Vous êtes sur la page 1sur 13

Modeling delamination growth in laminated composites

F. Shen, K.H. Lee, T.E. Tay*


Department of Mechanical Engineering, National University of Singapore, 10 Kent Ridge Crescent, Singapore 119260, Singapore
Received 2 May 2000; received in revised form 17 November 2000; accepted 13 February 2001
Abstract
This paper deals with the computational modeling of delamination and the prediction of delamination growth in laminated
composites. In the analysis of post-buckled delaminations, an important parameter is the distribution of the local strain-energy
release rate along the delamination front. A study using virtual crack closure technique is made for three-dimensional nite-element
models of circular delaminations embedded in woven and non-woven composite laminates. The delamination is embedded at dif-
ferent depths along the thickness direction of the laminates. The issue of symmetry boundary conditions is discussed. It is found
that bre orientation of the plies in the delaminated part play an important role in the distribution of the local strain-energy release
rate. This implies that the popular use of quarter models in order to save computational eort is unjustied and will lead to erro-
neous results. Comparison is made with experimental results and growth of the delamination front with fatigue cycling is predicted.
A methodology for the prediction of delamination areas and directions using evolution criteria derived from test coupon data is
also described. It is found that evolution criteria based on components of the strain-energy release rate predict the rate of delami-
nation growth much better than evolution criteria based on the total strain energy release rate. # 2001 Elsevier Science Ltd. All
rights reserved.
Keywords: B. Fatigue; B. Modeling; C. Delamination; C. Finite-element analysis; Strain-energy release rate
1. Introduction
Although post-buckled delamination growth in com-
posite laminates has been widely studied, many studies
have concentrated on through-width [14] or axisym-
metric delaminations [5,6]. The results of such simplied
analyses are sometimes assumed to reect the general
behaviour and characteristics of delaminations in com-
posite structures. However, in general, delaminations in
laminated structures are usually three-dimensional (3D)
in nature [7,8] because they are embedded within the
laminates. Typically, delaminations may occur during
low-velocity impact, around fastener holes, or as man-
ufacturing defects. Near-surface multiple delaminations
can result from impact damage. As the post-buckling
response of such delaminations under compressive fati-
gue loads is complicated, it is unlikely that simplied 2D
analyses will be adequate for describing all the char-
acteristics of real structural delaminations. Therefore, in
order to understand the mechanism of delamination
growth more fully, there is a need for more accurate and
realistic models. Furthermore, these models must be
validated, together with appropriately formulated
growth and fracture criteria, by comparison with experi-
mental results. Towards this end, some nite-element
(FE) analyses of delaminations in composite laminates,
with varying degrees of complexities, have been per-
formed. Although some are 2D plane-strain analyses
[9,10], many may be classied as quasi-3D analyses,
employing plate or shell elements in order to reduce
computational eort [1114]. Afeware truly 3Danalyses
[7,8]. In these models, releasing selected node pairs
advances the delamination crack front locally, so that the
nal shape of the delamination crack front is often dif-
ferent from its initial (usually assumed circular) shape.
The virtual-crack-closure technique (VCCT) [15,16] is
commonly used to determine the local strain-energy
release rates (SERRs) and the locations where the node
pairs are to be released. An advantage of using plate or
shell elements is that the laminate architecture can be
easily incorporated into plate and shell theories. Mesh
renement and adaptation, if and when desired, is also
0266-3538/01/$ - see front matter # 2001 Elsevier Science Ltd. All rights reserved.
PI I : S0266- 3538( 01) 00023- 9
Composites Science and Technology 61 (2001) 12391251
www.elsevier.com/locate/compscitech
* Corresponding author. Tel.: +65-874-2887; fax: +65-779-1459.
E-mail address: mpetay@nus.edu.sg (T.E. Tay).
less tedious [9,12]. Nevertheless, a shortcoming of using
plate and shell elements is that generally it does not
permit the separation of the SERR into its three com-
ponents (i.e. the opening and two shearing modes) [13].
Only the total strain-energy release rate (SERR) along
the delamination front can be calculated. Separation of
the fracture energies for the various modes is not possi-
ble except for some special cases [9,14]. In order to
develop a delamination growth criterion, it is necessary
to determine the relative contributions of the various
fracture modes (i.e. the opening Mode I, shearing Mode
II and tearing Mode III), since it is known that most
composite materials exhibit substantially higher fracture
resistance in Mode II than in Mode I [17,18]. The third
mode, Mode III, is usually not very signicant in most
practical applications. Currently, to obtain the separate
fracture mode energies at the delamination front, a fully
3D FE analysis has to be performed, although it is more
computationally intensive.
The direction and rate of delamination growth are
inuenced by the type of composite material, lay-up
sequence, position of the initial delamination and
applied external load. All these factors have to be taken
into account in the model. However, in attempts to fur-
ther reduce computational eort, some researchers have
often used one-quarter models of circular embedded
delamination with assumed symmetry boundary condi-
