Sie sind auf Seite 1von 19

An interface element for the simulation of delamination

in unidirectional ber-reinforced composite laminates


C. Balzani
*
, W. Wagner
Universita t Karlsruhe (TH), Institut fu r Baustatik, Kaiserstr. 12, 76131 Karlsruhe, Germany
Available online 13 March 2007
Abstract
Unidirectional ber-reinforced composite laminates are widely used in aerospace industry for a great variety of struc-
tural parts. In order to enhance the exploitation of material reserves, there is a need for the integration of progressive dam-
age scenarios in the design phase. Due to their hazardous eects on the load-carrying capacity of composite structures, this
work focusses on the simulation of delaminations. A nite element based on a cohesive zone approach is developed. Two
constitutive laws are proposed. One is characterized by linear degradation after delamination onset, the other is governed
by exponential softening response. The damage process is history-dependent leading to an irreversible stiness degradation
in damaged zones. The practicability of the proposed model and the assets and drawbacks of the two material laws are
shown by some numerical examples.
2007 Elsevier Ltd. All rights reserved.
Keywords: Delamination; Progressive damage; Finite element; Cohesive element; Cohesive zone; Interface element
1. Introduction
High-performance structures are often made of unidirectional ber reinforced plastics (UD-FRPs) due to
their high strength and stiness paired with low weight. The application area ranges from sports equipment to
airframe structures. The reinforcement direction can ideally be adapted to a particular stress state so this kind
of material is highly customizable. For maximum exploitation of material reserves, damage and failure scenar-
ios should be taken into consideration during the design phase.
Failure processes in UD-FRPs are very complex ranging from intra-laminar failure (e.g. ber fracture or
inter-ber fracture like transverse matrix cracking, bermatrix debonding etc.) to inter-laminar failure,
namely delamination [14]. Geometrical discontinuities, e.g. free edges, stiener terminations, ply drop-os,
access holes or bonded and bolted joints can provoke both intra- and inter-laminar failure due to arising high
stress gradients. Delamination is one of the most frequently appearing fracture modes in UD-FRPs due to
their lack of reinforcement in thickness direction. Once occurred, delaminations can be exceedingly dangerous
0013-7944/$ - see front matter 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engfracmech.2007.03.013
*
Corresponding author.
E-mail address: claudio.balzani@bs.uni-karlsruhe.de (C. Balzani).
Available online at www.sciencedirect.com
Engineering Fracture Mechanics 75 (2008) 25972615
www.elsevier.com/locate/engfracmech
Nomenclature
Abbreviations
ANS assumed natural strain
DCB double cantilever beam
ENF end notch exure
FM fracture mechanics
FPZ fracture process zone
MMB mixed mode bending
UD-FRP unidirectional ber-reinforced plastic
VCCT virtual crack closure technique
Latin symbols
a
0
length of pre-crack
b specimen width
B matrix containing derivatives of shape functions
c loading arm length
d damage parameter
C interconnection matrix
D elasticity matrix
e
i
Euclidean space basis vectors
E
11
, E
22
longitudinal and transverse Youngs moduli
F applied load
F
e
, F
m
applied load at the end and in the middle of specimen
G
12
, G
23
shear moduli
G
c
; G
c
mixed mode and eective fracture toughnesses
G
I
, G
II
, G
III
mode I, II and III energy release rates
G
Ic
, G
IIc
, G
IIIc
mode I, II and III fracture toughnesses
G
shear
, G
T
shear and total energy release rates
h half thickness of specimen
h
0
initial thickness of interface elements
I, K integer subscript for node number
I identity matrix
I
c
compression identier matrix
k subscript for number of load step
K penalty stiness
K
e
T
element tangent stiness matrix
l specimen length
l
i
length of interface element insertion
N tri-linear shape function
R
e
element residual vector
s, t, n local in-plane and normal basis vectors
s, t, n integer subscripts for local directions
u, du, Du real, virtual and incremental displacement vectors
V, oV volume and surface of interface element
w crack opening (DCB) or load-point deection (MMB)
w
e
, w
m
deection at the end and in the middle of the specimen
x position vector
2598 C. Balzani, W. Wagner / Engineering Fracture Mechanics 75 (2008) 25972615
because they can lead to a signicant reduction of the compressive load-carrying capacity without any visible
damage. Hence, this paper focusses on the numerical treatment of delaminations in the framework of the nite
element method.
Delamination analyses are normally subdivided into the prediction of delamination initiation and propa-
gation. Usually, delamination onset is predicted using stress-strength-based criteria, e.g. [5]. Due to high stress
gradients appearing at crack fronts, employing solely stress-based criteria is not useful. Thus, fracture mechan-
ics (FM) approaches are often used for the simulation of delamination propagation. The most prominent FM
method is the virtual crack closure technique (VCCT) which has been proposed by Rybicki and Kanninen [6].
In this method, energy-based criteria are used to predict propagating delaminations [710]. Usually, a crack
can only propagate at element boundaries so very ne meshes are required. Rinderknecht and Kro plin [1113]
use a complex moving mesh technique which provides good results with coarser meshes. The VCCT is com-
putational eective. However, FM does not incorporate the prediction of damage initiation and needs the exis-
tence of pre-dened cracks. For several geometries, these might be dicult to specify.
Another ecient method to simulate delamination is the meso-level cohesive zone approach which goes
back to Dugdale [14] and Barenblatt [15]. A cohesive law describes the non-linear interfacial softening behav-
ior during the delamination process by forming an extended crack tip. A cohesive zone model can easily be
incorporated in a nite element code by implementing so-called interface elements. Usually, interface elements
relate interfacial tractions to relative displacements. Therefore, many authors use surface-like zero-thickness
interface elements [11,13,1626]. Borg et al. [27] formulated interface elements similar to non-linear springs
located in the element nodes. Also interface elements written in stressstrain relationships with low initial
thickness have been proposed [28]. Needleman [29] rated cohesive interface elements peculiar attractive when
Greek symbols
a internal variable
b mode mixing ratio
e, de, De real, virtual and incremental strain vectors
e
c
eective strain at peak e. stress (exponential model)
e
m
eective strain
e
0
m
e. strain at delamination onset (linear softening model)
e
f
m
e. strain at compl. decohesion (linear softening model)
e
n
, c
sn
, c
tn
strain components corresponding to mode I, II and III
e
0
n
; c
0
sn
; c
0
tn
single mode strains at delam. onset (lin. soft. mod.)
e
f
n
; c
f
sn
; c
f
tn
single mode strains at compl. decohesion (lin. soft. mod.)
c
0
, s
0
shear strain at delamination onset and shear strength
c
shear
total shear strain
g curve tting parameter
j abbreviation factor
k weighting factor for shear strains
m
12
Poissons ratio
r stress vector
r
c
peak eective stress (exponential model)
r
m
eective stress
r
n
, s
sn
, s
tn
stress components corresponding to mode I, II and III
r
0
n
; s
0
sn
; s
0
tn
single mode I, II and III strengths
w cohesive free energy
Other symbols
C stress linearization matrix
F loading function
C. Balzani, W. Wagner / Engineering Fracture Mechanics 75 (2008) 25972615 2599
the interfacial strengths are relatively weak compared with the adjoining material, as is the case for interlam-
inar fracture in UD-FRPs. The softening behavior of cohesive laws overcomes the diculties related to high
stress gradients at the delamination front. If the area under the cohesive law equals the fracture energy, the
cohesive zone model makes use of fracture mechanics. Cohesive elements are peculiar suitable if the crack path
is expectable since it is restricted to the location of interface elements. Unlike FM, cohesive laws are able to
predict both delamination onset and propagation by combining stress-strength-based and energy-based
criteria.
Using an initial thickness a simple continuum mechanics approach can be used to describe the discontin-
uous fracture process of delaminations as also proposed in [28]. In Section 2, the governing equations of an
interface element are established. The element is very similar to standard continuum elements and is written in
natural stressstrain relationships. Two classes of irreversible cohesive laws are presented in Section 3. The
rst one is mainly adopted from Camanho and Davila [30]. The second one is an enhanced version of the
model proposed in [31,32]. The element is able to predict both delamination onset and propagation under
mixed mode loading conditions. In Section 4.1, the abilities and deciencies of the two cohesive laws are dis-
cussed in terms of single mode I loading. The performance of the two concepts for mixed mode loading con-
ditions is studied in Section 4.2. The conclusions, Section 5, close this work.
2. Interface element formulation
In this section, the formulation of a cohesive interface element is developed. The idea is to use a continuum
approach to model the discontinuous interlaminar fracture of composites. The formulation is based on an iso-
parametric hexahedral solid element. In the following, we refer to it as a solid-like interface element. Consider
a thin layer of continuum elements inserted between two adjacent laminate layers. A very thin initial thickness
h
0
of this layer e.g. about 1/100 of the laminate thickness is sucient that all bending moments which
might be produced by the eccentricities of the nodal forces tend against zero, see e.g. [28].
Mohr [33] stated that failure of a material is triggered by the stresses acting on the fracture plane. This
hypothesis has been adopted by Hashin [34] for UD-FRPs for both ber fracture and inter-ber fracture in
a way that if a failure plane can be identied, fracture is caused by the direct stress and the two shear stresses
acting on the failure plane. We assume that this hypothesis is also valid for the case of delamination when
these stresses are the interlaminar stresses. The idea now is to account only for these interlaminar stresses
in the element formulation and to set all other stresses to zero a priori. The two in-plane normal stresses
and the in-plane shear stress would be addressed by in-plane deformation of the laminate. This kind of defor-
mation is carried by the laminate layers themselves but not by the interface. An erroneous additional stiness
would be inserted into the structure if the interface elements would contribute to the mechanical response of
the laminate for in-plane loadings. Consequently, the only remaining stresses are the through-the-thickness
normal stress and the two out-of-plane shear stresses forming an interlaminar stress vector, see Fig. 1.
The interface layer is parameterized in the Euclidean space spanned by the rectangular global coordinate
system e
i
. Its volume and its surface are denoted by V and oV. A local coordinate system s, t, n can be obtained
using a procedure proposed by Taylor [35] when the elements midplane is used. The local coordinate vector n
is the through-the-thickness direction and corresponds to mode I failure (opening of the interface). The vectors
denoted by s and t span the midplane of the interface and correspond to mode II and III failure (shear failure
parallel and transverse to the ber direction). Since the interface is surrounded by the adjoining composite,
n
tn
sn 3
2
1
e
e
e
tn
sn
n
1
2
3
4
5
6
7
8

