
/0]
s
laminate. The normal stress s
90
xx
is same for the entire 90
m
@z
2
s
m
zz
=
@
2
m
@x
2
s
m
xz
=
@
2
m
@x@z
: (4:166)
Thus,
(a) in the 90
layer
s
90
xx
= s
90
xx0
1 c x ( ) c
1
x ( )
//
2
z ( ) [ [;
s
90
xz
= s
90
xx0
c
/
x ( )z c
/
1
x ( )
/
2
z ( ) [ [;
s
90
zz
= s
90
xx0
c
//
x ( )
z
2
2
A
+
_ _
c
//
1
x ( )
2
z ( )
_ _
;
(4:167)
(b) in the 0
layer
s
0
xx
= s
0
xx0
s
90
xx0
c x ( )
//
1
z ( ) c
1
x ( )
//
3
z ( ) [ [;
s
0
xz
= s
90
xx0
c
/
x ( )
/
1
z ( ) c
/
1
x ( )
/
3
z ( ) [ [;
s
0
zz
= s
90
xx0
c
//
x ( )
1
z ( ) c
//
1
x ( )
3
z ( ) [ [ :
(4:168)
The boundary and interface conditions remain the same as given in Eq. (4.120),
which, after using Eqs. (4.167) and (4.168), results in
/
1
1 ( ) = 0;
1
1 ( ) = A
+
1
2
;
/
1
h ( ) =
1
h ( ) = 0;
/
3
h ( ) =
3
h ( ) = 0;
/
2
1 ( ) =
/
3
1 ( ) = 1;
2
1 ( ) =
3
1 ( );
c r ( ) = 1; c
/
r ( ) = c
/
1
r ( ) = c
1
r ( ) = 0;
(4:169)
where h = h=t
90
. Clearly, this model is complex and requires determining the
constant A*, and the functions c x ( ); c
1
x ( ); and
1
z ( );
2
z ( )
3
z ( ). However, the
authors found that the following choice of functions
1
z ( );
2
z ( );
3
z ( ) is useful
1
z ( ) =
1 cosh
1
z
h ( )
1
sinh
1
l
;
2
z ( ) = A
z
2n
2n
;
3
z ( ) =
1 cosh
3
z
h ( )
3
sinh
3
l
;
(4:170)
with A a constant and D
1
, D
3
, and n arbitrary shape parameters, where n is an
integer. Using the boundary conditions described in Eq. (4.169), we obtain
A
+
=
1
2
1 cosh
1
l
1
sinh
1
l
; A =
1
2n
1 cosh
3
l
3
sinh
3
l
: (4:171)
Now, the total complementary energy of the cracked laminate system is given by
/
=
/
0
_
r
r
v c; c
/
; c
//
; c
1
; c
/
1
; c
//
1
) d x; ( (4:172)
99 4.6 2D stress analysis: variational methods
where
/
0
is the complementary energy of the virgin laminate, and v is the comple
mentary energy density due to perturbation stresses. Using the minimization
procedure as described above, we nally obtain the following system of ordinary
differential equations with constant coefcients
C
0
22
c
///
C
0
02
C
0
11
_ _
c
//
C
0
00
c C
01
22
c
///
1
C
R
c
//
1
C
01
00
c
1
= 0;
C
01
22
c
///
C
R
c
//
C
01
00
c C
1
22
c
///
1
C
1
02
C
1
11
_ _
c
//
1
C
1
00
c
1
= 0;
(4:173)
with the boundary conditions as in Eq. (4.169) and constants given by
C
0
00
= 1
E
2
E
1
I
1
; C
0
22
=
1
20
A
+
3
A
+
2
I
2
;
C
0
11
=
E
2
3G
23
E
2
G
12
I
3
; C
0
02
= 2n
23
A
+
1
6
_ _
2n
12
E
2
E
1
I
4
;
C
1
00
= I
T
1
E
2
E
1
I
L
1
; C
1
22
= I
T
2
I
L
2
;
C
1
11
=
E
2
G
23
I
T
3
E
2
G
12
I
L
3
; C
1
02
= 2n
23
I
T
4
2n
12
E
2
E
1
I
L
4
;
C
01
00
= 1
E
2
E
1
I
C
1
; C
01
22
= F
3
I
C
2
;
C
01
11
=
E
2
G
23
F
1
E
2
G
12
I
C
3
; C
01
02
= 2n
23
F
2
2n
12
E
2
E
1
I
C2
4
;
C
01
20
= 2n
23
F
4
2n
12
E
2
E
1
I
C1
4
; C
R
=
1
2
C
01
20
C
01
02
_ _
C
01
11
;
(4:174)
where
I
1
=
_
h
1
//
1
z ( ) [ [
2
dz; I
2
=
_
h
1
1
z ( ) [ [
2
dz; I
3
=
_
h
1
/
1
z ( ) [ [
2
dz;
I
4
=
_
h
1
1
z ( )
//
1
z ( )dz; I
T
1
=
_
1
0
//
2
z ( ) [ [
2
dz; I
T
2
=
_
1
0
2
z ( ) [ [
2
dz;
I
T
3
=
_
1
0
/
2
z ( ) [ [
2
dz; I
T
4
=
_
1
0
2
z ( )
//
2
z ( )dz; F
1
=
_
1
0
z
2
z ( )dz;
F
2
=
_
1
0
2
z ( )dz; F
3
=
_
1
0
2
z ( )
z
2
2
A
+
_ _
dz; F
4
=
1
2
A
+
2
1 ( ) F
2
;
I
L
1
=
_
h
1
//
3
z ( ) [ [
2
dz; I
L
2
=
_
h
1
3
z ( ) [ [
2
dz; I
L
3
=
_
h
1
/
3
z ( ) [ [
2
dz;
I
L
4
=
_
h
1
3
z ( )
//
3
z ( )dz; I
C
1
=
_
h
1
1
//
z ( )
//
3
z ( )dz; I
C
2
=
_
h
1
1
z ( )
3
z ( )dz;
I
C
3
=
_
h
1
1
/
z ( )
/
3
z ( )dz; I
C1
4
=
_
h
1
1
z ( )
//
3
z ( )dz; I
C2
4
=
_
h
1
//
1
z ( )
3
z ( )dz:
(4:175)
100 Microdamage mechanics
The minimum value of the complementary energy corresponding to the obtained
solution is given by [73] as
/
min
=
/
0
s
90
xx0
t
90
_ _
2
2E
2
C
01
22
///
1
r ( ) C
0
22
///
r ( )[ :
_
(4:176)
The stresses resulting from cracking can be calculated by solving the system of
coupled ODEs in Eq. (4.173) to get the perturbation functions, and then putting
them into Eqs. (4.167)(4.168). The average of the stresses will yield the average
stiffness properties of the cracked laminate. For this model, the longitudinal
Youngs modulus normalized with its virgin state value is given by [73]
E
x
E
x0
=
1
1
E
2
lE
1
f r ( )
r
; (4:177)
where
f r ( ) =
1
2
_
r
r
x ( )
1
x ( ) [ [ d x : (4:178)
The normalized effective Poissons ratio for the cracked crossply laminate is
given by
n
xy
n
0
xy
=
1 1
E
2
E
1
_ _
t
90
h
f r ( )
r
1
E
2
lE
1
f r ( )
r
: (4:179)
4.6.5 Related works
There have been some further developments based on the variational analysis
procedure. The analysis of transverse cracks in crossply laminates undergoing
shear loading was covered by Hashin in his original paper [67]. Later he utilized
the variational approach to predict the thermal expansion coefcients for the
cracked crossply laminate [74]; this is covered later in Section 4.11. In another
paper [75], he also analyzed the case of orthogonally cracked crossply lamin
ates. Another signicant analysis of crossply laminates was carried out by
Kuriakose and Talreja [76] where they analyzed crossply laminates subjected
to bending moments.
4.6.6 Comparison between 1D and 2D stressbased models
Comparison of stress predictions between 1D shear lag [27] and 2D
variational analysis [67] for a [0/90
2
]
s
carbon epoxy laminate are shown in
101 4.6 2D stress analysis: variational methods
Figure 4.15(a)(b). The laminate is subjected to an applied mechanical stress
of 100 MPa, and a temperature change of DT = 125
C. The normalized
crack spacing is r = 4.
As seen in Figure 4.15(b), s
90
xx
= 0 at x= r as the crack surfaces are traction
free and the transverse plies carry no axial load at the crack planes. As we move
away from the crack surfaces, s
90
xx
increases as stress is transferred back into the
90
layers in the
middle. The longitudinal modulus and Poissons ratio for the laminate are
given by
E
xx
=
s
0
e
S
xx
; n
xy
=
e
S
yy
e
S
xx
; (4:181)
where s
0
is the applied stress, the overbars denote volume averages, and
superscript S denotes sublaminate. The average COD for this laminate is given by
102 Microdamage mechanics
u = l e
S
xx
e
90
xx
_ _
: (4:182)
Following [79], the stiffness properties for the cracked laminate can be shown to
be given by
E
xx
E
xx0
=
1
1 arR l
_ _ ;
u
xy
u
xy0
=
1 crR l
_ _
1 arR l
_ _ ; (4:183)
where r = 1/2l is the crack density; a, c, and g are known functions of laminate
material and thickness; R
l ( ) is the average stress perturbation function;
and
l = l=t
90
is the halfcrack spacing normalized with the 90
layer thickness.
Different approaches model R
l ( ) differently:
1. Shear lag models:
R
l ( ) =
2
b
tanh b
l ( ); (4:184)
where b is the shear lag parameter. Different shear lag models use different
denitions for this parameter [78].
2. Variational methods:
R
l ( ) =
4a
1
a
2
a
2
1
a
2
2
cosh 2a
1
l ( ) cos 2a
2
l ( )
a
2
sinh 2a
1
l ( ) a
1
sin 2a
2
l ( )
; (4:185)
where the constants a
1
, a
2
are dened as per model.
The comparisons of different models for [S/90
4
]
s
laminates, as performed
by Joffe and Varna [79] are shown in Figures 4.164.19 for different values
of y. For the 2D0 model, the expression for the stress perturbation function
is the same as for Hashins variational analysis, except that the coefcients are
now given as [79]
C
00
=
1
E
2
1
E
S
x
I
1
;
C
02
= 2
n
32
E
2
1
6
A
+
_ _
2
n
S
xz
E
S
x
I
4
;
C
11
=
1
2G
23
1
G
S
xz
I
3
:
(4:186)
This comparison conrms that variational analyses consistently provide a
lower bound to stiffness properties and are signicantly better than shear lag
models. The variational analysis predictions are also very close to FE
predictions.
103 4.6 2D stress analysis: variational methods
4.7 Generalized plain strain analysis McCartneys model
These models may be seen as a further development of stress analysisbased
damage models for cracked laminates. The cracked crossply laminates, symmetric
about the laminate midplane, were considered in a generalized plane strain
condition, thereby reducing the directional dependence of essentially a 3D stress
eld in the cracked laminate. Assuming a regular array of fully grown parallel
cracks in 90
 and 90
 and 90
n
m
21
s
m
22
E
m
22
n
m
31
s
m
33
E
m
33
a
m
11
T;
e
m
22
=
n
m
12
s
m
11
E
m
11
s
m
22
E
m
22
n
m
32
s
m
33
E
m
33
a
m
22
T;
(4:187)
e
m
33
=
n
m
13
s
m
11
E
m
11
n
m
23
s
m
22
E
m
22
s
m
33
E
m
33
a
m
33
T;
e
m
12
=
s
m
12
2G
m
12
; e
m
13
=
s
m
13
2G
m
13
; e
m
23
=
s
m
23
2G
m
23
;
1.0
(a)
0.9
0.8
0.7
E
x
/
E
x
0
0.6
0.5
0.4
0.0 0.1 0.2
Sh.L.2
Hashin
FE
2D0
[
+
15/90
4
]
s
Experiment
0.3
Crack density (cr/mm)
0.4 0.5 0.6 0.7
_
1.0
(b)
0.9
0.8
0.7
V
x
y
/
V
x
y
0
0.6
0.5
0.4
0.0 0.1
Sh.L.2
Hashin
FE
2D0
Experiment
0.2 0.3
[15/90
4
]
s
Crack density (cr/mm)
0.4 0.5 0.6 0.7
Figure 4.17. Variation of elastic stiffness moduli with crack density for a [15/90
4
]
s
glass/
epoxy laminate: (a) normalized Youngs modulus; (b) normalized Poissons ratio versus
crack density. Reprinted, with kind permission, from Compos Sci Technol, Vol. 59, R. Joffe
and J. Varna, Analytical modeling of stiffness reduction in symmetric and balanced
laminates due to cracks in 90 degrees layers, pp. 164152, copyright Elsevier (1999).
105 4.7 Generalized plain strain analysis McCartneys model
where m = 0, 90 for 0
ply and 90
 and 90
 and 90
 and 90
2
15E
90
22
t
90
t
0
t
90
t
0
_ _
2
5
2
t
90
t
0
15
8
_ _
;
G =
1
3
1
G
0
12
1
G
0
23
t
90
t
0
n
0
12
E
0
11
n
90
23
E
90
22
2
t
90
t
0
3
_ _ _ _
;
H =
1
E
0
11
t
90
t
0
1
E
0
22
:
(4:197)
When the inner and outer plies are made of the same material the differential
equation, Eq. (4.196), has the same form as the EulerLagrange equation derived
by Hashin. It is worth noting that the Reissner energy functional has a stationary
value when the constitutive relations and equilibrium equations are satised in
an average sense. Thus, McCartneys approach is quite analogous to Hashins
approach: the former uses a displacement formulation (equivalent to minimiza
tion of Reissners energy functional), while the latter uses a stress formulation
(equivalent to minimization of complementary energy functional).
For a cracked crossply laminate, McCartneys 2D approach yields the
following relations for the effective longitudinal modulus
E
xx
=
E
xx0
1
t
90
l
E
0
22
E
0
11
; (4:198)
where
=
2pq
p
2
q
2
cosh
2
pl
t
0
cosh
2
ql
t
0
_ _
;
1
= q sinh
pl
t
0
cosh
pl
t
0
p sin
ql
t
0
cos
ql
t
0
;
p =
r s
2
_
; q =
r s [ [
2
_
;
r =
G
2F
> 0; s =
H
F
_
> 0 :
(4:199)
If the plies are thick, the averaging of stresses andstrains is not advisable as it canresult
in inaccurate predictions. To improve the approach, McCartney [8186] used the so
called plyrenement procedure, in which each ply is divided thicknesswise into N
segments, and the resulting stressstrain relations are satised, in an average sense,
over each ply segment. Consequently, the unknown functions C
i
(x) for the ith ply
segment (i = 1,2, . . . , N), satisfy the following N homogeneous differential equations
N
i=1
F
ij
C
IV
i
x ( )
N
i=1
G
ij
C
//
i
x ( )
N
i=1
H
ij
C
i
x ( ) = 0; j = 1; 2; : : : ; N; (4:200)
where the coefcients F
ij
, G
ij
, and H
ij
are calculated numerically using a suitable
ODE solver. McCartney [82, 84, 86] has later generalized this approach for multi
layered crossply laminates, and also for triaxial loading.
109 4.7 Generalized plain strain analysis McCartneys model
McCartney [82, 84, 86] has also shown that for cracked crossply laminates, the
moduli are interrelated. Similar interrelationships among the Poissons ratio,
transverse modulus, and longitudinal modulus have also been independently
derived by Nairn [87]. If we dene a damage parameter or normalized stiffness
parameter as
D =
1
E
xx
1
E
xx0
; (4:201)
then other moduli for the cracked crossply laminate are interrelated as
n
xy0
E
xx0
n
xy
E
xx
= k
1
D;
1
E
yy
1
E
yy0
= k
2
1
D;
n
xz0
E
xx0
n
xz
E
xx
= k
2
D;
1
E
zz
1
E
zz0
= k
2
2
D;
n
yz0
E
yy0
n
yz
E
yy
= k
1
k
2
D;
(4:202)
and
a
xx
a
xx0
= k
3
D; a
yy
a
yy0
= k
1
k
3
D; a
zz
a
zz0
= k
2
k
3
D
; (4:203)
where
k
1
=
E
xx0
E
yy0
B n
xy0
E
yy0
=E
xx0
_ _
1 n
xy0
B
;
k
2
=
E
xx0
A n
xz0
n
xz0
E
xx0
=E
yy0
_ _
B
1 n
xy0
B
;
k
3
=
E
xx0
s
xx0
Ba
yy0
C
_
1 n
xy0
B
;
(4:204)
with
A =
n
13
E
22
n
23
n
12
E
11
; B = n
12
; C = a
22
n
12
a
11
: (4:205)
These interrelationships can in principle reduce the burden of evaluating change in
each modulus subsequent to matrix cracking. Instead, the evaluation of degrad
ation in longitudinal modulus is sufcient for predicting other moduli. However,
this result has yet to be veried experimentally.
4.8 CODbased methods
One way to view the elastic response changes caused by the presence of cracks in a
medium is to consider the additional overall (global) strain of the RVE contrib
uted by the crack surface displacements of the individual cracks within the RVE.
Contrarily, if none of the cracks in the RVE conducts surface displacements, the
overall elastic response of the RVE will not change. This observation motivates
the focus on crack surface displacements. Although generally these displacements
110 Microdamage mechanics
can be described as crack opening displacement (COD) and crack sliding displace
ment (CSD), one commonly refers only to COD. We will in the following discuss
stiffness relations in terms of average COD that can be calculated either analytic
ally [8896] or numerically [97101].
4.8.1 3D laminate theory: Gudmundsons model
Consider a general threedimensional laminate consisting of N plies (see
Figure 4.20). Each ply is dened by its material properties, layup angle, and
thickness. For uncracked laminate the global average stresses s
ij
and strains e
ij
are dened as [102]
s
ij
=
N
k=1
V
k
s
k
ij
; e
ij
=
N
k=1
V
k
e
k
ij
; (4:206)
where s
k
ij
; e
k
ij
are the average stresses and strains, respectively, in the kth ply (k =
1, 2, . . ., N), and V
k
stands for the volume fraction of the kth ply such that
N
k=1
V
k
= 1. Note that for uncracked laminates the ply average stresses and
strains are the same as the individual (constant) ply stresses. For the sake of
simplicity, we will denote the tensors with a ~. Now, let us partition the stresses,
strains, and thermal expansion coefcients into inplane and outofplane parts
s
~
=
s
~
I
s
~
O
_ _
; e
~
=
e
~
I
e
~
O
_ _
; s
~
=
s
~
I
s
~
O
_ _
; (4:207)
where
s
~
I
=
s
xx
s
yy
s
xy
_
_
_
_
; e
~
I
=
e
xx
e
yy
2e
xy
_
_
_
_
; a
~
I
=
a
xx
a
yy
2a
xy
_
_
_
_
; (4:208)
denote the inplane stresses, strains, and thermal expansion coefcients, and
s
~
O
=
s
zz
s
xz
s
yz
_
_
_
_
; e
~
O
=
e
zz
2e
xz
2e
yz
_
_
_
_
; a
~
O
=
a
zz
2a
xz
2a
yz
_
_
_
_
; (4:209)
k
k+1
t
k
l
k
l
k
Figure 4.20. A general 3D laminate with cracks in some offaxis layers.
111 4.8 CODbased methods
denote the outofplane stresses, strains, and thermal expansion coefcients. The consti
tutive law between global average stress tensor and average strain tensor is given by
e
~
= S
~
s
~
a
~
T
=
e
~
I
e
~
O
_ _
=
S
~
II S
~
IO
S
~
IO
_ _
T
S
~
OO
_
_
_
_
s
~
I
s
~
O
_ _
a
~
I
a
~
O
_ _
T;
(4:210)
where S
~
is the compliance tensor, and the superscript T represents the trans
pose. Similarly, the relation between ply stress tensor and ply strain tensor can
be written as
e
~
k
= S
k
~
s
k
~
s
k(r)
~
_ _
a
~
T
=
e
~
I
k
e
~
O
k
_
_
_
_
=
S
~
I
k
S
~
IO
k
S
~
IO
k
_ _
T
S
~
OO
k
_
_
_
_
s
~
I
k
s
~
I
k(r)
s
~
O
k
_
_
_
_
s
~
I
k
s
~
O
k
_
_
_
_
T;
(4:211)
where s
~
k(r)
denote residual stresses present due to reasons other than thermal
mismatch, such as chemical shrinkage during the manufacturing process. The
global average inplane and outofplane residual stresses vanish due to equilib
rium. The compatibility and equilibrium conditions in the laminate give us
e
~
I
k
= e
~
I
; s
~
O
k
= s
~
O
: (4:212)
Substituting Eq. (4.212) into Eq. (4.210), one obtains
s
~
I
k
= S
~
II
k
_ _
1
e
~
I
S
~
IO
k
s
~
O
a
~
I
k
T
_ _
s
~
I
k(r)
: (4:213)
Using Eq. (4.206), and using the fact that the volume average of residual stresses
vanish, the above equation can be rewritten as
e
~
I
= S
~
II
s
~
I
S
~
IO
s
~
O
a
~
I
T; (4:214)
where
S
~
II
=
N
k=1
V
k S
~
II
k
_ _
1
_ _
1
;
S
~
IO
= S
~
II
N
k=1
V
k S
~
II
k
_ _
1
S
~
IO
k
_ _
;
a
~
I
= S
~
II
N
k=1
V
k S
~
II
k
_ _
1
a
~
I
k
_ _
:
(4:215)
112 Microdamage mechanics
In a similar way, substituting Eq. (4.212) into Eq. (4.211), and applying Eq.
(4.206), we get
e
~
O
= S
~
IO
_ _
T
s
~
I
S
~
OO
s
~
O
a
~
O
T; (4:216)
where
S
~
OO
= S
~
IO
_ _
T
S
~
II
_ _
1
S
~
IO
N
k=1
V
k S
~
OO
k
S
~
IO
k
_ _
T
S
~
II
k
_ _
1
S
~
IO
k
_ _
;
a
~
O
= S
~
IO
_ _
T
S
~
II
_ _
1
a
~
I
N
k=1
V
k
a
~
O
k
S
~
IO
k
_ _
T
S
~
k
II
)
1
a
~
I
k
_ _
:
_
(4:217)
Eqs. (4.215) and (4.217) relate the effective laminate properties to the local
ply level properties. The above equations form the basis of a 3D laminate
theory. The corresponding relations for a 2D laminate theory are obtained
by considering only the inplane tensors. To obtain the effective laminate
properties for a cracked laminate, dene the nondimensional crack density in
the kth ply as
r
k
=
t
k
l
k
; (4:218)
where t
k
and l
k
denote the thickness and average crack spacing, respectively, in the
kth ply. The process of transverse cracking reduces the elastic energy of the
laminate and the change in elastic energy is associated with the release of tractions
on the crack surfaces. It was shown by Gudmundson and Ostlund [92] that the
coupling terms between the energy of the uncracked laminate and the change in
elastic energy due to matrix cracking vanish (see also the proof by Hashin [67],
which is described earlier in the chapter, Eq. (4.135)).
It is noted that for a cracked laminate, the effective strains are different from the
average strains, whereas there is no distinction between global effective and
average stresses. The effective strains are the strains measured on a global scale,
whereas average strains come from averaging strains in individual plies over the
RVE. The difference between the effective strains and average strains is equal to
strain increment caused by crack opening displacements. Thus, the average
stresses are dened as (see Eq. (4.206))
s
(a)
ij
=
N
k=1
V
k
s
k(a)
ij
; (4:219)
where the superscript (a) denotes average variables. The global effective strains are
dened as
e
(e)
ij
=
1
2V
_
out
u
i
n
j
u
j
n
i
_ _
d; (4:220)
113 4.8 CODbased methods
where u
i
, i = 1, 2, 3 indicates the displacement vector, n
i
is the unit normal vector
on the outer boundary surface G
out
of the representative volume V, and the
superscript (e) denotes the effective variables. In the same way the effective ply
strains can be dened as [102]
e
k(e)
ij
=
1
2V
k
_
kout
u
k
i
n
k
j
u
k
j
n
k
i
_ _
d; (4:221)
where V
k
is the volume of the ply k and the surface integral is performed on the
outer boundary of ply k. Obviously,
1
2V
_
out
u
i
n
j
u
j
n
i
_ _
d =
N
k=1
V
k
1
2V
k
_
kout
u
k
i
n
k
j
u
k
j
n
k
i
_ _
d
_ _
: (4:222)
Hence, from Eqs. (4.220)(4.222), one gets
e
(e)
ij
=
N
k=1
V
k
e
k(e)
ij
: (4:223)
Applying the divergence theorem on Eq. (4.221), we get
e
k(e)
ij
=
1
2V
k
_
V
k
u
k
i;j
u
k
j;i
_ _
d
1
2V
k
_
kc
u
k
i
n
k
j
u
k
j
n
k
i
_ _
d; (4:224)
where the rst integral is equal to the average ply strain, given by
e
(a)
ij
=
1
2V
k
_
V
k
u
k
i;j
u
k
j;i
_ _
d; (4:225)
and the second integral is the strain increment due to matrix cracks e
k
ij
is
given by
e
k
ij
=
1
2V
k
_
kc
u
k
i
n
k
j
u
k
j
n
k
i
_ _
d =
r
k
2t
k
u
k
i
n
k
j
u
k
i
n
k
j
_ _
; (4:226)
where
u
k
i
=
1
t
k
_
t
k
0
u
k()
i
u
k()
i
_ _
dt
k
=
1
t
k
_
t
k
0
u
k
i
dt
k
(4:227)
is the average crack opening displacement for ply k. Thus, we can write effective
ply strains as
e
k(e)
ij
= e
k(a)
ij
e
k
ij
: (4:228)
Finally, after applying the proper boundary conditions, Gudmundson and
Zang [89] arrive at the following expressions for effective properties of cracked
laminate
114 Microdamage mechanics
S
~
II(c)
= S
~
II
_ _
1
N
k=1
n
k
r
k
A
~
k
_ _
T
N
i=1
b
~
ki
A
~
k
_ _
1
;
S
~
IO(c)
=S
~
II(c)
S
~
II
_ _
1
S
~
IO
N
k=1
n
k
r
k
A
~
k
_ _
T
N
i=1
b
~
ki
B
~
i
_ _
1
;
S
~
OO(c)
= S
~
IO(c)
_ _
T
S
~
II(c)
_ _
1
S
~
IO(c)
S
~
IO
_ _
T
S
~
II
_ _
1
S
~
IO
S
~
OO
N
k=1
n
k
r
k
B
k
~
_ _
T
N
i=1
b
~
ki
B
~
i
;
a
~
I(c)
=a
~
I
S
~
II(c)
N
k=1
n
k
r
k
A
~
k
_ _
T
N
i=1
b
~
ki
C
~
i
;
a
~
O(c)
=a
~
O
S
~
IO(c)
_ _
T
S
~
II(c)
_ _
1
a
~
I(c)
a
~
I
_ _
N
k=1
n
k
r
k
B
~
k
_ _
T
N
i=1
b
~
ki
C
~
i
;
(4:229)
where matrices S
~
II
; S
~
IO
; S
~
OO
; S
~
II
; a
~
I
; and a
~
O
are given in Eqs. (4.215) and (4.217),
b
~
is the matrix containing average crack opening displacements (COD), and
matrices A
~
; B
~
; C
~
are given by
A
~
k
= N
~
I
k
S
~
II
k
_ _
1
; B
~
k
= N
~
O
k
N
~
I
k
S
~
k
II
)
1
S
~
k
IO
; C
~
k
=
A
~
k
a
~
I
a
~
k
_ _
;
_
(4:230)
where the matrices N
~
k
I
and N
~
k
O
represent the unit normal vector on the crack
surfaces in ply k, i.e.,
N
~
I
k
=
n
k
1
0 n
k
2
0 n
k
2
n
k
1
0 0 0
_
_
_
_
N
~
k
O
=
0 0 0
0 0 0
0 n
k
1
n
k
2
_
_
_
_
: (4:231)
The expressions in Eq. (4.229) for stiffness of the cracked laminates are exact
assuming the homogenization procedure. However, the main problem is the
determination of average COD for the cracked laminate. The heterogeneity in
the composite laminates and constraint effects on crack surfaces from the sur
rounding uncracked plies make it impossible to derive exact analytical solutions.
Gudmundson and coworkers, however, made the following assumptions in order
to evaluate the average crack opening displacements:
1. The surface displacements of a ply crack in a nitethickness laminate are
equal to those of a crack in an innite, homogeneous transversely isotropic
medium. The stress intensity factors for an innite row of equidistant cracks
in an innite homogeneous isotropic medium under the action of uniform
tractions on crack surfaces, given by Benthem and Koiter [104] and Tada et al.
[105], are assumed to hold for the current case of transversely isotropic
medium.
2. There is no effect of orientation of a cracked ply. This means that the matrices b
~
ki
in Eq. (4.229) may not be accurate for offaxis cracks in laminates.
115 4.8 CODbased methods
3. The crack density is low (r 1).
4. There is no coupling between crack opening displacements of different plies.
Hence the matrices b
~
ki
are diagonal.
With these assumptions, the matrices b
~
ki
are given by
b
~
ki
=
0
~
; for all k ,= i;
or b
~
kk
=
b
k
1
0 0
0 b
k
2
0
0 0 b
k
3
_
_
_
_
;
(4:232)
where b
k
1
; b
k
2
; andb
k
3
are determined by Gudmundson using a numerical integra
tion and are given by
b
k
1
=
4
p
g
1
ln cosh
pr
k
2
_ _ _ _
r
k
( )
2
;
b
k
2
=
p
2
g
2
10
j=1
a
j
1 r
k
( )
j
;
b
k
3
=
p
2
g
3
9
j=1
b
j
1 r
k
( )
j2
;
(4:233)
for cracks in an internal ply, where
g
1
=
1
2G
12
; g
2
= g
3
=
1 u
12
u
21
E
2
; (4:234)
and a
j
and b
j
are numerical parameters given in Table 4.2. For surface cracks (or
cracks in external plies), we have
b
k(s)
1
=
8
p
g
1
ln cosh pr
k
_ _ _
2r
k
( )
2
;
b
k(s)
2
= 2 1:12 ( )
2
p
2
g
2
10
j=1
c
j
1 r
k
( )
j
_ _
;
(4:235)
Table 4.2 Numerical parameters used in calculation of COD matrix coefcients b
k
1
; b
k
2
; and b
k
3
j a b c
1 0.63666 0.63666 0.25256
2 0.51806 0.08945 0.27079
3 0.51695 0.15653 0.49814
4 1.04897 0.13964 8.62962
5 8.95572 0.16463 51.2466
6 33.0944 0.06661 180.9631
7 74.32002 0.54819 374.298
8 103.064 1.07983 449.5947
9 73.60337 0.45704 286.51
10 20.3433 73.84223
116 Microdamage mechanics
where c
j
are another set of numerical parameters also given in Table 4.2.
It is clear despite the assumptions used in COD calculation, the above approach
is quite complex and difcult to implement in a numerical scheme. Similar
relations are also available for the combined extension and bending loading
scenario [94, 95].
4.8.2 LundmarkVarna model
Lundmark and Varna [98, 99, 106, 107] have recently derived effective properties
using a homogenization very similar to that in Gudmundsons approach. How
ever, their relationships are simpler and they use the crack surface displacements
numerically obtained from FEM simulations, which makes their model much
more accurate. However, their model has been checked against experimental
data only for crossply laminates. Here, we provide the main relations derived
by them. Accordingly, the average strains in a ply, analogous to (4.228), are
dened as
e
ij
_ _
a
k
= e
ij
_ _
LAM
b
ij
_ _
k
; (4:236)
where the superscript a means average and
e
ij
_ _
a
k
=
e
11
e
22
2e
12
_
_
_
_
_
_
a
k
; e
ij
_ _
LAM
=
e
11
e
22
2e
12
_
_
_
_
_
_
LAM
;
b
ij
_ _
k
=
b
11
b
22
2
b
12
_
_
_
_
_
_
k
; (4:237)
where
b
ij
_ _
k
represents the VakulenkoKachanov tensor, dened by
b
ij
_ _
k
=
1
2V
k
_
kc
u
k
i
n
k
j
u
k
j
n
k
i
_ _
d; (4:238)
which is same as e
k
ij
in Eq. (4.226). In terms of crack surface displacements,
b
ij
_ _
k
for a cracked ply is derived as
b
_ _
k
=
r
kn
E
2
A [ [
k
U [ [
k
A [ [
k
Q
_
k
e
0
LAM
a
0
k
T
_ _
; (4:239)
where [A]
k
is the transformation matrix for ply k and [U]
k
is the displacement
matrix given by
U [ [
k
= 2 =
0 0 0
0 u
k
2an
0
0 0
E
2
G
12
u
k
1an
_
_
_
_
; (4:240)
where u
k
1an
and u
k
2an
are the normalized average crack face sliding and opening
displacements, respectively, and are given by
u
k
1an
= u
k
1a
G
12
t
k
s
k
120
; u
k
2an
= u
k
2a
E
2
t
k
s
k
20
; (4:241)
117 4.8 CODbased methods
with u
k
1a
and u
k
2a
being the average crack surface displacements, dened as
u
k
1a
=
1
2t
k
_
t
k
2
t
k
2
u
1
z ( ) dz; u
k
2a
=
1
2t
k
_
t
k
2
t
k
2
u
2
z ( ) dz; (4:242)
where Du
1
and Du
2
are the relative separation of the two crack faces along and
normal to the crack surface. Now, the average stressstrain relationships for the
kth ply in the global coordinate system are
s
a
k
=
Q
_
k
e
a
k
a
k
T
_
: (4:243)
Since, the average stresses for the laminate remain equal to the applied stresses,
we have
s
LAM
= s
a
=
N
k=1
s
a
k
t
k
H
; (4:244)
where H is the total thickness of the laminate. Substituting Eqs. (4.236) and (4.239)
into Eq. (4.244), we obtain
s
LAM
= Q
0
[ [
LAM
e
LAM
e
0
LAM
th
_ _
1
H
N
k=1
t
k
Q
_
k
b
_ _
k
; (4:245)
where [Q
0
]
LAM
is the stiffness matrix for the undamaged laminate, and
e
0
LAM
th
= 1=H ( )
N
k=1
t
k
a
k
T are the thermal strains in the undamaged
laminate. Substituting Eq. (4.239) into Eq. (4.245), the average thermomechanical
response of the damaged laminate is given by
s
LAM
= Q
0
[ [
LAM
e
LAM
e
0
LAM
th
_ _
1
HE
2
N
k=1
r
kn
t
k
Q
_
k
A [ [
k
U [ [
k
A [ [
k
Q
_
k
e
0
LAM
a
0
k
T
_ _
:
(4:246)
For purely mechanical response, the stiffness matrix of the damaged laminate is
given by
Q [ [
LAM
= I [ [
1
HE
2
N
k=1
r
kn
t
k
Q
_
k
A [ [
k
U [ [
k
A [ [
k
Q
_
k
S
0
[ [
LAM
_ _
1
Q
0
[ [
LAM
; (4:247)
where [S
0
]
LAM
= ([Q
0
]
LAM
)
1
is the compliance tensor for the undamaged
laminate.