tions. This is despite the fact that the material at each
ply level (which may be orthotropic) may not exhibit
property symmetry compatible with the boundary con-
ditions imposed on the model. It is implicitly assumed
(by invoking St Venants Principle) that stress distribu-
tions within the structure at distances far from the
boundaries are not signicantly aected by the spurious
symmetry assumptions. At rst glance, this argument may
appear reasonable, since the dimensions of the delamina-
tion and the composite structure are of several orders of
magnitude larger than the thickness of each individual
ply. However, the calculation of the SERR along the
delamination front involves stresses and displacements
close to the delamination front, which includes points near
the boundaries with assumed symmetry conditions. Fur-
thermore, it is known that if the postbuckled delaminated
portion consists of an unbalanced sublaminate, warpage
may occur, resulting in local crack closure. In such cases,
structural as well as material symmetry are clearly absent.
In this paper, the eect of the assumptions of boundary
symmetry conditions on the computed local distribution
of the SERR is studied with the aid of both one-quarter
and complete 3D models.
Finally, this paper also proposes a methodology (to
be used with the FE method) for predicting delamina-
tion growth in woven fabric composites, using fatigue
evolution criteria obtained from earlier experiments
[19,20]. Although other studies have been conducted to
simulate delamination growth, they are mostly qualita-
tive in nature. The numerical predictions of delamination
areas and shapes obtained in this paper are compared
with C-scan results. Reasonable agreement has been
found. It is shown that evolution criteria based on
components of the strain energy release rate predict the
rate of delamination growth much better than evolution
criteria based on the total strain energy release rate.
2. Computation of the local SERR
The problem of a single delamination between two
plies of dierent bre orientations within a laminated
composite can be considered as that of a crack embed-
ded between two dissimilar orthotropic materials. In such
interfacial crack problems, the near-tip stress singularity
exhibit an oscillatory nature [21]. It can be shown that
while the total SERR does not show oscillatory beha-
viour, the Mode I and Mode II components of the SERR
do [21,22]. It is generally recognized that the oscillatory
stress and displacement solutions are non-physical, as it
implies interpenetration of the crack surfaces. Some
researchers have, therefore, proposed alternative mea-
sures of SERR components that do not exhibit this
oscillatory behaviour [22,23]. Fortunately, the dom-
inance of this oscillatory behaviour of the stresses and
displacements is conned to a small region near the crack
tip (estimated to be about 0.25 of the ply thickness [24]).
Outside this region, the virtual crack closure technique
(VCCT) can be eectively applied to yield physically
reasonable values of Mode I and Mode II components
of the SERR [22,24]. The use of VCCT also avoids the
need to employ special crack tip elements, which entail
the shifting of mid-side nodes to the quarter point posi-
tion relative to the crack tip. These elements assume a
square-root stress singularity, but this does not represent
the true nature of the stress singularity for a delamination
front in composite structures. A brief description of the
VCCT and its extension to the analysis of 3D delamina-
tion follows.
The VCCT is an approximate method that is derived
from the more fundamental crack closure technique
(CCT). Suppose a crack extends by a small amount c;
in the CCT, it is assumed that the strain energy released
in the process is equal to the work required to close the
crack to its original length. When applied in a numerical
method, two analyses are needed to obtain the compo-
nents of SERR. The rst is performed for the case just
prior to crack growth, which yields the nodal forces at
the crack tip. In the second analysis, the relevant crack
tip nodes are released, providing the required corre-
sponding nodal displacements. The nodal forces
obtained in the rst step are the forces required to close
the crack. The work done during this process can then be
obtained by multiplying one-half of the nodal forces with
the corresponding displacements. The components of
1240 F. Shen et al. / Composites Science and Technology 61 (2001) 12391251
SERR can then be obtained by separating the total
work into its corresponding components.
The VCCT, however, is an approximate method that
dispenses with the need for two analysis steps (hence its
popularity). The nodal forces at the crack front and the
displacements behind the crack front are used to calcu-
late the SERR. This implies self-similarity in crack
growth, a fact generally not crucial for 2D delamination
problems (unless bre bridging or crack kinking out of
the initial delamination plane occurs), but not generally
true for 3D delamination growth. Fortunately, it has
been shown in a previous work [25] that the distribu-
tions of SERR calculated using CCT and VCCT are
similar. Since the VCCT enables considerable saving in
computational eort especially in non-linear analyses
such as the ones considered here, it will be used
throughout this work.
Using the VCCT for 2D crack problems (Fig. 1), the
SERR components G
I
and G
II
can be expressed as
G
I
=
1
2c
Y
5
v
1
v
2
( ) Y
6
v
3
v
4
( ) [ ] (1)
G
II
=
1
2c
X
5
u
1
u
2
( ) X
6
u
3
u
4
( ) [ ] (2)
The SERR (and its Modes I, II and III components
G
I
, G
II
and G
III
, respectively) for a 3-D crack front can
be similarly calculated in a single FE analysis by using
the nodal forces at the crack front and the displace-
ments behind the crack front (Fig. 1):
G
I
=
1
2A
n
Z
c
3
w
a
3
w
A
3