Fig. 1. Three-dimensional stress state of a solid-like interface element.


2600 C. Balzani, W. Wagner / Engineering Fracture Mechanics 75 (2008) 25972615
stress-like boundary conditions can be excluded. Under iso-thermal static conditions the principle of virtual
work for solid-like interface elements writes
dPu
_
V
de
T
rdV ; 1
where de is the virtual strain vector and r is the interlaminar stress vector which is non-linear with respect to
the strains. The displacement vector u, the strain vector e in a geometrically linear formulation and the inter-
laminar stress vector are dened by
u
u
s
u
t
u
n
_

_
_

_; e
c
sn
c
tn
e
n
_

_
_

_
u
s;n
u
n;s
u
t;n
u
n;t
u
n;n
_

_
_

_; r
s
sn
s
tn
r
n
_

_
_

_: 2
Corresponding to (2), the virtual and incremental strains write
de
dc
sn
dc
tn
de
n
_

_
_

_
du
s;n
du
n;s
du
t;n
du
n;t
du
n;n
_

_
_

_; De
Dc
sn
Dc
tn
De
n
_

_
_

_
Du
s;n
Du
n;s
Du
t;n
Du
n;t
DU
n;n
_

_
_

_; 3
where du = [du
s
, du
t
, du
n
]
T
and Du = [Du
s
, Du
t
, Du
n
]
T
are the virtual and incremental displacement vectors. The
linearization of (1) is given by
DdPu
_
V
de
T
CDe dV ; with C
or
oe
: 4
A nite element approach in an iso-parametric map yields the discretization of the interface through n
elem
2 N
solid-like interface elements. We employ trilinear shape functions N
I
and natural coordinates. With x
I
the
nodal position vectors, any position vector can be calculated by the expression
x

8
I1
N
I
x
I
; with x
I
x
Is
; x
It
; x
In
: 5
The real, virtual and incremental displacement vectors are approximated by
u

8
I1
N
I
u
I
; du

8
I1
N
I
du
I
; Du

8
I1
N
I
Du
I
; 6
where u
I
, du
I
and Du
I
are the nodal real, virtual and incremental displacement vectors. The real, virtual and
incremental strain vectors are dened by
e

8
I1
B
I
u
I
; de

8
I1
B
I
du
I
; De

8
I1
B
I
Du
I
; 7
with
B
I
:
N
I;n
0 N
I;s
0 N
I;n
N
I;t
0 0 N
I;n
_

_
_

_: 8
With this in hands, the principle of virtual work and its linearization on element level write
dP
e