The only remaining unknown in this model is [U]
k
, Eq. (4.240). To determine the
crack surface displacements, Lundmark and Varna [98] suggest using actual FE
calculations on a unit cell of cracked laminate. For crossply laminates, they
carried out a set of such FE calculations for varying values of ply thickness and
118 Microdamage mechanics
stiffness, and by curve tting they provided the following power lawtype expres
sions for the normalized average crack opening displacements
u
2an
= A B
E
2
E
s
x
_ _
p
; (4:248)
where A, B, and p are constants that depend on the thickness ratio of cracked to
uncracked plies, and the type of crack (in an internal or external ply). For
example, for an internal crack in a GFRP [S
n
/90
m
]
s
laminate,
A = 0:52; B = 0:3075 0:1652
t
90
2t
s
2t
s
_ _
;
p = 0:0307
t
90
2t
s
_ _
2
0:0626
t
90
2t
s
_ _
0:7037;
(4:249)
where t
90
and t
s
are the thicknesses of cracked 90
layer.
4.9 Computational methods
For simulating the effect of damage in composite laminates, in general,
many computational methods have been utilized, such as the nite element
method (FEM), the nite difference method (FDM), and the boundary element
method (BEM). For the particular problem of transverse cracking, the most
common numerical tool is FEM. Some simpler numerical tools have also been
devised, e.g., the nite strip method by Li et al. [108], and the layerwise theory of
Reddy [109]. In this subsection, we will briey describe some of these
developments.
0
Crack density (cracks/mm)
[30/30/90
4
]
s
+ experimental
model
0.2 0.4 0.6
0.6
0.7
0.8
0.9
E
x
/
E
x
0
1.0
(a) (b)
0
Crack density (cracks/mm)
[30/30/90
4
]
s
+ experimental
model
0.2 0.4 0.6
0.6
0.7
0.8
0.9
1.0
n
x
y
/
n
x
y
0
Figure 4.21. Reductions in longitudinal modulus (a) and Poissons ratio (b) for a graphite/
epoxy [30/90
4
]
s
laminate using the LundmarkVarna model [98]. Reprinted, with kind
permission, from P. Lundmark and J. Varna, Int J Damage Mech, Vol. 14, pp. 23559,
copyright # 2005 by Sage Publications.
119 4.9 Computational methods
4.9.1 Finite element method (FEM)
FEM is the most widely used numerical approach to analyze damage in composite
laminates. Its advantages are that it can model highly complex situations, and
provides quite accurate results. As can be seen from the preceding discussion, the
analytical models of cracking are severely limited in scope with respect to laminate
layup and loading scenario. Nonetheless, all numerical methods have a unique
limitation: every time the laminate geometry, loading, or material changes, simu
lations need to be carried out afresh, which might involve new mesh generation,
and computation. Hence, FE modeling of cracked laminates is time consuming
and does not by itself provide much insight into the damage mechanisms. Even
with these limitations it can be successfully used for calibration/verication of
analytical models as well as for computation of parameters or constants useful
in analytical methods (for example, it can be used to evaluate constants used in
Talrejas continuum damage model, see the next chapter for details). FE modeling
can also be utilized to simulate complex experimental situations, a task equivalent
to carrying out numerical experiments, and eliminating the need for cumber
some experimentation whenever possible.
The rst task in FE modeling is to dene a geometrical model of the cracked
laminate. Mostly, a representative unit cell is developed assuming periodic array
of selfsimilar cracks (see Figure 4.7(a)). For crossply laminates, a 3D repeating
unit can be reduced to a 2D plane stress/strain model (see Figure 4.7(b)).
Furthermore, the symmetry of the resulting boundary value problem can
reduce the size of the modeled unit cell. For example, cracking in a [0/90]
s
laminate can be modeled using a quarter of a representative unit cell as illus
trated in Figure 4.22.
For multidirectional laminates with cracks in more than one orientation, how
ever, a repeating unit cannot be dened uniquely because of differences in direc
tionality and mutual spacing of cracks in different plies. A repeating unit cell can
be dened for up to two offaxis cracking modes with cracked surfaces represented
on nonorthogonal boundaries [110]. The RVE in such cases is therefore 3D and
skewed (nonorthogonal in the xy plane; z being the thickness direction).
A reduction to two dimensions is often not possible in such cases.
As is common in micromechanics, the displacement, strain, and stress elds
must be periodic across repeating unit cells. Thus the following periodic boundary
conditions [111] are applied on the FE model
u
i
x
a
x
a
( ) = u
i
x
a
( ) x
b
@u
i
@x
b
_ _
;
e
ij
x
a
x
a
( ) = e
ij
x
a
( );
s
ij
x
a
x
a
( ) = s
ij
x
a
( );
(4:250)
where u
i
, @u
i
/@u
b
), and Dx
b
, i, b = 1,2,3 represent the displacements, the volume
averaged displacement gradients, and the vector of periodicity, respectively. For a
120 Microdamage mechanics
skewed RVE, periodic BC might be complicated to enforce. For a detailed
discussion about skewed RVE, periodic BCs, and symmetries in composites, the
reader is referred to [110, 112114].
Once the FE results are available, the overall stiffness properties of the damaged
laminate can be calculated by volume averaging of stresses and strains inside the
RVE. For example, for a cracked laminate undergoing inplane loading, the in
plane elasticity properties can be calculated by
E
x
=
s
xx
)
e
xx
)
; E
y
=
s
yy
_
e
yy
_ ; G
xy
=
s
xy
_
g
xy
_ ; n
xy
=
e
yy
_
e
xx
)
; (4:251)
where <g> denotes the volume average of a eld g. Important numerical
studies related to analysis of cracked composite laminates can be found in [110
113, 115, 116].
4.9.2 Finite strip method
To enable predictions for laminated composites with arbitrary layups [110,
117, 118], Li et al. [108] devised an approximate numerical approach based on
the generalized plain strain formulation. Although there can be more than
one cracked ply, all the cracked plies have to be of the same orientation.
Hence, this method cannot be used for laminates with cracks in multiple
orientations.
Consider rst an uncracked laminate. Given the generalized strains {e},
e = e
xx0
; e
yy0
; 2e
xy0
; k
xx
; k
yy
; 2k
xy
_
T
; (4:252)
the displacement eld based on the classical laminate theory without rigid body
displacements can be written as
u
0
= xe
xx0
ye
xy0
xzk
xx
yzk
xy
;
v
0
= ye
yy0
xe
xy0
yzk
yy
xzk
xy
;
w
0
=
x
2
2
k
xx
y
2
2
k
yy
xyk
xy
o z ( );
(4:253)
t
0
t
90
h
x
z
2l
90
0
0
(b)
(a)
s
c
s
c
Figure 4.22. Building FE model for a cracked crossply laminate: (a) 2D representative unit
cell; (b) FE model (1/4 cell).
121 4.9 Computational methods
where o is an arbitrary integration function which depends on boundary
conditions.
When transverse cracks appear, perturbations to this displacement eld are
induced. These perturbations can simply be superimposed on the displacements due
to the linearity of the problem. Assuming sufciently long cracks, the perturbations
are independent of y. Hence, the displacement eld for a cracked laminate reads
u = u
0
U x; z ( );
v = v
0
V(x; z);
w = w
0
W(x; z);
(4:254)
where U, V, and W denote changes due to the presence of cracks. The resulting
strains are calculated using innitesimal deformation theory. Our objective then is
to solve for U, V, and W. The approach in the nite strip method is to divide the
planar region of a typical segment of the cracked laminate into a nite number of
strip elements parallel to the xaxis. In each element, M nodal lines are introduced
along which displacements are functions of x only and the displacements in the
strip elements are then interpolated by polynomials in the zdirection. For a
typical element, the displacement eld can be expressed as
U
V
o W
_
_
_
_
_
_
e
=
3
i=1
N
i
z ( )
U
i
V
i
W
i
_
_
_
_
_
_
e
; (4:255)
where the superscript e denotes the element number, and N
i
are shape functions,
the same as those used in the onedimensional nite element analysis [119], which
are given by
N
1
=
1
2
z 1 z ( ); N
2
= 1 z
2
; N
1
=
1
2
z 1 z ( ); (4:256)
with 1 _ z _ 1 being the nondimensional coordinate along the zaxis
z =
3
i=1
N
i
z ( )z
i
; (4:257)
where z
i
is the zcoordinate of the ith nodal line in the element. Going through the
usual FEM formulation, Li et al. arrived at the following set of differential
equations for the resulting variational problem
K
2
[ [
y
_ _
K
1
[ [ K
1
[ [
T
_ _
_
y
_ _
K
0
[ [ y = F
0
; (4:258)
and a set of boundary conditions, where [K
2
], [K
1
], and [K
0
] are the matrices
dened in Li et al. [108] and
y =
U
V
W
_
_
_
_
_
_
: (4:259)
122 Microdamage mechanics
No direct solution to the differential Eq. (4.258) exists. An approximate solution
can be obtained by taking the nodal displacements as a Fourier series of unknown
constants, as
U
n
= U
h
n
x
l
K
n
k=1
U
k
n
sin
kpx
l
;
V
n
= V
h
n
x
l
K
n
k=1
V
k
n
sin
kpx
l
;
W
n
= W
h
n
x
l
_ _
2
W
0
n
K
n
k=1
U
k
n
cos
kpx
l
;
(4:260)
for n = 1,2, . . ., M, and the coefcients U
h
n
; V
h
n
; W
h
n
; U
k
n
; V
k
n
; V
k
n
; and W
0
n
are
unknown constants to be determined, K
n
is the order where the Fourier series is
truncated for the approximate solution, and M is the number of nodal lines in the
laminate. Substituting Eq. (4.260) into Eq. (4.258), and rearranging, the following
algebraic equations similar to FEM are obtained
A [ [ = : (4:261)
The solution of these algebraic equations along with the appropriate boundary
conditions gives the displacement eld, from which the strain eld can be
obtained.
4.9.3 Layerwise theory
The layerwise theory of Reddy [109] is motivated by a desire to develop a
computational model that is more efcient than the conventional 3D nite
element models [119] and can incorporate damage effects such as transverse cracks
and delaminations in a layered medium. It is based on 3D kinematics where the
displacement eld within each layer is expanded using the Lagrange family of
nite elements. For a comparable mesh, the layerwise theory typically takes less
computational time as compared to the conventional 3D FEM while providing
the same level of accuracy.
In the theory, the whole laminate is divided into a number of subdivisions
across its thickness. The displacement eld in the laminate is written as
u
i
x; y; z ( ) =
N
J=1
U
J
i
x; y ( )
J
z ( ); i = 1; 2; 3; (4:262)
where N is the number of subdivisions through the thickness of the laminate and F
J
are global interpolation functions dened in terms of the Lagrange interpolation
functions associated with the layers connected to the Jth interface through the
laminate thickness, and U
J
i
are the nodal displacements. Independent interpolation
functions for u
1
, u
2
, and u
3
can also be used whenever necessary. The strain eld is
determined using the von Karman nonlinear theory. Then the governing equations
123 4.9 Computational methods
for the nodal displacements are derived using the principle of virtual displacements
[120, 121]. The resulting system of partial differential equations can be converted to
a systemof equations analogous to the FEMin a procedure similar to that followed
in previous subsection. Solution of this system gives us the nodal displacement,
from which strain and stress elds can be determined. Further details on the
approach and its implementation can be found in [122127]. Recent papers by Na
and Reddy [128, 129] provide direct implementation of the layerwise theory for
transverse cracking and delaminations, respectively, in crossply laminates.
4.10 Other methods
There have been other developments in analyzing cracked laminates. The detailed
treatment is not covered here and we will highlight the important aspects of these
approaches. The readers are referred to the cited articles for further details. It
should be pointed out, however, that these approaches are mostly extensions of
the previously developed ideas, and do not really improve the predictions. More
over, many of these approaches are complex, semianalytical and thus difcult to
implement, and may sometimes need high computational times.
Close to the development of the variational analysis by Hashin, Aboudi and
coworkers [130, 131] developed a threedimensional semianalytical method. In
this approach, an approximate analytical solution for the displacement eld is
sought using a series expansion in the form of Legendre polynomials.
In a somewhat similar way, Lee and Hong [132] developed a renement of
the shear lag approach using a series polynomial expansion. The approach
accounted for the crack opening displacement, thermal stresses, and Poissons
effects. However, the improvement over the traditional shear lag methods was
insignicant.
Another similar effort by Gamby and Rebiere [133] used the transverse shear
stresses in the 0
 and 90
f; (4:263)
where a
c
is the longitudinal thermal expansion coefcient for an uncracked
composite laminate, and
f =
1
2l
_
l
l
f x ( ) dx;
B
0
= 1 l ( )
n
12
E
1
;
(4:264)
Lundmark and Varna [98] also developed a model for thermal expansion coef
cients. Following the analysis covered in Section 4.8.2 and noting that only
thermal loading is present, the strain eld in the laminate is given by (setting
mechanical loads to zero in Eq. (4.245))
e
LAM
= e
0
LAM
th
S
0
[ [
LAM
1
H
N
k=1
t
k
Q
_
k
b
_ _
k
: (4:265)
Substituting
b
_ _
k
from Eq. (4.239) and dividing by DT yields
a
LAM
= I [ [
N
k=1
t
k
r
kn
D [ [
k
_ _
a
LAM
0
1
H
N
k=1
t
k
r
kn
D [ [
k
a
k
; (4:266)
where
D [ [
k
= S
0
[ [
LAM
1
E
2
Q
_
k
T [ [
T
k
U [ [
k
T [ [
k
Q
_
k
: (4:267)
The model predictions for longitudinal thermal expansion coefcient are plotted
in Figure 4.23 for a [0/90]
s
carbon/epoxy laminate. For comparison, experimental
and model data from Kim et al. [141] are also shown.
125 4.11 Changes in thermal expansion coefcients
4.12 Summary
This chapter has provided an exposition of the main concepts and methods related to
evaluating the effects of multiple cracking in composite materials on their deform
ational response. Beginning with the early approaches, known as shear lag methods,
where onedimensional stress analysis is used, and progressing to computational
methods that accurately determine the local stress elds, the range is covered as much
as possible. While the eld is still evolving and more methods are appearing in the
literature, it is hoped that the treatments and discussions provided here are useful for
researchers in the eld to gain an appreciation of this class of approaches. A key
consideration in selecting an approach is the purpose at hand. For material selection
purposes a quick assessment of the elastic modulus may be needed, in which case the
shear lag approach may be adequate, while for structural analysis purposes an
evaluation of all properties and the suitability of the approach as an integral part
of a structural analysis scheme would be the factors of consideration. In any case, the
microdamage mechanics (MIDM) treated here is one part of the total damage
mechanics picture. The next chapter on macrodamage mechanics (MADM) pro
vides another perspective on the complex problemof damage in composite materials.
References
1. Z. Hashin, Analysis of damage in composite materials. In Yielding, Damage, and Failure
of Anisotropic Solids, ed. J. P. Boehler. (London: Mechanical Engineering Publications,
1990), pp. 331.
2. S. NematNasser and M. Hori, Micromechanics: Overall Properties of Heterogeneous
Materials, 2nd edn. (Amsterdam: North Holland, 1999).
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Crack density (cracks / mm)
0.2
0.4
0.6
0.8
1.0
a
x
/
a
x
0
[0/90]
s
experimental, Kim et al. (2000)
model
model, Kim et al. (2000)
Figure 4.23. Variation in longitudinal thermal expansion coefcient for a carbon/epoxy
[0
2
/90
2
]
s
laminate. The LundmarkVarna model is compared with model and
experiment from Kim et al. [141]. Reprinted, with kind permission, from P. Lundmark and
J. Varna, Int J Damage Mech, Vol. 14, pp. 23559, copyright #2005 by Sage Publications.
126 Microdamage mechanics
3. J. Aveston, G. A. Cooper, and A. Kelly, Single and multiple fracture. In The Properties
of Fiber Composites. (Surrey, UK: IPC Science and Technology Press, National Phys
ical Laboratory, 1971), pp. 1526.
4. A. Kelly, The 1995 Bakerian Lecture: composite materials. Phil Trans R Soc London
A, 354:1714 (1996), 184174.
5. G. A. Cooper and J. M. Sillwood, Multiple fracture in a steel reinforced epoxy resin
composite. J Mater Sci, 7:3 (1972), 32533.
6. J. Aveston and A. Kelly, Theory of multiple fracture of brous composites. J Mater Sci,
8:3 (1973), 35262.
7. S. W. Wang and A. ParviziMajidi, Experimental characterization of the tensile behav
iour of Nicalon brereinforced calcium aluminosilicate composites. J Mater Sci, 27:20
(1992), 548396.
8. J. E. Masters and K. L. Reifsnider, An investigation of cumulative damage develop
ment in quasiisotropic graphite/epoxy laminates. In Damage in Composite Materials,
ASTM STP 775, ed. K. L. Reifsnider. (Philadelphia, PA: ASTM, 1982), pp. 4062.
9. J. Tong, F. J. Guild, S. L. Ogin, andP. A. Smith, Onmatrix crackgrowthinquasiisotropic
laminates I. Experimental investigation. Compos Sci Technol, 57:11 (1997), 152735.
10. D. T. G. Katerelos, P. Lundmark, J. Varna, and C. Galiotis, Analysis of matrix
cracking in GFRP laminates using Raman spectroscopy. Compos Sci Technol, 67:9
(2007), 194654.
11. T. Yokozeki, T. Aoki, and T. Ishikawa, Consecutive matrix cracking in contiguous
plies of composite laminates. Int J Solids Struct, 42:910 (2005), 2785802.
12. T. Yokozeki, T. Aoki, T. Ogasawara, and T. Ishikawa, Effects of layup angle and ply
thickness on matrix crack interaction in contiguous plies of composite laminates.
Compos A, 36:9 (2005), 122935.
13. M. Kashtalyan and C. Soutis, Stiffness and fracture analysis of laminated composites
with offaxis ply matrix cracking. Compos A, 38:4 (2007), 12629.
14. M. Kashtalyan and C. Soutis, Modelling offaxis ply matrix cracking in
continuous brereinforced polymer matrix composite laminates. J Mater Sci, 41:20
(2006), 678999.
15. L. N. McCartney and G. A. Schoeppner, Predicting the effect of nonuniform ply
cracking on the thermoelastic properties of crossply laminates. Compos Sci Technol,
62:14 (2002), 184156.
16. V. V. Silberschmidt, Effect of microrandomness on macroscopic properties and frac
ture of laminates. J Mater Sci, 41:20 (2006), 676876.
17. V. V. Silberschmidt, Matrix cracking in crossply laminates: effect of randomness.
Compos A, 36 (2005), 12935.
18. L. E. Asp, L. A. Berglund, and R. Talreja, A criterion for crack initiation in glassy
polymers subjected to a compositelike stress state. Compos Sci Technol, 56:11 (1996),
1291301.
19. L. E. Asp, L. A. Berglund, andR. Talreja, Effects of ber andinterphase onmatrixinitiated
transverse failure in polymer composites. Compos Sci Technol, 56:6 (1996), 65765.
20. L. E. Asp, L. A. Berglund, and R. Talreja, Prediction of matrixinitiated transverse
failure in polymer composites. Compos Sci Technol, 56:9 (1996), 108997.
21. H. S. Huang and R. Talreja, Numerical simulation of matrix microcracking in short
ber reinforced polymer composites: initiation and propagation. Compos Sci Technol,
66:15 (2006), 274357.
127 References
22. H. Huang and R. Talreja, Effects of void geometry on elastic properties of
unidirectional ber reinforced composites. Compos Sci Technol, 65:13 (2005),
196481.
23. K. A. Chowdhury, R. Talreja, and A. A. Benzerga, Effects of manufacturinginduced
voids on local failure in polymerbased composites. J Eng Mater Tech, Trans ASME,
130:2 (2008), 021010.
24. K. A. Chowdhury, Damage initiation, progression and failure of polymer based
composites due to manufacturing induced defects. Ph.D. thesis, Texas A&M Univer
sity, College Station, TX (2007).
25. H. L. Cox, The elasticity and strength of paper and other brous materials. Br J Appl
Phys, 3 (1952), 729.
26. J. A. Nairn, On the use of shearlag methods for analysis of stress transfer unidirec
tional composites. Mech Mater, 26:2 (1997), 6380.
27. K. W. Garrett and J. E. Bailey, Multiple transverse fracture in 90 degrees crossply
laminates of a glass berreinforced polyester. J Mater Sci, 12:1 (1977), 15768.
28. A. Parvizi, K. W. Garrett, and J. E. Bailey, Constrained cracking in glass ber
reinforced epoxy crossply laminates. J Mater Sci, 13:1 (1978), 195201.
29. P. W. Manders, T. W. Chou, F. R. Jones, and J. W. Rock, Statistical analysis of
multiple fracture in 0
/90
/0
.S
Figure 5.2. The concept of effective stress for isotropic damage in uniaxial loading.
135 5.1 Introduction
to the rupture of the specimen. Following this approach, Robotnov [3] postulated
the creep strain rate as
de
dt
= B
s
1 o
_ _
n
; (5:4)
where B and n are constants. It should be noted here that Kachanovs
effective stress, as dened, increases to innity at failure, while Robotnovs effect
ive stress, as interpreted, would increase only by a few percent since the volume
fraction of voids or discontinuities at failure is found by microstructural studies
for metals to be small.
Robotvovs concept of effective stress was also used later by Murakami and
Ohno [4] for creep damage of polycrystalline metals and by Lemaitre and co
workers [5, 6] for effective elastic properties. Denoting the quantity in parenthesis
in Eq. (5.4) by
~ s =
s
1 o
; (5:5)
Hookes law can be expressed in two equivalent forms as
s =
~
Ee
e
; ~ s = Ee
e
; (5:6)
where e
e
is the elastic strain and
~
E is the effective Youngs modulus, which is
related to the damage parameter o as follows
~ s =
s
1 o
=
E
~
E
s = o = 1
~
E
E
: (5:7)
In a somewhat similar way, for brittle creep damage, Lemaitre and Chaboche [7]
dened the damage variable in terms of the strain rate. Using a power law, the
strain rate during secondary creep (undamaged state), _ e
s
, can be described by
_ e
s
=
s
l
_ _
N
; (5:8)
where l is a material constant and the exponent N is found from experimental
tests. The damage variable follows easily from the measurement of the strain rate
during tertiary creep, _ e, and the use of effective stress concept as
o = 1
_ e
_ e
s
_ _
1=N
: (5:9)
Similar to the effective stress concept, the strain equivalence principle has also
been proposed by Lemaitre and coworkers [6, 8], which states that any strain
constitutive equation for a damaged material may be derived in the same way as
for a virgin material except that the usual stress is replaced by the effective stress.
In creep of metals as well as in brittle cracking of ceramics, rocks, concrete, etc. the
voids and cracks usually form along some preferred orientations, e.g., on grain
136 Macrodamage mechanics
boundaries and in weak planes. It is obvious that a scalar damage variable cannot
account for the directional dependence of the effects of voids and cracks.
To characterize this effect, Murakami and Ohno [4] considered an arbitrarily oriented
plane in a solid containing arbitrarily distributed voids and dened a secondorder
tensor to describe the anisotropic net area reduction. This tensor was assumed to
represent the directional nature of discontinuities and was called the damage tensor.
The MurakamiOhno damage tensor V relates the elemental area dA of an
arbitrarily oriented plane with the unit normal n lying in the damaged congur
ation to the elemental area dA* of the same plane with the unit normal n* lying in
a ctitious conguration with reduced net load carrying area. Thus,
n
+
dA
+
= I V ( )n dA; (5:10)
where I is the identity tensor of second order.
The effective stress tensor s* then is the magnied (net area) stress in the ctitious
conguration which is related to the stress in the damaged conguration by
s
+
= I V ( )
1
s: (5:11)
Following (5.7), Chaboche [911] generalized the relation between elastic con
stants and the damage variable to the threedimensional case, replacing o by a
damage tensor D. This damage tensor, in general, is a nonsymmetric fourthorder
tensor, and is dened by the following transformation
E
~
= I D ( ) : E; (5:12)
where I is the identity tensor of fourth order, E and E
~
are the elasticity tensor for
the undamaged and damaged materials, respectively, and (:) stands for the second
order contraction. Thus, the damage tensor is given by
D = I E
~
E
1
; (5:13)
and the effective stress tensor is given by
s
+
= I D ( )
1
: s: (5:14)
It is worth noting that the secondorder damage tensor dened by Murakami and
Ohno [4], Eq. (5.10), is based on the notion of effective net area reduction, while
the fourthorder damage tensor dened by Chaboche [911], Eq. (5.13), uses the
notion of the effective stress dened by Eq. (5.6).
The damage tensor D dened by (5.13) has been the focus of much attention in
several works on continuum damage mechanics. Ju [12], for instance, has dis
cussed the notion of isotropic and anisotropic damage variables and has demon
strated that isotropic damage does not necessarily imply a scalar damage variable.
From a mechanics study of changes in the elasticity compliance tensor due to
microcracks he found that the fourthorder tensor D is isotropic for the case of
isotropic damage, e.g., when the cracks are perfectly randomly distributed in all
137 5.1 Introduction
directions. However, for preferred directions in crack orientations the damage
tensor will, in general, be anisotropic.
5.2 Continuum damage mechanics (CDM) of composite materials
The damage variables dened on the basis of the effective stress concept, or
the associated notion of the reduced net area, do not contain any specic details
of the damage entities. In the LemaitreChaboche damage tensor, for instance,
the components of the damage tensor are determined from the elastic compli
ance changes. Thus, e.g., two sets of damage entities of different characteristic
sizes and concentrations leading to the same elastic compliance changes will be
represented by the same variables. Such equivalency of different sets of damage
entities cannot, in general, be expected to hold for all effects of damage.
Furthermore, in media with, say, two levels of microstructure, e.g., in compos
ites, the damage entities forming at the lower sizescale may be affected by the
higher sizescale microstructure as well as by the difference in the symmetry
properties of the two microstructures. To illustrate this aspect, consider an
ellipsoidal particle embedded in a compliant matrix and surrounded by a
parallel array of stiff bers (Figure 5.3). Let the particle debond from the
matrix under appropriate loading such that an ellipsoidal damage entity is
formed. The local stress perturbation induced by the surface displacements of
this damage entity, and the resulting overall elastic compliance, will depend on
the local microstructure details such as inclination of the bers with respect to
the principal planes of the damage entity and the ber stiffness. Thus the same
microstructure arranged differently can produce a different effect on the overall
elastic response. This example serves to illustrate that a characterization of
damage in terms of the elastic compliance changes will not necessarily be
unique.
Figure 5.3. Schematic illustration of anisotropic constraint effect in composites. An
ellipsoid entity is shown surrounded by bers or plies in a composite. The axes of the
ellipsoid are inclined with respect to the axes of the constraining elements. Reprinted, with
kind permission, from Damage Mechanics of Composite Materials, R. Talreja, Damage
characterization by internal variables, pp. 5378, copyright Elsevier (1994).
138 Macrodamage mechanics
In the alternative characterization of damage, adopted here, the damage
entities are represented by appropriate specic variables, and averages of these
variables over an RVE then give the damage variables. The degree of charac
terization, i.e., to what extent the details of damage entities are represented,
depends on the type of variables chosen. Thus, a scalar variable will only
represent the size but not the shape or orientation characteristics of the damage
entities. A vector variable can account for the size and orientation, but the
positive and negative senses of a vector introduce ambiguities in the damage
characterization that cause difculties in the description of damage evolution.
With two vectors simultaneously associated with a damage entity these difcul
ties are overcome. At the same time, the resulting secondorder tensor gives
mathematical convenience of formulating response functions, which are com
monly expressed in terms of the secondorder stress and strain tensors, as
discussed below.
Talrejas original paper on CDM used a vectorial description of damage [13].
Although the ambiguities concerning the sense of damage vector could be remed
ied by imposing an additional condition, it was found that using a secondorder
tensor instead avoided this step [1417]. Here we will follow the tensorial
description.
5.2.1 RVE for damage characterization
Any continuum description of a solid entails homogenization since materials
are inherently heterogeneous. For polycrystalline metals, for instance, the scale
of heterogeneity (e.g., grain size) is often small compared to the scale at which
material response characteristics (e.g., the elastic constants) are measured,
allowing the stress and strain states to be dened as continuous elds. For
commonly used berreinforced solids, such as glass/epoxy and carbon/epoxy,
the ber diameter of approximately 10 micrometers allows treating these
materials as a homogeneous continuum with good accuracy. When internal
surfaces in composite materials form, their characteristic dimensions and
mutual distances between them can be orders of magnitude larger than the
scale of heterogeneities underlying the homogenized pristine composite.
Furthermore, on application of external loads the internal surfaces are subject
to evolution (enlargement and multiplication), as discussed above. This war
rants a separate (and different) homogenization of the composite with internal
surfaces (collectively called damage).
Figure 5.4 depicts a homogenization procedure for a composite solid contain
ing damage. The heterogeneities in the pristine (undamaged) composite are
referred to as stationary microstructure and are homogenized rst. This may
also be called classical homogenization. Textbooks on mechanics of composite
materials usually begin with this homogenization. In fact, the classical laminate
theory goes one step further by developing homogeneous constitutive relations
for laminates consisting of stacked layers of homogenized unidirectionally
139 5.2 Continuum damage mechanics (CDM) of composite materials
reinforced composite (ply or lamina). Returning to Figure 5.4, the second hom
ogenization pertains to the internal surfaces, collectively named as damage or
evolving microstructure to highlight their ability to permanently change by
processes of energy dissipation. Homogenization of the evolving microstructure
necessitates employing the notion of a representative volume element (RVE),
which will be discussed next.
A general and thorough exposition of the RVE notion in the context of micro
mechanics is given in [18]. Here, we shall apply this notion to the particular case
of composite materials with damage. With reference to Figure 5.4 again, a generic
point P in the homogenized composite with damage has associated with it a
damage state (in addition to stress and strain states), which is given by an
appropriate volumeaveraged measure of the presence of internal surfaces that
affect the constitutive behavior (stressstrain relations) at the point P. The volume
over which the averaging is performed must be representative of the neighborhood
of point P that can be associated with P. This neighborhood is the RVE, whose
volume is not xed but depends on the geometrical conguration (size, spacing,
etc.) of the internal surfaces around P. As this conguration changes under
applied loading, the RVE size changes.
With the notion of RVE at hand, the damage state at P can be dened by a
set of variables obtained by averaging over the RVE. The choice of the
variables is guided by the type of internal surfaces formed, and in this respect
P
a
i
n
j
Homogenization of
stationary microstructure
Homogenization of
evolving microstructure
Characterization of a
damage entity
RVE for damage characterization
Fully homogenized continuum
P
V
Step 2
Step 1
Stationary microstructure
Evolving microstructure
Continuum after homogenizing
the stationary microstructures
Figure 5.4. Homogenization of a continuum body with heterogeneous stationary
structure and evolving damage entities. A tensorial characterization of a damage entity
is depicted on the right.
140 Macrodamage mechanics
the knowledge of damage mechanisms discussed above is useful. In general,
the variables can be scalars, vectors, or tensors of second order or higher.
Settling on which variables to employ is a matter of nding a balance between
capturing sufcient physics of the damage process and usefulness of the ensu
ing formulation of constitutive relations. In the following the secondorder
characterization of damage in composite materials adopted by Talreja [14, 15,
17] is described.
5.2.2 Characterization of damage
As shown in Figure 5.4, a single internal surface within a RVE, called a damage
entity from now on, can be characterized by two vectors: a unit outward normal n
at a point on the surface, and an inuence vector a at the same point. A dyadic
product of the two vectors, integrated over the surface S, is denoted the damage
entity tensor, and given by
d
ij
=
_
S
a
i
n
j
dS; (5:15)
where the components of the vectors are with reference to a Cartesian coordinate
system. The dyadic product assures consistency of the signs of the two vectors.
A characterization of this type was rst proposed by Vakulenko and Kachanov
[19] for at cracks, where the inuence vector a represented the displacement jump
across the crack surface.