Z
d
2
w
b
2
w
B
2

1
2
Z
c
2
w
a
2
w
A
2

Z
c
4
w
a
4
w
A
4

o
(3)
G
II
=
1
2A
n
X
c
3
u
a
3
u
A
3

X
d
2
u
b
2
u
B
2

1
2
X
c
2
u
a
2
u
A
2

X
c
4
u
a
4
u
A
4

o
(4)
G
III
=
1
2A
n
Y
c
3
v
a
3
v
A
3

Y
d
2
v
b
2
v
B
2

1
2
Y
c
2
v
a
2
v
A
2

Y
c
4
v
a
4
v
A
4

o
(5)
In the above equations, X, Y and Z are nodal force
components, and u, v and w are nodal displacement
components in the x, y and z directions, respectively.
The subscripts denote the corresponding nodes in Fig. 1.
The delaminated area is given by A = cl.
3. Three-dimensional analysis of post-buckled
delamination
All FE analyses are carried out using an in-house
nonlinear code specially designed for solving delamina-
tion problems. Two types of models are used: one-
quarter models and full models (Fig. 2). The composite
laminate has a quasi-isotropic lay-up [(0/45/-45/90)
3
]
S
and a centrally located through hole of 5 mm. The cir-
cular delamination has a diameter of 20 mm. Since the
delaminated portion (referred to as the sublaminate) is
expected to buckle under compressive load and thus
experience larger deformations and stresses relative to
the substrate, greater mesh renement is used for the
sublaminate. Isoparametric twenty-node brick elements
are used throughout the models. Around the crack tip, a
locally symmetric square mesh S2, shown in Fig. 3, is
used to assure convergence of the SERR [25]. Naturally,
the question of the adequacy of the local near-tip mesh
is a signicant one and has been investigated in greater
Fig. 1. Schematic of crack front region in 2D and 3D cases.
Fig. 2. One-quarter and full nite element models.
Fig. 3. Local near-tip meshes considered, from Ref. [25].
F. Shen et al. / Composites Science and Technology 61 (2001) 12391251 1241
detail in a recent work by the authors in Ref. [25], where
three dierent near-tip meshes (Fig. 3) were used in 3D
nite element models of delamination in a woven fabric
composite. In the gure, S1 is a rather crude mesh, with
the length of the side of the near-tip element equal to the
thickness of one ply of the composite. S2 and S3 are
renements of the local mesh, with the length of the side
of the smallest near-tip element equal to half and a
quarter of the thickness of one ply, respectively. In Ref.
[25], it was shown that the Modes I and II components
of the SERR calculated using S2 and S3 near-tip
meshes did not show appreciable dierence in values.
Only the mesh S1 showed a signicantly lower predic-
tion of the Mode I component under certain conditions
(such as when the aspect ratio with the circumferential
length of the near-tip element was greater than 10).
Hence, the near-tip mesh S2 is deemed adequate and,
therefore, used for the nite element models reported in
the current paper. It is perhaps worthwhile to point out
that further renement of the near-tip meshes with ele-
ment sizes less than a quarter of the ply thickness does
not imply improved accuracy of the computed compo-
nents of the SERR. Indeed, an extremely and increas-
ingly ne mesh would result in non-convergence since
the region would have entered the zone of oscillatory
stress elds. The work in Ref. [25] showed that in the
near-tip region, if the smallest element size is between a
quarter and one ply thickness, the components of the
SERR are unique and well behaved. Although this
implies the existence of a length scale (which, admittedly
is still controversial and poses conceptual diculties), it
at least assures stable and single-valued components
over a reasonable range.
Symmetric boundary conditions are applied to the xz
and yz planes of the one-quarter models (Fig. 2). A
uniform prescribed displacement is imposed at one
edge. The sum of all the nodal reaction forces at the
opposite edge is equivalent to the applied external load.
The prescribed displacement is increased incrementally
until the applied external load reaches the peak com-
pressive fatigue load (30 kN). No symmetric boundary
conditions are necessary for the full model, of course,
but the same method of applying load through pre-
scribed displacement is used. In order to initiate the
buckling of the sublaminate, a small initial out-of-plane
displacement (equivalent to about 7% of a single ply
Table 1
Description of FE models and experimental specimens
One-quarter
FE models
Full FE
models
Experimental
specimens
Position of
delamination
Ply angles adjacent
to delamination.
Sublaminate
lay-up
Q1 F1 E1 Between layers 1 and 2 0/45 [0]
Q2 F2 E2 Between layers 2 and 3 45/45 [0/45]
Q3 F3 E3 Between layers 3 and 4 45/90 [0/45/45]
Q4 F4 E4 Between layers 4 and 5 90/0 [0/45/45/90]
Fig. 4. Applied load vs maximum transverse displacement.
1242 F. Shen et al. / Composites Science and Technology 61 (2001) 12391251
thickness) is prescribed near the center of the circular
sublaminate. This initial displacement does not sig-
nicantly aect the postbuckling behaviour of the sub-
laminate [25]. Four cases are analyzed for each set of
one-quarter models (Q series) and full models (F series),
depending on the location of the delamination in the
laminate (Table 1). In the rst case (Q1 and F1), the
delamination is located between the rst and second
plies, i.e. between the 0 and 45