8
I1
du
T
I
_
V
B
T
I
rdV

8
I1
du
T
I
R
e
I
;
DdP
e

8
I1

8
K1
du
T
I
_
V
B
T
I
CB
K
dV Du
K

8
I1

8
K1
du
T
I
K
e
T
IK
Du
K
;
9
where R
e
and K
e
T
are the element residual vector and element tangent stiness matrix.
C. Balzani, W. Wagner / Engineering Fracture Mechanics 75 (2008) 25972615 2601
The kind of integration scheme used in (9) deserves further consideration. Gauss integration can lead to
spurious oscillations of the stress eld [36]. For example, NewtonCotes integration can improve the element
performance [3741]. Alfano and Criseld [42] examined the performance of linear cohesive elements using a
NewtonCotes integration scheme. Here, a NewtonCotes integration scheme is used with 3 3 integration
points in the mid-surface of the element. This leads to a stress distribution which is constant over the elements
thickness as is the case for an interface between two composite layers.
3. Constitutive equations
For a reliable and ecient prediction of the interface response, a suitable constitutive law has to be incor-
porated in the solid-like interface element. This section describes two classes of irreversible cohesive laws: (i) a
model proposed by Camanho and Davila [30] which has been slightly modied and (ii) an enhanced version of
the model developed by de-Andres et al. [31] and Ortiz and Pandol [32]. Both models allow for the prediction
of delamination onset and propagation for mixed mode loading conditions. The model in [30] is characterized
by linear softening, Fig. 2, and is thus referred to as the linear softening model in the following. It shows a
discontinuity in the stressstrain relationships at delamination onset which can lead to numerical diculties
as will be shown in Section 4. In [31,32], a cohesive free energy function of SmithFerrante type [43] is pro-
posed which is governed by exponential response without any point of discontinuity, Fig. 4. Henceforth, this
model is called the exponential model.
Compressive direct stresses do not generate delamination. They might rather complicate the initiation of
damage. This has to be accounted for in the constitutive equations. However, friction eects are neglected
in this work. Camanho and Davila [30] include a criterion for mode I compression in order to avoid the inter-
penetration of the crack faces. Such a criterion is not incorporated in the original exponential model. Thus, we
introduce an additional (penalty) term in the free energy function which accounts for mode I compression. In
both models, the irreversibility of the damage process is accounted for by including an unloading/reloading
criterion which is governed by the assumption that the interface unloads linearly to the origin, cf. e.g. Cam-
acho and Ortiz [44] among many others.
The two models have in common their formulation in terms of a scalar-valued eective strain e
m
and a cor-
responding eective stress r
m
. An internal variable a is introduced which governs the inelastic softening
response. The internal variable is dened by
a
k
: maxfe
m
; a
k1
g; 10
where the subscript k and k 1 denote the current and the preceding load steps. Eq. (10) tracks the maximum
eective strain observed in the loading history. In order to distinguish loading from unloading, a loading func-
tion is introduced which is dened by
Fig. 2. Cohesive law for mixed mode delamination with linear softening (r
m
: eective stress; e
m
: eective strain; r
0
m
: eective strength;
e
0
m
; e
f
m
: eective strain at delamination onset or total decohesion; K: penalty stiness; d: damage variable; G
c
: mixed mode fracture
toughness; h
0
: initial interfacial thickness).
2602 C. Balzani, W. Wagner / Engineering Fracture Mechanics 75 (2008) 25972615
F
he
m
a
k1
i
e
m
a
k1
; 11
where hi : (jj + )/2 are the Macauley brackets and hi/ : 0 if = 0. Loading is identied if (11) is equal
to 1. Unloading/reloading is dened when there is no loading or, in a more mathematical sense, when (11) is
equal to 0. In the following, the subscript k will be omitted when there is no possibility of misunderstanding.
3.1. Linear softening model
In this section, the mixed mode cohesive law proposed by Camanho and Davila [30] is reviewed and a slight
modication is pointed out. Fig. 2 shows the stressstrain relationship considered here with loading and
unloading paths. The eective strain e
m
is dened by
e
m
:

he
n
i
2
c
2
sn
c
2
tn
_
: 12
In case of direct tension, (12) is the norm of the interlaminar strain vector. Otherwise, the eective strain is the
total shear strain
c
shear
:

c
2
sn
c
2
tn
_
: 13
This accounts for the assumption that compressive direct stresses and, as a consequence, compressive direct
strains do not provoke delamination. The constitutive equations have to distinguish three dierent cohesive
states: (i) the initially linear elastic state (no damage), (ii) the softening state (partly damaged) and (iii) the del-
aminated state (total decohesion). The internal variable dened by (10) permits the identication of the par-
ticular states. Delamination initiation is governed by the eective strain at delamination onset denoted by e
0
m
.
We assume that delamination initiation can be predicted by a quadratic failure criterion expressed by
hr
n
i
r
0
n
_ _
2

s
sn
s
0
sn
_ _
2

s
tn
s
0
tn
_ _
2
1; 14
cf. [34,45,46]. Cui et al. [47] showed that the quadratic failure criterion is superior in the prediction of delam-
ination onset compared to maximum stress criteria since it allows for arbitrary mode interactions. In addition,
(14) takes into account only tensile normal stresses as postulated earlier. In (14), r
0
n
; s
0
sn
and s
0
tn
denote the
mode I, II and III strengths. If the interlaminar mode I strength is equal to the intra-laminar normal tensile
strength its value can be determined according to ASTM standard [48]. We assume the same penalty stiness
K for all delamination modes in the initially linear elastic state. Furthermore, same strengths for shear modes
are assumed, s
0
sn
s
0
tn
s
0
. The shear strength can also be determined according to ASTM standard [49]. The
single mode strains at delamination onset are obtained through
e
0
n

r
0
n
K
; c
0
c
0
sn
c
0
tn

s
0
K
: 15
In case of positive direct strains we introduce a mode mixing ratio b dened by the expression
b :
c
shear
e
n
: 16
With (12)(16) the eective strain at delamination onset writes
e
0
m

e
0
n
c
0

1b
2
c
0

2
be
0
n

2
_
e
n
> 0;
c
0
e
n
6 0:
_
_
_
17
An eective fracture toughness is introduced which is dened by
G
c
:
G
c
h
0
; 18
C. Balzani, W. Wagner / Engineering Fracture Mechanics 75 (2008) 25972615 2603
where G
c
is the mixed mode critical energy release rate. The eective fracture toughness represents the area
under the (eective) stressstrain curve. For the calculation of G
c
, Camanho and Davila [30] recommend a
criterion proposed by Benzeggagh and Kenane [50] (BK criterion) which is originally given by
G
c
G
Ic
G
IIc
G
Ic

G
II
G
T
_ _
g
: 19
Here, G
Ic
and G
IIc
are the single mode I and II fracture toughnesses which can be obtained e.g. by double can-
tilever beam (DCB) or end notch exure (ENF) tests [51,52]. The mode I and II energy release rates are de-
noted by G
I
and G
II
, respectively. The total energy release rate is dened by G
T
= G
I
+ G
II
. Experimentally,
mixed mode I/II fracture toughnesses can be determined by mixed mode bending (MMB) tests [53] for dier-
ent mode ratios G
II
/G
T
. The mixed mode fracture energies are not linearly depending on the mode ratio.Thus,
the parameter g allows for curve tting with experimental data. If mode III occurs the BK criterion is mod-
ied in [30], so (19) becomes
G
c
G
Ic
G
IIc
G
Ic