The physical signicance of this characterization is that it represents the
oriented nature of the presence of internal surfaces. As illustrated by the examples
of damage in Chapter 3, common internal surfaces are cracks (at or curved)
generated by interface debonding and matrix failure. The unit normal vector at a
point on the damage entity carries the information on orientation of the surface
(with respect to the frame of reference), while the other vector represents an
appropriate inuence induced by activation of the considered point on the surface.
This inuence is generally also directed in nature. For the case of mechanical
response, the appropriate inuence would be the displacement of the activated
point on the damage entity surface. For a nonmechanical response, such as
thermal or electrical conductivity, the perturbation induced by an internal surface
can also be cast as a vectorvalued quantity.
Integrating the dyadic product in Eq. (5.15) over the damage entity surface
provides the total net effect of the entity. For example, if the entity is a at crack,
then taking a as the displacement vector in the integral gives the crack surface
separation times the crack surface area. This product may be viewed as an
affected volume associated with the crack. For a pennyshaped crack with the
two surfaces separating symmetrically about the initial crack plane, the sole
surviving term of the damage entity tensor represents an ellipsoidalshaped
volume.
141 5.2 Continuum damage mechanics (CDM) of composite materials
Referring once again to Figure 5.4, the RVE associated with a generic point P
carries a sufciently large number of the discrete damage entities to represent
the collective effect on the homogenized constitutive response at the point. The
number of damage entities needed for this representation, and the consequent
RVE size, depend on the distribution of the entities. For instance, if the entities
are sparsely distributed, then the RVE size would be large, while for densely
distributed case a small RVE would sufce. Furthermore, for uniformly distrib
uted entities of the same geometry, a repeating unit cell containing a single
entity can replace the RVE, while for the cases of nonuniform distribution of
unequal entities, the RVE size will increase until a statistically homogeneous
representation is attained. This implies that further increasing the RVE size will
have no impact on the averages of the selected characteristics. As an example, if
the selected characteristic is the affected matrix volume by a damage entity, as
mentioned above, then the average value of this quantity will vary as the RVE
size increases and will approach a constant value at a certain RVE size. The
minimum RVE size beyond which no appreciable change in the considered
average is found may be taken as the needed RVE. It is apparent that the
RVE is not unique but is subject to the choice made for the particular formula
tion of the constitutive response of a continuum with damage. Consequently,
there is no unique constitutive theory of a continuum with damage; however,
the use of the concept of an internal state in a given theory requires specifying
RVE in a consistent manner and assuring that the conditions for its existence
are present.
From the cases of damage mechanisms reviewed in Chapter 3 it can be noted
that in composite laminates the damage tends to occur as sets of parallel cracks
within the plies, each oriented along bers in the given ply. It is therefore
convenient to separate each set of ply cracks according to its orientation,
referred to a xed frame of reference, and assign it a damage mode number.
Denoting damage mode by a = 1, 2, . . ., n, a damage mode tensor can be
dened as
D
(a)
ij
=
1
V
k
a
d
ij
_ _
k
a
; (5:16)
where k
a
is the number of damage entities in the ath mode, and V is the RVE
volume. As noted above, if the ply cracks of a given orientation are uniformly
spaced, then the RVE will reduce to the unit cell containing one crack. For
nonuniform distribution of ply cracks, V must be large enough to provide a steady
average of the damage mode tensor components.
As dened by Eq. (5.16) the damage mode tensor will in general be asymmet
rical. Decomposing the inuence vector a along directions normal and tangential
to the damage entity surface S gives,
a
i
= an
i
bm
i
; (5:17)
where n and m are unit normal and tangential vectors on S such that n
i
m
i
= 0.
142 Macrodamage mechanics
Using Eq. (5.17) in Eq. (5.16) the damage entity tensor can be written in two
parts as
d
ij
= d
1
ij
d
2
ij
; (5:18)
where
d
1
ij
=
_
S
an
i
n
j
dS; d
2
ij
=
_
S
bm
i
n
j
dS: (5:19)
Here a and b are the magnitudes of the normal and tangential projections,
respectively, of vector a
i
and vectors n
i
and m
j
are unit normal and tangential
vectors, respectively. The damage mode tensor for a given mode can now be
written as
D
(a)
ij
= D
1(a)
ij
D
2(a)
ij
; (5:20)
where
D
1(a)
ij
=
1
V
k
a
d
1
ij
_ _
k
a
; D
2(a)
ij
=
1
V
k
a
d
2
ij
_ _
k
a
: (5:21)
This separation of the damage mode tensor in two parts allows the analysis to
be simplied to avoid having to deal with asymmetric tensors. For instance,
for damage entities consisting of at cracks, the two parts of the damage mode
tensor represent the two crack surface separation modes. If an assumption can
be made that only the symmetric crack surface separation (known as mode I or
crack opening mode in fracture mechanics) is signicant, then the second term
in Eq. (5.21) can be neglected. This will render the damage mode tensor
symmetrical and it can then be written as
D
(a)
ij
= D
1(a)
ij
=
1
V
k
a
_
S
an
i
n
j
dS
_ _
k
a
: (5:22)
The consequence of this assumption was examined by Varna [20] for one class of
laminates and it was found that not including the crack sliding displacement
(CSD) for ply cracks inclined to the laminate symmetry directions resulted in
errors in estimating the degradation of the average elastic properties of laminates.
However, these errors were found to be small in absolute values while being
signicant in percentages. In fact for those ply crack orientations where CSD
dominates, the cracks are difcult to initiate until high loads close to failure load
are applied.
Some damage mechanisms such as crystalline slip may require only the
tangential part. For other situations the sliding between the crack faces can
be negligible, e.g., for intralaminar cracks constrained by stiff plies, and ber/
matrix debonds. Hence D
ij
2(a)
= 0 under such conditions. For cases where the
143 5.2 Continuum damage mechanics (CDM) of composite materials
damage entity surfaces conduct tangential displacements only (e.g., CSD by at
cracks), it is possible to formulate the damage mode tensor as a symmetric tensor.
One example of this is sliding of the ber/matrix interface in ceramic matrix
composites [16].
With stress, strain, and damage, all expressed as symmetric secondorder
tensors, a constitutive theory can now be formulated to have a convenient, usable
form. Such a formulation is described next.
5.2.3 A thermodynamics framework for materials response
Referring once again to Figure 5.4, a formulation of the constitutive response
of a homogenized continuum with damage will now be discussed. In view of the
observed behavior of common composite materials such as glass/epoxy and
carbon/epoxy, only elastic response will be considered. Theoretical treatment of
elastic response of solids is classical and can be found in textbooks. Incorpor
ating damage is, however, not a simple extension of the classical theory of
elasticity. The CDM approach to be described here is based on thermodynam
ics and is naturally suited for thermomechanical response. It can be extended to
incorporate nonmechanical effects, such as electrical and magnetic, as well as
chemical. Every extension, however, comes with the price of having to deter
mine associated response coefcients (material constants) by a certain identi
cation procedure. In the treatment presented here, the task of determining
material constants is reduced by use of selected micromechanics. This way of
combining micromechanics with CDM generates useful synergism, justifying
the characterization of the combined approach as synergistic damage mechan
ics (SDM), to be discussed later. We begin with the conventional CDM frame
work rst.
At the foundation of CDM are the rst and second laws of thermodynamics.
Additionally, use is made of the concept of an internal state, which is identied here
as the evolving microstructure depicted in Figure 5.4. As discussed before, this
microstructure is homogenized into a damage eld characterized by the set of damage
mode tensors D
ij
(a)
. The collectionof all variables resulting fromthermodynamics with
internal state can now be placed in two categories: state variables and response
functions (see Table 5.1). The thermodynamical response of a composite body, limited
to small strains, is given by a set of ve response functions: the Cauchy stress tensor s
ij
,
Table 5.1 Thermodynamics variables for damage analysis
Thermodynamic state variables Thermodynamic response variables
1. Strain tensor e
ij
=
1
2
u
i; j
u
j;i
_ _
1. Cauchy stress tensor s
ij
2. Temperature T 2. Specic Helmholtz free energy c
3. Temperature gradient g
i
= T
,i
3. Specic entropy
4. Damage tensors D
ij
(a)
4. Heat ux vector q
i
5. Damage rate tensors
_
D
(a)
ij
144 Macrodamage mechanics
the specic Helmholtz free energy c, the specic entropy , the heat ux vector q
i
, and
a set of damage rate tensors
_
D
(a)
ij
, a =1,2, ..., n. The thermodynamic state of the body is
givenby the straintensor e
ij
= 1=2 ( ) u
i; j
u
j;i
_ _
, withdisplacement vector u
i
, absolute
temperature T, temperature gradient g
i
= T
,i
, and a set of damage tensors D
ij
(a)
.
Following Truesdells principle of equipresence, which states that all state
variables should be present in all response functions unless thermodynamics or
other relevant considerations preclude their dependency, we write
s
ij
= s
ij
e
kl
; T; g
k
; D
(a)
kl
_ _
;
c = c e
kl
; T; g
k
; D
(a)
kl
_ _
;
= e
kl
; T; g
k
; D
(a)
kl
_ _
;
q = q e
kl
; T; g
k
; D
(a)
kl
_ _
;
_
D
(a)
kl
=
_
D
(a)
kl
e
kl
; T; g
k
; D
(b)
kl
_ _
:
(5:23)
The following balance laws should hold for the continuum body:
v Balance of linear momentum:
s
ij;j
rb
j
= r x
j
; (5:24)
where b
j
are the components of the body force per unit mass and r is the mass
density.
v Balance of angular momentum:
s
ij
= s
ji
: (5:25)
v Balance of energy:
r _ u s
ij
_ e
ij
q
i;i
= rr; (5:26)
where u is the specic internal energy per unit mass and r is the heat supply per
unit mass.
v Second law of thermodynamics in the form of the ClausiusDuhem
inequality:
s
ij
_ e
ij
r
_
c r
_
T
q
i
g
i
T
_ 0; (5:27)
where c = u T.
Time differentiation of c in (5.23) gives,
_
c =
@c
@e
kl
_ e
kl
@c
@T
_
T
@c
@g
i
_ g
i
@c
@D
(a)
kl
_
D
(a)
kl
: (5:28)
Substitution of Eq. (5.28) into Eq. (5.27) gives
145 5.2 Continuum damage mechanics (CDM) of composite materials
s
ij
r
@c
@e
kl
_ _
_ e
ij
r
@c
@T
_ _
_
T r
@c
@g
i
_ g
i
r
a
@c
@D
(a)
kl
_
D
(a)
kl
q
i
g
i
T
_ 0; (5:29)
where the summation sign is shown on a to include all damage modes. Now
requiring (5.29) to hold for the independently varying strain, temperature, and
temperature gradient gives the following results
s
ij
= r
@c
@e
kl
; (5:30)
=
@c
@T
; (5:31)
@c
@g
i
= 0 : (5:32)
Equation (5.32) states that the Helmholz free energy function does not depend on
the temperature gradient and, consequently, Eqs. (5.30) and (5.31) eliminate this
dependency from stress and entropy. The last two equations in the response
function set (Eq. (5.23)) remain unaffected.
Equations (5.30)(5.32) lead to the following functional dependencies
s
ij
= s
ij
e
kl
; T; D
(a)
kl
_ _
;
c = c e
kl
; T; D
(a)
kl
_ _
;
= e
kl
; T; D
(a)
kl
_ _
;
q = q e
kl
; T; g
k
; D
(a)
kl
_ _
;
_
D
(a)
kl
=
_
D
(a)
kl
e
kl
; T; g
k
; D
(b)
kl
_ _
:
(5:33)
From Eq. (5.29) the following restriction (also known as internal dissipation
inequality) results
a
R
(a)
kl
_
D
(a)
kl
q
i
g
i
T
_ 0; (5:34)
where R
kl
(a)
are the thermodynamic forces conjugate to D
kl
(a)
and are given by
R
(a)
kl
= r
@c
@D
(a)
kl
: (5:35)
Each of these forces is analogous to the crack extension force (i.e., energy release
rate) for a single crack. As an example, for a damage mode component, say
D
11
of mode a = 1, the quantity R
11
(1)
can be interpreted as the force causing
an innitesimal change in the internal state represented by D
11
(1)
. Equation (5.34)
146 Macrodamage mechanics
expresses the condition these forces must satisfy as damage evolves under thermo
mechanical impulses.
The complete thermomechanical response for the composite body is governed
by the set of functions Eq. (5.33) and the internal dissipation inequality Eq. (5.34).
In view of the experimental data, which are mostly available for polymer matrix
composites at room temperature, the thermomechanical framework will be
developed further for the purely mechanical response. Thus, for isothermal condi
tions (T = 0, g
i
= 0) the set of response functions is reduced to the following
s
ij
= s
ij
e
kl
; D
(a)
kl
_ _
;
c = c e
kl
; D
(a)
kl
_ _
;
_
D
(a)
kl
=
_
D
(a)
kl
e
kl
; g
k
; D
(b)
kl
_ _
;
a
R
(a)
kl
_
D
(a)
kl
_ 0 :
(5:36)
Since s
ij
are derivable fromc, according to Eq. (5.30), it sufces to formulate c and
_
D
(a)
kl
for a purely mechanical response. Dealing with this scalarvalued function (c) as
the sole response function for a given internal state of damage provides a favorable
situation for further development of the theory. The form of the Helmholz free
energy function can be chosen in different ways. A powerful way is possible by use
of the theory of invariants for polynomial functions [21]. In the following it is
illustrated for one case of orthotropic composites containing one damage mode.
To derive the rate equations for s
ij
, we differentiate Eq. (5.30) with respect to
time and use the functional dependency given in Eq. (5.36), which yields
_ s
ij
= r
@
2
c
@e
ij
@e
kl
_ e
kl
r
a
@
2
c
@e
ij
@D
(a)
mn
_
D
(a)
mn
: (5:37)
Substitution of Eq. (5.35) gives
_ s
ij
= r
@
2
c
@e
ij
@e
kl
_ e
kl
a
@R
(a)
mn
@e
ij
_
D
(a)
mn
; (5:38)
which may be rewritten as
_ s
ij
= C
ijkl
_ e
kl
a
K
(a)
ijmn
_
D
(a)
mn
; (5:39)
where
C
ijkl
= r
@
2
c
@e
ij
@e
kl
(5:40)
and
147 5.2 Continuum damage mechanics (CDM) of composite materials
K
(a)
ijmn
=
@R
(a)
mn
@e
ij
= r
@
2
c
@e
ij
@D
(a)
mn
: (5:41)
The matrix C
ijkl
contains stiffness coefcients of the composite in its virgin mater
ial state, whereas coefcients of K
ijmn
(a)
are functions determining the change of state
caused by the internal dissipative mechanisms. The rate shown in Eq. (5.39)
requires formulation of the scalar function c and the tensor components
_
D
(a)
kl
.
The discussion of damage evolution is left for later; the next section will treat the
stressstrain relations at a xed damage state.
5.2.4 Stiffnessdamage relationships
The stiffness coefcient matrix C
ijkl
of the composite in a given state of damage
is derivable from the Helmholtz free energy function c according to Eq. (5.40).
The scalar valued function c can be written as a polynomial in its variables, i.e.,
c = c
P
e
ij
; D
(a)
ij
_ _
; (5:42)
where c
P
stands for the polynomial function.
Let us now focus on the particular case of intralaminar cracking in composite
laminates. Figure 5.5 shows an RVE illustrating one set of intralaminar cracks in
an offaxis ply of a composite laminate. Although for clarity of illustration the
cracking is shown only in one lamina, it is understood that in general it exists in
multiple plies of the laminate. The thickness of the cracked plies is denoted by t
c
, s
is the average crack spacing, t is the total laminate thickness, and W and L
stand for the width and the length, respectively, of the RVE. The volume of the
RVE, the surface area of a crack, S, and the inuence vector magnitude, a, are
specied as
V = L:W:t;
S =
W:t
c
sin y
;
a = kt
c
;
(5:43)
where k, called the constraint parameter, is an unspecied constant of (assumed)
proportionality between a and the crack size t
c
(also the crackedply thickness).
Here, y is a positive quantity so that the surface area is always positive. Assuming
a to be constant over the crack surface S, one gets from Eq. (5.22)
D
(a)
ij
=
kt
2
c
st sin y
n
i
n
j
; (5:44)
where n
i
= sin y; cos y; 0 ( ).
Expansion of the polynomial function in Eq. (5.42) can in general have innite
terms, which will obviously present an impractical situation. One way to restrict
the functional form is by expanding the polynomial in terms that account for the
148 Macrodamage mechanics
initial material symmetry. This is done in the polynomial invariant theory by using
the socalled integrity bases [21]. Such bases have been developed for scalar
functions of various vector and tensor variables. For the case of two symmetric
secondorder tensors, such as in Eq. (5.42), the integrity bases for orthotropic
symmetry are given by Adkins [22]. Considering a single damage mode, a = 1, we
have the following set of invariant terms
e
11
; e
22
; e
33
; e
2
23
; e
2
31
; e
2
12
; e
23
e
31
e
12
;
D
11
; D
22
; D
33
; D
2
23
; D
2
31
; D
2
12
; D
23
D
31
D
12
;
e
23
D
23
; e
31
D
31
; e
12
D
12
;
e
31
e
12
D
23
; e
12
e
23
D
31
; e
23
e
31
D
12
;
e
23
D
31
D
12
; e
31
D
12
D
23
; e
12
D
23
D
31
:
(5:45)
For the sake of applying the constitutive theory to thin laminates where only
inplane strains are of interest, and for small strains, the expansion of the function
c (Eq. (5.42)) can be restricted to no more than quadratic terms in strain compon
ents e
11
, e
22
, and e
12
To what extent the damage tensor components are to be
taken in the expansion depends on the nature and amount of information that can
Figure 5.5. A representative volume element illustrating intralaminar multiple cracking in
a general offaxis ply of a composite laminate.
149 5.2 Continuum damage mechanics (CDM) of composite materials
be acquired for evaluation of the material constants that will appear in the
polynomial function. This issue will be discussed later. To begin with the simplest
possible case, we will include only linear terms in D
11
, D
22
, and D
12
, which are the
nonzero components for intralaminar cracks. Thus the set of invariant terms
reduces to
e
1
; e
2
; e
2
6
;
D
1
; D
2
; D
2
6
;
e
6
D
6
;
(5:46)
where e
1
= e
11
; e
2
= e
22
; e
6
= e
12
; D
1
= D
11
; D
2
= D
22
; D
6
= D
12
. The most general
polynomial form for the Helmholtz free energy, restricted to secondorder terms
in the strain components and rstorder terms in damage tensor components, is
given by
rc = P
0
c
1
e
2
1
c
2
e
2
2
c
3
e
2
6
c
4
e
1
e
2
_ _
c
5
e
2
1
D
1
c
6
e
2
1
D
2
_ _
c
7
e
2
2
D
1
c
8
e
2
2
D
2
_ _
c
9
e
2
6
D
1
c
10
e
2
6
D
2
_ _
c
11
e
1
e
2
D
1
c
12
e
1
e
2
D
2
c
13
e
1
e
6
D
6
c
14
e
2
e
6
D
6
P
1
(e
p
; D
q
) P
2
(D
q
);
(5:47)
where P
0
and c
i
, i = 1, 2, . . ., 14 are material constants, P
1
is a linear function of
strain and damage tensor components, and P
2
is a linear function of damage
tensor components. Setting the free energy to zero for unstrained and undamaged
material, we have P
0
= 0, and assuming the unstrained material of any damaged
state to be stress free, we get P
1
= 0. The stress components in the Voigt notation
are now given by (from Eq. (5.30))
s
p
= r
@c
@e
p
; (5:48)
where p = 1, 2, 6. A differential in stress can now be written as
ds
p
= r
@c
@e
p
@e
q
de
q
r
@c
@e
p
@D
r
dD
r
= C
pq
de
q
K
pr
dD
r
; (5:49)
where
C
pq
= r
@c
@e
p
@e
q
(5:50)
is the stiffness matrix when dD
r
= 0, i.e., at constant damage. This is illustrated for
the uniaxial stressstrain response in Figure 5.6. As seen there, the elastic modulus
at any point on the stressstrain curve is the secant modulus, not the tangent
modulus.
Combining Eqs. (5.47), (5.48), and (5.50), one obtains
C
pq
= C
0
pq
C
(1)
pq
(5:51)
150 Macrodamage mechanics
where
C
0
pq
=
2c
1
c
4
0
2c
2
0
Symm 2c
3
_
_
_
_
=
E
0
x
1 n
0
xy
n
0
yx
n
0
xy
E
0
y
1 n
0
xy
n
0
yx
0
E
0
y
1 n
0
xy
n
0
yx
0
Symm G
0
xy
_
_
_
_
(5:52)
represents the orthotropic stiffness matrix for virgin composite material, in which
E
0
x
; E
0
y
; n
0
xy
; G
0
xy
are effective moduli for the undamaged laminate, and
C
(1)
pq
=
2c
5
D
1
2c
6
D
2
c
11
D
1
c
12
D
2
c
13
D
6
2c
7
D
1
2c
8
D
2
c
14
D
6
Symm 2c
9
D
1
2c
10
D
2
_
_
_
_; (5:53)
represents the stiffness change brought about by the damage entities of damage
mode 1.
It can be noted here that Eqs. (5.51)(5.53) show linear dependence of the
stiffness properties on damage tensor components. This is the consequence of
including only linear terms in these components in the polynomial expansion of
the free energy function, Eq. (5.47). Including higherorder terms will add additional
constants c
i
, which will need to be evaluated. The evaluation procedure is described
below, but it is remarked here that the formulation of constitutive response is in
no way restricted only to linear dependence on the chosen damage measure.
E
0
E
s
e
Figure 5.6. Stressstrain curve of a composite with damage. The secant modulus E varies
with the state of damage.
151 5.2 Continuum damage mechanics (CDM) of composite materials
Case 1: Cracking in one offaxis orientation
From Eqs. (5.51)(5.53), it can be seen that the presence of damage entities, even
for one orientation, removes the initial orthotropic symmetry. For the case of
intralaminar cracks in one orientation, as illustrated in Figure 5.5, the nonzero
components of the damage tensor D
ij
(1)
are given from Eq. (5.44) by
D
1
= D
(1)
11
=
kt
2
c
st
sin y;
D
2
= D
(1)
22
=
kt
2
c
st
cos
2
y
sin y
;
D
6
= D
(1)
12
=
kt
2
c
st
cos y:
(5:54)
Inserting these into Eq. (5.51) and using Eq. (5.52) and Eq. (5.54) we obtain
the stiffness matrix of the damaged composite laminate for a xed state of
damage as
C
pq
=
E
0
x
1 n
0
xy
n
0
yx
n
0
xy
E
0
y
1 n
0
xy
n
0
yx
0
E
0
y
1 n
0
xy
n
0
yx
0
Symm G
0
xy
_
_
_
kt
2
c
st
sin y
2c
5
2c
6
cot
2
y c
11
c
12
cot
2
y c
13
cot y
2c
7
2c
8
cot
2
y c
14
cot y
Symm 2c
9
2c
10
cot
2
y
_
_
_
_
:
(5:55)
Case 2: Crossply laminates
For the special case of crossply laminates, y = 90
, and hence
C
pq
=
E
0
x
1 n
0
xy
n
0
yx
n
0
xy
E
0
y
1 n
0
xy
n
0
yx
0
E
0
y
1 n
0
xy
n
0
yx
0
Symm G
0
xy
_
_
_
kt
2
c
st
2a
1
a
4
0
2a
2
0
Symm 2a
3
_
_
_
_ (5:56)
where a
1
= c
5
; a
2
= c
7
; a
3
= c
9
; and a
4
= c
11
. It can be observed that the ortho
tropic symmetry is retained for intralaminar cracking in crossply laminates.
The engineering moduli can be derived from the following relationships
152 Macrodamage mechanics
E
x
=
C
11
C
22
C
2
12
C
22
; E
y
=
C
11
C
22
C
2
12
C
11
;
n
xy
=
C
11
C
22
; G
xy
= C
66
:
(5:57)
Thus, for crossply laminates with 90
ply cracks,
E
x
=
E
0
x
1 n
0
xy
n
0
yx
2
kt
2
c
st
a
1
n
0
xy
E
0
y
1 n
0
xy
n
0
yx
kt
2
c
st
a
4
_ _
2
E
0
y
1 n
0
xy
n
0
yx
2
kt
2
c
sin y
st
a
2
;
E
y
=
E
0
y
1 n
0
xy
n
0
yx
2
kt
2
c
st
a
2
n
0
xy
E
0
y
1 n
0
xy
n
0
yx
kt
2
c
st
a
4
_ _
2
E
0
x
1 n
0
xy
n
0
yx
2
kt
2
c
st
a
1
;
n
xy
=
n
0
xy
E
0
y
1 n
0
xy
n
0
yx
kt
2
c
st
a
4
E
0
y
1 n
0
xy
n
0
yx
2
kt
2
c
st
a
2
;
G
xy
= G
0
xy
2
kt
2
c
st
a
3
:
(5:58)
Evaluation of material constants
In the damagestiffness relations Eq. (5.56) and Eq. (5.58), a
i
,i = 1,2,3,4 are a set of
four phenomenological constants, which need to be determined to predict stiffness
degradation. As seen from Eq. (5.58), the shear modulus is uncoupled from the
other three moduli and thus can be treated independently. These phenomeno
logical constants are material and laminate conguration dependent and can be
evaluated for a selected laminate by using data generated either experimentally or
by an analytical or a computational model. As an example, let the moduli of a
given damaged crossply laminate at a xed state of damage, s = s
1
, be given as
E
x
; E
y
; G
xy
; and n
xy
, then by using Eq. (5.56), we obtain
a
1
=
s
1
t
2kt
2
c
E
x
1 n
xy
n
yx
E
0
x
1 n
0
xy
n
0
yx
_ _
;
a
2
=
s
1
t
2kt
2
c
E
y
1 n
xy
n
yx
E
0
y
1 n
0
xy
n
0
yx
_ _
;
a
3
=
s
1
t
2kt
2
c
G
xy
G
0
xy
_ _
;
a
4
=
s
1
t
kt
2
c
n
xy
E
y
1 n
xy
n
yx
n
0
xy
E
0
y
1 n
0
xy
n
0
yx
_ _
:
(5:59)
153 5.2 Continuum damage mechanics (CDM) of composite materials
In the above expressions it should be noted that while the values of a
i
are xed for
a given composite laminate (that has been homogenized), the parameter k depends
on the ability of the cracks to perform surface displacements under an applied
mechanical impulse. Thus this parameter may be viewed as a measure of the
constraint to the crack surface separation imposed by the material surrounding
the crack. One way to view this is by considering a crack of a given size embedded
in an innite isotropic material, in which case the crack surface separation is
unconstrained and can be calculated by fracture mechanics methods. When the
laminate geometry is nite and its symmetry is different from isotropic, the k
parameter will take a value less than that for the innite isotropic medium. This
consideration allows us to assign k an undetermined value, say k
0
, for a reference
laminate under reference loading conditions, and evaluate a change from this
value for another crack orientation. This procedure will be discussed in more
detail later.
The CDM approach described here has been used to successfully predict
degradation in the longitudinal and transverse moduli and the Poissons ratio
for a variety of laminates, e.g., [0/90
3
]
s
, [90
3
/0]
s
, and [0/45]
s
as reported in [1315,
17, 23]. Figure 5.7 shows the degradation of the longitudinal Youngs modulus
in a [0/90
3
]
s
glass/epoxy laminate with the applied tensile stress. The predictions
agree with the observed values fairly well in the entire range of cracking.
The predictions by the ply discount method are found to overestimate the total
modulus reduction. The stressstrain curve constructed from the reduced modulus
is shown in Figure 5.8.
0
0.5
0.6
0.7
0.8
0.9
1.0
E
1
/E
1
0
100 200 300
Gl./Ep. [0/90
3
]
s
Calculated
Measured
Ply discount prediction
s
(MPa)
x
x
x
x
x
x
x
x
x x
x
x
x
x
Figure 5.7. Variation of the longitudinal Youngs modulus with applied stress for a glass/
epoxy [0/90
3
]
s
laminate. Source: [23].
154 Macrodamage mechanics
5.3 Synergistic damage mechanics (SDM)
The observation that the kparameter (hitherto referred to as constraint param
eter) may be viewed as a carrier of the local effects on damage entities within a
RVE, while the a
i
constants are material constants, led to a number of studies to
explore prediction of elastic property changes due to damage in different modes.
To be sure, the elastic properties are the averages over appropriate RVEs.
At rst it was found that from changes in E
x
and n
xy
due to transverse cracking
in [0/90
3
]
s
glass/epoxy laminates reported in [24] and assuming no changes in E
y
,
changes in E
x
for the same glass/epoxy of [0/90]
s
conguration could be predicted
with good accuracy. Also, in [0/45]
s
laminates of the same glass/epoxy, the
change in E
x
could be predicted by setting D
1
= D
2
(a good approximation,
supported by crack density data). These results have been reported in [14].
Later, a systematic study of the effect of constraint on the constraint parameter
was done by experimentally measuring the crack opening displacement (COD)
in [y/90
2
]
s
laminates [25] for different yvalues. By relating these values to the
COD at y = 90
as the reference,
the kparameter for other ply orientations was evaluated from the COD values
and E
x
and n
xy
for different y were then predicted [33].
While the experimental studies supported the idea of using the constraint
parameter as a carrier of local constraints, the scatter in test data and the cost of
0 0.5 1.0 1.5 2.0 2.5 e (%)
0
50
100
150
200
250
s
(MPa)
Gl./Ep. [0/90
3
]
s
Figure 5.8. Longitudinal stressstrain response for a glass/epoxy [0/90
3
]
s
laminate.
Source: [23].
155 5.3 Synergistic damage mechanics (SDM)
testing do not make the experimental approach attractive. Therefore, another
systematic study of [0
m
/y
n
/0
m/2
]
s
laminates was undertaken [28, 29] where com
putational micromechanics was employed instead of physical testing. An elaborate
parametric study of the constraint parameter allowed developing a master curve
for elastic property predictions. The most recent study [30] examines damage
modes consisting of transverse ply cracks as well as inclined cracks of different
orientations in [0
m
/y
n
/90
r
]
s
and [0
m
/90
r
/y
n
]
s
laminates. The reference value of k
for y = 90
_
_
_
C
(2)
pq
=
2c
7
D
(2)
1
2c
8
D
(2)
2
c
19
D
(2)
1
c
20
D
(2)
2
c
22
D
(2)
6
2c
11
D
(2)
1
2c
12
D
(2)
2
c
24
D
(2)
6
Symm 2c
15
D
(2)
1
2c
16
D
(2)
2
_
_
_
_
:
(5:64)
157 5.3 Synergistic damage mechanics (SDM)
In general, for N damage modes, Eq. (5.63) can be written as
C
pq
= C
0
pq
N
a=1
C
(a)
pq
: (5:65)
Let us now consider a special case of a general laminate undergoing damage in two
symmetrically placed damage modes, such as 0
m
=y
n
=
p
_
s
, with restricted to
angles that do not cause ply cracking. In such laminates, an inplane tensile
loading will produce an inplane stress state in each offaxis ply consisting
of normal stresses along and perpendicular to bers in that ply and a shear stress
in the plane of the ply. Depending on the values of y, , and ply properties,
the stress perpendicular to the bers could be tensile or compressive. Thus,
on loading, an offaxis ply may or may not develop intralaminar cracks. When
y = 90
to 90
t
2
c
s
y
n
t
sin
2
y; D
(1)
2
=
k
y
t
2
c
s
y
n
t
cos
2
y; D
(1)
6
=
k
y
t
2
c
s
y
n
t
sin y cos y;
a = 2 : n
(2)
i
= sin y; cos y; 0 ( );
D
(2)
1
=
k
y
t
2
c
s
y
n
t
sin
2
y; D
(2)
2
=
k
y
t
2
c
s
y
n
t
cos
2
y; D
(2)
6
=
k
y
t
2
c
s
y
n
t
sin y cos y;
(5:67)
where the superscripts y
+
and y
= k
y
= k
y
; s
y
n
= s
y
n
= s
y
n
: (5:68)
Substituting Eq. (5.67)(5.68) into Eq. (5.64), we obtain
C
(1)
11
C
(2)
11
= 2
k
y
t
2
c
s
y
n
t
c
5
c
7
( )sin
2
y c
6
c
8
( )cos
2
y
_
;
C
(1)
22
C
(2)
22
= 2
k
y
t
2
c
s
y
n
t
c
9
c
11
( )sin
2
y c
10
c
12
( )cos
2
y
_
;
C
(1)
66
C
(2)
66
= 2
k
y
t
2
c
s
y
n
t
c
13
c
15
( )sin
2
y c
14
c
16
( )cos
2
y
_
;
C
(1)
12
C
(2)
12
=
k
y
t
2
c
s
y
n
t
c
17
c
19
( )sin
2
y c
18
c
20
( )cos
2
y
_
;
C
(1)
16
C
(2)
16
=
k
y
t
2
c
s
y
n
t
sin y cos y c
21
c
22
[ [ = 0;
C
(1)
26
C
(2)
26
=
k
y
t
2
c
s
y
n
t
sin y cos y c
23
c
24
[ [ = 0 :
(5:69)
Thus,
C
(1)
pq
C
(2)
pq
=
2a
1
D
1
2b
1
D
2
a
4
D
1
b
4
D
2
0
2a
2
D
1
2b
2
D
2
0
Symm 2a
3
D
1
2b
3
D
2
_
_
_
_; (5:70)
where the superscripts for denoting damage mode have beendroppedfor convenience,
and a
i
and b
i
, i = 1, 2, 3, 4 are the two sets of four material constants, given by
a
1
= c
5
c
7
; a
2
= c
9
c
11
; a
3
= c
13
c
15
; a
4
= c
17
c
19
;
b
1
= c
6
c
8
; b
2
= c
10
c
12
; b
3
= c
14
c
16
; b
4
= c
18
c
20
:
(5:71)
Here, a
i
and b
i
are functions of y. Denote
159 5.3 Synergistic damage mechanics (SDM)
a
1
y ( ) = a
1
sin
2
y b
1
cos
2
y;
a
2
y ( ) = a
2
sin
2
y b
2
cos
2
y;
a
3
y ( ) = a
3
sin
2
y b
3
cos
2
y;
a
4
y ( ) = a
4
sin
2
y b
4
cos
2
y :
(5:72)
Then,
C
(1)
pq
C
(2)
pq
=D
y
2a
1
y ( ) a
4
y ( ) 0
2a
2
y ( ) 0
Symm 2a
3
y ( )
_
_
_
_; (5:73)
where
D
y
=
k
y
t
2
c
s
y
n
t
: (5:74)
Rewrite Eq. (5.72) as
a
i
y ( ) = a
i
sin
2
y b
i
cos
2
y = a
i
sin
2
y 1
b
i
a
i
cot
2
y
_ _
: (5:75)
Consider for the moment the case when a
i
_ b
i
. Then,
b
i
a
i
cot
2
y _ 1 for
p
4
_ y _
p
2
: (5:76)
Also, it can be expected that
b
i
a
i
cot
2
y 1 for
p
3
_ y _
p
2
;
i:e:; a
i
y ( ) ~ a
i
for
p
3
_ y _
p
2
:
(5:77)
For this case, we have laminate stiffness matrix as
C
pq
=
E
0
x
1 n
0
xy
n
0
yx
n
0
xy
E
0
y
1 n
0
xy
n
0
yx
0
E
0
y
1 n
0
xy
n
0
yx
0
Symm G
0
xy
_
_
_
_
D
y
2a
1
a
4
0
2a
2
0
Symm 2a
3
_
_
_
_
; (5:78)
which is of identical form to the expression for 90
cracking in a crossply
laminate, see Eq. (5.56). The moduli can be nally obtained using the relations
in Eq. (5.57). The resultant expressions will be of the same form as in Eq. (5.58),
except that kt
c
2
/st is now replaced with D
y
.