plies. In the second case


(Q2 and F2), the delamination is located between the
second and third plies, i.e. between the 45 and 45

plies. Similarly, in the third case (Q3 and F3), the dela-
mination is located between the third (45

) and fourth
(90

) plies, and in the last case (Q4 and F4), the dela-
mination is located between the fourth (90

) and fth
(0

) plies. It is noted that in the fourth case, the sub-


laminate has a quasi-isotropic lay-up.
In order to compare the FE results with experiment,
quasi-isotropic [(0/45/-45/90)
3
]
S
laminates were fabri-
cated from unidirectional graphite-epoxy Fibredux 924
prepregs. Pre-programmed delaminations in the form of
20-mm diameter circular Teon inserts were positioned
within the laminates at four dierent locations during
fabrication, as shown in Table 1. The laminates were cut
Fig. 5. Variation of SERR components with applied strain.
Fig. 6. Distribution of G
I
along delamination front (1-ply delamination) for F1 and Q1.
F. Shen et al. / Composites Science and Technology 61 (2001) 12391251 1243
into specimens (E series) of length 250 mm and width 50
mm, and a hole of 5 mm diameter is drilled through
each specimen at the center of the Teon insert. The
specimens were bonded with glass bre-epoxy end tabs
and compression loaded in fatigue (R = 1, peak load
30 kN). An anti-buckling device was used during the test
to prevent global buckling of the laminates. The growth
of delamination was monitored by C-scan at regular
intervals of fatigue cycles. The material properties used
are: E
1
= 134 GPa, E
2
= E
3
= 10:2 GPa, G
12
= G
13
=
5:5 GPa, G
23
= 3:4 GPa,
12
=
13
= 0:3 and
23
= 0:49.
Fig. 4 shows a typical postbuckled load versus dis-
placement curve for the sublaminate. It clearly shows an
increase in the out-of-plane displacement when the cri-
tical load is reached. The stiness of the sublaminate in
the postbuckling stage has been greatly reduced. Similar
behaviour can be observed from experiments [13].
Although the sublaminate has buckled and is thus sub-
jected to nonlinear deformation, the substrate remains
linearly elastic. In the gure, d is the initial small out-of-
plane displacement, h is the sublaminate thickness and
P the applied compressive load. Fig. 5 shows a typical
relationship between the computed SERR and the
external applied load. A marked increase in both Modes
I and II components of the SERR is observed for loads
greater than the critical buckling load. Although both
Fig. 7. Distribution of G
II
along delamination front (1-ply delamination) for F1 and Q1.
Fig. 8. Distribution of G
I
along delamination front (2-ply delamination) for F2 and Q2.
1244 F. Shen et al. / Composites Science and Technology 61 (2001) 12391251
components increase rapidly and nonlinearly with load,
G
I
increases at a faster rate than G
II
.
Figs. 6 and 7 show the distribution of G
I
and G
II
respectively, for the full model F1 and the one-quarter
model Q1. In the gures, is the angle measured along
the delamination front from the direction of the applied
load. The results of both models agree very well with
each other, although the maximum value of G
II
from Q1
is slightly higher than that of F1. Here, the sublaminate
is a single layer of unidirectional graphite epoxy, i.e. [0],
and the boundaries of the one-quarter model coincide
with the material symmetry planes. It is therefore not
surprising that there should be agreement between the
results of the two models. During the analysis, it is
found that there is substantial crack closure over parts
of the delamination front, and therefore constraints
must be applied at each step to ensure that non-physical
interpenetration of delaminated surfaces is not allowed.
Unfortunately in this case, the experimental results for
specimen E1 could not be used for comparison because
there was extensive transverse cracking in the [0] sub-
laminate, which was not modeled in the FE analysis.
Figs. 8 and 9 show the distribution of SERR compo-
nents along the delamination front for the full model F2
and the one-quarter model Q2. In this case, the delami-
nated part is an unbalanced [0/45] sublaminate; hence