G
shear
G
T
_ _
g
; 20
where G
shear
= G
II
+ G
III
is the shear energy release rate and the total energy release rate is dened by
G
T
= G
I
+ G
shear
. The single mode strains at complete decohesion are given by
e
f
n

2G
Ic
h
0
r
0
; c
f
sn

2G
IIc
h
0
s
0
; c
f
tn

2G
IIIc
h
0
s
0
; 21
where G
IIIc
is the single mode III fracture toughness. The next aim is an expression for the eective strain at
complete decohesion. Let us rst consider mixed mode loading in absence of tensile normal strains. According
to [30] the eective strain at complete decohesion for this load case is dened by
e
f
m

c
f
sn
_ _
2
c
f
tn
_ _
2
_
e
n
6 0: 22
Obviously, the latter circular condition overestimates the single mode values, see (21) and Fig. 3. Thus, the
solutions are not conservative which is a postulation for safe design. To the knowledge of the authors neither
a standardized test method for the determination of the mode III fracture toughness nor studies for mixed
mode behavior incorporating mode III are available. For simplicity it seems adequate to use a circular con-
dition but it should guarantee safe design. Thus, we introduce a modied version of (22) which is given by
e
f
m
c
f
sn
; 23
see Fig. 3. Eq. (23) yields the exact value for single mode II. Admittedly, it implies same fracture energies for
mode II and mode III, G
IIc
= G
IIIc
. This assumption is contrary to Lees work [54] who showed that the mode
III fracture energy is usually higher than the respective mode II value. However, it yields an underestimated
value for single mode III delamination, an attribute which agrees with the postulation of safe design. The
Fig. 3. Eective strain at complete decohesion for mixed mode II/III loadings.
2604 C. Balzani, W. Wagner / Engineering Fracture Mechanics 75 (2008) 25972615
assumption of equal behavior for mode II and mode III is automatically incorporated for tensile normal
strains since Eq. (20) contains solely G
Ic
and G
IIc
. Using (18), (20), (21) and (23) yields
e
f
m

2
Kh
0
e
0
m
G
Ic
G
IIc
G
Ic

b
2
1b
2
_ _
g
_ _
e
n
> 0;
c
f
sn
e
n
6 0;
_
_
_
24
an expression for the eective strain at complete decohesion for any mixed mode loading condition. For the
calculation of the extent of damage a scalar-valued damage parameter
d :
e
f
m
a e
0
m

ae
f
m
e
0
m

25
is introduced. The damage parameter governs the softening behavior of the interface. It increases from 0 (no
damage) to 1 (complete decohesion) for monotonous load progression. The constitutive equations are
r De: 26
The elasticity matrix D contains the distinction of the particular cohesive states inside of the interface and is
expressed by
D
KI a 6 e
0
m
;
1 dKI dKI
c
e
0
m
< a < e
f
m
;
KI
c
e
f
m
6 a;
_

_
27
where I is the identity matrix and
I
c

0 0 0
0 0 0
0 0
he
n
i
e
n
_

_
_

_ 28
is a compression identier matrix. Using (4)
2
yields the linearization matrix of the stresses which writes
C D jFI
c
Iee
T

he
m
e
0
m
i
e
m
e
0
m
he
f
m
e
m
i
e
f
m
e
m
29
with
j : K
e
0
m
e
f
m
a
3
e
f
m
e
0
m

: 30
It should be recognized that the second term in (29) vanishes in case of unloading/reloading, no damage or
complete decohesion. In these cases the constitutive equations are linear and the stress linearization matrix
equals the elasticity matrix.
3.2. Exponential model
The cohesive law proposed in this section is basically inspired by de-Andres et al. [31] and Ortiz and Pan-
dol [32]. They introduced a potential description of SmithFerrante type [43] which is dominated by expo-
nential response. Fig. 4 depicts the considered (eective) stressstrain relationships for loading and
unloading/reloading. Han et al. [55] used this model for delamination analyses of honeycomb panels. Hence,
the model is considered to be applicable for the simulation of delamination in UD-FRPs as well. In the ori-
ginal model [31,32] an eective strain
e
m

e
2
n
k
2
c
2
sn
c
2
tn

e
T
Ce
p
with C :
k
2
0 0
0 k
2
0
0 0 1
_

_
_

_ 31
C. Balzani, W. Wagner / Engineering Fracture Mechanics 75 (2008) 25972615 2605
is introduced. This denition does not distinguish compression from tension. Also compressive normal stresses
provoke delamination. Hence, we redene the eective strain
e
m

he
n
i
2
k
2
c
2
sn
c
2
tn

e
T
Ce
p
with C :
k
2
0 0
0 k
2
0
0 0
he
n
i
e
n
_

_
_

_: 32
Recall that hi/ : 0 if = 0. The parameter k applies dierent weights to mode I and shear modes. For k = 1,
(32) equals (12). In contrast to the original model, the Macauley brackets in (32) consider only tensile direct
strains so damage can not be triggered by mode I compression. The cohesive free energy function proposed in
[31,32] is given by
w
^
we
m
; a er
c
e
c
1 1
a
e
c
_ _
e
a=e
c
_ _
; 33
a function which does not take into account any contact constraints within the interface element. In this work,
the potential formulation is enhanced by a penalty term. The additional term penalizes compressive direct
strains in order to avoid the interpenetration of the crack faces for both intact and damaged interfaces.
The cohesive free energy function used here is dened by
w
^
we
m
; a
^
w
1
e
m
; a
^
w
2
e
m
; a 34
with
w
1

^
w
1
e
m
; a er
c
e
c
1 1
a
e
c
_ _
e
a=e
c
_ _
;
w
2

^
w
2
e
m
; a
1
2
Ke
2
n
he
n
i
e
n
;
35
where K is the penalty stiness applied for mode I compression. In (33) and (35)
1
, r
c
and e
c
are the eective
limit stress and the corresponding eective strain. The free energy determines the response at loading. As a
consequence of the rst and second law of thermodynamics, e.g. [56], the cohesive stresses in case of loading
write
r
o
^
we
m
; a
oe

o
^
w
1
e
m
; a
oe
m
oe
m
oe

o
^
w
2
e
m
; a
oe
: 36
With
0
= o/oe
m
the eective stress can be dened by the expression
r
m
w
0
1