160 Macrodamage mechanics
The overall SDM procedure for this laminate is sketched in Figure 5.10.
As illustrated, it combines micromechanics with CDM for complete evaluation
of the structural response. Micromechanics involves analysis to determine CODs
in cracked plies within a RVE (or unit cell, if applicable) of [0
m
/y
n
/0
m/2
]
s
layup,
from which the constraint effect is evaluated for different y and/or m.
The constraint effect is carried over in the CDM formulation through the con
straint parameter. In a separate step, the damage constants a
i
are determined
from experimental data for a reference laminate, which is chosen here to be
[0/90
8
/0
1/2
]
s
. A crossply laminate of the same class as the ply layup and material
in consideration is a good choice for the reference laminate because either the
experimental data are often available or can be obtained by using any of the
damage models for 90
EXPERIMENTAL/
COMPUTATIONAL
Determine COD and constraint parameter(s)
; = = +
(
( )
(
(
)
u
2
u
2
u
2
n
=
q
90
q
n q
n +q
n
q
b
Evaluate damage constants
using available data
for reference laminate
configuration [0/90
8
/0
1/2
]
s
90
8
( )
u
2 ( )
u
2
Figure 5.10. Flowchart showing the multiscale synergistic methodology for analyzing
damage behavior in a class of symmetric laminates with layup [0
m
/y
n
/0
m/2
]
s
containing
ply cracks in the +y and y layers.
161 5.3 Synergistic damage mechanics (SDM)
with boundary conditions and coordinate systems, is shown in Figure 5.11 [29, 31,
32]. As can be seen from Figure 5.12, the computed CODs averaged over the
thickness of the cracked plies agree quite well with the experimental data over
Figure 5.11. Representative unit cell for COD computation for laminates. Reprinted, with
kind permission, from Int J Solids Struct, Vol. 45, C.V. Singh and R. Talreja, Analysis of
multiple offaxis ply cracks in composite laminates, pp. 457489, copyright Elsevier (2008).
20 30 40 50 60 70 80 90
Ply orientation (q)
0
2
4
6
8
C
O
D
(
m
)
FEM
Experiment
Figure 5.12. Variation of average COD with change in ply orientation for a [0
m
/y
n
/0
m/2
]
s
glass/epoxy laminate. Reprinted, with kind permission, from Int J Solids Struct, Vol. 45,
C.V. Singh and R. Talreja, Analysis of multiple offaxis ply cracks in composite laminates,
pp. 457489, copyright Elsevier (2008).
162 Macrodamage mechanics
the whole range of ply orientations considered. This suggests that the 3DFEanalysis
is an accurate tool to evaluate the constraint parameter. The proles of COD nor
malized with t
c
through the ply thickness are shown in Figure 5.13. As expected, for
crossply laminates, the prole is symmetric about midplane of the cracked ply and
consequently the maximum COD occurs at the midplane of the cracked layer.
However, due to the difference in constraint from the surrounding material, this
COD prole is different from an elliptical prole for a single crack in an innite
isotropic elastic medium subjected to a uniform fareld stress (see Figure 5.13 (b)).
Thus, the magnitude of the average COD for a 90
) and maximumCODdoes
not occur midway through the thickness of the cracked y layers. The aspect ratio of
COD proles, g = u
2
( )
max
_
u
2
, varies from 1.33 for y = 90
to 1.40 for y = 40
(Figure 5.13 (a)), which is different from the aspect ratio of 1.273 ( = 4/p) for an
(a)
1
0
3
u
2
/
t
c
(b)
1
0
3
u
2
/
t
c
(c)
1
0
3
u
2
/
t
c
(d)
1
0
3
u
2
/
t
c
Figure 5.13. COD proles for cracked plies in [0/y
4
/0
1/2
]
s
laminates: (a) CODs averaged
over +y and yplies; (b) Crack prole for 90ply crack compared with an elliptic prole
for an isotropic medium; (c), (d) COD proles for +y and y separately: (c) y = 70,
(d) y = 40. Parts (c), (d) depict the asymmetry of opening displacements for offaxis
laminates, especially at a ply orientation farther from y = 90
and 55
plies, respectively.
Figure 5.15 shows the variation of average COD for changes in these parameters.
The computed average CODs are given by the following parametric equation
u
2
_ _
y
n
= U:f
1
y ( ):f
2
r ( ):f
3
m ( ):f
4
n ( ); (5:79)
where U is the average COD for the reference laminate [0/90
8
/0
1/2
]
s
, and
f
1
y ( ) = sin
2
y;
f
2
r ( ) = r
c
1
;
f
3
m ( ) =
c
2
m
c
3
;
f
4
n ( ) = c
4
n
c
5
;
(5:80)
are tted functions where tting constants (unitless) c
1
c
5
are given as: c
1
=
0.0871; c
2
= 0.1038; c
3
= 0.8949; c
4
= 0.247; c
5
= 0.99 for [0
m
/y
n
/0
m/2
]
s
glass/
epoxy laminates.
The parametric study described above enables us to predict stiffness degrad
ation in offaxis laminates with different geometry and stiffness values. For
example, one can consider a laminate with stiffer outer plies. The variation of
engineering moduli E
x
and n
xy
for different stiffness ratios (r) for a [0/70
4
/0
1/2
]
s
laminate is shown in Figure 5.16(a). As expected, stiffer outer plies cause less
severe degradation in the moduli. In contrast to changing the stiffness of outer
plies, one can vary the number of constraining plies (m) or cracked plies (n). The
effect of number of cracked plies on the change in stiffness moduli is shown in
Figure 5.16(b). These results indicate that the cracking ply thickness, i.e., crack
size, has signicant effect on stiffness degradation while the thickness of the
constraining plies as well as the change in axial stiffness ratio r have small effect.
5.3.2 Three damage modes
We will now consider a special case of cracking in three offaxis plies, of which two
are symmetrically opposite of each other, and the third is placed along the
transverse direction. Thus cracking is in the +y, y, and 90
ply are
165 5.3 Synergistic damage mechanics (SDM)
(b)
(a)
(c)
Figure 5.15. Variation of average COD for [0
m
/y
n
/0
m/2
]
s
glass/epoxy laminates with
(a) axial stiffness ratio, r (for m = 1, n = 4); (b) number of cracked plies, n; and
(c) number of constraining plies, m. Reprinted, with kind permission, from Int J Solids
Struct, Vol. 45, C.V. Singh and R. Talreja, Analysis of multiple offaxis ply cracks in
composite laminates, pp. 457489, copyright Elsevier (2008).
166 Macrodamage mechanics
D
(3)
1
=
k
90
t
2
90
s
90
t
; D
(3)
2
= D
(3)
6
= 0: (5:82)
The integrity bases (Eq. (5.61)) have additional components for D
(3)
1
. The free
energy function thus gets the following terms added to Eq. (5.62)
rc a = 3 ( ) = a
/
1
e
2
1
D
(3)
1
a
/
2
e
2
2
D
(3)
1
a
/
3
e
2
6
D
(3)
1
a
/
4
e
1
e
2
D
(3)
1
; (5:83)
where a
/
i
; i = 1; 2; 3; 4 are additional material constants. Substituting Eq. (5.83)
into Eq. (5.48), we obtain
C
(3)
pq
=D
(3)
1
2a
/
1
a
/
4
0
2a
/
2
0
Symm 2a
/
3
_
_
_
_; (5:84)
where the contribution to the shear components is zero.
It is important to emphasize here that the relative location of different
damage modes in the whole laminate will cause different losses in stiffness due to
damage in the laminate. To illustrate this let us consider two specic examples of
laminates with damage modes consideredinthe present section, viz., +y, y, and 90
.
Figure 5.17(a)(b) shows representative unit cells for laminates with [0
m
/y
n
/90
r
]
s
(a)
(c) (d)
(b)
1.2
1
0.8
0.6
0.4
0.2
0 0.2 0.4 0.6
r =0.5
r =1
r =2
n=1
n=2
n=4
n=1
n=2
n=4
r =5
r =0.5
r =1
r =2
r =5
0.8
Crack density (1/mm)
Crack density (1/mm)
Crack density (1/mm)
1 1.2 1.4 1.6 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
0
1.2
1
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
1.2
1
0.8
0.6
0.4
0.2
0
Crack density (1/mm)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
1.2
1
0.8
0.6
0.4
0.2
0
E
x
n
xy
n
xy
n
xy
E
x
0
E
x
E
x
0
0
n
xy
0
Figure 5.16. Effect of change in axial stiffness ratio (r): (a), (b), and cracked ply thickness (n):
(c), (d), on stiffness degradation in [0
m
/70
n
/0
m/2
]
s
glass/epoxy laminates. Reprinted, with
kind permission, from Int J Solids Struct, Vol. 45, C.V. Singh and R. Talreja, Analysis of
multiple offaxis ply cracks in composite laminates, pp. 457489, copyright Elsevier (2008).
167 5.3 Synergistic damage mechanics (SDM)
and [0
m
/90
r
/y
n
/+y
n
]
s
congurations, respectively. The boundary conditions are
shown for the 3D FE computation of average COD values. The global laminate
(X, Y, Z) and local crack plane (x
1
, x
2
, x
3
) coordinate systems are also shown.
Considering rst the case of a [0
m
/y
n
/90
r
]
s
laminate, we note that y modes occur
twice in the whole laminate, above and belowthe midplane of the laminate, whereas
90
N
a=1
C
(a)
pq
is given by
C
pq
= 2 C
(1)
pq
y ( ) C
(2)
pq
y ( )
_ _
C
(3)
pq
90 ( ) : (5:85)
Collecting terms from Eq. (5.73) and Eq. (5.84), while assuming the a
i
s to be
independent of y, we get
C
pq
= 2D
y
2a
1
a
4
0
2a
2
0
Symm 2a
3
_
_
_
_
D
90
2a
/
1
a
/
4
0
2a
/
2
0
Symm 2a
/
3
_
_
_
_
; (5:86)
where, for the laminate conguration considered,
D
y
=
k
y
2nt
0
( )
2
s
y
n
t
; D
90
=
k
90
2rt
0
( )
2
s
90
t
; (5:87)
where t
0
denotes the thickness of a single ply. The special case when y = 90
, and
DC
pq
as given by Eq. (5.86) should be equal to that given by a single 90
mode with
crack size of (4n+2r) plies. Consider, for example, the DC
11
term. If we assume
that the normal crack spacing is the same in all cracked plies, then DC
11
for a
[0
m
/y
n
/90
r
]
s
laminate at y = 90
, [0
m
/y
n
/90
r
]
s
is equivalent to [0/90
2n+r
]
s
, we can consider their
stiffness changes to be the same. Thus, DC
11
can also be written, using Eq. (5.84), as
C
11
= C
(3)
11
= 2a
/
1
D
1
=
k
90
4n2r
4n 2r ( )t
0
[ [
2
s
90
t
:2a
/
1
; (5:89)
where the subsubscript 4n+2r in k
90
4n2r
represents the crack size for the 90
mode
in a [0/90
2n+r
]
s
laminate. Equating DC
11
from Eq. (5.88) and Eq. (5.89), we have
2n
2
k
y
[
y=90
a
1
90 ( ) r
2
k
90
a
/
1
= k
90
4n2r
2n r ( )t
0
[ [
2
a
/
1
; (5:90)
i.e.,
a
1
90 ( ) =
k
90
4n2r
2n r ( )
2
r
2
k
90
2n
2
k
y
[
y=90
a
/
1
: (5:91)
Generalizing, we can write the interrelationship between two sets of constants as
a
i
=
k
90
4n2r
2n r ( )
2
r
2
k
90
2n
2
k
y
[
y=90
a
/
i
i = 1; 2; 3; 4: (5:92)
Substituting Eq. (5.92) into Eq. (5.86), DC
pq
for a damaged [0
m
/y
n
/90
r
]
s
laminate
is given by
C
pq
=
D
2a
/
1
a
/
4
0
2a
/
2
0
Symm 2a
/
3
_
_
_
_; (5:93)
where
D =
4t
2
0
t
1
s
y
n
k
y
k
y
[
y=90
2n r ( )
2
k
90
4n2r
r
2
k
90
_ _
r
2
k
90
s
90
_ _
; (5:94)
where the constraint parameters are given by
k
y
=
u
y
_ _
y
2n
2nt
0
; k
90
4n2r
=
u
y
_ _
90
4n2r
4n 2r ( )t
0
; k
90
=
u
y
_ _
90
2r
2rt
0
: (5:95)
Consider now the case of the [0
m
/90
r
/y
n
]
s
laminate conguration. It must be
noted that, unlike [0
m
/y
n
/90
r
]
s
laminates, y damage modes in this case are
centrally placed, thereby the corresponding equivalent crack size is 4nt
0
(with
averaging over two +y and two y layers). On the other hand, the crack size for
the 90
damage mode is rt
0
. The derived stiffnessdamage relationships retain the
form of Eq. (5.93). However,
D in this case is given by
D =
2t
2
0
t
1
s
y
n
k
y
k
y
[
y=90
2 2n r ( )
2
k
90
4n2r
r
2
k
90
_ _
r
2
k
90
s
90
_ _
(5:96)
and the corresponding constraint parameters are now given by
169 5.3 Synergistic damage mechanics (SDM)
k
y
=
u
y
_ _
y
2n
4nt
0
; k
90
4n2r
=
u
y
_ _
90
4n2r
4n 2r ( )t
0
; k
90
=
u
y
_ _
90
r
rt
0
: (5:97)
The SDM procedure for this laminate conguration is quite similar to that
described earlier for the [0
m
/y
n
/0
m/2
]
s
layup in Figure 5.10. The damage constants
are evaluated from Eq. (5.59) after replacing kt
c
2
/st by
D by using the stiffness
degradation data for the reference laminate, which is chosen as [0/90
3
]
s
in this case.
Depending on the specic 90
, (b) y = 55
f s ( ) =
_
0
e
st
f t ( ) dt; (5:101)
and
~
C
ijkl
= sC
ijkl
: (5:102)
The constitutive equations (5.100) in the Laplace domain are similar to
the relationships of linear elasticity, except for the dependency on the trans
formation parameters. Additionally, the equilibrium equations and straindis
placement equations also retain their form in the Laplace domain. This
correspondence between linear elasticity and linear viscoelasticity is the so
called Correspondence Principle [39], and it applies to solving boundary
value problems as long as the boundary conditions do not change in time.
Thus, if ply cracks are present in a composite laminate, then the Correspond
ence Principle can still be applied by viewing the cracked laminate as a homo
geneous solid in which the internal boundaries (cracks) retain the same
boundary conditions in time. This implies that the cracks do not grow or heal
in time. We shall therefore address the linear viscoelastic response of laminates
172 Macrodamage mechanics
at a given (xed) damage state. Specically, we will express the relaxation
modulus tensor in Eq. (5.99) as a function of damage.
Following [38] a pseudo energy density function is dened in the Laplace
domain in terms of transformed strain and damage variables, such that
s
ij
=
@
W e
ij
;
D
(a)
ij
_ _
@e
ij
: (5:103)
It is noted that the above equation is valid for a xed ply crack density. The time
dependent deformation of the plies themselves will result in timevarying con
straint on the cracks, leading to the crack surface separation changing in time.
For illustration, let us consider a crossply laminate with transverse cracks,
Figure 5.5, with y = 90
D
11
=
k s ( )t
2
90
s
t
t
T
: (5:104)
where k s ( ) represents the constraint parameter in the Laplace domain, t
90
is
the thickness of the cracked layer, t
T
is the total laminate thickness, and s
t
is the
crack spacing. Using the same procedure as in Section 5.1.4, Kumar and Talreja
[38] derived the following transformed constitutive relationships (in the Voigt
notation) as
s
1
s
2
s
6
_
_
_
_
=
~
C
11
~
C
12
0
~
C
12
~
C
22
0
0 0
~
C
66
_
_
_
_
e
1
e
2
e
6
_
_
_
_
; (5:105)
where
~
C
pq
=
~
C
0
pq
~
C
(1)
pq
: (5:106)
Here
~
C
0
pq
is the transformed relaxation modulus of the crossply laminate without
cracks and
~
C
(1)
pq
, given by the following, is transverse cracking (one mode of
damage, a = 1)
~
C
(1)
pq
=
D
11
2g
11
g
12
0
g
12
2g
22
0
0 2g
66
_
_
_
_; (5:107)
where g
11
, g
12
, g
22
, and g
66
are material constants that appear as coefcient terms
in the polynomial expansion of the pseudo strain energy density function W. The
damage function D
11
is given by Eq. (5.104).
The timedependency of the crack opening displacement has been handled in
two ways. In Kumar and Talreja [38] the function W was assumed to depend on
173 5.4 Viscoelastic composites with ply cracking
the initial (t = 0) value of the damage tensor, thereby absorbing all time depend
ency in the presence of damage into the coefcient terms g
11
, etc. Later Varna et al.
[41] chose to retain the time dependency in the damage tensor and dealt with the
timevarying constraint to the COD explicitly.
Taking the inverse Laplace transform of Eq. (5.106), one obtains
C
pq
t ( ) = C
0
pq
t ( )
2r
n
t
90
t
T
k
ij
t ( ); (5:108)
where
r
n
=
t
90
2s
t
(5:109)
is the normalized crack density, and
k
11
t ( ) = 2L
1
kg
11
=s ( ); k
12
t ( ) = L
1
kg
12
=s ( );
k
22
t ( ) = 2L
1
kg
22
=s ( ); k
66
t ( ) = 2L
1
kg
66
=s ( ) ;
(5:110)
where L
1
(*) represents the inverse Laplace transform. Note that k = k
0
, its value
at time t = 0, if all time dependency is assumed to be in g
11
, etc. in the above
equations.
The functions k
11
, k
12
, and k
22
in Eq. (5.110) can be determined by procedures
similar to those described above for elastic constants. Here these three unknown
functions, which are decoupled from k
66
, can be evaluated from the timevarying
differences of the axial modulus E
x
(t) and Poissons ratio n
xy
(t) from their initial
values, and by assuming no change in the transverse modulus E
y
(t), as done in
Kumar and Talreja [38]. The timevariation of the material constants can be
determined from experimental data, if available, or by a micromechanics approxi
mation. The micromechanics can be analytical, if possible, or numerical, e.g., by a
nite element model.
In Kumar and Talreja [38] crossply laminates of given linear viscoelastic ply
properties were considered for validation of the CDM approach described above.
The functions k
11
(t), k
12
(t), and k
22
(t) were determined from the calculated visco
elastic response of a [0/90
2
]
s
laminate at a xed crack density of 0.4 cracks/mm.
These functions were then used to predict the time variations of the axial modulus
and Poissons ratio at other crack densities and for other crossply laminate
congurations of the same material. The predictions were compared with inde
pendently calculated time variations of the properties by a nite element model
and an analytical micromechanics model reported in Kumar and Talreja [40].
Predictions agreed well in all cases.
Varna et al. [41] demonstrated the use of SDM for linear viscoelastic response
predictions by explicitly treating the COD variations in time. The time dependency
of COD, i.e., the constraint parameter k and its counterpart in the Laplace
domain k, were calculated by a FE model. These were then inserted in
174 Macrodamage mechanics
Eq. (5.110) to determine k
11
(t), k
12
(t), and k
22
(t). Predictions thus made agreed
well with independently calculated viscoelastic response at different crack densities
and different crossply laminate congurations. A parametric study was also
performed in Varna et al. [41] to determine a master function for COD variation.
That function has the following form
k t ( ) = a b c
t
c
2t
s
1
_ _ _ _
E
2
t ( )
E
1
t ( )
_ _
d
; (5:111)
where a, b, c, and d are constants, E
1
and E
2
are the axial and transverse values,
respectively, of the Youngs modulus of the ply material, and t
s
here is the
thickness of the 0
N
a=1
C
(a)
ijkl
r
(a)
_ _
; (6:3)
where C
0
ijkl
is the stiffness tensor for undamaged laminate, and C
(a)
ijkl
represents
the stiffness changes brought about by damage mode a.
2. Describe the evolution of crack density as a function of applied loading:
r
(a)
= r
(a)
e
kl
( ): (6:4)
Combining the solution to the two subproblems above, we obtain
C
ijkl
= C
0
ijkl
N
a=1
C
DAM
ijkl
r
(a)
e
mn
( )
_ _
: (6:5)
The solution to the rst subproblem has been described in Chapters 4 and 5. This
chapter concerns the second subproblem.
To take advantage of directional properties of plies, composite laminates are
made from a mix of longitudinal, transverse, and angle plies. Since the unidirec
tional lamina has low strength in the transverse direction, it is prone to cracking
along bers. When the applied load is increased beyond the strain (or stress) at
which initiation of cracking occurs, new cracks form in the cracked ply in between
existing cracks. Initially these cracks are far apart and do not interact with each
other. However, quickly they form a roughly periodic array of parallel cracks.
Figure 6.1 shows the increase in density of transverse cracks (i.e., number of
cracks per unit length normal to the crack plane) for typical congurations of
crossply laminates when loaded under axial tension. Prediction of such curves has
been an extensive subject of study. The approaches used to model crack initiation
and progression (multiplication) can be divided into two categories: strength
based models and energybased models. As the names suggest, the rst category
of models involves use of strength (failure) criteria, while energy balance concepts
underlie the second category.
6.2 Experimental techniques
Before discussing experimental observations and measurements in Section 6.3
below, a brief description of each of the techniques is given next. This is to help
provide a background against which to interpret the observations. No attempt is
made here to go into any depth on these techniques, as that would distract from
the focus of this chapter. For further details references [24] are suggested.
Over the past forty years many nondestructive evaluation (NDE) techniques
have been developed to detect, monitor, and observe ply cracking damage in
40
30
20
10
0
n=4
[0/90
n
/0],T300/934
Cracks/In.
n=3
n=2
n=1
20 40 60 80 100 120 Ksi
Figure 6.1. Transverse crack density vs. applied axial stress in [0
m
/90
n
]
s
laminates [1].
Reprinted, with kind permission, from Composites Technology Review, copyright ASTM
International, 100 Barr Harbor Drive, West Conshohocken, PA 19428.
180 Damage progression
composite laminates. The quantities targeted for observations and measurements
concerning ply cracking include the following:
1. Crack initiation strains (or stresses).
2. Increase in number of cracks with applied loading or with number of cycles in
case of fatigue experiments.
3. Changes in stiffness properties with damage.
4. Crack opening displacements (COD) and crack proles.
5. Final failure strains and damage leading to the nal failure.
Brief overviews of the main techniques are provided here.
Edge replication
Although direct optical microscopy can be used [5] for surface observation to evaluate
damage, it requires in most cases specimens to be removed at regular intervals during
loading. If feasible, anoptical microscope canbe mountedonthe testing machine itself,
but it often requires the microscope to be brought close to the specimen surface for
enough resolution. This limitation can be overcome by using edge replication instead.
This is a simple and easy technique if the purpose is to monitor damage (count cracks)
in composites, as was shown early by Stalnaker, Stinchcomb, and Masters [6, 7]. The
approach involves the microscopic examination of surface replicas, which are prepared
by pressing softened rubber tape (or tape with an adhesive, e.g., acetate) against the
specimen edge. With this procedure it is possible to quickly obtain permanent records
of the specimen edge while the specimen itself is held in loading. The popularity of the
technique is evidenced by its utilization by many researchers in the eld [2, 814]. Two
examples of photomicrographs from edge replicas are shown in Figure 6.2.
Acoustic emission
In this technique, a sensor is used to monitor acoustic emission (AE) signals from
stress waves generated due to some local failure in the material, such as formation of
a ply crack. A major limitation of the approach is its inability to distinguish between
damage types and to provide information concerning crack location and orientation
[14]. However, if the damage mechanismis known based on prior experience with the
material system, this technique can indicate the applied load at which the damage
initiated. Using this to monitor damage and to take edge replicas accordingly has
been found useful. A combination of ultrasonic polar backscattering scans and AE,
which yields accurate crack location, has also been suggested [15, 16]. Figure 6.3
shows the variation of cumulative count of acoustic events and the corresponding
stressstrain response of a ceramic composite subjected to tensile loading.
Xray radiography
Xray radiography is a useful technique in nding internal cracks that are not visible
by optical microscopy. For clarity of observationa penetrant uidthat absorbs Xrays
is often necessary to provide sufcient contrast between exposed and unexposed
regions. Pictures of the developed Xray lms can then be enlarged for further clarity.
181 6.2 Experimental techniques
Transverse Cracking
Transverse Cracking
Longitudinal Cracking
(a) (b)
Figure 6.2. Photomicrographs of edge replicas showing details of damage development in
fatigue loading of quasiisotropic laminates: (a) [0/45/90]
s
and (b) [0/90/45]
s
laminate
[2]. Reprinted, with kind permission, from Damage in Composite Materials, copyright
ASTM International, 100 Barr Harbor Drive, West Conshohocken, PA 19428.
1200
(a)
900
600
300
0
0 20 40
Stress (MPa)
C
u
m
u
l
a
t
i
v
e
A
E
c
o
u
n
t
s
60 80 100
100
80
60
40
20
0
(b)
0
Strain (%)
S
t
r
e
s
s
(
M
P
a
)
0.05 0.1 0.15 0.2
Figure 6.3. Variation of cumulative acoustic event (AE) counts (a), and stressstrain curve
(b) for a ceramic composite loaded in tension [14]. Reprinted, with kind permission, from
Damage Detection in Composite Materials, copyright ASTM International, 100 Barr Harbor
Drive, West Conshohocken, PA 19428.
182 Damage progression
The pictures, however, only provide 2D images of cracks and therefore cannot be
used to nd the depth information needed to separate cracks of one layer from
another. For that purpose penetrant enhanced Xray stereo radiography can be
used. The standard procedure is to make two Xray radiographs of the object at
slightly different orientations, which is usually done by rotating the Xray source
relative to the object. The spatial locations of cracks can be observed with a
microscope but cannot be recorded. The approach has been successfully used to
detect ber fractures, delaminations and matrix cracks [1719]. To enable imaging
and ner precision detection of internal defects, the technique employed is Xray
tomography, which uses a medical scanner and yields a threedimensional image of
an object [20]. An Xray radiograph of damage in a quasiisotropic laminate is
shown in Figure 6.4.
Ultrasonic Cscan
Ultrasonic Cscan is a nondestructive inspection technique in which a short pulse
of ultrasonic energy is made incident on a sample. The attenuation of the pulse is
inuenced by voids, delaminations, state of resin cure, the ber volume fraction,
Figure 6.4. An Xray radiograph of damage in a quasiisotropic laminate [19]. Reprinted,
with permission, from Effects of Defects in Composite Materials, copyright ASTM
International, 100 Barr Harbor Drive, West Conshohocken, PA 19428.
183 6.2 Experimental techniques
the condition of the ber/matrix interface, and any foreign inclusions present. The
technique is, however, limited by its inability to detect very small defects or cracks
with their planes parallel to the wave direction [14].
In some cases a damaged laminate material may contain a large number of
small and partially grown internal cracks. On increase in loading, these cracks can
grow further and coalesce to form larger cracks. The common NDE techniques
described above cannot provide accurate information on the formation, number,
size, and progression of these internal cracks. For such cases a throughtransmis
sion ultrasonic Cscan imaging with inclined focusing transducers in confocal
conguration has been recently suggested [21, 22]. Other methods based on
vibrations and lamb waves have also been suggested [23, 24].
Technique for COD measurement
In addition to the observations of damage certain measurements also need to be
made as dictated by the models. For instance, the COD of ply cracks is a quantity
that enters the SDM approach described in Chapter 5. Since there are no standard
methods for measuring this quantity, Varna et al. [2627] developed a setup for
this purpose. To observe an individual ply crack, the setup uses a miniature
materials tester (MINIMAT) for loading a thin strip cut out from the cracked
laminate. The open crack is observed under an optical microscope equipped with a
video camera. The video signal transmitted to a TV monitor displays the crack
prole at sufcient magnication (~ 210
3
) to measure the COD. Figure 6.5
presents COD as a function of position along the cracked 90
ply thickness
(zdirection) and compares it with the theoretical shape prediction obtained using
Linear Elastic Fracture Mechanics (LEFM).
Raman spectroscopy
Recently Katerelos and coworkers [28] have developed an experimental technique
based on Raman spectroscopy, which uses the property that the Raman
vibrational wave numbers (frequencies) of certain chemical groups of commercial
reinforcing bers, such as aramid or carbon, are stress and strain dependent [16].
Thus, the wave number shift along an embedded ber can be utilized to determine
stress or strain. The technique can provide a high spatial resolution of ~ 1mm.
However, for the approach to work the matrix needs to be translucent. Also, the
data acquisition can take longer than milliseconds for a single measurement.
Moreover, certain amorphous bers such as glass have a weak Raman response.
Aramid bers, on the other hand, scatter the waves well. Therefore, a small
amount of aramid ber placed in strategic positions within a glassber laminate
can act as Raman sensors of stress and strain. The micromechanical strain
mapping results are then used to derive the properties, i.e., the longitudinal
modulus of elasticity and the magnitude of the residual strains caused by cracking
[29]. The approach has been successfully applied to evaluate ply cracking damage
evolution and resulting stiffness changes [3032].
184 Damage progression
6.3 Experimental observations
Experimental studies on initiation and growth of intralaminar cracking in composite
laminates have been extensive. Most of the work has focused on 90
ply cracking in
crossply and quasiisotropic laminates. Chapter 3 reviewed some of those observa
tions to illustrate the nature of composite damage. A book chapter by Nairn [33] is
also recommended for a good overview of the topic. In the following we review
quantitative data related to the initiation and progression of ply cracking in laminates.
6.3.1 Initiation of ply cracking
The applied loading (stress or strain) at which ply cracking in laminates rst
occurs is of interest from materials selection as well as design points of view.
A good laminate conguration (orientation, thickness, and sequence of plies) will
delay initiation of ply cracking to as high a load as possible. Experimental
observations have indicated that all laminate conguration parameters inuence
ply crack initiation. Most early studies examined ply cracking in crossply lamin
ates of glasspolyester or glass/epoxy under axial tension. The ply cracking in the
90
plies could in most cases be observed by looking at the specimen surface under
an optical microscope at low magnication. The neartransparency of composites
25
20
15
[O2/908]s
Hybrid CF/GF
(Different sections)
LEFM theory
10
5
0
0.0 0.2 0.4 0.6 0.8 1.0
Relative position in 90layer
C
r
a
c
k
o
p
e
n
i
n
g
d
i
s
p
l
a
c
e
m
e
n
t
u
(
m
m
)
zt
90
2t
90
Figure 6.5. Crack opening displacement measured using MINIMAT along the 90
layer
thickness of a hybrid CF/GF crossply laminate at various sections through the specimen
width. The solid line depicts predictions from LEFM. Source: J. Varna, L. Berglund,
R. Talreja, and A. Jakovics, A Study of crack opening displacement of transverse cracks in
crossply laminates, Int J Damage Mech, Vol. 2, pp. 27289 (1993).
185 6.3 Experimental observations
of glass bers made this possible. In carbon ber composites, however, one had to
rely on edge observations until Xray radiography allowed imaging of the interior
cracks. The axial strain values at initiation of transverse cracking in crossply
laminates of glass/epoxy and carbon/epoxy fall typically in the range 0.41.0%.