the symmetry boundary conditions imposed on the one-
quarter model are incorrect, since the boundaries of the
one-quarter model do not coincide with the material
symmetry planes. Clearly, the distributions of SERR
components from the two models are very dierent. The
maximum value of G
II
is only about half the maximum
value of G
I
. The positions of maximum G
I
and G
II
are
superimposed on the C-scan image of the delamination
growth for specimen E2 in Fig. 10. The direction of
delamination growth coincides well with the positions of
maximum G
I
and G
II
.
Figs. 11 and 12 show the distribution of G
I
and G
II
respectively, for the full model F3 and the one-quarter
model Q3. Since the sublaminate has the unsymmetrical
lay-up [0/45/-45], the results of F3 and Q3 are not
Fig. 9. Distribution for G
II
along delamination front (2-ply delamination) for F2 and Q2.
Fig. 10. C-scan image of delamination growth of specimen E2.
F. Shen et al. / Composites Science and Technology 61 (2001) 12391251 1245
expected to agree. While the results for G
I
are very dif-
ferent, the results for G
II
look surprisingly similar. This
is, however, almost certainly fortuitous and highlights
the need for caution when analyzing the results of FE
models. The C-scan image of the delamination growth
for the corresponding specimen E3 is shown in Fig. 13.
In this case, the direction of delamination growth
appears to agree more with the direction of maximum
G
I
rather than maximum G
II
.
Finally, the distribution of SERR components for F4
and Q4 are shown in Figs. 14 and 15, respectively. The
sublaminate is quasi-isotropic with the lay-up [0/45/-45/
90]. Interestingly, the results of the full model show a
similar distribution along the delamination front with
those of the one-quarter model. The magnitudes, how-
ever, dier considerably; the one-quarter model con-
sistently over-predicts the values of G
I
and G
II
. This case
is particularly interesting because if this problem were
to be analyzed using plates or shell elements, one might
reasonably suppose that the quasi-isotropy of the sub-
laminate justies the use of one-quarter models, since
the symmetry boundary conditions appear to be com-
patible with material symmetry planes. This analysis
shows that such an assumption would be erroneous,
since the sublaminate is still unsymmetrical (albeit
quasi-isotropic), and boundary eects are still sig-
nicant. The comparison with experimental results from
specimen E4 is shown in Fig. 16. The direction of
Fig. 11. Distribution of G
I
along delamination front (3-ply delamination) for F3 and Q3.
Fig. 12. Distribution of G
II
along delamination front (3-ply delamination) for F3 and Q3.
1246 F. Shen et al. / Composites Science and Technology 61 (2001) 12391251
growth is transverse to the load direction, which coin-
cides with the direction of maximum G
I
. In this case, G
II
does not appear to inuence the direction of growth.
4. Prediction of delamination growth rate
In this section, the prediction of delamination growth
rate and direction is investigated with the aid of FE
analysis of a woven fabric composite laminate with a
centrally located hole and circular delamination. The
local SERR components calculated from Eqs. (3)(5)
are used in a suitable fatigue propagation criterion.
Equations similar in form to the Paris Equation are
commonly used to characterize delamination propaga-
tion in composites. However, instead of stress intensity
factors, the SERR components are usually used. These
criteria can be based either on the total SERR [19] or
the SERR components [20,26], which are obtained from
standard composite test pieces such as the cracked lap
shear specimen (mixed mode, and Mode II), end not-
ched exure specimen (Mode II) and the double canti-
lever beam (Mode I). Their chief use is for comparison
of fracture toughness and fatigue resistance of various
composite material systems, not for predicting growth
of delamination in composite structures [27]. Two such
criteria are used in this paper:
4.1. Propagation criterion based on the total SERR
[19]
dA
dN
= 3014
G
T
1000