^
w
0
1
e
m
; a er
c
a
e
c
e
a=e
c
37
Fig. 4. Cohesive law for mixed mode delamination with exponential behavior (r
m
, r
c
: eective and peak stress; e
m
, e
c
: eective strain and
eective strain at peak stress). SmithFerrante loading and linear elastic unloading/reloading.
2606 C. Balzani, W. Wagner / Engineering Fracture Mechanics 75 (2008) 25972615
so the interlaminar stress vector becomes
r
r
m
e
m
Ce KI
c
e; 38
where I
c
is the compression identier matrix dened by (28). We use a quadratic interaction criterion accord-
ing to (14). As in the linear softening model, we assume equal strengths for mode II and mode III. Further-
more, the critical eective stress shall be equal to the mode I strength, r
c
r
0
n
. Then the eective stress also
writes
r
m

r
T
C
1
r
_

hr
n
i
2
k
2
s
2
sn
s
2
tn

_
: 39
Now the parameter k denes the ratio between the shear and the direct critical stress, s
0
and r
0
n
, respectively. It
has now physical meaning and can be determined through material testing. We assume unloading linearly to
the origin and reloading on the same path so the eective stress in this case is dened by
r
m

r
m
a
a
e
m
F 0 40
In accordance with (4)
2
, the stress linearization matrix C is
C
^
w
00
1
e
m
; a
^
w
0
1
e
m
; a
a
_ _
a
2
FCeCe
T

^
w
0
1
e
m
; a
a
C KI
c
: 41
Recall that the eective stress w
0
1
is given by (37). The derivative of the eective stress with respect to the eec-
tive strain, w
00
1
, is dened by
w
00
1

^
w
00
1
e
m
; a e
r
c
e
c
1
a
e
c
_ _
e
a=e
c
: 42
Substitution of (37) and (42) in (38) and (41) yields the stress vector and the stress linearization matrix
r e
r
c
e
c
e
a=e
c
Ce KI
c
e; 43
C Fe
r
c
e
2
c
a
1
e
a=e
c
CeCe
T
e
r
c
e
c
e
a=e
c
C KI
c
: 44
The choice of the penalty stiness K which has to be applied for mode I compression is discussed in the fol-
lowing. Consider that shear strains are equal to zero (single mode I) so the stress inside of the interface in case
of loading is
r
n

Ke
n
e
n
< 0;
ee
n
r
c
e
c
e
e
n
=e
c
e
n
P0:
_
45
Postulating the same tangent for e
n
! 0 for tension and compression yields a relation for the penalty stiness
given by
or
n
e
n
< 0
oe
n

e
n
0

or
n
e
n
P0
oe
n

e
n
0
) K e
r
c
e
c
: 46
In analogy to Section 3.1, the area under the (eective) stressstrain-curve from origin to total decohesion
shall dene the mixed mode fracture energy G
c
divided by the initial thickness of the interface h
0
, Fig. 4.
De-Andres et al. [31] and Ortiz and Pandol [32] seem to assume same fracture energies for mode I and shear
modes. This assumption does not hold for UD-FRPs. Thus, the modied BK criterion, Eq. (20), is used for
the determination of the mixed mode fracture toughness G
c
which is then given by
G
c
G
Ic
G
IIc
G
Ic

b
g
: 47
C. Balzani, W. Wagner / Engineering Fracture Mechanics 75 (2008) 25972615 2607
Here, we have used the relationship

b :
G
shear
G
T

c
shear
e
T
_ _
2
; 48
where c
shear
is dened by (13) and e
T
:

he
n
i
2
c
2
shear
_
. The parameters G
shear
and G
T
are dened analogous to
Section 3.1. Thus, the eective fracture toughness is dened by
G
c
:
G
c
h
0

1
h
0
lim
a!1
f
^
w
1
e
m
; ag
er
c
e
c
h
0
; 49
a function bringing together the fracture energy G
c
and the eective strain e
c
so Eq. (46) becomes
K
er
c

2
G
c
: 50
In order to track the extent of damage progression we introduce a damage parameter which is given by the
expression
d
^
w
1
e
m
; a
G
c
; 51
cf. [31,32]. The damage parameter denes the ratio between dissipated energy and eective fracture energy, the
energy dissipated at complete decohesion. It takes values from 0 to 1. A damage parameter d = 0 indicates no
damage whereas d = 1 means complete decohesion and formation of a new surface.
4. Numerical examples
The interface element and the two cohesive laws have been implemented in an extended version of the nite
element code FEAP [57]. In order to limit computation time the composite strips are discretized with quad-
rilateral shell elements of ReissnerMindlin type. Shear locking eects are reduced by using an assumed natural
strains (ANS) approach [58]. The osets between nodal sites and shell midplane are accounted for in the shell
element formulation. Geometrical non-linearity does not aect the examples presented in this work. Thus,
geometrical linear calculations have been carried out. For the solution of the non-linear equilibrium equations
a displacement-controlled arc-length method has been employed. Due to symmetry in transverse direction, the
computations have been performed on the half of the specimen with taking into account symmetry boundary
conditions.
4.1. Double cantilever beam test
A standard testing method for the determination of mode I fracture toughness is the DCB test [51]. An
experiment with AS-4/3501-6 graphite/epoxy material which has been conducted by Aliyu and Daniel [59]
is the basis for this example. Wagner et al. [28] also simulated this experiment and specied the material prop-
erties listed in Table 1. The crack opening velocity was _ w 0:85 mm=s so quasi-static conditions are applied.
The fracture mode is pure mode I delamination. Hence, the mode I fracture toughness and strength are the
governing material parameters for the softening response inside of the interface elements. The geometry, load-
ing and boundary conditions as well as the FE mesh of the DCB test are depicted in Fig. 5.
The specimen has a length of l = 150 mm, a width of b = 25.4 mm and a thickness of 2h = 3.05 mm. A pre-
delaminated region with a length of a
0
= 31.75 mm is inserted in the specimens midplane. At the free end of
Table 1
Inter- and intra-laminar material properties of AS-4/3501-6
E
11
(N/mm
2
) E
22
(N/mm
2
) G
12
(N/mm
2
) G
23
(N/mm
2
) m
12
G
Ic
(N/mm) r
0
n
(N/mm
2
) s
0
(N/mm
2
)
138,000 8960 7100 3446 0.3 0.222 N/mm 51.7 91.7
2608 C. Balzani, W. Wagner / Engineering Fracture Mechanics 75 (2008) 25972615
the cantilever the specimen edges are loaded with opposite loads F/b. The bers are oriented in longitudinal
direction of the specimen. Interface elements are inserted in the extension of the pre-delaminated region
(region 2, black area in Fig. 5). Here, delamination is expected to propagate. Preliminary observations showed
that an initial thickness of the interface of h
0
= 2h/100 fullls the postulation of suciently thin interface lay-
ers made in Section 2, see also [28]. The specimen is discretized with ve elements along the half of the width.
The FE mesh is staggered in longitudinal direction. In the pre-delaminated region (region 1), 15 shell elements
are used. Subsequently, region 2 is discretized with 120 elements in the coarse version and 240 elements in the
ne version (shells and interface elements). This region has a length of l
i
= 60 mm. The remaining region 3 is
discretized with 15 shell elements. In this region crack propagation is not possible. In the transition line
between region 2 and region 3, the nodes are linked pair-wise so the nodal degrees of freedom receive the same
values. This methodology restricts the interface nodes located there to stay together and simulates a perfect
bond in region 3.
Analytical solutions based on Bernoulli beam theory are available for the DCB test based on standard for-
mulations of continuum mechanics. For the initially linear elastic part it is given by
F w
E
11
bh
3
8a
3
0
w; 52
where w is the opening at the free end of the specimen. The analytical solution for delamination propagation
writes
F w