As the offaxis ply angles decrease from 90
plies cracking barely initiates at 1.0% axial strain [34]. Figure 6.6
illustrates the effect of offaxis angle on crack initiation strain for (a) GFRP [35],
and (b) CFRP [36]. Data for notched samples are also indicated to illustrate the
effect of local stress enhancement. In Figure 6.6 the data for unnotched samples
with and without polished edges illustrate the effect of machininginduced aws in
initiating cracking. Also of note is the range of strain and the lowest strain for
GFRP versus CFRP. In the latter case, the lowest strain to crack initiation is higher
due to the higher elastic modulus of carbon bers. In both cases the initiation strain
at low offaxis angles is limited by the ber failure strain.
The thickness of the cracking ply also has a large effect on e
init
. Garret and
others [3744] performeda systematic study of thickness effect on cracking in [0
m
/90
n
]
s
laminates. They used glassreinforced polyester [37, 38] and glassreinforced epoxy
[3941] as the laminate materials andvariedthe thickness of the 90
plies
increases, e
init
decreases. When the thickness of the 90
plies [47].
The actual ply layup in the laminate may also play a role in determining e
init
.
For instance, [90
2
/0]
s
laminates develop cracks sooner than [0/90
2
]
s
laminates [48].
It is because the 90
plies in [90/0]
s
laminates are on the outer surface and may not
experience much support for crack suppression from the inner 0
plies. For
multidirectional laminates, the situation is even more complex and e
init
will depend
on the thickness ratio and the stiffness properties of the cracking and supporting
plies. The laminate preparation and processing method also may affect crack
initiation; lamentwound laminates may be more prone to cracking than those
made from prepregs using an autoclave.
6.3.2 Crack growth and multiplication
The growth of ply cracks is usually unstable in composites such as GFRP and
CFRP. The initial transverse cracks are found to grow through the lamina
thickness quickly but are usually arrested at interfaces with adjacent plies of
different orientation, e.g., at the 90/0 interface for crossply laminates [38, 40]
and at the 90/45 interface in [0/90/45/+45]
s
laminates [34]. Continued loading
usually leads to formation of progressively more cracks between the already
formed cracks. Most experimental studies point out that once the ply cracks have
grown through the lamina thickness, they often grow unstably along the ber
direction through the laminate width, and are thus described as tunneling
cracks. In some cases, however, cracks in plies other than 90
plies can accumulate numerous cracks. The evolution of average crack density
with respect to applied load for the same specimens is shown in Figure 6.9. It can be
observed here that the saturation crack density is roughly proportional to 1/t
90
. The
damage evolution curves in Figure 6.10 also show that thicker crossply laminates
typically show lower crack density at saturation. The damage evolution in outer
and inner 90
plies are
Figure 6.8. Transverse cracking in GFRP specimens with transverseply thickness of (a) 0.75
mm, (b) 1.5 mm, and (c) 2.6 mm, strained to 1.6%. Reprinted, with kind permission,
from Springer Science+Business Media: J Materials Science, Multiple transverse fracture in
90 degrees crossply laminates of a glass berreinforced polyester, Vol. 12, pp. 15768,
K. W. Garrett and J. E. Bailey.
188 Damage progression
observed to have lower saturation crack densities. A typical damage evolution
curve for ply cracking based on these studies is shown in Figure 6.11.
6.3.3 Crack shapes
When cracks are widely spaced, the maximum principal stress occurs on the plane
midway between existing cracks. Thus, at low crack density there is a tendency for
new cracks to form midway and develop into a periodic array. However, at large
crack density cracks interact causing the maximum principal stress to shift towards
the 0/90 interface close to an existing crack. This may result in curved or oblique
microcracks forming near the 0/90 interface [54, 56, 57]. These cracks make an angle
of 4060
with existing straight cracks. Lundmark and Varna [58] have recently
reported that curved microcracks form more readily in the lowtemperature regime
than at room temperature. In fact, at low temperatures complex crack trajectories
are observed to form (see Figure 6.12). This may result in a highly damaged region
in the laminate encompassing multiple crack types (Figure 6.13).
6.3.4 Effect of cracking
The most direct effect of ply cracking is the reduction of the thermomechanical
properties of the laminate, including changes in the effective values of Youngs
moduli, Poissons ratios, and thermal expansion coefcients. The changes in
stiffness properties can in turn lead to change in the behavior of the whole
structure, e.g., the magnitude of its deection and vibrational frequencies, some
times making the structure unable to carry out its intended design function. Even
if it does not cause the structure to fail, substantial ply cracking may give rise to
1.2
3.2mm
2t
90
2mm
1.5mm
0.75mm
1
0.8
0.6
0.4
0.2
0
0 50 100 150 200 250
Applied stress (MPa)
A
v
e
r
a
g
e
c
r
a
c
k
d
e
n
s
i
t
y
(
1
/
m
m
)
Figure 6.9. Average crack density as a function of applied stress for GFRP crossply laminates
with different transverse ply thicknesses (2t
90
). The experimental data are from [54].
189 6.3 Experimental observations
Stage I
Applied load (strain or stress)
C
r
a
c
k
d
e
n
s
i
t
y
(
n
o
.
o
f
c
r
a
c
k
s
/
m
m
)
Stage II Stage III
Crack initiation &
propagation
through
laminate width
Multiple crack
formation
Saturation of
progressive cracking
Figure 6.11. A typical damage evolution curve for ply cracking in laminates.
1.4
(a)
1.2
1.0
0.8
0.6
0.4
0.2
0.0
0 200 400
[0/90
4
]
s
[0/90
2
]
s
[0/90]
s
600 800
Stress (MPa)
C
r
a
c
k
d
e
n
s
i
t
y
(
1
/
m
m
)
1.0
(b)
0.8
0.6
0.4
0.2
0.0
0 100 200 300
[90/0/90]
T
[90/0]
s
[90/0
2
]
s
[90/0
4
]
s
400 500
Stress (MPa)
C
r
a
c
k
d
e
n
s
i
t
y
(
1
/
m
m
)
600 700 800 900 1000
Figure 6.10. Damage evolution curves for (a) [0/90
m
]
s
and (b) [90/0
m
]
s
laminates. Reprinted,
with kind permission, from Polymer Matrix Composites, J. A. Nairn, Matrix microcracking
in composites, pp. 40332, copyright Elsevier (2000).
190 Damage progression
more deleterious forms of damage such as delamination and longitudinal splits, or
provide pathways for the entry of moisture and corrosive liquids.
6.3.5 Loading and environmental effects
Most experiments are performed using uniaxial tension, but ply cracks will also
form under other loading conditions, such as fatigue, biaxial, or shear loading.
Biaxial loading of [0
m
/90
n
]
s
laminates may show cracks in both 0
 and 90
plies. If
the material response remains linearly elastic after cracking, then neglecting crack
interaction effects between plies the effect of biaxial loading of [0
m
/90
n
]
s
laminates
may be seen as equivalent to two uniaxial loading cases, one on [0
m
/90
n
]
s
and the
other on [90
m
/0
n
]
s
laminates. On thermal loading, the differential shrinkage
between the 0
 and 90
C). Reprinted, with kind permission, fromEng Fract Mech, Vol. 75, P. Lundmark and
J.Varna, Damage evolution and characterization of crack types in CF/EP laminates loaded
at low tempratures, pp. 263141, copyright (2008), with permission from Elsevier.
191 6.3 Experimental observations
thermal effects can be accounted for in the analysis and predictions of classical
laminate theory. Bailey et al. [41] studied the effect of thermal stresses and Poissons
contraction on ply cracking in CFRP and GPRP crossply laminates. Thermal
residual stresses typically lower crack initiation strains. Thermal effects are larger
in CFRPs than in GFRPs due to larger differences in both thermal expansion
coefcients and Youngs moduli in directions parallel and perpendicular to bers
[41]. For instance, a [0/90]
s
laminate with a ply thickness of 0.5mm shows 0.322%
thermal strains for CFRPs and 0.094% for GFRPs. The Poissons effect was found
to be greater in GFRPs because of their larger failure strains. Poissons strains can
sometimes be so high that they can induce transverse cracking of the 0
plies.
The mismatch between the thermal expansions of different plies can also result
in a highdensity form of microcracking known as stitch cracking. Lavoie and
Adolfsson [66] studied this type of cracking in [+y
n
/y
n
/90
2n
]
s
laminates (see
Figure 6.14). Stitch cracks appear to form instead of interply delamination at
the tip of a 90
ply matrix cracks run from left to right, and trigger the formation of
many short, stitchlike 45
, 45
, and 90
orientations, exemplied by
the [0/90/45]
s
conguration. In the offaxis plies of such laminates, cracks
usually initiate at much higher applied axial strains than in 90
the dominant
damage was ply cracking [31, 35], while in [90/30/30]
s
a combination of edge
delamination and ber fracture led to nal failure. Crocker et al. [35] also found
that in [0/45/0] laminates, delamination ensued as soon as ply cracks initiated in
45
plies.
In case of multidirectional laminates containing 90
ply. Experiments by
Yokozeki and coworkers [50, 51] on [0/y
2
/90]
s
laminates also point out that
the angle of intersection between the 90
cracks pro
mote enhanced cracking in the 90
1
lE
0
x0
_ _
; (6:7)
where G
90
xz0
is the initial inplane shear modulus of the 90
ply, l = t
0
/ t
90
is the
ply thickness ratio, and E
0
x0
and E
90
x0
are the longitudinal moduli for the 0
 and
90
bl
N
2e
bl
2N
: (6:11)
The evolution of crack density thus can be generated using the above iterative
scheme. Figure 6.16 shows the model prediction of crack spacing variation with
increasing loading for a glasspolyester specimen of length 130 mm and transverse
ply thickness of 3.2 mm. Although this model represents the general trends of the
experimental results it generally underestimates the average crack spacing. This is
supposedly due to the constant strength of the 90
ply strength.
The corresponding strains can be calculated from the stressstrain relations for the
cracked laminate derived in [83] as
0
C
r
a
c
k
s
p
a
c
i
n
g
(
m
m
)
30
0.2 0.4 0.6 0.8 1.0
Strain (%)
1.2 1.4 1.6 1.8 2.0 2.2
20
10
0
Figure 6.17. Variation of average crack spacing for a [0/90/0] glass/epoxy composite laminate
as predicted using the probabilistic shear lag model by Manders et al. [92] for. Solid and
open circles represent experimental data from two specimens, while the solid line represents
the model prediction.
197 6.4 Modeling approaches
e
c
=
s
c
E
c
1
t
90
Q
22
t
0
Q
11
1
Q
12
A
12
Q
22
A
22
_ _
tanh al
al
_ _
t
90
s
90
xx R
t
0
Q
11
tanh al
al
: (6:16)
The crack initiation strain can be obtained by setting tan al / al = 0 (as l ) in
Eq. (6.16), and 2l = L
0
, d
1
= 1 in Eq. (6.15). Therefore, the applied strain at rst
cracking in the 90
ply is given by
e
0
=
1
Q
22
1
Q
12
A
12
Q
22
A
22
_ _ s +
t
1
t
90
_ _
1=v
s
90
xx R
_ _
: (6:17)
These statistical descriptions of ply strength yielded good results for crossply
laminates. But these models cannot account properly for effects of changes in ply
thickness. Also, as mentioned before, shear lag analysis is a onedimensional stress
analysis and therefore cannot be accurate. More recently, 2D shear lag analysis
based on Steifs parabolic displacement variation in 90
plies to
transverse cracking in [0/90]
s
laminates and took account of it in the energy
balance during cracking. Recalling from ACK theory, Section 4.2, a crack does
not form in a specimen loaded in constant tension until
W _ U
S
U
D
2g
m
V
m
; (6:19)
where V
m
is the matrix volume fraction and, dened per unit crosssectional area
of the composite, DW is the work done by applied stress, DU
S
is the increase in
energy stored in the composite volume, U
D
is the energy lost by some dissipative
processes during cracking (sliding friction between debonded bers and matrix),
and g
m
is the matrix surface energy per unit surface area. For the present case of
cracking in a transverse ply, the above inequality becomes
W _ U
S
2g
t
t
90
h
; (6:20)
where g
t
is the surface energy of the transverse ply for cracking in a direction
parallel to the bers, and h = t
0
+ t
90
. For a linear elastic body the work of
external forces equals half the stored strain energy, i.e.,
U
S
=
1
2
W: (6:21)
Substituting Eq. (6.21) into Eq. (6.20), the cracking will occur when
W _ 4g
t
t
90
h
: (6:22)
Oncracking anadditional stress is thrownontoouter uncrackedplies, andthe laminate
increases in length. The work done during this process can be derived as [81]
W =
2E
90
x0
E
c
e
2
tu
lE
0
x0
b
; (6:23)
where b is as dened earlier in Eq. (6.7). Combining Eqs. (6.22) and (6.23), the
strain to initiate cracking in transverse ply is given by
e
0
= e
min
tu
=
2t
0
E
0
x0
g
t
b
hE
90
x0
E
c
: (6:24)
The model predictions for various transverse ply thicknesses are shown in Figure
6.18 against experimental data for glass/epoxy crossply laminates.
199 6.4 Modeling approaches
A more sophisticated energybased analysis of progressive ply cracking using
onedimensional stress analysis was performed by Laws and Dvorak [100]. (Refer
to Section 4.4.1 for the shear lag analysis.) Consider a cracked crossply laminate
in state 1 where cracks are a distance 2l apart, and the ligament AB between
the cracks is as yet uncracked (Figure 6.19(a)). When the applied load reaches a
critical value s
c
, a new crack appears in this segment at some location C (Figure
6.19(b)). Assuming that the load is kept xed during formation of this additional
crack, Laws and Dvorak calculated the energy released during cracking of lamin
ate of width w as
=
2t
2
90
whE
c
bt
0
E
0
x0
E
90
x0
s
90
xxR
E
90
x0
E
0
x0
s
c
_ _
2
tanh
bl
1
2t
90
tanh
bl
2
2t
90
tanh
bl
t
90
_ _
; (6:25)
0
2.5
1 2
Transverse ply thickness 2d (mm)
T
r
a
n
s
v
e
r
s
e
c
r
a
c
k
i
n
g
s
t
r
a
i
n
e
t
u
,
e
t m
u
i
n
(
%
)
3 4
2.0
1.5
1.0
0.5
0.0
Experimental e
tu
Theoretical e
tu
min
Figure 6.18. Values of crack initiation strain e
min
tu
as a function of ply thickness 2t
90
= 2d
(from Eq. (6.24)) and experimental data for various ply thicknesses in glass ber/epoxy
crossply laminates. The horizontal line depicts the limiting values of e
min
tu
for large innerply
thicknesses [81]. Reprinted, with kind permission, from Springer Science+Business
Media: J Mater Sci, Constrained cracking in glass berreinforced epoxy crossply laminates,
Vol. 13, 1978, pp. 195201, A. Parvizi, K. W. Garrett and J. E. Bailey.
A B
z
x
State 1
(a) (b)
State 2
A
s=2l l
1
l
2
B C
Figure 6.19. Progressive multiplication of ply cracks in transverse layer of crossply laminate:
(a) state 1 with crack spacing, s = 2l; (b) state 2 with an additional crack in the ligament
AB at location C.
200 Damage progression
where b
2
= Kt
90
1=E
90
x0
_ _
1=lE
0
x0
_ _ _
is the shear lag parameter (see Eq. (4.67)),
and s
90
xxR
is the thermal residual stress in the transverse ply. Substituting Eq. (6.25)
in Eq. (6.18), and using DA = 2wt
90
, a new crack will form if
t
90
hE
c
bt
0
E
0
x0
E
90
x0
s
90
xxR
E
90
x 0
E
0
x0
s
c
_ _
2
tanh
bl
1
2t
90
tanh
bl
2
2t
90
tanh
bl
t
90
_ _
_ g: (6:26)
The rst ply failure stress (the applied stress at crack initiation) will be given by the
limit l , i.e.,
s
fpf
c
=
bt
0
E
0
x0
E
c
g
t
90
hE
90
x0
_ _
1=2
E
c
E
90
x0
s
90
xxR
: (6:27)
The thermal residual stress, s
90
xxR
, and other parameters in the expression above
are known from property data for an undamaged laminate. Therefore,
Eq. (6.27) provides a relation among s
fpf
c
, b, and g. Now s
fpf
c
and g can
be determined from experimental data. Therefore, Laws and Dvorak regard
Eq. (6.27) as the relation that determines the shear lag parameter b. Once
b is known, the applied stress needed to cause cracking at location C is given
by
s
c
(l
1
) = s
fpf
c
E
c
E
90
x0
s
90
xxR
_ _
tanh
bl
1
2t
90
tanh
bl
2
2t
90
tanh
bl
t
90
_ _
1=2
E
c
E
90
x0
s
90
xxR
:
(6:28)
In a practical scenario, the location C is random due to spatial variation of the
resistance to crack formation. Let p be the probability density function for the
next crack to occur at a given location. In a laminate which already contains
cracks with normalized crack density, r
c
= t
90
/ l, the expected value of the applied
stress to cause additional cracking is then
E s
c
r
c
( ) [ [ =
_
2l
0
p x ( )s
c
x ( ) dx: (6:29)
Three possible choices for p(x) are:
Case 1: The next crack occurs midway, so
p x ( ) = d(x l); (6:30)
where d(x) is the Dirac delta function.
Case 2: All locations are equally likely. Therefore,
p x ( ) =
1
2l
: (6:31)
Case 3: p(x) is proportional to the stress at the location. For this case, p(x) is
given by [100] as
201 6.4 Modeling approaches
p x ( ) = s
90
xxR
E
90
x0
E
0
x0
s
c
_ _
1
cosh
bx
t
90
cosh
bl
t
90
_
_
_
_
: (6:32)
For case 1, the solution is explicitly given by
E s
c
r
c
( ) [ [ = s
fpf
c
E
c
E
90
x0
s
90
xxR
_ _
2 tanh
b
2r
c
tanh
b
r
c
_ _
1=2
E
c
E
90
x0
s
90
xxR
: (6:33)
For cases 2 and 3, the integral in Eq. (6.29) must be evaluated numerically.
The model predictions for these cases of p(x) are shown in Figure 6.20 (with
g = 193 J/m
2
and b = 0.9). Based on comparison with the experimental data,
Laws and Dvorak argue that fracture mechanicsbased p(x) (case 3) is the most
promising choice. With this choice, model predictions also compare well with
another set of experimental data from [1] for graphite/epoxy laminates.
Variational analysis
Nairn [99] used the variational approach [101] for cracked crossply laminates,
including thermal residual stresses, in conjunction with the energy release rate
criterion to predict crack densities in cracked crossply laminates. His predictions
showed good agreement with experiments when the critical energy release rate for
matrix cracking, dened as such, was deduced from test data rather than evaluat
ing it independently.
Another damage evolution model for crossply laminates is by Vinogradov and
Hashin [97, 98, 102]. It uses variational analysis [101] for stress computation and
nite fracture mechanics [97] for cracking criterion. Recalling from Section 4.6,
Figure 6.20. Crack density evolution in a glass/epoxy [0/90]
s
laminate for three choices of
probability distribution function p(x) [100]. The experimental data are from [8]. Note: [8]
has additionally three data points at the high load end that fall away from model
predictions. Reprinted, with kind permission, from N. Laws and G. J. Dvorak, J Compos
Mater, Vol. 23, pp. 90016, copyright #1988 by Sage Publications.
202 Damage progression
where the stress calculations are described in detail, complementary energy change
due to presence of N transverse cracks can be derived as [98]
+
=
N
n=1
+
n
= s
90
xx0
_ _
2
t
2
90
C
22
N
n=1
w r
n
( ); (6:34)
where C
22
= l 1 ( ) 3l
2
12l 8
_ _
1
60E
2
(see Eq. (4.151)), and
w(r
n
) =
d
3
f
n
dx
3
r
n
= 2a
1
a
2
a
1
2
a
2
2
_ _
cosh (2a
1
r
n
) cos(2a
2
r
n
)
a
1
sin (2a
2
r
n
) a
2
sinh (2a
1
r
n
)
; (6:35)
and the summation in Eq. (6.34) is over all blocks bounded by adjacent cracks. It
is noted that r
n
= l
n
/t
90
is the normalized crack spacing and should not be
confused with the crack density. For clarity of notations, the reader is referred
to Section 4.6. Initially when the cracks are far apart (r
n
), the function
achieves its asymptotic value given by
w() = 2a
1
a
1
2
a
2
2
_ _
: (6:36)
Now consider the state of a laminate with N cracks. When a new crack appears,
the energy release can be expressed as [97]
n
=
+
s
N1
_ _
+
s
N
_ _
; (6:37)
where s
N
and s
N+1
are the stress elds before and after formation of a new (N+1)th
crack. The stresses include both mechanical and thermal effects (if present).
Assuming that the new crack appears instantly, both stress elds are evaluated
at the same external load. Equation (6.37) can be rewritten as
n
=
+
s
N1
_ _
+
s
0
_ _ _
+
s
N
_ _
+
s
0
_ _ _
; (6:38)
where s
0
is the stress eld in the undamaged material at the same external load.
Assuming Eq. (6.34) is a good approximation of the energy, the energy release
due to the (N+1)th crack is given by
n
= s
90
xx0
_ _
2
t
2
90
C
22
N1
i=1
w r
N1
i
_ _
s
90
xx0
_ _
2
t
2
90
C
22
N
i=1
w r
N
i
_ _
; (6:39)
where s
90
xx0
is the stress in the undamaged 90
ply, r
N
i
is the nondimensional crack
spacing for the ith block for the Nth cracking step (going from N to N+1 cracks).
Using the energy release rate criterion, Eq. (6.18), taking averages and express
ing variables as continuous variables, the energy released during cracking can be
expressed as (see [98] for full derivation)
g = s
90
xx0
_ _
2
t
90
C
22
d
d r
w
r
_ _
r
2
: (6:40)
203 6.4 Modeling approaches
Using the lower bound of the laminate longitudinal modulus (see Section 4.6) by
Hashin [101]
1
E
+
x
=
1
E
0
k
2
1
t
90
h
C
22
w
r
; (6:41)
where k
1
= s
90
xx 0
=s
0
xx 0
if the temperature change is absent. Using A = L=2 r, we
nally obtain the cracking criterion as
g =
1
2
s
90
xx0
_ _
2 d
dA
1
E
+
x
_ _
V; (6:42)
where V is the laminate volume. Equation (6.42) represents a particular homo
thermal case of the general fracture criterion derived by Hashin [103] given by
g =
1
2
s
@S
+
@A
s
@a
+
@A
sT
1
2
@c
+
p
@A
T
2
T
r
_ _
V; (6:43)
where S* is the effective elastic compliance tensor, a* is the effective thermal
expansion tensor, c
+
p
is the effective specic heat of a composite, and T
r
is the
reference temperature.
The damage evolution predictions using this approach are very accurate when
the probabilistic distribution of the energy release rate is utilized. This will be
discussed in the next section.
Nairns original result [48, 99] was quite similar to the fracture criterion
stated above. However, in place of the usual energy release rate (2g) he used
the matrix fracture toughness (G
m
) and suggested that it could be obtained
through tting experimental data for ply cracking. If we assume that the
new crack forms midway between the existing cracks, Nairns fracture criter
ion is
G
m
= s
2
c
E
2
2
E
2
c
aT
2
C
2
00
_ _
t
90
C
22
2w r=2 ( ) w r ( ) [ [; (6:44)
where C
00
= (1/E
2
) + (1/lE
1
) (see Eq. (4.151)). An alternative formulation is to
allow the new crack to form anywhere between two existing cracks. If the prob
ability of crack formation at any position is proportional to the tensile stress, the
energy release rate is given by
G
m
= s
2
c
E
2
2
E
2
c
aT
2
C
2
00
_ _
t
90
C
22
w d ( ) w r d ( ) w r ( ) [ [; (6:45)
where
w d ( ) w(r d) w(r) [ [ =
_
r=2
0
w(d) w(r d) [ [ 1 f(r 2d) [ [dd
_
r=2
0
1 f(r 2d) [ [dd
w(d): (6:46)
204 Damage progression
For cracking in [90
m
/0
n
]
s
laminates, the expressions for energy release rate remain
the same except that the constant C
22
is now given as C
22
= (l+1)(3+12l+8l
2
)
(1/60E
2
) (see Eq. (4.164)). It is noted that Nairns analysis included residual thermal
stresses, which mainly change the crack initiation strain (discussed later in detail).
The parameter G
m
is evaluated by tting the model to the experimental data. The
model predictions for a [0/90
3
]
s
glass/epoxy laminate with G
m
= 330 J/m
2
and a
thermal residual stress of 13.6 MPa are shown in Figure 6.21.
Plainstrain formulation
McCartney [104109] developed a model based on the Gibbs free energy, instead
of complementary strain energy as described above. He used a plane strain
formulation for the estimation of elastic moduli of the damaged laminate, which
has been covered in Section 4.7. Consider the damage progression from a state of
m cracks to n cracks in the 90
1
~
E o
m
( )
m
A
o
n
( ) g o
n
( ) ( )
2
m
A
o
m
( ) g o
m
( ) ( )
2
2 o
n
( ) o
m
( ) [ [ > 0;
(6:50)
where ^e o ( );
~
E o ( ); m
A
o ( ); g o ( ), and G(o) denote axial strain, axial Youngs modu
lus, inplane axial shear modulus, applied inplane shear strain, and the energy
absorption per unit volume for length 2L of laminate, respectively, for given
damage state o. G(o) is here given by
o ( )
h
(90)
hL
M
j=1
d
(90)
j
g
(90)
j
; (6:51)
where 2h
(90)
and 2h represent the total thickness of 90
layer) a new
crack develops midway between two existing cracks and the total number of
cracks becomes 2N with spacing l (Figure 6.19(b)). According to the crack closure
concept the released energy due to these N new cracks is equal to the work needed
206 Damage progression
to close them. If we denote this work by W
2NN
and the work to close all cracks
simultaneously by W
2N0
, energy balance requires
W
2N0
= W
2NN
W
N0
; (6:52)
where the work to close N cracks with spacing s is
W
N0
= N 2
1
2
_
t
90
t
90
s
90
xx0
u(z) dz = 2Ns
90
xx 0
t
90
u
n
(l) = N
s
90
xx0
_ _
2
E
2
t
2
90
~ u
n
(l); (6:53)
where unit width is assumed and u(z), u
n
; and u
n
represent the variation of the
normal crack opening displacement (COD) along the thickness direction, its
average value, and its average value normalized with respect to the remote
stress and transverse modulus for the ply, respectively. u
n
; and u
n
are thus
dened as
u
n
=
1
t
90
_
t
90
0
u(z) dz; ~ u
n
=
u
n
s
90
xx0
=E
2
_ _
t
90
: (6:54)
Similarly, the work to close 2N cracks with spacing l is given by
W
2N0
= 4N
s
90
xx0
_ _
2
E
2
t
2
90
~ u
n
(l=2): (6:55)
Substituting Eqs. (6.53) and (6.55) into Eq. (6.52), the energy released by forma
tion of a crack midway between two existing cracks of spacing s is
W
2NN
= 2N
s
90
xx0
_ _
2
E
2
t
2
90
2~ u(l=2) ~ u(l) [ [: (6:56)
The cracks form when this work is greater than or equal to the cumulative surface
energy of newly created surfaces, i.e.,
W
2NN
_ 2 N 2t
90
G
c
: (6:57)
From Eqs. (6.56) and (6.57), the criterion for crack formation is
s
90
xx0
_ _
2
2E
2
t
90
2~ u(s=2) ~ u(s) [ [ _ G
c
: (6:58)
To analyze cracking in an arbitrary position between two preexisting cracks,
a new crack is introduced in an arbitrary position between the cracks (Figure 6.19
(b)), which leads to a new damage state with one crack spacing equal to s
1
and the
second one equal to s
2
= s s
1
. The cracking criterion in this case is
s
90
xx0
_ _
2
2E
2
t
90
2~ u(l
1
=2) ~ u(l
1
) 2~ u(l
2
=2) ~ u(l
2
) [ [ _ G
c
: (6:59)
207 6.4 Modeling approaches
The authors applied their analysis for the prediction of crack density evolution
in glass/epoxy [y/90
4
]
s
laminates for the case of varying crack spacing.
A Weibull distribution for G
c
was utilized and the CODs were calculated
using FE analysis. The predictions are shown against experimental data in
Figure 6.22.
Adolfsson and Gudmundson [11] also developed an energybased damage
evolution approach using their stress analysis. Basic stress analysis using this
approach is covered in Section 4.9.1, although they updated their analysis to
include bending loads, which can be found in [11, 115]. The energy model for
crack density evolution is based on changes in strain energy due to cracking. From
[11], the strain energy per unit inplane area of the damaged laminate with n plies
may be written as
w
(c)
=
1
2
e a
(c)
T e
(R)
_ _
T
C
(c)
e a
(c)
T e
(R)
_ _
n
k=1
h
k
T; s
k(R)
_ _
; (6:60)
where boldface letters represent matrices; C
(c)
, a
(c)
, and e
(R)
are the stiffnesses,
thermal expansion coefcients and residual stressinduced eigenstrains, respect
ively, of the cracked laminate. Expressions for these crack density quantities were
derived in [115]. In Eq. (6.60), DT is the temperature difference between the curing
and service temperature and h
k
are the functions containing energy stored in the
laminate due to interlaminar constraints and thermal residual stresses (their
expressions are given in Appendix A of [11]). From the strain energy, the energy
release rate for the ith cracked ply is derived as
0.8
[q/90
4
]
s
FEM model, energy approach
0.6
0.4
0.2
0.0
q=0
q=15
q=30
q=40
0
C
r
a
c
k
d
e
n
s
i
t
y
(
c
r
/
m
m
)
50 100
Stress s
0
(MPa)
150 200 250
Figure 6.22. Evolution of crack density as a function of applied stress using energy model
for [y/90
4
]
s
laminates. Symbols represent experimental data and lines represent the
average of four runs using the energy model [113]. Reprinted, with kind permission, from
Compos Sci Technol, Vol. 61, R. Joffe, A. Krasnikovs, and J. Varna, CODbased simulation
of transverse cracking and stiffness reduction in [S/90n]s laminates, pp. 63756, copyright
Elsevier (2001).
208 Damage progression
G
i
=
@U
@A
i
=
@ Aw
(c)
_ _
@A
i
; (6:61)
where A
i
is the crack surface area in ply i. The area A
i
is given by A
i
= Ar
i
with
normalized crack density r
i
= t
i
/ l
i
, where l
i
is the average spacing of cracks in the
ith ply and A is the laminate inplane area. From Eqs. (6.60) and (6.61) the energy
release rate G
i
for cracking in the ith ply is given by
G
i
=
@a
(c)
@r
i
T
@e
(R)
@r
i
_ _
T
C
(c)
e a
(c)
T e
(R)
_ _
1
2
e a
(c)
T e
(R)
_ _
T
@C
(c)
@r
i
e a
(c)
T e
(R)
_ _
N
k=1
@h
k
@r
i
:
(6:62)
The above expression contains the derivatives of the effective thermoelastic
properties of the damaged laminate with respect to the ply crack densities.
Calculating these quantities is a more complex task than determining the proper
ties themselves. For this purpose, either FEM or the approximate analytical
expressions given in Appendix A of [11] have to be utilized. The model predictions
for the stressstrain response of graphite/epoxy crossply laminates are compared
with experimental data in Figure 6.23.
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Strain (%)
0
100
200
300
400
500
600
S
t
r
e
s
s
(
M
P
a
)
(0/90/0)
(0/90
2
/0)
(0/90
4
/0)
(0/90
8
/0)
Simulations
Figure 6.23. Stressstrain response of graphite/epoxy crossply laminates. Symbols
represent experimental data whereas the solid lines represent predictions from the model by
Adolfsson and Gudmundson [11]. Reprinted, with kind permission, from Int J Solids Struct,
Vol. 36, E. Adolfsson and P. Gudmundson, Matrix crack initiation and progression
in composite laminates subjected to bending and extension, pp. 313169, copyright
Elsevier (1999).
209 6.4 Modeling approaches
QuHoiseth analysis
The evaluation of moduli for damaged crossply laminates using the approach
proposed by Qu and Hoiseth [116] is covered in Section 4.5. The cracking criterion
for this model is derived as
G
c
=
2e
c
2
t
90
E
c
E
2
(E
1
E
2
)r
exp
dr
2t
90
_ _
exp
dr
t
90
_ _ _ _
; (6:63)
where G
c
is the inplane mode I fracture toughness of transverse ply; e
c
is the applied
strain; E
c
is the plainstrain Youngs modulus of the undamaged crossply laminate
in the longitudinal direction; E
1
and E
2
are longitudinal and transverse moduli of
the ply, respectively; d is the average crack opening displacement of the 90
crack;
and r = t
90
/ l is the normalized crack density. The threshold strain at which the
transverse matrix cracking initiates can be obtained by setting r 0 in Eq. (6.63) as
e
0
=
G
c
E
1
E
2
( )
dE
c
E
2
: (6:64)
The model predictions for a crossply laminate made of AS/350106 material for
two values of G
c
are shown in Figure 6.24. The experimental data are from [117].
6.4.3 Strength vs. energy criteria for multiple cracking
There is a fundamental difference between strength and energybased criteria
when applied to multiple ply cracking in laminates. Strength essentially represents
the failure at a material point when a specied stress component or function of
0.0
C
r
a
c
k
d
e
n
s
i
t
y
0.0
0.5
s
c
= E
c
e
c
Prediction with G
c
=130J/m
2
1.0
1.5
2.0
0.5
Applied stress
1.0
AS/350106 [0
2
/90
2
]
s
1.5 2.0
Experimental data
Prediction with G
c
=104J/m
2
Figure 6.24. Comparison between model predictions and experimental data for the COD
model of Qu and Hoiseth [116] for a crossply laminate, s
c
represents the applied stress on
the composite. Reprinted, with kind permission, from Fatigue Frac Eng Mater, Vol. 24,
J. Qu and K. Hoiseth, Evolution of transverse matrix cracking in crossply laminates,
pp. 451464, copyright Wiley (1998).