7:43
(6)
4.2. Propagation criterion based on the SERR
components [20]
dA
dN
= 0:7188
G
I
103

8
6:5938
G
II
456

6
(7)
The units of delaminated area A and G are mm
2
and
J/m
2
, respectively.
The analysis is carried out for an FE model of a
woven fabric composite plate with an embedded delami-
nation and fatigue loaded under compression. In this
case, only a quarter of the plate is modeled (Fig. 1), since
the use of symmetry boundary conditions is justied Fig. 13. C-scan image of delamination growth of specimen E3.
Fig. 14. Distribution of G
II
along delamination front (4-ply delamination) for F4 and Q4.
F. Shen et al. / Composites Science and Technology 61 (2001) 12391251 1247
because they are compatible with material property
symmetry (all the plies are identically oriented). Dela-
mination growth is modeled by releasing node pairs
held together by multi-point constraints (MPC) between
the surfaces of the delaminated part and the substrate.
The procedure for node release, with consideration for
local damage accumulation at each node pair, is as fol-
lows.
(a) For each step k (to determine increment in dela-
minated area), rst perform a nonlinear FE analysis and
determine G
I
, G
II
and G
III
for all non-midside nodes i
along the crack front (keeping the external applied load
constant).
(b) From Eqs. (6) or (7), for a given increment in
delaminated area A
i
, calculate N
(k)
i
for these nodes,
i = 1; . . . ; n. Select the node j with the smallest N
(k)
j
1 D
(k1)
j
h i
, where D
(k1)
j
is the accumulated damage
coecient for point j.
(c) The increment in damage coecient for each node
i is dened by:
D
(k)
i
=
N
(k)
j
N
(k)
i
1 D
(k1)
j
h i
; i = 1; . . . ; n: (8)
where the accumulated damage coecient D
(k1)
j
= 0
for the rst step.
(d) Calculate the accumulated damage coecient for
all nodes
D
(k)
i
= D
(k1)
i
D
(k)
i
i = 1; . . . ; n: (9)
Select the node j with the largest value of D
(k)
j
(e) Calculate accumulated number of fatigue cycles
N
(k)
t
= N
(k1)
t
N
(k)
j
1 D
(k1)
j
h i
(10)
and the total delaminated area
A
(k)
= A
(k1)
A
j
(11)
(f) Release the node j and the related mid-side nodes;
set D
(k)
j
= 0.
(g) Set k=k+1 and repeat procedure from (a).
Fig. 15. Distribution of G
II
along delamination front (4-ply delamination) for F4 and Q4.
Fig. 16. C-scan image of delamination growth of specimen E4.
1248 F. Shen et al. / Composites Science and Technology 61 (2001) 12391251
In order to compare the FE results with experiment,
12-ply woven fabric laminates were fabricated from
ECS002 plain-weave fabric carbon prepregs, and fatigue
loaded (R = 1, peak load 28 kN). A 20 mm circular
pre-delamination was introduced for each specimen by
inserting a circular Teon piece between the rst and
second plies during casting. This initial delamination
was used to simulate delamination by impact damage.
Circular holes of 5 mm diameter were drilled at
approximately the center of the Teon inserts. The
dimensions of the composite coupons are similar to
those of the previous specimens. The average thickness
of the cured specimens is 3.37 mm. The delamination
growth was recorded by C-scan at regular intervals and
the cumulated area determined. The material properties
for the woven fabric laminates are: E
1
= E
2
= 42:5
GPa, E
3
= 14:5 GPa, G
12
= 17:4 GPa, G
23
= G
13
= 0:85
GPa, and
12
=
13
=
23
= 0:22.
Fig. 17 shows the distributions of SERR components
along the delamination front. The perimeter coordinate
S is used here. The maximum values of G
I
and G
II
are
located at the position S = 0, indicating that the dela-
mination will propagate in the direction perpendicular
to the load direction. This is conrmed by the C-scan
image of the delamination growth (Fig. 18). G
III
, the
tearing mode component of the SERR, is found to be
negligible. Another observation is that the magnitude of
G
I
is signicantly larger than that of G
II
, suggesting the
dominance of Mode I. In this problem, contact con-
straints must also be imposed to prevent interpenetra-
tion of delaminated surfaces. This is particularly so
when the applied load becomes large relative to the
Euler critical buckling load of the sublaminate.
The predicted delaminated areas using Eqs. (6) and
(7) are plotted in Fig. 19 with the experimentally deter-
mined values, which were obtained by superimposing
ne square grids over the C-scan images. In the gure, A
is the total delamination area and N is the number of
fatigue cycles. Although both criteria under-predict the
extent of delamination growth, the results using Eq. (7)
appear to agree much closer with the experimental
values. Eq. (6) produces particularly poor prediction. It
appears therefore that predictions based on the total
SERR [Eq. (6)] are not accurate, and that the compo-
nents of the SERR [Eq. (7)] are needed. It is also
observed that during the simulation, the mode mixity
varies as the delamination propagates. Before the dela-
mination grows, the SERR component G
III
is initially
insignicant. Subsequently, as the delamination grows
and changes shape, the value of G
III
becomes larger.
The maximum G
III
values are located between - 45