2b
2
3w

G
3
Ic
E
11
h
3
12

_
: 53
Fig. 6 shows loaddeection curves of the FE simulation using the linear softening model. The penalty stiness
was set to K = 10
6
N/mm
2
. An equivalent value has also been suggested in [30]. Normally, the penalty stiness
has to be the highest value which leads to converged solutions. Unlike in [30] the linear softening model fails to
converge with a realistic strength after reaching the limit load. A reason for that might be that there a line
search algorithm was employed which has not been done in this work. Additionally, Camanho and Davila
[30] used only a 2 2 NewtonCotes integration scheme which seems to be doubtful. Since the cohesive
law is highly non-linear a 3 3 NewtonCotes integration has been used here. It can be seen that for dierent
Fig. 5. Sketch and FE mesh of the DCB test (only 60 elements plotted in region 2, FE mesh mirrored in transverse direction so the
complete specimen is plotted).
C. Balzani, W. Wagner / Engineering Fracture Mechanics 75 (2008) 25972615 2609
strength values dierent mesh sizes are necessary to get converged solutions. This does not lead to a mesh
dependency. Only slight dierences occur in the undamaged region due to strength variation whereas very
good results are obtained for delamination propagation. Linear unloading to the origin can also be observed.
Numerical solutions of the exponential model are presented in Fig. 7. As also with the linear softening
model the results for delamination propagation are coincident with the analytical solution and very close
to the experimental results. Again, linear unloading to the origin can be observed. In contrast to the linear
softening model the exponential model converges for both mesh versions with realistic material properties.
The analytical solution is slightly underestimated in the undamaged linear part because the rst elements begin
to fail before reaching the limit load in the numerical results which is not accounted for in the analytical
solution.
Fig. 8 shows a contour plot of the damage variable at a crack opening of w = 10.0 mm. The clamped end is
on the right-hand side. The horizontal line is the symmetry plain. The dashed vertical line is the location of the
original straight crack front and the solid vertical line is the end of region 2. Damage variables dierent from 0
or 1 signify the fracture process zone (FPZ). All calculations have been stopped before the FPZ reached the
Fig. 6. Loaddeection curves of a DCB test using the linear softening model.
Fig. 7. Loaddeection curves of a DCB test using the exponential model.
2610 C. Balzani, W. Wagner / Engineering Fracture Mechanics 75 (2008) 25972615
end of region 2. Otherwise, the perfect bond in region 3 inuences crack propagation and corrupts the results.
It can be observed that the FPZ is still far away from region 3. The magnied view in Fig. 8 shows a curvi-
linear shape of the delamination front. This is a typical eect which is caused by transverse bending.
4.2. Mixed mode bending test
The following example is the simulation of an MMB test [53]. Fig. 9 shows a sketch of the experimental
setup and a simplied geometry for the FE-approximation. A key benet of the MMB test is that any mode
mixture can be applied by calibration of the loading arm length c. DCB and ENF tests are borderlines for
MMB test applications. Furthermore, one can use the same specimens as for DCB or ENF tests.
Camanho and Davila [30] performed experiments ranging from DCB to ENF for UD AS4/PEEK carbon-
ber reinforced composites which build the basis for this numerical example. Fig. 10 shows the geometry of
the specimens and the FE discretization. The bers are oriented in longitudinal direction of the specimens. The
material properties can be found in Table 2. The specimens have a length of l = 102 mm, a width of
b = 25.4 mm and a height of 2h = 2 1.56 mm. The number of plies and the ply thickness are not given in
[30] so we use two (articial) plies with thickness 1.56 mm with the pre-crack and the interface elements
located between them. The initial thickness of the interface is set to h
0
= 2h/100. The specimens are discretized
by 5 elements along the half of their widths. In the longitudinal direction, the discretization pattern is stag-
gered. In the pre-delaminated region (region 1) 15 elements are used followed by 50 elements in the delami-
nation propagation region (region 2) which extends up to the middle of the specimen. The remaining
region where no delamination shall occur (region 3) is discretized by 20 shell elements. In the transition line
between region 2 and region 3 the nodes are linked pair-wise so the nodal degrees of freedom receive the same
values. This imitates a perfect bond in region 3. In order to avoid the interpenetration of the upper and lower
specimen arms in the pre-cracked area interface elements with modied constitutive equations are used in
region 1. Thus it holds for the constitutive equations
Fig. 8. Contour plot of damage variable d in top view at w = 10.0 mm; clamped end is on the right-hand side.
Fig. 9. Experimental setup of the MMB test and simplied geometry for FE approximation.
C. Balzani, W. Wagner / Engineering Fracture Mechanics 75 (2008) 25972615 2611
r De KI
c
e;
C D KI
c
:
54
The tests were performed for dierent mode mixing ratios G
II
/G
T
. The pre-delamination lengths a
0
and the
mixed mode fracture toughnesses obtained experimentally for the particular mode mixing ratios are given
in Table 3. For DCB loadings the simulations are according to Fig. 5. For ENF and mixed mode loading,
the bottom shell elements at the pre-damaged end of the specimen are supported in vertical direction. At
the opposite end the nodes are supported in longitudinal and vertical direction, Fig. 10.
Camanho and Davila [30] produced loaddeection curves during their experiments. The applied load F
has been recorded versus the load-point deection w at the end of the loading arm. The specimens are modeled
like beams which are subjected to dierent loads at their ends (F
e
) and in their center (F
m
). A procedure of how
to compute these forces is given in [30]. The values of the applied loads are listed in Table 4. The load-point
deection can be calculated after the FE-analysis from the nodal displacements of the loaded nodes. The posi-
tive direction of the load-point displacement and the deections in the middle and at the end of the specimen
Fig. 10. Sketch and FE-mesh of the MMB test (only 25 elements plotted in region 2, FE-mesh mirrored in transverse direction, so the
complete specimen is plotted).
Table 2
Inter-and intra-laminar material properties of AS-4/PEEK
E
11
(N/mm
2
) E
22
(N/mm
2
) G
12
= G
13
(N/mm
2
) G
23
(N/mm
2
) m
12
G
Ic
(N/mm) G
IIc
(N/mm) r
0
n
(N/mm
2
) s
0
(N/mm
2
)
122,700 10,100 5500 3700 0.25 0.969 1.719 80 100
Table 3
Mixed mode fracture energies and pre-delamination lengths
G
II
/G
T
0.0 (DCB) 0.2 0.5 0.8 1.0 (ENF)
G
c
(N/mm) 0.969 1.103 1.131 1.376 1.719
a
0
(mm) 32.9 33.7 34.1 31.4 39.3
Table 4
Loads on the MMB specimen
G
II
/G
T
0.0 (DCB) 0.2 0.5 0.8 1.0 (ENF)
F
m
0.00 3.15F 1.87F 1.56F 1.00F
F
e
1.00F 2.15F 0.87F 0.56F 0.00
2612 C. Balzani, W. Wagner / Engineering Fracture Mechanics 75 (2008) 25972615
(w
m
and w
e
) is the direction of the load F, Fig. 9. We assume rigid body motion of the loading arm so the load-
point displacement is obtained by
w
2c l
l
w
m