210 Damage progression
stress components reaches a critical value. This approach is a legacy of homo
geneous materials such as metals and ceramics where yielding at a point (in a
metal) is assumed to occur according to, e.g., the von Mises criterion, or brittle
failure (in a ceramic) is assumed to occur when the maximum tensile principal
stress reaches a critical value. In case of yielding there is no ambiguity in terms of
the stress components attaining critical value at a point since this type of failure
can spread spatially from one point to another as indicated by the contour of the
yield criterion. However, brittle failure represents the instability of crack growth,
and unless a crack exists its growth is meaningless. This difculty has been
conventionally overcome (or bypassed) by assuming that brittle failure according
to a pointfailure (strength) criterion is the concurrent formation and instability of
a crack. In an unconstrained failure case, such as the brittle failure of an unre
inforced ceramic, this approximating assumption causes little difculty since the
initiation of cracking and its unstable growth are usually not far apart, i.e., they
occur at roughly the same applied load. However, when the constraint to crack
growth is imposed by the presence of reinforcements, or stiff elements in the matrix
generally, then the formation of a crack and the instability of its growth are
determined by different conditions. This fact was realized in [91] when it was found
that the strength criterion was inadequate to predict the formation of multiple
cracking in unidirectional brittle matrix composites. Energy considerations were
then made by recognizing the dissipation of energy in crack surface formation.
Multiple cracking in composite materials is an inherent feature of the failure
process due to the presence of directed interfaces (ber/matrix and interlami
nar) that impart mechanisms of stress transfer from the cracking elements to
the noncracking elements, which in turn provide a constraint to the cracks.
Thus, the inadequate incorporation of constrained cracking in a multiple
cracking process is bound to induce error. One example, unfortunately not
uncommon, is using solutions of crack opening displacements in an innite
medium in models of multiple cracking.
6.5 Randomness in ply cracking
Physical observations of ply cracking indicate that in the early stages of the cracking
process randomness exists in the location of cracks, their size, and how the evolu
tion (growth and multiplication) of cracking occurs. As the cracking process
evolves, randomness tends to decrease, and as the process approaches saturation,
uniformity in crack spacing results. The causes of randomness are many, most
induced by the manufacturing process. For instance, the ber volume fraction can
vary spatially. Image analysis reported in [118] showed that in a T300/914 carbon/
epoxy composite with an average ber volume fraction of 55.9% the local volume
fraction ranged between 15 and 85%. Other common defects are voids and inclu
sions in the matrix, partially cured regions of matrix, broken bers, ber waviness,
unbonded regions of interfaces at ber surfaces, and between plies.
211 6.5 Randomness in ply cracking
Several attempts exist in the literature to treat random variations of microstruc
ture. Silberschmidt [119, 120] suggested a lattice scheme which incorporates the
effects of the initial microstructural randomness as well as a dispersed evolution of
damage and its transition to spatially localized matrix cracking. The scheme
involves mapping a dynamic matrix of stressrenormalizing coefcients onto the
lattice of elements covering the cracked (90
(
M
P
a
)
Figure 6.26. Crack distribution (a) and corresponding variation (b) of the average
longitudinal stress in the 90
2p
_ exp
(a m
a
)
2
2s
2
a
_ _
;
f (S) =
1
S
2p
_ exp
(S m
S
)
2
2s
2
S
_ _
;
(6:65)
where 2a is the average aw size, S is the average distance between two adjacent
aws, and m
a
, m
S,
s
a
, and s
S
are tting parameters. The worst of the aws causes
the rst ply cracking. With increased loading, smaller aws cause further trans
verse cracking. The rst transverse crack forms when
G s
c
; a
0
( ) = G
c
; (6:66)
where s
c
is the longitudinal stress applied to the composite, 2a
0
is the initial aw
size, and G
c
is the critical energy release rate, which is assumed to be constant
along the laminate length. The propagation of the aw will be stable if
G s
c
; a
0
a ( )<G
c
; (6:67)
and unstable if
G(s
c
; a
0
a) > G
c
: (6:68)
Progressive cracking will ensue when there is enough energy available for multiple
aws to propagate into fully grown transverse cracks. For crack formation after
the rst crack the energy release rate for aw propagation depends on its relative
distance S from the existing crack and can be expressed as
G(s
c
; a) = R(S)G
0
(s
c
; a); (6:69)
where G
0
is the energy release rate when no crack is present, and R(S) is the energy
retention factor, with a value between 0 and 1, accounting for the presence of a
neighboring crack. Similarly, for a aw to propagate between two existing trans
verse cracks, the energy release rate is
G(s
c
; a) = R(S
L
)G
0
(s
c
; a)R(S
R
) (6:70)
where S
L
and S
R
are the distance of the aw from the left and right cracks,
respectively. Chou et al. [126] implemented the approach using a Monte Carlo
scheme. The results showed a fair agreement with experimental data. Essentially,
213 6.5 Randomness in ply cracking
this approach predicts the event when a microaw develops into a fully grown
transverse crack, thereby predicting the multiplication of ply cracks. However, the
approach has not gained wide usage because it requires many unknown param
eters which are found by tting to experimental data. As mentioned earlier,
experimental observations show that the transverse cracks usually grow quickly
through the 90
0
rp r ( )dr; w =
_
0
w r ( )p r ( ) dr; (6:71)
where p(r) is the probability density function (PDF) of distances between adjacent
cracks. The criterion for rst crack formation can be found by substituting r
in Eq. (6.40) and using Eq. (6.36), to obtain
g = s
90
xx0
_ _
2
t
90
C
22
w ( ) = 2 s
90
xx0
_ _
2
t
90
C
22
a a
2
b
2
_ _
: (6:72)
In fact the criterion in Eq. (6.72) is expected to predict the initial stage of the
damage evolution curve. For any material block between two existing adjacent
cracks, the cracking criterion in Eq. (6.40) can be rewritten as [98]
g = s
90
xx0
_ _
2
t
90
C
22
w
r x
2
_ _
w
r x
2
_ _
w r ( )
_ _
; (6:73)
where x denotes the nondimensional coordinate of the new crack location between
the two existing cracks. Equation (6.73) is a local criterion for crack formation
because it deals with the location of the next crack.
The physical nature of damage evolution can be achieved by having a
probabilistic variation of material property g. Thus,
g = G(x): (6:74)
The parameter G(x) can be thought of as local toughness of the material by
arguing that it is easy to form a crack at a section which contains many aws and
has a weak interface. The variation of g is usually described using a Weibull
distribution, i.e., the PDF of g can be expressed as
214 Damage progression
p
r
(g) =
g
0
g g
min
g
0
_ _
1
exp
g g
min
g
0
_ _ _ _
; g _ g
min
; (6:75)
where g
min
is the minimum possible value of g, and and g
0
are parameters of the
distribution, usually evaluated by tting experimental data.
For different laminate systems, i.e., for different mixes of 0 and 90
plies,
e.g., [0
n1
/90
m1
]
s
and [0
n2
/90
m2
]
s
laminates, the distribution parameters
may not be the same. If the parameters for the rst laminate conguration,
1
, g
01
, are known (through tting of experimental data), the parameters
2
, g
02
for the second laminate conguration can be found from the following
relations
2
2
2
_ _
2
1
2
2
_ _
2
1
2
2
_ _ =
m
2
m
1
2
1
1
_ _
2
1
1
1
_ _
2
1
1
1
_ _ ;
g
02
= g
01
1
1
1
_ _
1
2
2
_ _;
(6:76)
where G(x) represents the standard gamma function for the random variable x.
The derivation can be found in the original article [98].
The simulation procedure for this model can be summarized as follows:
1. Choose a ply material or a laminate conguration of a ply material.
2. Distribute random points for possible crack locations along the laminate
length.
3. Generate a random value of g at each point according to the Weibull
distribution.
4. Fit the model predictions to the experimental data for crack density evolution
to deduce the parameters of the Weibull distribution.
5. Calculate Weibull parameters for other laminate systems using Eq. (6.76), and
predict the crack density evolution for these laminates.
Some examples of the numerical simulation results with the tted and calculated
parameters of the distribution are shown in Figure 6.27.
A recent strengthbased analysis by Berthelot and Le Corre [121] has revealed
that the choice of probabilistic distribution should account for weakness areas in
the material. This analysis for [0/90
2
]
s
carbon/epoxy laminate shows that a prob
abilistic distribution of strength which accounts for weakness areas properly
corrects the crack density evolution in the beginning stage (see Figure 6.28).
A divergence between models with and without consideration of weakness areas
is always observed at low crack densities. This is because, initially, cracking is
preferred at weakness areas where the fracture toughness of the material is low
215 6.5 Randomness in ply cracking
(due to inherent defects) as compared to its average value in the whole laminate.
For glass/epoxy laminates, Berthelot and Le Corre found that delamination
occurs at high crack densities and is the cause of data deviating from the model
prediction.
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
(a)
200
C
r
a
c
k
d
e
n
s
i
t
y
(
1
/
m
m
)
[0/90
3
]
s
[0/90
2
]
s
400 600
Stress (MPa)
800 1000 1200 1400
Avimid K Polymer/IM6
0
1.2
(b)
100
C
r
a
c
k
d
e
n
s
i
t
y
(
1
/
m
m
)
[0/90
4
]
s
[0/90
2
]
s
[0/90]
s
200 300 400
Stress (MPa)
Hercules AS4/35016
500 600 700 800
1
0.8
0.6
0.4
0.2
0
Figure 6.27. Prediction of crack density evolution in crossply laminates using the
Vinogradov and Hashin model [98] for two material systems: (a) Avimid K polymer/IM6,
(b) Hercules AS4/35016. The experimental data are from [127]. Reprinted, with kind
permission, from Int J Solids Struct, Vol. 42, V. Vinogradov and Z. Hashin, Probabilistic
energybased model for prediction of transverse cracking in crossply laminates, pp. 365
392, copyright Elsevier (2005).
216 Damage progression
6.6 Damage evolution in multidirectional laminates
Although many generic features of ply cracking are evident in crossply laminates,
this class of laminates is used only in limited cases. Most applications require a mix
of lamina orientations in the laminate conguration to generate properties to carry
combinations of normal loads, bending moments and torsion. In a multidirectional
laminate the ply cracking in any ply will generally take place under stresses normal
to and parallel to the bers, as well as inplane shear. Experimental investigations
[27, 34, 35, 50, 51, 128] have claried some of the complexities relating to mode
mixity of crack growth and interactions between cracks within and among plies.
Consider now a [0/90/y
1
/y
2
]
s
laminate where ply cracks can appear in the 90
,
y
1
, and y
2
plies, assuming loading in the 0
1
)
200
0
250
300 350 400
Average stress (MPa)
450 500 550
1400
(b)
1200 Experimental results
Without weakness areas
With weakness areas
1000
800
C
r
a
c
k
d
e
n
s
i
t
y
(
m
1
)
600
400
200
0
250 300 350 400
Average stress (MPa)
450 500 550
Figure 6.28. Evolution of crack density as a function of applied stress for a [0/90
2
]
s
carbon/
epoxy laminate [121]. The experimental data are from [1]. Reprinted, with kind permission,
from Compos Sci Technol, Vol. 60, J. M. Berthelot and J. F. Le Corre, Statistical analysis
of the progression of transverse cracking and delamination in crossply laminates,
pp. 265969, copyright Elsevier (2000).
217 6.6 Damage evolution in multidirectional laminates
begin cracking (assuming y
1
>y
2
), and witha further increase in the imposed load, an
interactive cracking process continues inthe 90
 andy
1
ply orientations. At straine
y
2
0
cracking initiates in the y
2
plies, and eventually all offaxis plies conduct interactive
crack multiplication. The crack initiation strains and crack multiplication rates
dependonthe constraint imposedby the neighboring plies tothe cracks ina givenply.
For predicting the evolution of ply cracking the authors of this book have
developed an energybased approach, which is capable of dealing with cracking
in offaxis plies of orthotropic laminates. In [129] the approach is described and
applied to several ply cracking cases. A brief description of the approach follows.
As illustratedinFigure 6.30, twodamage states are considered: state 1 withNparallel
offaxis cracks spaced at distance s, and state 2 where the cracks have multiplied to 2N
and attained the spacing s/2. Evolution of cracking damage is assumed when the work
requiredingoing fromstate 1 tostate 2 (whichis the same as the workneededtoclose N
cracks in going from state 2 to state 1) exceeds a critical value, i.e., if
W
2NN
_ N:G
c
:
1
sin y
t
c
; (6:77)
where y is the offaxis angle and G
c
is the critical (threshold) value of the energy
required for multiple ply crack formation within the given laminate (more discus
sion about this later). The work required to form N additional cracks in going from
state 1 to state 2 (the same as the work required to close those cracks) is given by
W
2NN
= W
2N0
W
N0
; (6:78)
90 q
1
q
2
0
0 0
Figure 6.29. Cracking process in a [0/90/y
1
/y
2
]
s
halflaminate.
State 1: N cracks, crack spacing = s State 2: 2N cracks, crack spacing = s/2
t
q
s
s/2
x
3
x
2
Figure 6.30. Progressive multiplication of offaxis ply cracks in a multidirectional laminate
[129]. Reprinted, with kind permission, from Int J Solids Struct, Vol. 47, C. V. Singh
and R. Talreja, Evolution of ply cracks in multidirectional composite laminates,
pp. 133849, copyright Elsevier (2010).
218 Damage progression
where W
N0
and W
2N0
represent the work required to close N cracks in state 1,
and 2N cracks in state 2, respectively, and the two quantities are calculated as (see
[130] for detailed derivation)
W
N0
= N
1
sin y
t
c
( )
2
1
E
2
s
y
20
_ _
2
:~ u
y
n
s ( ) s
y
120
_ _
2
~ u
y
t
s ( )
_ _
; (6:79)
W
2N0
= 2N
1
sin y
t
c
( )
2
1
E
2
s
y
20
_ _
2
:~ u
y
n
s
2
_ _
s
y
120
_ _
2
~ u
y
t
s
2
_ _ _ _
; (6:80)
where ~ u
y
n
; ~ u
y
t
are the normalized average crack opening and sliding displacements
(COD and CSD), respectively. These are given by
~ u
y
n
=
u
y
n
t
c
s
y
20
=E
2
_ _ =
1
t
c
s
y
20
=E
2
_ _
_
t
y
=2
t
y
=2
u
n
(z)dz;
~ u
y
t
=
u
y
t
t
c
s
y
120
=E
2
_ _ =
1
t
c
s
y
120
=E
2
_ _
_
t
y
=2
t
y
=2
u
t
(z) dz;
(6:81)
where u
n
and u
t
represent the relative opening and sliding displacement of the
cracked surfaces, respectively, and overbars represent averages. For the special
case of cracking in the 90
plies in
their model and normalize the average COD with half the ply thickness (t
c
/2).
For cracking in a general offaxis ply, one can use a multimode criterion
given as
w
I
G
Ic
_ _
M
w
II
G
IIc
_ _
N
_ 1; (6:83)
where
w
I
=
s
y
20
_ _
2
t
c
E
2
2:~ u
y
n
s
2
_ _
~ u
y
n
s ( )
_ _
; w
II
=
s
y
120
_ _
2
t
c
E
2
2:~ u
y
t
s
2
_ _
~ u
y
t
s ( )
_ _
; (6:84)
where G
IIc
is the critical energy release rate in mode II (crack sliding mode), and
the exponents M and N depend on the material system, e.g, for a glass/epoxy
system, M = 1, N = 2 [130].
In our work [129] we interpret the critical material parameters G
Ic
and G
IIc
not
in the usual linear elastic fracture mechanics sense where they are dened as the
resistance to advancement of the crack front at the point of unstable crack growth.
219 6.6 Damage evolution in multidirectional laminates
Instead, we postulate that the work required to go from state 1 to state 2 involves a
range of dissipative processes that all depend on the material condition in a
cracking ply within the given laminate. The material parameter representing the
dissipated energy per unit of ply crack surface is, therefore, not what is obtained in
a standard fracture toughness test for determining G
Ic
or G
IIc
. To emphasize that
the critical energy terms used here are not the usual fracture toughness values G
Ic
or G
IIc
, we will henceforth use the symbols W
Ic
and W
IIc
instead. These new
quantities are not to be obtained by independent tests, but are to be evaluated
by tting model predictions (Eq. (6.83)) to the experimental data for a reference
laminate. This way the values obtained will be representative of the energies
associated with multiple cracking within a laminate. A reference laminate is
chosen from the class of laminates (material etc.) for which predictions are to be
made, and for which experimental data are readily available [129, 131]. Further
more, as described in [129], it is argued that a ply crack within a laminate cannot
be formed unless sufcient energy is available to open its surfaces (i.e., in mode I
cracking). In other words, a pure sliding action will not by itself generate the set of
parallel cracks illustrated in Figure 6.30. This will imply that the second term in
Eq. (6.84) is negligible in comparison to the rst term. With these assumptions and
approximations the predictions of crack density evolution agree well with experi
mental data [129].
The complete procedure to implement the described energy model for micro
crack initiation and evolution in an offaxis ply of a general symmetric laminate is
outlined below. The procedure is in two parts:
Part I: Estimate W
Ic
1. From FE simulations, determine the variation of normalized COD and CSD
(Eq. (6.81)) with crack spacing.
2. Assume a value for W
Ic
. Plot the damage evolution for the reference laminate as
follows:
(a) Divide the specimen length into small intervals of length, dX = t
y
/10 is
chosen here.
(b) Find the multiple crack initiation strain, Eq. (6.82) with the COD value
calculated with a very large spacing (s ).
(c) Assume a small initial crack density, e.g., r
initial
= 1/50t
y
is chosen here.
(d) Choose a random length interval and check for cracking. A new crack forms
when the criterion set in Eq. (6.82) is satised. Increase the crack density and
eliminate the crackedlengthinterval fromfurther considerationfor plycracking.
(e) Choose another length interval and repeat the previous step until the
fracture criterion is satised.
(f) Increase the applied strain. Repeat steps (d) and (e) using this strain value.
3. Iterate step 2 by varying W
Ic
so that the resulting evolution curve ts the
experimental data for the reference laminate. For example, for predicting the
damage evolution in [0/y
4
/0
1/2
]
s
laminates, we chose [0/90
8
/0
1/2
]
s
as the refer
ence laminate.
220 Damage progression
Part II: Predict the damage evolution for other offaxis plies:
1. From FE simulations, determine the variation of COD and CSD (Eq. (6.81))
with crack spacing for a given offaxis laminate.
2. Using the value for W
Ic
obtained above, predict the damage evolution by
following steps 2(a)(f) described in Part I.
The above semianalytical model is coded in a MATLAB program. The input
data include the following laminate properties: ply material (elastic moduli), ply
thicknesses and orientations (i.e., laminate layup), and variation of COD with
respect to crack density (which can be obtained from independent 3D FE
analysis).
The energy model described above was applied to predict damage evolution
in glass/epoxy [0/y
4
/0
1/2
]
s
, quasiisotropic ([0/90/45]
s
), and [0
m
/90
n
/y
p
]
s
laminates [129]. Figure 6.31 shows the variation of crack initiation strains with
offaxis ply orientation (y) for [0/y
4
/0
1/2
]
s
laminates. As expected, the crack
initiation strain increases as y decreases and it may exceed 1.5% if y < 45
. In
fact, the experiments by Varna et al. [27] for this laminate revealed that ply
cracks did not form fully for y < 40
and 55
and 70
and 55
plies. Reprinted,
with kind permission, from Int J Solids Struct, Vol. 47, C. V. Singh and R. Talreja, Evolution
of ply cracks in multidirectional composite laminates, pp. 133849, copyright Elsevier (2010).
222 Damage progression
enhanced cracking in the transverse ply. The model predictions for crack density
evolution in this case are compared with experimental data in Figure 6.34. No
modier for crack spacing was needed for this case, i.e., f = 1.
A parametric study performed for [0
m
/90
n
/y
p
]
s
laminates reveals that inter
actions between the crack systems of different orientations may have a signicant
effect on damage evolution. If y is close to 90
cracks, a multimode
scenario is used in FE modeling. The 60
, 60
, and +60
 and +60
layer. If y < 45
layer, (b) 60

plies. Thus to use this solution we must retain the symmetry and periodicity of the
cracks assumed in obtaining the solution. This condition eliminates analysis of the
cyclic growth of transverse cracks from partial to full extent. Obviously, a numer
ical analysis of this case is possible, but that would not provide the analytic fatigue
damage model we intend to develop.
With the analytical stress solution to the cracked crossply laminate at hand
we see that unless some irreversible mechanism is included no change in the
crack density can be predicted since any repetition of load in an elastic solid
cannot change the stress eld. Therefore, in order to have an analytical stress
solution, i.e., keeping the symmetry of laminate geometry and periodicity of
cracks, and to incorporate irreversibility, a novel idea was proposed in [132].
According to that, all irreversibility leading to damage accumulation is lumped
into delamination surfaces emanating from the transverse crack fronts. Figure
6.36(a) illustrates the resulting model geometry of the cracked laminate while
Figure 6.36(b) shows the repeating unit cell. As shown, the preexisting trans
verse cracks are spaced at distance 2a and the delamination on either side of the
cracks is of distance d. The idea behind the model is that the delamination grows
under applied cyclic loading, the same way as a crack does, imparting changes to
the stress eld in the region between transverse ply cracks. In this way the model
captures cycledependent irreversibility, thereby allowing the fatigueinduced
multiplication of transverse ply cracks to be modeled. Although the formation
of new cracks between preexisting cracks can be modeled by different criteria, in
[132] a maximum stress criterion for cracking is used, supported by previous
work [133]. It can, however, be shown that if the delamination surfaces are
traction free, then the maximum axial stress between the ply cracks goes down
as the delamination length d increases. This suggests that the irreversibility
captured in such delamination growth is inadequate for the purpose. As argued
225 6.7 Damage evolution under cyclic loading
in [132], the delamination surfaces are indeed under compressive stress [134],
making it plausible that frictional contact exists between those surfaces. The
ensuing frictional sliding was modeled by an interfacial shear stress, which
provided the needed increase in the axial stress for crack formation.
A description of the model now follows.
Referring to Figure 6.36(a) and (b), the stress analysis is performed by a
variational approach along the lines in [101], conducting the minimization of
complementary energy separately for region I and region II. First, an admissible
stress system in the xz plane of region II is expressed as
s
L(m)
ij
= s
0(m)
ij
s
(m)
ij
; (6:86)
where s
L(m)
ij
and s
0(m)
ij
are the stress components in the cracked laminate and in the
virgin laminate, respectively, and s
(m)
ij
are the perturbations; m = 1 and 2 indicate
the 90
 and 0
/90
plies is obtained as
s
xx
0; z ( ) = s
1
1 f
1
0 ( ) ( ); t
1
<z<t
1
: (6:92)
The two constants, A
1
and A
2
, are found using the traction continuity conditions at
x =(a d) if the stress in region I is known.
To obtain the stress state in region I, a very similar variational approach is
applied. For an admissible stress system, the perturbation stresses are assumed
rst as
s
(1)
xx
= s
1
c
1
(x); s
(2)
xx
= s
2
c
2
x ( ) A
z [ [: (6:93)
A cubic variation of shear stress along the interface (z = t
1
) is enforced to account
for the effect of frictional sliding along delamination
s
(1)
xx
(x; t
1
) =
t
a
3
(x a)[x (a d)[
2
; (6:94)
227 6.7 Damage evolution under cyclic loading
where t is an unknown. After integrating the equilibrium equation, and applying
continuity of stress across the interface at z = t
1
, as well as tractionfree boundary
conditions at the crack surfaces, the admissible stress system is built up as
functions of the only unknown, t. By minimizing the corresponding complemen
tary energy functional,
dU
c
dt
= 0; (6:95)
the only unknown t is solved, so the stress eld in region I is obtained. By applying
the traction continuity conditions at the boundary between region I and region II,
A
1
and A
2
in Eq. (6.91) are determined. Substituting A
1
and A
2
into Eq. (6.92),
nally, the axial normal stress at the midplane in 90
l
_ _
m
; (6:96)
where t = t
max
t
min
( )=s
max
and l = l=t
1
, and where l is the delamination
length, denoted as d in the above stress analysis.
Integration of Eq. (6.96) yields the relationship between delamination length
and the number of cycles, N. From Figure 6.37 we see that after the initial
drop the axial normal stress increases as delamination grows. When the
maximum stress criterion s
xx
(x = 0, t
1
< z < t
1
) = s
c
is satised, a new
crack forms midway between the preexisting cracks, and the corresponding
crack spacing (or crack density) is updated. In this way, the multiplication of
matrix crack under fatigue loading is actually controlled by the growth of
delamination, and therefore determined by the number of cycles through
integration of Eq. (6.96). The damage evolution as a function of the number
0.0004 0.0002
s
x
x
(
x
=
0
,
z
=
0
)
(
P
a
)
3.7510
7
4.2510
7
4.510
7
4.7510
7
510
7
5.2510
7
0.0006
d (m)
Figure 6.37. Typical variationof the axial normal stress with delamination at a xed crack density.
228 Damage progression
of cycles in crossply laminates is nally quantitatively described in the model.
Figure 6.38 shows the variation of crack density as a function of the number of
cycles under a given cyclic tension.
6.8 Summary
Initiation of cracks within the plies of a laminate, and their growth and multipli
cation, are part of the eld of damage evolution in composites that constitutes a
key element in the performance assessment of structures made of these materials.
This chapter has focused on various stress and failure analysis methods associ
ated with the prediction of initiation and progression of ply cracks. Since cracks
form from material defects that can be considered random in their size and
spatial distributions, statistical considerations have been included in the
analyses.
For the formation of cracks the criteria used are based either on strength or on
the energy associated with fracture. Both criteria have been treated and compared.
Experimental data, wherever available, have been used to assess the predictions.
Most of the damage evolution work in the literature has been for transverse
cracking in crossply laminates. While this cracking mode has been amply dealt
with, more recent work on oblique cracks, i.e., cracks in offaxis plies of multi
directional laminates, has also been treated.
While the fatigue of composite laminates is the focus of a separate chapter,
where a broad treatment of the subject is given, a section on ply cracking under
cyclic loading has been included here to illustrate how damage accumulation
under repeated loads is modeled.
2.0
1.8
1.6
1.4
C
r
a
c
k
d
e
n
s
i
t
y
(
/
m
m
)
1.2
1.0
0 200 000 400 000 600 000
Model prediction
Experiment
Number of cycles
800 000 1000 000
Figure 6.38. Transverse crack evolution with cycles for a [0/90
2
]
s
carbon/epoxy laminate at a
maximum stress of 482.633 MPa and a stress ratio of 0.1.
229 6.8 Summary
References
1. A. S. D. Wang, Fracture mechanics of sublaminate cracks in composite materials.
In Composites Technology Review. (Philadelphia, PA: ASTM, 1984), pp. 4562.
2. J. M. Masters and K. L. Reifsnider, An investigation of cumulative damage develop
ment in quasiisotropic graphite/epoxy laminates. In Damage in Composite Materials,
ASTM STP 775, ed. K. L. Reifsnider. (Philadelphia, PA: ASTM, 1982), pp. 4062.
3. K. L. Reifsnider, ed. Damage in Composite Materials, ASTM STP 775, ed. K. L.
Reifsnider. (Philadelphia, PA: ASTM, 1982).
4. J. E. Masters, ed. Damage Detection in Composite Materials, ASTM STP 1128.
(Philadelphia, PA: ASTM, 1992).
5. J. A. Nairn, S. F. Hu, and J. S. Bark, A criticalevaluation of theories for predicting
microcracking in composite laminates. J Materials Sci, 28:18 (1993), 5099111.
6. D. O. Stalnaker and W. W. Stinchcomb, Load historyedge damage studies in
two quasiisotropic graphite epoxy laminates. In Composite Materials: Testing and
Design (Proc. 5th Conference), ASTM STP 674. (Philadelphia, PA: ASTM, 1979), pp.
62041.
7. J. E. Masters, An experimental investigation of cumulative damage development in
graphite epoxy laminates. Ph.D. dissertation, Materials Engineering Science, Depart
ment, Virginia Polytechnic Institute, Blacksburg, VA, 1980, p. 4062.
8. A. L. Highsmith and K. L. Reifsnider, Stiffnessreduction mechanisms in composite
laminates. In Damage in Composite Materials, ASTM STP 775, ed. K. L. Reifsnider.
(Philadelphia, PA: ASTM, 1982), pp. 10317.
9. K. L. Reifsnider and R. Jamison, Fracture of fatigueloaded composite laminates. Int
J Fatigue, 4:4 (1982), 18797.
10. K. L. Reifsnider and A. Talug, Analysis of fatigue damage in composite laminates. Int
J Fatigue, 2:1 (1980), 311.
11. E. Adolfsson and P. Gudmundson, Matrix crack initiation and progression in composite
laminates subjected to bending and extension. Int J Solids Struct, 36:21 (1999), 313169.
12. B. F. Sorensen and R. Talreja, Analysis of damage in a ceramic matrix composite. Int
J Damage Mech, 2:3 (1993), 24671.
13. B. F. Sorensen, R. Talreja, and O. T. Sorensen, Micromechanical analysis of damage
mechanisms in ceramicmatrix composites during mechanical and thermal cycling.
Composites, 24:2 (1993), 12940.
14. T. J. Dunyak, W. W. Stinchcomb, and K. L. Reifsnider, Examination of selected NDE
techniques for ceramic composite components. In Damage Detection in Composite
Materials, ASTM STP 1128, ed. J. E. Masters. (Philadelphia, PA: ASTM, 1992), pp.
324.
15. K. V. Steiner, Defect classications in composites using ultrasonic nondestructive
evaluation techniques. In Damage Detection in Composite Materials, ASTM STP 1128,
ed. J. E. Masters. (Philadelphia, PA: ASTM, 1992), pp. 7284.
16. K. V. Steiner, R. F. Eduljee, X. Huang, and J. W. Gillespie, Ultrasonic NDE tech
niques for the evaluation of matrix cracking in composite laminates. Compos Sci
Technol, 53:2 (1995), 193198.
17. G. E. Maddux and G. P. Sendeckyj, Holographic techniques for defect detection in
composite materials. In Nondestructive Evaluation and Flaw Criticality for Composite
Materials, ASTM STP 696. (Philadelphia, PA: ASTM, 1979), pp. 2644.
230 Damage progression
18. G. P. Sendeckyj, G. E. Maddux, and E. Porter, Damage documentation in composites
by stereo radiography. In Damage in Composite Materials, ASTM STP 775, ed. K. L.
Reifsnider. (Philadelphia, PA: ASTM, 1982), pp. 1626.
19. R. D. Jamison, K. Schulte, K. L. Reifsnider, and W. W. Stinchcomb, Characterization
and analysis of damage mechanisms in tensiontension fatigue of graphite/epoxy
laminates. In Effects of Defects in Composite Materials, ASTM STP 836. (Philadelphia,
PA: ASTM, 1984), pp. 2155.
20. C. Bathias and A. Cagnasso, Application of xray tomography to the nondestructive
testing of highperformance polymer composites. In Damage Detection in Composite Mater
ials, ASTM STP 1128, ed. J. E. Masters. (Philadelphia, PA: ASTM, 1992), pp. 3554.
21. K. Maslov, R. Y. Kim, V. K. Kinra, and N. J. Pagano, A new technique for the
ultrasonic detection of internal transverse cracks in carbonbre/bismaleimide compos
ite laminates. Compos Sci Technol, 60:1213 (2000), 218590.
22. V. K. Kinra, A. S. Ganpatye, and K. Maslov, Ultrasonic plybyply detection of matrix
cracks in laminated composites. J Nondestr Eval, 25:1 (2006), 3951.
23. Y. Zou, L. Tong, and G. P. Steven, Vibrationbased modeldependent damage (dela
mination) identication and health monitoring for composite structures a review.
J Sound Vib, 230:2 (2000), 35778.
24. S. K. Seth, et al., Damage detection in composite materials using lamb wave methods.
Smart Mater Struct, 11:2 (2002), 269.
25. R. Talreja, Transverse cracking and stiffness reduction in composite laminates.
J Compos Mater, 19:4 (1985), 35575.
26. J. Varna, N. V. Akshantala, and R. Talreja, Crack opening displacement and the associ
ated response of laminates with varying constraints. Int J Damage Mech, 8 (1999), 17493.
27. J. Varna, R. Joffe, N. V. Akshantala, and R. Talreja, Damage in composite laminates
with offaxis plies. Compos Sci Technol, 59:14 (1999), 213947.
28. D. G. Katerelos, L. N. McCartney, and C. Galiotis, Local strain redistribution and
stiffness degradation in crossply polymer composites under tension. Acta Mater, 53:12
(2005), 333543.
29. D. T. G. Katerelos, P. Lundmark, J. Varna, and C. Galiotis, Analysis of matrix
cracking in GFRP laminates using Raman spectroscopy. Compos Sci Technol, 67:9
(2007), 194654.
30. D. G. Katerelos, M. Kashtalyan, C. Soutis, and C. Galiotis, Matrix cracking in
polymeric composite laminates: Modelling and experiments. Compos Sci Technol,
68:12 (2008), 231017.
31. D. G. Katerelos, L. N. McCartney, and C. Galiotis, Effect of offaxis matrix cracking
on stiffness of symmetric angleply composite laminates. Int J Fract, 139:34 (2006),
52936.