and 60

. Therefore the local mode mixity is dependent


Fig. 17. Distribution of SERR components along delamination front for a woven fabric composite laminate.
Fig. 18. C-scan image of delamination growth in a woven fabric
laminate.
F. Shen et al. / Composites Science and Technology 61 (2001) 12391251 1249
on the shape of the delamination front. However, most
published evolution criteria such as Eq. (7) do not con-
tain G
III
. Thus presently, it is dicult to assess the sig-
nicance of G
III
.
Finally, typical shapes of the delamination growth
predicted by Eqs. (6) and (7) are shown in Figs. 20 and
21, respectively. The delaminated area in Fig. 20 is 72
mm
2
and in Fig. 21 is 140 mm
2
. The predicted shape
using the criterion based on SERR components [Eq. (7)]
is slightly narrower than the delamination predicted by
using the criterion based on the total SERR [Eq. (6)].
5. Conclusions
In this paper, both woven fabric and non-woven fab-
ric laminates with post-buckled embedded delamination
have been studied using nonlinear nite elements. Fati-
gue analyses are also carried out based on a proposed
numerical scheme for advancing the delamination front
using fatigue evolution criteria. The numerical results
are validated with experiments. The following conclu-
sions are obtained:
Fig. 19. Prediction of delamination growth rate for circular delamination in a woven fabric laminate.
Fig. 20. Predicted delamination shape using Eq. (6) (delaminated area
72 mm
2
).
Fig. 21. Predicted delamination shape using Eq. (7) (delaminated area
72 mm
2
).
1250 F. Shen et al. / Composites Science and Technology 61 (2001) 12391251
1. The direction of delamination growth generally
coincides with the direction of maximum strain
energy release rate.
2. Sublaminate lay-up sequence plays an important
role in the distribution of local strain energy
release rate components along the delamination
front. The use of one-quarter models is popular
because of signicant reduction in computational
eort, but will generally lead to erroneous results,
particularly when material property symmetry is
incompatible with symmetry conditions imposed
on the boundaries. Even in the case of quasi-iso-
tropic sublaminates, the use of one-quarter models
can lead to over-prediction of the values of SERR
components. In cases where it is not possible to
impose correct boundary conditions in one-quarter
models, full models must be used instead.
3. A method for predicting delamination fatigue
growth is proposed and illustrated, taking into
account the accumulated damage along the delami-
nation front. It is found that propagation criteria
employing strain energy release rate components
rather than total strain energy release rate give closer
predictions when compared to experimental data.
References
[1] Chai H, Babcock CD, Knauss WG. One-dimensional modeling
of failure in laminated plates by delamination buckling. Int J
Solids Struct 1981;17(11):106983.
[2] Whitcomb JD. Finite element analysis of instability related dela-
mination growth. J Composite Mater 1981;15(9):40326.
[3] Kardomateas GA. The initial post-buckling and growth behavior
of internal delaminations in composite plates. J Appl Mech
1993;60:90310.
[4] Gaudenzi P, Perugini P, Spadaccia F. Post-buckling analysis of a
delaminated composite plate under compression. Composite
Struct 1998;40(3-4):2318.
[5] Kachanov LM. Failure of composite materials due to delamina-
tion. Mech Polymers 1976;5:91822.
[6] Yin WL. Axisymmetric buckling and growth of a circular dela-
mination in a compressed laminate. Int J Solids Struct 1985;21(5):
503514.