2c
l
w
e
: 55
In the calculations performed here the experimentally determined mixed mode fracture energies have been
used for G
c
in (49). Fig. 11 shows the loaddeection curves of the experiments, the FE-simulations extracted
from [30] and the numerical results using the derived formulations with both the linear softening and the expo-
nential model. The calculations have been stopped before the delamination front approached region 3.
As also in Section 4.1 the linear softening model exhibits convergence problems with realistic properties.
Additionally, the convergence behavior has been shown to be very sensitive to the initial penalty stiness
K. The curves shown in Fig. 11 were obtained using half strengths and a penalty stiness of K = 10
3
N/
mm
2
. The reduced strength and the relatively small penalty stiness lead to a slight under-estimation of the
experimental data. Again it is strange that Camanho and Davila [30] obtained converged solutions although
they used an equivalent penalty stiness of K = 10
6
N/mm
2
and the real strengths. Possible reasons are already
pointed out in Section 4.1. The exponential model converges also with the realistic strength. Compared with
the linear softening model, the stiness of the undamaged linear part and the predicted limit loads are closer to
the experimental data since the strength has not been reduced. In general, very good agreement with the
experiments can be observed.
5. Conclusions
A nite element which describes three-dimensional mixed mode delamination has been developed. It is
based on the formulation of standard solid elements but carries only interlaminar stresses. The constitutive
equations are introduced in the framework of a cohesive zone approach. Two modied cohesive laws have
been shown. The rst one is based on the development of Camanho and Davila [30] which is governed by lin-
ear softening after delamination onset. This model has been edited since the original version leads to uncon-
servative solutions for pure shear loading or mode I compression. The second one is based on publications of
Fig. 11. Results of the MMB test.
C. Balzani, W. Wagner / Engineering Fracture Mechanics 75 (2008) 25972615 2613
de-Andres et al. [31] and Ortiz and Pandol [32] and derives from an exponential free energy function. The
exponential model has been modied such that only tensile normal stresses provoke delamination. In addition,
a penalty term has been added to the free energy function which accounts for contact in order to avoid the
interpenetration of the crack faces. Some numerical examples have been carried out which demonstrate the
performance of both models to simulate propagating mixed mode delaminations. The exponential model
showed a superior convergence behavior since the linear softening model failed to converge with realistic
material parameters for the chosen FE mesh. In general both cohesive laws obtained good agreement with
experimental results and analytical solutions. However, due to the better convergence behavior the exponen-
tial model is peculiarly suitable for delamination analyses.
Acknowledgements
This work is supported by the European Commission, Priority Aeronautics and Space, Contract AST3-
CT-2003-502723. The information in this paper is provided as is and no guarantee or warranty is given that
the information is t for any particular purpose. The user thereof uses the information at its sole risk and
liability.
References
[1] Garg AC. Delamination a damage mode in composite structures. Engng Fract Mech 1988;29(5):55784.
[2] Bolotin VV. Delaminations in composite structures: its origin, buckling, growth and stability. Compos Part B: Engng
1996;27B:12945.
[3] OBrien TK. Fracture mechanics of composite delamination. In: ASM handbook 21, Composites, ASM International, 2001. p. 2415.
[4] Tay TE. Characterization and analysis of delamination fracture in composites: an overview of developments from 1990 to 2001. Appl
Mech Rev 2003;56(1):132.
[5] Camanho PP, Matthews FL. Delamination onset prediction in mechanically fastened joints in composite laminates. J Compos Mater
1999;33:90627.
[6] Rybicki EF, Kanninen MF. A nite element calculation of stress intensity factors by a modied crack closure integral. Engng Fract
Mech 1977;9:9318.
[7] Kru ger R. Delaminationswachstum in Faserverbundlaminaten. PhD thesis, University of Stuttgart. Institute for Statics and
Dynamics of Aerospace Structures, Report-No. 13-96, 1996 [in German].
[8] Wang GS. A probabilistic damage accumulation solution based on crack closure model. Int J Fatigue 1999;21:53147.
[9] Tessmer J. Theoretische und algorithmische Beitrage zur Berechnung von Faserverbundschalen. PhD thesis. University of Hannover,
Institute for Construction Mechanics and Numerical Mechanics, Report-No. F 00/3, 2000 [in German].
[10] Kru ger R. Virtual crack closure technique: history, approach and applications. Appl Mech Rev 2004;57:10943.
[11] Rinderknecht S, Kro plin B. Calculation of delamination growth with fracture and damage mechanics. In: Hughes TJR, On ate E,
Zienkiewicz, editors. Recent developments in nite element analysis, Barcelona, 1994. p. 18190.
[12] Rinderknecht S, Kro plin B. Delamination growth simulation with a moving mesh technique. In: Topping BHV, Papadrakakis M,
editors. Advances in non-linear nite element methods. Edinburgh, Scotland: CIVIL-COMP Ltd.; 1994. p. 18797.
[13] Rinderknecht S, Kro plin B. A nite element model for delamination in composite plates. Mech Compos Mater Struct 1995;2:1947.
[14] Dugdale DS. Yielding of steel sheets containing slits. J Mech Phys Solids 1960;8:1004.
[15] Barenblatt GI. Mathematical theory of equilibrium cracks in brittle failure. Adv Appl Mech 1962;7.
[16] Needleman A. An analysis of tensile decohesion along an interface. J Mech Phys Solids 1990;38:289324.
[17] Needleman A. An analysis of decohesion along an imperfect interface. Int J Fract 1990;42:2140.
[18] Tvergaard V, Hutchinson JW. The relation between crack growth resistance and fracture process parameters in elasticplastic solids.
J Mech Phys Solids 1992;40:137797.
[19] Xu XP, Needleman A. Void nucleation by inclusion debonding in a crystal matrix. Modell Simul Mater Sci Engng 1993;1:11132.
[20] Criseld M, Mi Y, Davies GAO, Hellweg HB. Finite element methods and the progressive failure-modelling of composites structures.
In: Owen DRJ, Onate E, Hinton E, editors. Computational plasticity, Barcelona, 1997. p. 23954.
[21] Geubelle PH, Baylor J. The impact-induced delamination of laminated composites: a 2D simulation. Compos Part B: Engng