32. D. T. G. Katerelos, J. Varna, and C. Galiotis, Energy criterion for modelling
damage evolution in crossply composite laminates. Compos Sci Technol, 68:12 (2008),
231824.
33. J. A. Nairn, Matrix microcracking in composites. In Polymer Matrix Composites, ed. R.
Talreja and J. A. E. Manson. (Amsterdam: Elsevier Science, 2000), pp. 40332.
34. J. Tong, F. J. Guild, S. L. Ogin, and P. A. Smith, On matrix crack growthinquasiisotropic
laminates I. Experimental investigation. Compos Sci Technol, 57:11 (1997), 152735.
35. L. E. Crocker, S. L. Ogin, P. A. Smith, and P. S. Hill, Intralaminar fracture in angle
ply laminates. Compos A, 28:910 (1997), 83946.
231 References
36. N. Balhi, et al., Intralaminar cracking in CFRP laminates: observations and model
ling. J Mater Sci, 41:20 (2006), 6599609.
37. K. W. Garrett and J. E. Bailey, Effect of resin failure strain on tensile properties of glass
berreinforced polyester crossply laminates. J Mater Sci, 12:11 (1977), 218994.
38. K. W. Garrett and J. E. Bailey, Multiple transverse fracture in 90
crossply laminates
of a glass brereinforced polyester. J Mater Sci, 12:1 (1977), 15768.
39. A. Parvizi and J. E. Bailey, Multiple transverse cracking in glassber epoxy crossply
laminates. J Mater Sci, 13:10 (1978), 21316.
40. A. Parvizi, K. W. Garrett, and J. E. Bailey, Constrained cracking in glass bre
reinforced epoxy crossply laminates. J Mater Sci, 13:1 (1978), 195201.
41. J. E. Bailey, P. T. Curtis, and A. Parvizi, On the transverse cracking and longitudinal
splitting behavior of glass and carbonber reinforced epoxy cross ply laminates and
the effect of Poisson and thermally generated strain. Proc R Soc London A, 366:1727
(1979), 599623.
42. M. G. Bader, J. E. Bailey, P. T. Curtis, and A. Parvizi, eds., The mechanisms of
initiation and development of damage in multiaxial brereinforced plastic laminates.
In Proceedings of the Third International Conference on the Mechancial Behaviour in
Materials (ICM3), Cambridge, UK, Vol. 3. (1979), pp. 22739.
43. J. E. Bailey and A. Parvizi, On ber debonding effects and the mechanism of trans
verseply failure in crossply laminates of glass berthermoset composites. J Mater Sci,
16:3 (1981), 64959.
44. F. R. Jones, A. R. Wheatley, and J. E. Bailey, The effect of thermal strains on the
microcracking and stress corrosion behaviour of GRP. In Composite Structures 1st
International Conference, ed. I. H. Marshall. (Barking, UK: Applied Science Publishers,
1981), pp. 41529.
45. D. L. Flaggs and M. H. Kural, Experimentaldetermination of the insitu transverse
lamina strength in graphite epoxy laminates. J Compos Mater, 16 (1982), 10316.
46. J. A. Nairn and S. Hu, Micromechanics of damage: a case study of matrix micro
cracking. In Damage Mechanics of Composite Materials, ed. R. Talreja. (Amsterdam:
Elsevier, 1994), pp. 187243.
47. R. Talreja, Transverse cracking and stiffness reduction in composite laminates.
J Compos Mater, 19:4 (1985), 35575.
48. J. A. Nairn and S. F. Hu, The formation and effect of outerply microcracks in cross
ply laminates a variational approach. Eng Fract Mech, 41:2 (1992), 20321.
49. J. Tong, F. J. Guild, S. L. Ogin, andP. A. Smith, Onmatrix crackgrowthinquasiisotropic
laminates II. Finite element analysis. Compos Sci Technol, 57:11 (1997), 153745.
50. T. Yokozeki, T. Aoki, T. Ogasawara, and T. Ishikawa, Effects of layup angle and ply
thickness on matrix crack interaction in contiguous plies of composite laminates.
Compos A, 36:9 (2005), 122935.
51. T. Yokozeki, T. Aoki, and T. Ishikawa, Consecutive matrix cracking in contiguous
plies of composite laminates. Int J Solids Struct, 42:910 (2005), 2785802.
52. J. E. Masters and K. L. Reifsnider, An investigation of cumulative damage deve
lopment in quasiisotropic graphite/epoxy laminates. In Damage in Composite
Materials, ASTM STP 775, ed. K. L. Reifsnider. (Philadelphia, PA: ASTM, 1982),
pp. 4062.
53. N. V. Akshantala and R. Talreja, A micromechanics based model for predicting fatigue
life of composite laminates. Mater Sci Eng A, 285:12 (2000), 30313.
232 Damage progression
54. K. W. Garrett and J. E. Bailey, Multiple transverse fracture in 90 degrees crossply
laminates of a glass berreinforced polyester. J Materials Sci, 12:1 (1977), 15768.
55. J. M. Berthelot, Transverse cracking and delamination in crossply glassber and
carbonber reinforced plastic laminates: static and fatigue loading. Appl Mech Rev,
56 (2003), 11147.
56. S. E. Groves, et al., An experimental and analytical treatment of matrix cracking in
crossply laminates. Exp Mech, 27:1 (1987), 739.
57. S. F. Hu, J. S. Bark, and J. A. Nairn, On the phenomenon of curved microcracks in
[(S)/90(n)](s) laminates their shapes, initiation angles and locations. Compos Sci
Technol, 47:4 (1993), 3219.
58. P. Lundmark and J. Varna, Damage evolution and characterisation of crack types in
CF/EP laminates loaded at low temperatures. Eng Fract Mech, 75:9 (2008), 263141.
59. H. L. McManus, D. E. Bowles, and S. S. Tompkins, Prediction of thermal cycling
induced matrix cracking. J Reinf Plast Compos, 15:2 (1996), 12440.
60. H. L. McManus and J. R. Maddocks, On microcracking in composite laminates under
thermal and mechanical loading. Polym Compos, 4:5 (1996), 30514.
61. C. H. Park and H. L. McManus, Thermally induced damage in composite laminates:
predictive methodology and experimental investigation. Compos Sci Technol, 56:10
(1996), 120917.
62. T. G. Reynolds and H. L. McManus, Understanding and accelerating environmentally
induced degradation and microcracking. In Proceedings of 39th AIAA Structures, Struc
tural Dynamics and Materials Conference. (Reston, VA: AIAA, 1998, pp. 210316.
63. D. S. Adams, D. E. Bowles, and C. T. Herakovich, Thermally induced transverse
cracking in graphiteepoxy crossply laminates. J Reinf Plast Compos, 5:3 (1986),
15269.
64. C. T. Herakovich and M. W. Hyer, Damageinduced property changes in composites
subjected to cyclic thermal loading. Eng Fract Mech, 25:56 (1986), 77991.
65. R. G. Spain, Thermal microcracking of carbon bre/resin composites. Compos, 2:1
(1971), 337.
66. J. A. Lavoie and E. Adolfsson, Stitch cracks in constraint plies adjacent to a cracked
ply. J Compos Mater, 35:23 (2001), 207797.
67. X. Huang, J. W. Gillespie, and R. F. Eduljee, Effect of temperature on the transverse
cracking behavior of crossply composite laminates. Compos B, 28:4 (1997), 41924.
68. M. H. Han and J. A. Nairn, Hygrothermal aging of polyimide matrix composite
laminates. Compos A, 34:10 (2003), 97986.
69. J. E. Lundgren and P. Gudmundson, Moisture absorption in glassbre/epoxy lamin
ates with transverse matrix cracks. Compos Sci Technol, 59:13 (1999), 198391.
70. E. Vauthier, J. C. Abry, T. Bailliez, and A. Chateauminois, Interactions between
hygrothermal ageing and fatigue damage in unidirectional glass/epoxy composites.
Compos Sci Technol, 58:5 (1998), 68792.
71. A. Tounsi, K. H. Amara, and E. A. AddaBedia, Analysis of transverse cracking and
stiffness loss in crossply laminates with hygrothermal conditions. Comput Mater Sci,
32:2 (2005), 16774.
72. P. A. Smith and S. L. Ogin, On transverse matrix cracking in crossply laminates loaded
in simple bending. Compos A, 30:8 (1999), 10038.
73. P. A. Smith and S. L. Ogin, Characterization and modelling of matrix cracking in a (0/90)
2s GFRP laminate loaded in exure. Proc R Soc London A, 456:2003 (2000), 275570.
233 References
74. S.R. Kimand J. A. Nairn, Fracture mechanics analysis of coating/substrate systems: part
I: analysis of tensile and bending experiments. Eng Fract Mech, 65:5 (2000), 57393.
75. S.R. Kim and J. A. Nairn, Fracture mechanics analysis of coating/substrate systems:
part II: experiments in bending. Eng Fract Mech, 65:5 (2000), 595607.
76. L. N. McCartney and C. Pierse, Stress Transfer Mechanics for Multipleply Laminates
Subject to Bending. NPL Report CMMT(A) 55, February 1997.
77. L. N. McCartney and C. Pierse, Stress transfer mechanics for multiple ply laminates for
axial loading and bending. Proc ICCM11, 5 (1997), 66271.
78. S. A. Salpekar and T. K. Obrien, Analysis of matrix cracking and local delamination in
(0/y/y)s graphiteepoxy laminates under tensile load. J Compos Tech Res, 15:2 (1993),
95100.
79. P. Johnson and F. K. Chang, Characterization of matrix crackinduced laminate
failure part I: experiments. J Compos Mater, 35:22 (2001), 200935.
80. P. Johnson and F. K. Chang, Characterization of matrix crackinduced laminate
failure part II: analysis and verications. J Compos Mater, 35:22 (2001), 203774.
81. A. Parvizi, K. W. Garrett, and J. E. Bailey, Constrained cracking in glass ber
reinforced epoxy crossply laminates. J Mater Sci, 13:1 (1978), 195201.
82. H. T. Hahn and S. W. Tsai, Behavior of composite laminates after initial failures.
J Compos Mater, 8 (1974), 288305.
83. H. Fukunaga, T. W. Chou, P. W. M. Peters, and K. Schulte, Probabilistic failure
strength analysis of graphite epoxy crossply laminates. J Compos Mater, 18:4 (1984),
33956.
84. H. Fukunaga, T. W. Chou, K. Schulte, and P. W. M. Peters, Probabilistic initial failure
strength of hybrid and nonhybrid laminates. J Mater Sci, 19:11 (1984), 354653.
85. P. W. M. Peters, The strength distribution of 90deg plies in 0/90/0 graphiteepoxy
laminates. J Compos Mater, 18 (1984), 54556.
86. P. W. M. Peters, The ber/matrix bond strength of CFRP deduced from the strength
transverse to the bers. J Adhes, 53 (1995), 79101.
87. P. W. Manders, T. W. Chou, F. R. Jones, and J. W. Rock, Statistical analysis of multiple
fracture in [0/90/0] glass ber/epoxy resin laminates. J Mater Sci, 19 (1983), 287689.
88. F. W. Crossman, W. J. Warren, A. S. D. Wang, and J. G. E. Law, Initiation and
growth of transverse cracks and edge delamination in composite laminates: part 2.
Experimental correlation. J Compos Mater Suppl, 14 (1980), 89108.
89. P. W. M. Peters, Strength distribution of 90 degree plies in 0/90/0 graphiteepoxy
laminates. J Compos Mater, 18:6 (1984), 54556.
90. N. Takeda and S. Ogihara, Insitu observation and probabilistic prediction of micros
copic failure processes in CFRP crossply laminates. Compos Sci Technol, 52:2 (1994),
18395.
91. J. Aveston and A. Kelly, Theory of multiple fracture of brous composites. J Mater Sci,
8:3 (1973), 35262.
92. P. W. Manders, T. W. Chou, F. R. Jones, and J. W. Rock, Statistical analysis of
multiple fracture in 0
/90
/0
c
e
max
(%)
e
m
Figure 7.12. Fatiguelife diagram of a unidirectional carbon ber/epoxy composite with
a distinct region I (data courtesy of P. T. Curtis, RAE, UK).
0
0.0
0.5
1.0
1.5
2.0
1 2 4 3 5 6 7
log N
f
e
max
(%)
e
c
e
m
Figure 7.13. Fatiguelife diagramof a unidirectional Kevlar ber/epoxy composite (data from[6]).
246 Damage mechanisms and fatiguelife diagrams
polymer is an amorphous polyamide thermoplastic [6]. The generality of the
fatiguelife diagrams is that they lend themselves to description and interpretation
of any kind of fatiguelife data for composites. A comparison of fatigue perform
ances between different material candidates can then be made. Commercial mater
ial systems of unknown constituent composition can thus be compared, and the
best contender for an intended application can be identied.
7.5.1.1 Experimental studies of mechanisms
The fatiguelife diagrams developed in [1] and described above for PMCs were
based on deductive reasoning in the absence of any systematic investigation
of such mechanisms. Since then an investigation was conducted by Gamstedt
and Talreja [7], which, through direct observations of onaxis tensiontension
fatigue of unidirectional carbon/epoxy, gathered evidence to support the diagrams.
For instance, the presence of berbridged cracking postulated in region II was
0 1 2 3 4 5 6 7
log N
f
0.0
0.5
1.0
2.0
1.5
e
max
(%)
Figure 7.14. Fatiguelife diagram of a unidirectional Kevlar ber/J2 polymer composite
(data from [6]).
Figure 7.15. A crack on the surface of a carbon/epoxy unidirectional composite subjected to
tensiontension fatigue. The crack tips are squeezed by the bridging bers, as illustrated
schematically by the bottom gure. Reprinted, with kind permission, from Springer Science
+Business Media: J Mater Sci, Fatigue damage mechanism in unidirectional carbon
brereinforced plastics, Vol. 34, 1999, pp. 253546, E.K. Gamstedt and R. Talerja.
247 7.5 Unidirectional composites loaded parallel to the bers
found as exemplied by an image of a surface replica shown in Figure 7.15.
The crack opening prole shows squeezing of the crack tips by bers.
A mechanism for arresting matrix crack growth in region III was also dis
covered in [7]. Its evidence is displayed in Figure 7.16. As seen there, a matrix
crack growing transverse to the applied load is arrested by extensive ber/matrix
debonding. Without the energy dissipation in the debonding process the crack will
have a greater driving force for extension.
In [8] the fatigue propagation of berbridged cracks in unidirectional PMCs was
analyzed. As expected, in comparison to an unreinforced matrix crack, the fatigue
growth rate of berbridged cracks was found to be lower, and even showing
decelerating growth. In another study, Gamstedt [9] examined the role of ber/
matrix interfaces prone to debonding on the composite fatigue under onaxis
tensiontension loading. If the interfaces resist debonding, which may be viewed
as the baseline (normal) case, the fatigue mechanisms described above in developing
the fatiguelife diagram would apply. As debonding increases, the ber strength
variability becomes increasingly important in determining the damage progression,
and in signicant debonding cases, the berbridged crack growth gives way to
debondingassisted ber breakage as the progressive mechanism. The slope and
extent of region II of the fatiguelife diagram is then determined by the debonding
assisted ber breakage. The at region I, which is a consequence of the unconnected
ber breakage, then diminishes, and can eventually become part of region II.
Figure 7.17 illustrates the cases of little or no debonding vs. extensive debonding.
The former case is typical of relatively strong interfaces in carbon/epoxy composites.
When ber breaks occur, initially at randomly located weak points, they remain
isolated and unconnected due to the lack of debonding. As discussed above in
Fiber/matrix debonding
Matrix crack
Figure 7.16. A surface replica image showing a matrix crack arrested by ber/matrix
debonding. Reprinted, with kind permission, from Springer Science+Business Media:
J Mater Sci, Fatigue damage mechanism in unidirectional carbon brereinforced plastics,
Vol. 34, 1999, pp. 253546, E.K. Gamstedt and R. Talerja.
248 Damage mechanisms and fatiguelife diagrams
describing region I, any of the initial breaks are likely to form a criticalsize crack that
could grow unstably, leading to failure. Thus there is no progressive mechanism and
composite failure results essentially at any number of cycles, resulting in a horizontal
scatter band. In the case of interfaces that are prone to debond, the inuence of an
initial ber break extends to other bers, depending on the rate at which the debond
crack elongates, and a progressive ber breakage results. The rate of this damage
progression depends on the resistance to the debond crack growth given by the
interface region and the likelihood of the debond cracks connecting with other cracks.
The overall path to nal failure may be described in terms of an average progression
rate. Modeling of this progression is far from an easy task. Attempts at this were
made by Gamstedt [9], who conducted numerical simulation of the ber breakage
process and claried the roles of interfacial debonding and ber strength variability.
Figure 7.18 compares the fatiguelife diagrams of unidirectional carbon/epoxy and
carbon/PEEK (polyetheretherketone) composites loaded in tensiontension loading.
The former material typies the strong ber/matrix bonding case (left in Figure 7.17)
while the latter is prone to debonding. The PEEK matrix is a semicrystalline thermo
plastic that develops a socalled transcrystalline structure extending from the ber
surface. This structure provides weak planes parallel to bers for debonding to occur.
Althoughotherwise a toughmaterial, PEEKbecomes a source of brittle cracking of the
interfaces. The extensive debonding in carbon/PEEK removes region I of the fatigue
life diagram by inducing progressive ber breakage, as explained above.
(a)
(b)
Figure 7.17. Two cases of the role of ber/matrix debonding in fatigue damage progression
under onaxis tensiontension loading of unidirectional berreinforced PMCs. Case 1 (a)
illustrates unconnected ber breaks when little or no debonding occurs while case 2 (b) is
for extensive debonding that connects ber breaks and causes the ber breakage to be
progressive. Reprinted, with kind permission, from Springer Science+Business Media:
J Mater Sci, Fatigue damage mechanism in unidirectional carbon brereinforced plastics,
Vol. 34, 1999, pp. 253546, E.K. Gamstedt and R. Talerja.
249 7.5 Unidirectional composites loaded parallel to the bers
7.5.2 Metal matrix composites (MMCs)
The fatiguelife diagram discussed above should be viewed as a baseline diagram
for polymer matrix composites, as the mechanisms considered for its construction
were motivated by observations and conjectures related to this material system.
We shall now discuss what changes in the diagram are plausible when the matrix
material is a metal. Note that the role of bers is in modifying the fatigue
mechanisms taking place in the matrix, which is the material conducting irrevers
ible deformation (plasticity). The three regions of the fatiguelife diagram for
MMCs treated by Talreja [10] will be discussed next.
Region I: As discussed earlier, this region is manifested by ber failures.
For polymer matrix composites we argued that the ber failures occurred in a
manner that did not have signicant progressiveness (accumulation of ber
failures in a localized zone), leading to the nal (composite) failure from
random sites at random number of cycles. An important factor in this deduction
was insufcient irreversible deformation in the matrix to allow cycledependent
stress enhancement and failure of bers. This scenario will change when a
relatively brittle polymer matrix is replaced by a more ductile metal matrix.
The cyclic plastic deformation of a metal matrix around a broken ber will
redistribute stresses in the neighboring bers, allowing an accumulative ber
failure process to occur. This will introduce a localized progressive degradation,
which on reaching a critical level will cause composite failure. Thus, the hori
zontal scatter band of region I, characteristic of polymer matrix composites, will
now have a downward slope.
Region II: The mechanism of berbridged matrix cracking is also expected to
be the primary progressive mechanism for metal matrix composites. The
0
0.6
0.9
1.2
1.5
M
a
x
.
s
t
r
a
i
n
(
%
)
1 2 3 4 5 6 7
Log cycles
Carbon/PEEK
Carbon/epoxy
Figure 7.18. Comparison of the fatiguelife diagram for unidirectional carbon/epoxy and
carbon/PEEK composites loaded in tensiontension cycles along bers. Note the absence
of region I in carbon/PEEK.
250 Damage mechanisms and fatiguelife diagrams
difference with respect to polymer matrix composites lies in the role of ber/
matrix debonding. The ber/matrix bond in metal matrix composites is generally
stronger, leading to shorter extent of the interface failure. Furthermore,
increased ductility of the matrix is expected to provide more crack opening
displacement via crack tip blunting, causing increased failure of the bridging
bers. The matrix ductility effect will thus be as conjectured above in describing
the trend in region II of the fatiguelife diagram as illustrated in Figure 7.5.
Region III: This region is expected to be the same as in polymer matrix
composites. The composite fatigue limit is also expected to be related to the
matrix fatigue limit.
Let us now examine some test data for metal matrix composites. Figure 7.19
plots fatiguelife at room temperature for a SCS6/Ti153 unidirectional composite
under cyclic tension at two Rratios. Static failure strain is plotted on the vertical
axis at the rst cycle. As discussed above, region I displays some progressiveness by
deviating from a horizontal scatter band. Region II shows greater degradation
rates and tends to the fatigue limit gradually. Note that the matrix fatigue curve,
plotted as a broken line, tends to approximately the same fatigue limit as the
composite. The data for the same composite tested at a high temperature are
shown in Figure 7.20. The general features of the fatiguelife diagram at room
temperature are retained with different quantitative characteristics. Comparisons
Figure 7.19. Fatiguelife diagram of SCS6/Ti153 at room temperature.
Figure 7.20. Fatiguelife diagram of SCS6/Ti153 at high temperature (540550
C).
251 7.5 Unidirectional composites loaded parallel to the bers
of the two regions are displayed in Figures 7.21 and 7.22. Figure 7.21 shows the
data in region I at room and high temperatures on loglog scales. The bestt lines
for the data show that higher fatigue degradation (more progressiveness in the
ber failure mechanism) occurs at high temperature than at room temperature.
Similar region II comparison at the two temperatures is shown in Figure 7.22.
A leftward shift of region II, indicative of more fatigue degradation, is seen with
an increase in temperature. This is likely due to higher matrix ductility (lower yield
stress) at the high temperature, in accordance with the trends in the fatiguelife
diagram described above.
7.5.3 Ceramic matrix composites (CMCs)
In comparison with polymers and metals, ceramics have insignicant irreversi
bility of deformation. This suggests that there should be little incentive for fatigue
mechanisms to occur in ceramics. However, when ceramics are reinforced by
Figure 7.21. Region I for SCS6/Ti153 at high (540550
C).
Figure 7.22. Region II for SCS6/Ti153 at high (540550
C).
252 Damage mechanisms and fatiguelife diagrams
bers with which they bond weakly, a dissipative mechanism becomes available
from reversed (i.e., backandforth, not reversible) frictional sliding at interfaces.
This mechanism is believed to play a major role in causing fatigue of CMCs. In the
following a discussion of the fatiguelife diagram for unidirectional CMCs under
cyclic axial tension is presented.
Before getting into a discussion of fatigue of CMCs it would be useful to briey
reviewthe damage and accompanying stressstrain response observed under mono
tonic axial tension. Several works in the literature have reported such data, and we
shall mainly draw upon Srensen and Talreja [11] to illustrate some characteristics
of interest. Figure 7.23 shows three surface replicas of a SiC/CAS (calcium alu
minosilicate) glass specimen taken at three applied axial strains at room tempera
ture. For reference, the failure strain is approximately 1.0%. Note that at 0.15%
strain a few irregularly spaced cracks transverse to bers are seen. At this low strain
level the cracks are partially developed, i.e., they do not extend across the entire
width of the specimen. As the applied strain increases, the cracks span the width
fully and acquire more regular spacing. At some point the crack spacing approaches
a minimum value, usually described as saturation, as seen in the replica at 0.8%
strain. The stressstrain response in the axial as well as transverse directions is
shown in Figure 7.24, alongside a plot of the AE(acoustic emission) events recorded
during testing. The double stressstrain curves are due to strain gages used on both
sides of the specimen to monitor axiality of loading. Figure 7.25 summarizes the
different stages of damage progression, indicating the ranges in which they operate,
based on microscopy observations and AE event counts.
When considering fatigue, the rst question to ask is: what happens in the
material in the second and subsequent load cycles that is different from the
Fiberbridged
matrix crack
Fibers
500 m
e =0.15% e =0.5% e =0.8%
Figure 7.23. Micrographs of surface replicas of a unidirectional CMC with increasing strain:
015%, left; e = 05%, middle; e = 08% right. Source: [11].
253 7.5 Unidirectional composites loaded parallel to the bers
damage caused, if any, in the rst load cycle? The answer to this question lies in
knowing the source of irreversibility in the material. For instance, if a material
is cracked (singly or multiply) in the rst load cycle, it would not be cracked
further in the next and subsequent cycles unless an irreversible mechanism such as
plasticity or friction is present. As stated earlier, ceramic materials have insigni
cant irreversible (plastic) deformation to enable crack growth in cyclic loading.
The likely source of irreversibility is frictional sliding at the interface. Keeping this
in mind we shall examine below the existence and nature of the three regions of the
fatiguelife diagram for CMCs (see Figure 7.26).
0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
Strain (%)
0
100
200
300
400
S
t
r
e
s
s
(
M
P
a
)
500
600
Monotonic tension
IV Fiber
fracture
III Fiber
bridging
(Matrix fully
cracked)
IV
III
II
I
I No
damage
II Matrix crack
development
(initiation of
matrix crack)
D
a
m
a
g
e
d
e
v
e
l
o
p
m
e
n
t
SiC
f
/CASII
RT, 100 MPa/s
s
pl
=285 MPa
s
mc
=120 MPa
Figure 7.25. Schematic drawing of damage mechanisms at different stages in monotonic
tensile loading of a unidirectional CMC loaded in tension parallel to the bers.
2.0 1.5 1.0 0.0
0
100
200
300
400
500
600
0.0 0.25 0.50 0.75 1.00
AE events (10
6
) e (%)
e
L
e
T
s
(
M
P
a
)
Figure 7.24. Measured acoustic emission events and stressstrain response. Source: [11].
254 Damage mechanisms and fatiguelife diagrams
Region I: When a rst load is applied up to a maximum value within the scatter
band of the composite failure strain, the matrix cracks develop beyond their
saturation stage (see Figure 7.25). This means that a set of fully grown ber
bridged cracks exist at the end of the rst load cycle. It is likely that a signicant
portion of the bers will be broken due to the high load applied. When the
second and subsequent cycles are applied, the ber stress will increase due to the
increased debond length caused by the irreversible frictional sliding, supplying
the increased shear stress at the interfaces. Also, the cyclic grinding of the ber
surfaces by the debris generated by the interfacial wear will damage bers and
degrade their strength. All this will fail the bers and when any of the cross
sections going through the crack planes has insufcient number of intact bers
to bear the applied load, composite failure will occur. Now the critical question
is whether this failure mechanism is progressive or nonprogressive. If failure
were to come from only one cross section, i.e., if only one crack existed, or if one
crack was somehow the favored one, then the accumulative ber breakage will
result in a progressive failure. However, that is not characteristic of the damage
here. We have multiple cracks, and failure could potentially result from any one
of those cracks. Although each berbridged crack has progressive (accumula
tive) ber failure, the rate of progression would likely be discontinuous and
differ from that of other cracks. Thus the nal failure would not necessarily
result from the weakest of the cross sections produced by the rst load. We
postulate, therefore, that the number of cycles for composite failure in this case
cannot be derived from a single rate equation, and consequently, the failure
process is nonprogressive.
Region II: Consider now the case of loading where the rst cycle maximum load
is in a range such that the matrix cracks are either partially developed or are
Figure 7.26. Mechanisms in the three different regions of the fatiguelife diagram for
unidirectional CMCs loaded in tension parallel to the bers.
255 7.5 Unidirectional composites loaded parallel to the bers
fully developed but the bridging bers are not broken extensively. The partial
cracks will grow in subsequent cycles by debonding and/or breaking bers at the
crack tips (or fronts). The cyclic frictional sliding at the interfaces will debond
the bridging bers, causing crack surface separation to increase as the debond
length increases and as more bridging bers break. With cyclic stressing, there
fore, partial cracks will grow to their full extent. As more debonding occurs,
stress will be transferred from bers to the matrix increasingly, resulting in
matrix cracking between two preexisting cracks. A state of crack saturation
will be approached eventually, unless composite failure results from failure of
all bers bridging any of the cracks. The damage development in this region of
the fatiguelife diagram has welldened progressiveness from partial cracking
to full cracking until crack saturation. The terminal point is when at least one
crack loses all its bridging bers. This point may occur before crack saturation
(toward the lower end of region II) or beyond crack saturation (toward the
higher end of region II). The main difference between this region and region I is
that in the latter saturation cracking with extensive ber breakage already
occurs in the rst load application.
Region III: We have dened this region to be a range of loading (maximum
strain in the rst cycle) in which damage may develop but does not lead to
a critical state (failure) in a preselected large number of cycles (typically 10
6
).
In a unidirectional composite, failure (separation) will occur when all the
bridging bers of at least one crack fail, or, alternatively, broken ber ends
are interlinked by interfacial debonds and matrix cracks to form two separating
surfaces. In either case, interfacial frictional sliding plays an important role
by increasing debonding and thereby stressing bers to failure. What has been
found, however, is that the wear of the debonded interfaces (breakage of
asperities resulting in smoothing) can make the frictional sliding less effective.
This could slow down the cyclic development of damage to the extent of not
reaching failure in the prespecied number of cycles. Srensen et al. [12] found
that the temperature rise of SiC/CAS specimens caused by frictional heating at
debonded interfaces increased sharply just before failure, while for specimens
that did not fail in 10
8
cycles the temperature rise stopped at some cycles and
then took a downward turn (Figure 7.27). These authors have challenged the
notion of a true fatigue limit in CMCs, arguing that damage continues to
develop even after 10
8
cycles, as long as interfacial debonding occurred in the
rst cycle. They contend that the condition for no damage development is no
initiation of berbridged crack, providing the true fatigue limit as the stress
(or strain) at which such crack forms.
The discussion concerning fatigue limit in CMCs is somewhat academic because
the energy dissipating mechanism of frictional sliding at debonded interfaces
depends on the loading frequency and specimen geometry. This is because the
time rate at which conduction of the heat generated by frictional sliding occurs
affects the rate per cycle of energy dissipation by frictional sliding.
256 Damage mechanisms and fatiguelife diagrams
The fatigue data on CMCs reported in the literature are inadequate for a good
illustration of the usefulness of the fatiguelife diagram. Since the testing was
done with a traditional SN curve in mind the data generated were too few in
region I to validate its existence. Also, because of the material cost, the data
generated were at too few load levels to have all three regions of the diagram
appear with clarity.
7.6 Unidirectional composites loaded inclined to the bers
We consider the simple case of a single ber orientation in a composite placed at
an angle to the applied cyclic tension load as shown in Figure 7.28. As illustrated
in the gure, cracks initiate along the bers, either at interfaces or in the matrix.
At rst the cracks may initiate from defects in the matrix or at interfaces and grow
with cycles through the thickness as well as along the bers. At some stage in the
evolution of this damage, a single crack may grow to the extent of attaining
unstable growth in the next application of load and separate the composite in
two pieces, as illustrated in the gure.
In comparison to the onaxis loading case, i.e., when the bers are aligned with
the loading direction, the damage mechanism of the offaxis loading case is
drastically different and dramatically simple. The single progressive mechanism
depicted in Figure 7.28 will produce a continuous curve (and the associated scatter
band) in the fatiguelife diagram starting at the rstcycle failure strain and ending
asymptotically in the fatigue limit. For each offaxis angle the curve will be different.
A schematic depiction of the fatiguelife diagram is shown in Figure 7.29, which
illustrates the fatiguelife dependence on the offaxis angle. The onaxis fatiguelife
diagram is shown for reference in the gure in broken lines.
1
MPa
0
10
20
30
40
10
2
10
4
10
6
10
8
Cycles N
T
e
m
p
e
r
a
t
u
r
e
r
i
s
e
(
K
)
X
X
220 MPa
212 MPa
Run
out
s
max
=260 MPa
s
min
=10
[0
16
] SiC
f
/ CAS II
200 Hz, RT
Figure 7.27. Temperature increase due to frictional heating during cyclic loading of
a unidirectional SIC/CAS II composite.
257 7.6 Unidirectional composites loaded inclined to the bers
As Figure 7.29 illustrates, region I (horizontal scatter band) of the fatiguelife
diagram does not exist when the applied cyclic load is inclined to the ber
direction. The single scatter band starts at the composite failure strain and
asymptotically ends at the fatigue limit. The failure strain as well as the fatigue
limit strain depend on the offaxis angle, y. From the SN data reported for glass/
epoxy in [13] the fatiguelife diagrams for a few angles are shown in Figure 7.30.
The curves drawn by visual t to the data show the trends depicted in Figure 7.29.
The fatigue limits extracted from the data are shown plotted in Figure 7.31 where
the fatigue limits deduced from [14] for angleply laminates of glass/epoxy
are also plotted. The signicant improvement in the angleply laminates over
unidirectional composites for the same offaxis angle is discussed in the next
section on fatigue of laminates.
e
max
e
c
e
c
(0<q<90)
(q=0)
logN
q
Figure 7.29. Fatiguelife diagram for unidirectional composites under offaxis loading.
The diagram in broken lines is for the onaxis loading case.
Figure 7.28. Cracking in a unidirectional composite under offaxis cyclic tension (left)
and failure from growth of a crack (right).
258 Damage mechanisms and fatiguelife diagrams
7.7 Fatigue of laminates
Composite structures are built by placing bers in different orientations to
effectively carry multiaxial loading. We will here consider how the mechanisms
of fatigue damage are affected by multidirectional ber placement in a
0
q(degrees)
0.004
0.002
0.006
30 60 90
e
f
l
Figure 7.31. The offaxis fatigue limit from data in Figure 7.30 (dashed line) is shown
for comparison with the fatigue limit of angleply laminates of glass/epoxy taken from data
reported in [14], plotted against the offaxis ply angle.