[7] Whitcomb JD, Shivakumar KN. Strain-energy release rate ana-
lysis of plates with postbuckled delaminations. J Composite
Maters 1989;23:71434.
[8] Whitcomb JD. Three-dimensional analysis of a postbuckled
embedded delamination. J Composite Maters 1989;23:86289.
[9] Nilsson KF, Thesken JC, Sindelar P, Giannakopoulos AE, Storakers
B. A theoretical and experimental investigation of buckling induced
delamination growth. J Mech Phys Solids 1993;41(4): 749782.
[10] Nilsson KF, Giannakopoulos AEA. Finite element analysis of
congurational stability and nite growth of buckling driven
delamination. J Mech Phys Solids 1995;43(12):19832021.
[11] Nilsson KF, Asp LE, Alpman JE. Delamination buckling and
growth at global buckling. In: Rossmanith, editor. Damage and
failure of interfaces, Rotterdam: Balkema, 1997.
[12] Klug J, Wu XX, Sun CT. Ecient modeling of postbuckling
delamination growth in composite laminates using plate elements.
AIAA J 1996;34(1):17884.
[13] Lachaud F, Lorrain B, Michel L, Barriol R. Experimental and
numerical study of delamination caused by local buckling of ther-
moplastic and thermoset composites. Comp Sci Technol 1998;
58:72733.
[14] Chai H. Three-dimensional fracture analysis of thin-lm
debonding. Int J Fract 1990;46:23756.
[15] Raju IS. Calculation of strain-energy release rates with higher
order and singular nite elements. Eng Fract Mech 1987;
28(3):25174.
[16] Shivakumar KN, Tan PW, Newman Jr JC. A virtual crack-clo-
sure technique for calculating stress intensity factors for cracked
three dimensional bodies. Int J Fract 1988;36:R430R50.
[17] Jones R, Paul J, Tay TE, Williams JF. Assessment of the eect of
impact damage in composites: some problems and answers. Theo
Appl Fract Mechs 1988;9:8395.
[18] Sela N, Ishai O. Interlaminar fracture toughness and toughening
of laminated composite materials: a review. Composites 1989;
20(5):42335.
[19] Mohlin T, Blom AF, Carlsson LA, Gustavsson AI. Delamination
growth in a notched graphite/epoxy laminate under compression
fatigue loading. In: Johnson WS, editor. Delamination and
debonding of materials, ASTM STP 876. Philadelphia: American
Society for Testing and Materials, 1985.
[20] Ramkumar RL, Whitcomb JD. Characterization of mode I and
mixed-mode delamination growth in T300/5208 graphite/epoxy.
In: Johnson WS, editor. Delamination and debonding of materi-
als, ASTM STP 876. Philadelphia: American Society for Testing
and Materials, 1985.
[21] Hutchinson JW, Suo Z. Mixed mode cracking in layered materi-
als. In: Advances in applied mechanics, New York: Academic
Press, 1992, vol. 28.
[22] Beuth JL. Separation of crack extension modes in orthotropic
delamination models. Int J Fract 1996;77:30521.
[23] Qian W, Sun CT. Methods for calculating stress intensity factors
for interfacial cracks between two orthotropic solids. Int J Solids
Struct 1998;35(25):331730.
[24] Raju IS, Crews Jr JH, Aminpour MA. Convergence of strain
energy release rate components for edge-delaminated composite
laminates. Eng Fract Mech 1988;30(3):38396.
[25] Tay TE, Shen F, Lee KH, Scaglione A, Di Sciuva M. Mesh
design in nite element analysis of post-buckled delamination in
composite laminates. Composite Struct 1999;47:60311.
[26] Kenane M, Benzeggagh ML. Mixed-mode delamination fracture
toughness of unidirectional glass/epoxy composites under fatigue
loading. Comp Sci Technol 1997;57:597605.
[27] Tay TE, Williams JF, Jones R. Characterization of pure and
mixed mode fracture in composite laminates. Theo Appl Fract
Mechs 1987;7:11523.
F. Shen et al. / Composites Science and Technology 61 (2001) 12391251 1251

Vous aimerez peut-être aussi