1998;29B:589602.
[22] de Borst R, Schipperen JHA. Computational methods for delamination and fracture in composites. In: Allix O, Hild F, editors.
Continuum damage mechanics of materials and structures. Elsevier Science Ltd.; 2002. p. 32552.
[23] Shet C, Chandra N. Analysis of energy balance when using cohesive zone models to simulate fracture processes. ASME J Engng
Mater Technol 2002;124:44050.
[24] Zou Z, Reid SR, Li S, Soden PD. Modelling interlaminar and intralaminar damage in lament wound pipes under quasi-static
indentation. J Compos Mater 2002;36:47799.
[25] Zou Z, Reid SR, Li S. A continuum damage model for delaminations in laminated composites. J Mech Phys Solids 2003;51:33356.
2614 C. Balzani, W. Wagner / Engineering Fracture Mechanics 75 (2008) 25972615
[26] Hashagen F. Numerical analysis of failure mechanisms in bre metal laminates. PhD Thesis. Technical University of Delft, 1998.
ISBN 90-407-1753-3.
[27] Borg R, Nilsson L, Simonsson K. Modeling of delamination using a discretized cohesive zone and damage formulation. Compos Sci
Technol 2002;62:1299314.
[28] Wagner W, Gruttmann F, Sprenger W. A nite element formulation for the simulation of propagating deleminations in layered
composite structures. Int J Numer Methods Engng 2001;51:133759.
[29] Needleman A. A continuum model for void nucleation by inclusion debonding. ASME J Appl Mech 1987;54:52531.
[30] Camanho PP, Davila CG. Mixed-mode decohesion nite elements for the simulation of delamination in composite materials. NASA,
Report-No. TM-2002-211737, 2002.
[31] de-Andres A, Perez JL, Ortiz M. Elastoplastic nite element analysis of three-dimensional fatigue crack growth in aluminium shafts
subjected to axial loading. Int J Solids Struct 1999;36:223158.
[32] Ortiz M, Pandol A. Finite-deformation irreversible cohesive elements for three-dimensional crack-propagation analysis. Int J
Numer Methods Engng 1999;44:126782.
[33] Mohr O. Welche Umstande bedingen die Elastizitatsgrenze und den Bruch eines Materials? J VDI 24 1900;45:152430. and 46, p.
15727 [in German].
[34] Hashin Z. Failure criteria for unidirectional ber composites. J Appl Mech 1980;47:32934.
[35] Taylor RL. Finite element analysis of linear shell problems. In: Whiteman JR, editor. The mathematics of nite elements and
applications VI (MAFELAP 1987). London, UK: Academic Press; 1988. p. 191204.
[36] de Borst R, Rots JG. Occurrance of spurious mechanisms in computation of strain-softening solids. Engng Comput 1989;6:27280.
[37] Goncalves JP, de Moura MF, de Castro PT, Marques AT. Interface element including point-to-surface constraints for three-
dimensional problems with damage propagation. Engng Comput 2000;17(1):2847.
[38] Mi Y, Criseld MA, Davies GAO, Hellweg HB. Progressive delamination using interface elements. J Compos Mater
1998;32:124673.
[39] Schellekens JCJ, de Borst R. Numerical simulation of free edge delamination in graphite-epoxy laminates under uniaxial tension. In:
Proceedings of the sixth international conference on composite structures, 1991. p. 64757.
[40] Schellekens JCJ. Computational strategies for composite structures. PhD Thesis, The Netherlands, Technical University of Delft,
1992.
[41] Schellekens JCJ, de Borst R. On the numerical integration of interface elements. Int J Numer Methods Engng 1993;36:4366.
[42] Alfano G, Criseld MA. Finite element interface elements for the delamination analysis of laminated composite structures:
mechanical and computational issues. J Numer Methods Engng 2001;50:170136.
[43] Rose JH, Ferrante J, Smith JR. Universal binding energy curves for metals and bimetallic interfaces. Phys Rev Lett 1981;47(9):6758.
[44] Camacho GT, Ortiz M. Computational modeling of impact damage in brittle materials. Int J Solids Struct 1996;33:2899938.
[45] Mohammadi S, Owen DRJ, Peric D. A combined nite/discrete element algorithm for delamination analysis of composites. Finite
Elem Anal Des 1998;28(4):32136.
[46] Puck A. A failure criterion shows the direction. Kunststoe 1992;82:60710 [in German].
[47] Cui W, Wisnom MR, Jones M. A comparison of failure criteria to predict delamination of unidirectional glass/epoxy specimens
waisted through the thickness. Composites 1992;23(3):15866.
[48] Test Method D 3039/D 3039 M-00. Standard test method for tensile properties of polymer matrix composite materials. West
Conshohocken, PA, USA: American Society for Testing and Materials; 2000.
[49] Test Method D2344-84. Standard test method for apparent interlaminar shear strength of parallel ber composites by short-beam
method. West Conshohocken, PA, USA: American Society for Testing and Materials; 1995.
[50] Benzeggagh ML, Kenane M. Measurement of mixed-mode delamination fracture toughness of unidirectional glass/epoxy composites
with mixed-mode bending apparatus. Compos Sci Technol 1996;56:43949.
[51] Test Method D5528-01. Standard test method for mode i interlaminar fracture toughness of unidirectional ber-reinforced polymer
matrix composites. West Conshohocken, PA, USA: American Society for Testing and Materials; 2002.
[52] Test Method JIS 7086. Testing methods for interlaminar fracture toughness of carbon ber reinforced plastics. Tokyo, Japan: Japan
Industrial Standards, Japanese Standards Association; 1993.
[53] Test Method D6671-01. Standard test method for mixed mode Imode II interlaminar fracture toughness of unidirectional ber
reinforced polymer matrix composites. West Conshohocken, PA, USA: American Society for Testing and Materials; 2002.
[54] Lee SM. An edge crack torsion method for mode III delamination fracture testing. J Compos Technol Res 1993;15(3):193201.
[55] Han T-S, Ural A, Chen C-S, Zehnder AT, Ingraea AR, Billington SL. Delamination buckling and propagation analysis of
honeycomb panels using a cohesive element approach. Int J Fract 2002;115:10123.
[56] Hutter K. The foundations of thermodynamics, its basic postulates and implications. Acta Mech 1977;27:154.
[57] Zienkiewicz OC, Taylor RL. The nite element method, fth ed., vol. 13. Oxford: Butterworth-Heinemann; 2000.
[58] Dvorkin E, Bathe KJ. A continuum mechanics based four node shell element for general nonlinear analysis. Engng Comput
1984;1:7788.
[59] Aliyu AA, Daniel IM. Eects of strain rate on delamination fracture toughness of graphite/epoxy. In: Johnson WS, editor.
Delamination and debonding of materials, ASTM STP, 876, Philadelphia, 1985. p. 33648.
C. Balzani, W. Wagner / Engineering Fracture Mechanics 75 (2008) 25972615 2615

Das könnte Ihnen auch gefallen