0 2 4 6
lg N
e
db
e
c
e
c
e
c
e
m
e
m
a
x
e
c
0.008
0.006
0.004
0.002
(q=60)
(q=30)
(q=10)
q/deg
(q=5)
5
10
30
60
Figure 7.30. Fatiguelife data for glass/epoxy unidirectional composites from [13] plotted
as fatiguelife diagrams for different offaxis angles of tensiontension loading. The fatigue
limit of onaxis loading is indicated at 0.6% strain for reference. Reprinted, with kind
permission, from Z. Hashin and A. Rotem, J Compos Mater, Vol. 7, pp. 44864, copyright
# 1973 by Sage Publications.
259 7.7 Fatigue of laminates
laminate subjected to a cyclic tensile load and how that translates into changing
the basic fatiguelife diagram described above.
7.7.1 Angleply laminates
These laminates have two ber orientations placed symmetrically about a princi
pal direction, e.g., [y
n
]
s
illustrated in Figure 7.32. When loaded in the axial,
i.e., 0
plies.
Since multiple cracking is a basic feature of damage in composite materials, this
conguration and loading combination has been, and continues to be, a testbed
for modeling studies of multiple cracking and its consequences on the material
response.
The observations of multiple cracking in crossply laminates under monotonic
ally increasing load do not, however, reveal the key feature underlying progres
siveness of damage under cyclic loads. Since most studies typically observed the
cracking process on the free edges of a at specimen, the details of damage
accumulation were missed. The rst study to examine the interior of a laminate
with painstaking patience and using Xray radiography, combined with stereo
radiography, was by Jamison et al. [15]. The key Xray picture, along with a
schematic to depict interior details, is shown in Figure 7.33.
Other than the transverse cracks, the details seen in Figure 7.33 are not found to
develop sufciently under monotonic loading. However, these details hold the key
to the further progression of transverse cracking under cyclic loading. As we shall
261 7.7 Fatigue of laminates
discuss below in the section on modeling, the irreversible changes needed from one
load cycle to the next can be lumped into frictional sliding between the surfaces of
the propagating delamination.
In constructing the fatiguelife diagram of a crossply laminate under cyclic
axial tension, one needs to ask certain basic questions. First, are conditions
available for region I to be present? As discussed above in explaining the reasoning
behind this region, a statistical nonprogressive ber breakage mechanism must be
available for this region to exist. The presence of 0
, 45
, 45
, and
90
plies
provides region I, which lies as a scatterband about the ber failure strain, as in
the unidirectional onaxis loading case as well as in the crossply laminate case.
The fatigue limit for any laminate is determined by the rst cracking mechanism.
Therefore, if the 90
ply along which the cyclic tension load is applied, then region I will exist,
providing a at scatterband around the failure strain of the composite (which
equals the failure strain of bers). The next thing to look for is the ply orientation
in the laminate that makes the largest acute angle with the loading axis. In the two
examples above, that angle was 90
1
)
85%
66%
53%
35%
28%
Figure 7.37. Density of transverse cracks in a [0/90
2
]
s
laminate plotted against the log number
of tensiontension cycles in the axial direction. The data points are for different maximum
load values indicated as percentages of the ultimate tensile strength of the composite [20].
Reprinted, with kind permission, from Springer Science+Business Media: Appl Compos
Mater, Vol. 3, 1996, pp. 391406, X.X. Diao, L. Ye and Y.W. Mai.
Figure 7.38. Transverse cracks produced by the rst application of load (left); generation
of a new crack midway between the preexisting cracks on repeated application of the load
(middle); and completely grown crack at certain number of load applications (right).
267 7.8 Fatiguelife prediction
calculated by variational analysis [21, 22]. These stresses result from interaction
between cracks. As seen in Figure 7.39, the axial normal stress attains a maximum
midway between cracks and its value reduces from its precrack (constant) value.
Thus, in the absence of aws, a new crack can form midway between cracks if the
maximum stress there exceeds a critical value. However, since this stress is lower
than the critical value at which the previous cracks formed, new cracks can only
form if the applied load is increased. In a cyclic load of constant amplitude, new
cracks are therefore not possible, if the assumption of no aws still holds.
The conclusion has to be that in this scenario of transverse cracks in an elastic
composite, damage progression under cyclic loads cannot be achieved.
If the modeling of damage progression in crossply laminates is pursued without
entering aws in the model, then the only plausible place for irreversible mechan
isms is the interface between the 90
9
0
2 1 2 3
4
x
s
xz
(interfacial)
(1)
s
zz
(interfacial)
(1)
s
xx
(1)
z
x
50
1
Figure 7.39. Axial distribution of stresses in the 90
plies between
the transverse cracks. A cubic variation of the shear stress along the delamination
length 2l was assumed and the calculated axial normal stress then showed an
increase in its maximum value with l. Figure 7.41 shows the axial stress maximum
value for tractionfree delamination and when a shear stress acts between the
delamination surfaces.
As seen in Figure 7.41, if delamination surfaces are assumed traction free,
then the right conditions would not be present for new cracks to form between
cracks formed in the rst load cycle. The incentive for crack formation would
indeed decline as the delamination crack propagates under cyclic loading. On
2s
2l
Figure 7.40. Transverse cracks of spacing 2s in a crossply laminate with delamination
emanating from crack fronts and extending a distance l on either side of the transverse
crack. The delamination growth is assumed to be caused by the cyclic axial tension applied
to the laminate.
0
0
5
10
15
20
25
30
35
M
a
x
.
s
t
r
e
s
s
i
n
9
0
d
e
g
r
e
e
p
l
y
(
M
P
a
)
0.2 0.4 0.6 0.8 1
l / s
s l l
s
s l l
s
0
0
50
100
150
200
250
300
M
a
x
.
s
t
r
e
s
s
i
n
9
0
d
e
g
r
e
e
p
l
y
(
M
P
a
)
0.2 0.4 0.6 0.8 1
l / s
Figure 7.41. The variation of the maximum axial normal stress between two transverse cracks
is shown when delamination at crack fronts exists. The stress reduces with increase in the
delamination halflength l when the delamination surfaces are traction free (left) and it
increases when a shear stress acts between the delamination surfaces (right).
269 7.8 Fatiguelife prediction
the contrary, if delamination surfaces are engaged by asperities and/or by
compressive normal stress on the surfaces, then frictional sliding between the
surfaces will result, giving rise to a shear stress. This stress alters the redistri
bution of stresses between the axial and transverse plies, resulting in the
increase of the maximum axial normal stress as indicated in Figure 7.41. The
number of cycles beyond the rst cycle that will elevate this stress to a critical
value for crack formation is then the cycles needed to increase the delamination
length by fatigue growth.
Based on the assumption of frictional sliding of the delamination surfaces,
Akshantala and Talreja [24] developed a procedure by which the transverse crack
density under cyclic loading could be determined. Figure 7.42 shows examples of
the predicted crack density variation with the number of cycles. Note the crack
density variation displays saturation to different levels depending on the maximum
stress in accordance with the experimental data in Figure 7.37, except the initial
exponential rise in the crack density. Since the stress analysis model is for inter
acting transverse cracks the initial exponential increase in noninteracting cracks
cannot be predicted by the model.
Having predicted the crack density increase with load cycles, Akshantala
and Talreja [24] proceeded to use these data to predict fatiguelife. The
assumption was made that in the progressive damage represented by region
II of the fatiguelife diagram for crossply laminates (see Figure 7.34), a certain
maximum crack density is attained at failure of the laminate. This crack
density denoted
f
lies between the maximum achievable crack density
c
under
static loading and the minimum crack density
fpf
at the initiation of multiple
cracking, commonly called the rst ply failure (a misnomer since the ply cracks
0.0
0.0
0.5
1.0
1.5
2.0
0.2 0.4 0.6 0.8
Number of cycles (10
6
)
C
r
a
c
k
d
e
n
s
i
t
y
(
/
m
m
)
350 MPa
400 MPa
500 MPa
Figure 7.42. Transverse crack density in a crossply laminate predicted by Akshantala and
Talreja [24] at different cyclic load levels. Note the tendency for saturation at different levels
depending on the load level. Reprinted, with kind permission, from Mater Sci Eng A,
Vol. 285, A micromechanicsbased model for predicting fatiguelife of composite laminates,
pp. 30313, copyright Elsevier (2000).
270 Damage mechanisms and fatiguelife diagrams
but does not fail). The variation of
f
with failure load cycles was assumed to
be as depicted in Figure 7.43. The equation describing this variation,
f
= A
log N
f
+ B, has empirical constants A and B, which are determined by using
the minimum and maximum values of crack densities and their corresponding
number of cycles. The extreme values of the fatigue cycles are the beginning
and end of the progressive damage, i.e., region II. In Figure 7.43, these values
are shown as 10
2
and 10
6
cycles for illustration. As suggested at the end of
Section 7.7, a good approximation of the beginning of region II is 10
2
cycles
for glass/epoxy and 10
3
for carbon/epoxy composites. Note that the assumed
process of fatigue failure is not that it comes from the transverse cracks of a
certain density, but that the crack density increases from its rstload value to
that crack density. The failure of the laminate must involve delamination
growth and ber failures. The transverse crack density is simply assumed to
scale with fatiguelife in the assumed way.
Figure 7.44 depicts the procedure for fatiguelife prediction using the calcu
lated crack density from the stress analysis model [23] and the assumed
relationship of the crack density to failure cycles as displayed in Figure 7.43.
At a given cyclic load, the fatiguelife is given by the minimum number of
cycles until the crack density equals the limiting value. Thus, the graphical
method for determining this value is to nd the point of intersection of the
crack density increase curve and the straight line describing the limiting
crack density (see Figure 7.44). The number of cycles corresponding to the
intersection point is then the fatiguelife at that load level. The fatiguelife
prediction thus obtained is compared with the test data for a crossply lamin
ate in Figure 7.45.
Cycles to failure (log N
f
)
10
2
10
6
C
r
a
c
k
d
e
n
s
i
t
y
(
h
f
)
h
c
h
fpf
h
f
=A logN
f
+B
A
1
e
e
c
e
fpf
Figure 7.43. The assumed variation of the transverse crack density at failure in fatigue plotted
against the fatiguelife of a crossply laminate. The upper limit to this crack density is the
maximum saturation crack density under static load and the lower limit is the minimum
crack density at initiation of multiple cracking. The composite strains corresponding to the
two limits are also indicated. Reprinted, with kind permission, from Mater Sci Eng A,
Vol. 285, A micromechanicsbased model for predicting fatiguelife of composite laminates,
pp. 30313, copyright Elsevier (2000).
271 7.8 Fatiguelife prediction
10
0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
10
1
10
2
10
3
10
4
10
5
10
6
C
r
a
c
k
d
e
n
s
i
t
y
(
/
m
m
)
0.6%
0.65%
0.7%
0.76%
0.9%
1.0%
1.2%
h
fpf
h
c
h
f
= 0.425 N
f
+3.35
Number of cycles
Figure 7.44. The crack density data points represent the initial (rst load) values and their
increase with cyclic loading at different load levels (indicated by the corresponding rst
cycle peak strains). The crack density relationship to failure cycles, schematically described
in Figure 7.43, is also shown with the calculated endvalues of crack density. Fatiguelife
at a given load level is given by the cycles at intersection of the calculated crack density
with its failure value.
Cycles to failure
10
0
10
1
10
2
10
3
10
4
10
5
10
7
10
6
0.0
0.5
1.0
1.5
2.0
2.5
M
a
x
i
m
u
m
a
p
p
l
i
e
d
s
t
r
a
i
n
(
%
)
Fatigue limit
Experiment data
Model
Region II
Region 1
Figure 7.45. The predicted fatiguelife of a glass/epoxy [0/90]
s
laminate compared with test
data from [25].
272 Damage mechanisms and fatiguelife diagrams
7.8.2 General laminates
In Chapter 6 we treated the damage progression of a broad class of laminates
consisting of plies in multiple orientations subjected to axial tensile loads. From
the models described there one can calculate the load at initiation of multiple
cracking and crack densities resulting at the rst application of the maximum load
in the given cyclic load. For further progression of damage under repeated
application of that load one must ask: what is the mechanism responsible for
irreversibility that accumulates damage with each load cycle? For crossply lamin
ates we addressed this question and described the model that incorporates the
answer to the question. There is every reason to believe that in laminates with
multiple offaxis ply orientations the delamination occurring at the offaxis crack
fronts conducts frictional sliding and thereby provides the needed irreversibility
(energy dissipation). For crossply laminates the stress analysis conducted in [21]
for transverse cracks with delamination cannot simply be extended to multiple
crack orientations. In fact even without delamination the stress analysis of cross
ply laminates is not easily extended to the multiple crack orientation case.
7.9 Summary
This chapter has provided a systematic conceptual framework for interpretation
of the fatigue process in composite materials. No effort has been made to give
a comprehensive exposition of the vast literature on the subject. Instead, emphasis
has been placed on understanding of the physical mechanisms underlying fatigue
and incorporating this in a systematic way in the representation called fatigue
life diagrams. These diagrams provide a healthy departure from the empiricism
dominant in the fatigue literature. They also generate bases for material selection
and give useful guidelines for mechanismsbased modeling.
The eld of fatigue damage modeling and life prediction needs a great deal of
further work. The challenges of addressing the mechanisms of fatigue damage
accumulation have discouraged most research efforts that have instead taken the
path of empirical approach. In fact even in the century old eld of metal fatigue
empirical approaches are common. The socalled Paris Law of cyclic crack growth
is basically a curve t to the observed data. However, its simplicity has attracted
practical use but has impeded advances in fundamental understanding. It is
therefore not surprising that the composites community has so far also opted
for empiricism. It is worth noting that because of the large number of parameters
in composite materials (constituent properties, ply orientations, woven and other
complex architectures, etc.) the empirical path is highly inefcient. As demon
strated by the fatiguelife diagrams, focusing on the essential mechanisms allows
one to transcend the apparent complexity of the composite microstructure and
ber architecture.
273 7.9 Summary
In the next chapter we shall outline the areas in composite fatigue that must
receive attention by the composites community and must be supported by govern
ment and industry.
References
1. R. Talreja, Fatigue of composite materials: damage mechanisms and fatiguelife
diagrams, Proc R Soc London A, 378 (1981), 46175.
2. C. K. H. Dharan, Fatigue failure mechanisms in a unidirectionally reinforced compos
ite material. In Fatigue in Composite Materials, ASTM STP 569. (Philadelphia, PA:
ASTM, 1975), pp. 17188.
3. J. B. Sturgeon, Fatigue and creep testing of unidirectional carbon ber
reinforced plastics. In Proceedings of the 28th Annual Technical Conference of the
Society of the Plastics Industry. (Washington, DC: Reinforced Plastics Division, 1973),
pp. 1213.
4. J. Awerbuch and H. T. Hahn, Fatigue and prooftesting of unidirectional graphite/
epoxy composite. In Fatigue of Filamentary Composite Materials, ASTM STP 636, ed.
K. L. Reifsnider and K. L. Lauraitis. (Philadelphia, PA: ASTM, 1977), pp. 24866.
5. J. B. Sturgeon, Fatigue Testing of Carbon Fibre Reinforced Plastics, Technical Report,
Royal Aircraft Establishment, Farnborough (1975).
6. R. B. Croman, Tensile fatigue performance of thermoplastic resin composites
reinforced with ordered Kevlar aramid staple. In Proceedings of the Seventh Inter
national Conference on Composite Materials, vol. 2, ed., Y. Wu, Z. Gu, and R. Wu.
(Oxford: International Academic Publishers, 1989), pp. 5727.
7. E. K. Gamstedt and R. Talreja, Fatigue damage mechanisms in unidirectional carbon
brereinforced plastics. J Mater Sci, 34 (1999), 253546.
8. E. K. Gamstedt and S. O
EXPERIMENTAL/
COMPUTATIONAL
Determine COD and constraint parameter(s)
; = = +
(
( )
(
(
)
u
2
u
2
u
2
n
=
q
90
q
n q
n +q
n
q
b
Evaluate damage constants
using available data
for reference laminate
configuration [0/90
8
/0
1/2
]
s
90
8
( )
u
2 ( )
u
2
Figure 8.7. Flowchart showing the multiscale synergistic methodology for analyzing
damage behavior in a class of symmetric laminates with layup [0
m
/y
n
/0
m/2
]
s
containing ply
cracks in the +y and y layers.
288 Future directions
properties. The conventional material state description consists of constituent
properties and their relative proportions, e.g., volume fractions, and of the ber
architecture, e.g., ply thickness, orientation, and stacking sequence in a laminate,
or fabric type (e.g., 5 or 8harness satin), thickness, and layup in a woven fabric
composite, etc. In addition to this, the material state needs to be described by
certain defect descriptors. As we shall see, homogenized descriptions of the
constituents and defects do not sufce for the costeffectiveness assessment of a
manufacturing process.
The defect descriptors needed would depend on the manufacturing process.
Examples of such descriptors are distributions (or other statistics) of ber mis
alignment angles, of void size and location, of ber/matrix interfacial disbonds, of
delamination size and location, etc. The appropriate set of defect descriptors,
along with the conventional material state descriptors, makes up a complete
characterization of the manufactured composite material. Depending on the service
environment in which the composite structure is to function, i.e., the design
requirements imposed on the structure, the costeffectiveness assessment will
consider the necessary material properties and their relationships with the material
and defect descriptors. A cost/performance tradeoff exercise, and any iteration on
it, will result in an optimized costeffective product. In most applications, where
longterm performance is the critical design consideration, one needs to look at
the degradation of initial (endofmanufacturing) properties of interest under the
service environment. Thus the cost/performance tradeoff will consider the
residual properties. A common approach is to consider only the initial (i.e., pre
service) properties even for the longterm case, assuming implicitly that the
residual properties will relate to the defects (and the cost) the same way as the
initial properties do. This assumption is in fact questionable since the initial
properties may not show sensitivity to some of the material defects that may turn
out to be signicant in governing the longterm properties.
There is a variety of manufacturing processes used for composite structures,
e.g., autoclave molding, liquid compression molding, resin transfer molding,
Figure 8.8. Procedure for the costeffectiveness assessment of composite structures.
289 8.3 Costeffective manufacturing and defect damage mechanics
lament winding, chemical vapor deposition, etc. Each of these processes pro
duces defects in the manufactured part that are usually characteristic of that
process. The machining, joining, and assembly methods used for composite struc
tures produce defects that are generally different from those produced during
molding, winding, and vapor deposition. For instance, the defects in the inter
facial region between two parts will be different depending on whether the parts
were cocured or adhesively bonded. Signicant differences in the fatigue lives of
cocured vs. bonded joints have been reported [24].
In recent years, many methods have been developed to observe manufacturing
defects in composite materials and structures by nondestructive evaluation, based
mainly on ultrasound and radiography [25] and to some extent on thermal wave
imaging [26]. In the conventional approach these methods are utilized primarily for
quality control of the manufactured product. The premise of the quality control is
often that the presence of defects is undesirable. If defects of more than certain
threshold values are found, then the part is rejected and one strives to improve the
manufacturing process. This inevitably increases the cost and can result in the
composite part not being competitive with a metallic alternative. It is important to
realize that the presence of defects is not undesirable in all cases. In fact, if some defects
are allowed, the part can be produced at a lower cost. Figure 8.9 illustrates the
dependence of strength per dollar of manufacturing cost on the defect density. As
seen in the gure, the strength decreases gradually with defect density for low
densities, and drops rapidly at high densities, while the manufacturing cost increases
rapidly at low defect densities and falls off as more defects are allowed. Thus the
strength of the part achieved per dollar of manufacturing cost increases with defect
density, up toa point, beyond which the benet of allowing more defects decreases. It
must be remembered, however, that this situation is typical of the static strength. The
dependence of residual strength in longtermloading on the initial defect density may
show different behavior. This aspect has not been investigated sufciently.
Figure 8.9 also suggests that we should get away from the accept/reject
approach and advance to what may be called defect engineering. More specically,
we should engineer the components to have a certain amount of defects in order to
bring down the manufacturing cost while still having the performance require
ments satised. To achieve this higher level of engineering we need certain cap
abilities to be in place. Referring again to Figure 8.8, the connection between
manufacturing (box at top) and the material state achieved (box at left) requires a
capability to predict the defect structure along with the material composition
attained from the employed manufacturing process. Some attempts in this direc
tion have been made. As an example, see references [2731] for prediction of voids
in a liquid compression molding process.
Our efforts are focused on the connection between defect structures and the
mechanical properties, as well as their degradation in service. This type of activity
may be viewed as an extension of damage mechanics, which in its conventional
form deals with initiation and evolution of damage and the consequent changes in
mechanical properties. Thus our starting point in the extended damage mechanics
290 Future directions
is not a homogenized continuum, but a composite with bers and matrix as
constituents, and in addition, defects. The defects in our analyses are reallife
defects with their geometry and distribution as given by actual observations.
A broader strategy for durability assessment that includes analysis of defects
was coined as defect damage mechanics [32]. It is discussed next.
8.3.2 Defect damage mechanics
To illustrate the mechanics of damage incorporating defects, we take two
examples in the following.
8.3.2.1 Autoclave processing voids
The rst example deals with voids in composite laminates manufactured by
autoclave molding where we describe the observed characteristics of voids and
their incorporation in the modeling (for more details, see [33]).
Figure 8.10 (upper part) shows two crosssectional views, parallel and normal to
bers, of a unidirectional carbon/epoxy composite made by the autoclave process.
The voids seen are generally not spherical and are largely trapped between the
prepreg layers. In the lower part of the gure are two cross sections, a short
distance (1.2 mm) apart, showing the voids more closely.
Figure 8.11 summarizes numerous observations and measurements reported in
the literature [3537] concerning voids in composites made by autoclave molding.
The shape can be described as elongated cylinders of elliptical cross section capped
at the ends. The volume fraction of the voids is found to be less than 3% for a well
controlled autoclave process, which is safely below the 5% value taken for rejec
tion of parts in the aerospace industry.
The process by which voids form suggests that the voids must displace the bers
around them as they settle down in their equilibrium positions. Most continuum
models that homogenize the composite and embed voids do not account for this
fact. Such models essentially replace bers, not displace them. Reference [33]
accounted for the ber displacement as schematically illustrated by Figure 8.12.
Strength/dollar
Strength
Manufacturing
cost
Defect density
Figure 8.9. Illustration of the dependence on defect density of strength, manufacturing
cost, and strength per unit cost.
291 8.3 Costeffective manufacturing and defect damage mechanics
The predictions of the elastic moduli by the HuangTalreja procedure [33] are
compared with experimental data in Figure 8.13. Note the large change in the
throughthickness modulus (E
zz
) due to the voids.
8.3.2.2 Interlaminar voids
We now consider the effect of the presence of voids in an interlaminar plane
(layer) on the growth of a crack in that plane. As described above, most voids in
1.2 mm
2.5 mm
2.5 mm
Figure 8.10. Observed voids in unidirectional carbon/epoxy composites made by autoclave
process. Upper pictures [34]: cross section parallel to bers (left) and across bers (right).
Lower pictures [35]: two cross sections 1.2 mm apart showing voids. Upper pictures
reprinted, with kind permission, from K. J. Bowles and S. Frimpong, J Compos Mater,
Vol. 26, pp. 1487509, copyright # 1992 by Sage Publications. Lower pictures reprinted,
with kind permission, from Springer Science+Business Media: Review of Progress in
Quantitative Nondestructive Evaluation, A morphological study of porosity defects in
graphiteepoxy composite, Vol. 6B, 1986, pp. 117584, D. K. Hsu and K. M. Uhl.
Figure 8.11. Observed characteristics of voids in carbon/epoxy composites made by autoclave
molding.
292 Future directions
Figure 8.12. Modeling of voids accounting for displacement of bers. Reprinted, with
kind permission, from Compos Sci Technol, Vol. 65, H. Huang and R. Talreja, Effects
of void geometry on elastic properties of unidirectional berreinforced composites,
pp. 196481, copyright Elsevier (2005).
Figure 8.13. Elastic moduli predicted by [33] compared with experimental results. Data for
E
x
and E
y
are from [37] and for E
z
from [36]. Reprinted, with kind permission, from
Compos Sci Technol, Vol. 57, C. A. Wood and W. L. Bradley, Determination of the effect
of seawater on the interfacial strength of an interlayer Eglass/graphite/epoxy composite
by in situ observation of transverse cracking in an environmental SEM, pp. 103343,
copyright Elsevier (1997).
293 8.3 Costeffective manufacturing and defect damage mechanics
layered composites tend to place themselves between layers when manufactured
with a compression molding process. These voids can have different shapes, sizes,
and spacing in the interlaminar plane. This plane is also a plane that is prone to
cracking under service or may have preexisting aws due to insufcient adhesion.
A convenient way to make assessment of the effects of voids on interlaminar
fracture is to consider the geometry used for evaluation of interlaminar fracture
toughness. Ricotta et al. [38] conducted a systematic study of the voids inuence
on crack growth by considering this geometry. In the following some results from
that study are discussed to illustrate the effects.
Figure 8.14 shows a woven fabric composite where voids are found in the resin
rich regions between the ber bundles. These regions are likely to develop cracks
under service environment such as fatigue or fail under inplane compression,
leading to delamination. A representation of the effect of such voids on crack
growth is illustrated in the gure where a double cantilever beam (DCB) specimen
with voids ahead of the crack tip is shown. This geometry for mode1 crack
growth has been systematically analyzed in [38] considering various parameters
such as void shape (circular and elliptical), void size, and distance of void from of
the crack tip.
The approach described in detail in [38] consists essentially of rst validating
an analytical method by nite element analysis and then using the method to
conduct a parametric study of the effects of voids. The method uses a beamon
elasticfoundation analysis accounting for shear compliance and material
orthotropic symmetry. The voids are simulated as regions without support
from the elastic foundation. The strain energy release rate with voids present
(G
I,v
) is calculated for different cases. Figure 8.15 shows the effect of a single
circular void of different radius R placed at different distance D from the crack
tip. The G
I,v
normalized by the value without void (G
I
) shows increasing
enhancement as the void approaches the crack tip, and this enhancement
increases with increasing void radius. Figure 8.16 shows a similar effect for
elliptical voids.
The effects of multiple circular voids on the energy release rate are shown in
Figures 8.17 and 8.18. Figure 8.17 shows the effects of two and three voids of xed
radius and xed mutual spacing located at different distances from the crack tip.
In Figure 8.18 an interesting effect of void interaction is shown. As seen there, for
multiple circular voids where the nearest void is kept at a xed distance from the
crack tip, while the mutual void spacing is varied, the energy release rate does not
show a monotonic dependence on the void spacing. Instead, the void interaction
increases the energy release rate up to a certain void spacing, beyond which the
effect of having multiple voids decreases.
Finally, Figure 8.19 illustrates how the energy release rate increases with
crack propagation when a void exists ahead of the crack tip. The increase of the
energy release rate with crack length when no void exists is plotted for refer
ence. Thus the enhancement of the energy release rate is seen as the crack tip
approaches the void.
294 Future directions
Figure 8.14. Voids in resinrich areas between bundles in a woven fabric laminate and a DCB
specimen representation of crack growth in the presence of the voids.
1.3
1.2
1.1
G
I
,
v
/
G
I
1
0 5 10
R=0.2 mm
R=0.1 mm
R=0.08 mm
R=0.05 mm
Distance from crack tip D (mm)
15
D
20
Figure 8.15. Effects of a circular void of radius R and of distance D from crack tip on the
energy release rate.
1.9
1.8
1.7
1.6
1.5
1.4
1.3
1.2
1.1
1
0 2
a/b=1
a/b=1.5
a/b=2
a/b=4
b=0.1 mm
D
2b
2a
Distance from crack tip D (mm)
4 6
G
I
,
v
/
G
I
Figure 8.16. Effects of an elliptical void of different aspect ratio a/b, b = 0.1 mm, and of
distance D from crack tip on the energy release rate [38]. Reprinted, with kind permission,
from Compos Sci Technol, Vol. 68, M. Ricotta, M. Quaresimin and R. Talreja,
ModeI strain energy release rate in composite laminates in the presence of voids,
pp. 261623, copyright Elsevier (2008).
295 8.3 Costeffective manufacturing and defect damage mechanics
8.4 Final remarks
The analyses and methods presented in this book have been mostly directed at
composite materials having continuous ber reinforcement in individual layers
that are stacked to form laminates. These material systems with polymers as
matrix materials and with stiff bers such as carbon have spurred the development
of lightweight structures in the aeronautics industry. Today, new aircraft such as
Boeing 787 and Airbus 380 are products of those developments. It is arguable,
however, how much of the advancement in damage modeling presented in this
book is embedded in the design of these aircraft. While this situation is under
standable due to the stringent and costly airworthiness certication requirements,
it is hoped that eventually the output of research efforts in damage and failure will
transfer to designing safer and more costeffective structures.
1
0 1
3 voids R =0.1 mm
2 voids R =0.1 mm
1 void R =0.1 mm
2
Distance from crack tip D (mm)
3 4
1.1
1.2
1.3
G
I
,
v
/
G
I
Figure 8.17. Effect on the energy release rate of circular voids of radius R = 0.1 mm and of
2.0 mm mutual spacing placed ahead of the crack tip for varying distance D from the
crack tip.
1.11
1.10
1.09
1.08
0 2 4
3 voids R = 0.1 mm
2 voids R = 0.1 mm
6
Distance from first void c (mm)
8 10
G
I
,
v
/
G
I
Figure 8.18. Effect on the energy release rate of multiple voids of xed radius R = 0.1 mm of
varying mutual spacing c placed at a xed distance from the crack tip.
296 Future directions
The use of composites has over the years expanded beyond the aerospace
applications to other areas of structures. Carbon ber composites have experi
enced an explosive growth in recent decades with an annual growth rate ranging
from 10 to 15%. The emerging applications of composite materials in automotive
and wind energy sectors place different challenges on design of these materials
than what has been the case in the aerospace industry. Although the affordability
of aerospace vehicles, even in the defense industry, has been a consideration, cost
effectiveness is a prime factor in design of automotive and wind turbine structures.
The defect damage mechanics discussed above is bound to be an integral part of
the design process for these structures in the future. Incorporating this approach
in computational design methodologies will be a crucial next step.
For wind energy applications the key factor is longterm durability, other than
low cost. The design life of these structures is currently at 20 years (earlier it was
30 years!). For fatigue this translates to 10 million load cycles, or more. Most
composites fatigue testing has traditionally been done until 10
6
cycles, motivated
by metal fatigue where steels typically have a fatigue limit, which is revealed by the
SN curve attening out before this number of load cycles. For composite mater
ials the fatigue limit is not as easily determined. As discussed in Chapter 6,
considerations of damage mechanisms are necessary to deduce this property. This
is a signicant challenge for a highly complex composite construction in wind
turbine rotor blades. A much greater challenge is to determine the fatigue life at
a large number of cycles under multiaxial loading conditions typical for these
structures.
The eld of multiaxial fatigue in composites must be given the support it
deserves. A largescale activity that is comprehensive in its approach is needed.
It must include testing and evaluation at scales where damage initiates, to
scales of damage progression (crack multiplication), and failure criteria that
properly represent the mechanisms. The activity so far has been limited in
scope and mostly focused on empirical and semiempirical approaches. Most
540
520
500
480
G
I
(
J
/
m
2
)
460
440
420
0 1 2
Da
Circular void, R=0.1 mm
Without void
Crack propagation (mm)
3 4
Figure 8.19. Increase in the energy release rate as the crack tip approaches the void is
illustrated. The lower curve shows the energy release rate for comparison when no void
exists.
297 8.4 Final remarks
work continues to emulate metal fatigue despite fundamental differences in the
underlying mechanisms [39].
This book has not specically dealt with nanoscale reinforcements and compos
ite systems. Although many advances have recently occurred, it is not yet clear
what advantages this area holds for improving durability at low cost in load
bearing composite structures. The area of multifunctionality of composites, such
as conducting polymers and their composites, has on the other hand shown clear
promise for applications in structural health monitoring.
References
1. O. O. Ochoa and J. N. Reddy, Finite Element Analysis of Composite Laminates.
(Dordretchet, The Netherlands: Kluwer Academic Publishers, 1992).
2. M. J. Hinton and P. D. Soden, Predicting failure in composite laminates: the back
ground to the exercise. Compos Sci Technol, 58:7 (1998), 100110.
3. M. J. Hinton, A. S. Kaddour, and P. D. Soden, Evaluation of failure prediction in
composite laminates: background to part B of the exercise. Compos Sci Technol,
62:1213 (2002), 14818.
4. M. J. Hinton, A. S. Kaddour, and P. D. Soden, Evaluation of failure prediction
in composite laminates: background to part C of the exercise. Compos Sci Technol,
64:34 (2004), 3217.
5. Y. W. Kwon, D. H. Allen, and R. Talreja, eds., Multiscale Modeling and Simulation of
Composite Materials and Structures. (New York: Springer, 2008).
6. S. NematNasser and M. Hori, Micromechanics: Overall Properties of Heterogeneous
Materials, 2nd edn. (Amsterdam: North Holland, 1999).
7. R. Talreja, Multiscale modeling in damage mechanics of composite materials. J Mater
Sci, 41:20 (2006), 680012.
8. R. Talreja, On multiscale approaches to composites and heterogeneous solids with
damage. Philos Mag, 90:3132 (2010), 433348.
9. R. Pyrz, K. Anthony, and Z. Carl, Morphological characterization of microstructures.
In Comprehensive Composite Materials. (Oxford: Pergamon, 2000), pp. 46578.
10. S.