Académique Documents
Professionnel Documents
Culture Documents
www.elsevier.com/locate/compscitech
Received 27 November 2001; received in revised form 25 February 2002; accepted 26 March 2002
Abstract
Delamination initiation and growth are analyzed by using a discrete cohesive crack model. The delamination is constrained to
grow along a tied interface. The model is derived by postulating the existence of a maximum load surface which limits the adhesive
forces in the process zone of the crack. The size of this maximum load surface is made dependent on the amount of dissipated crack
opening work, such that the maximum load surface shrinks to zero as a predefined amount of work is consumed. A damage for-
mulation is used to reduce the adhesive forces. Mode I, II and III loading or any combined loading is possible. An analytical
solution is obtained for a single mode opening and the implications of this result on the governing equations is discussed. The
delamination model is implemented in the finite element code LS-DYNA and simulation results are shown to be in agreement with
experimental results.
# 2002 Elsevier Science Ltd. All rights reserved.
Keywords: A. Carbon Fibers; C. Laminates; C. Delamination; C. Crack
then many authors have considered crack and delami- 1.1. Notation
nation growing in one single mode, cf. Refs. [5–9].
Mixed mode I and mode II growth has also been The index notation is not fully adopted. Latin indices
studied by several authors, see e.g. Refs. [10–15]. How- take on the numbers 1, 2, 3 but the summation conven-
ever, there is no direct correlation between the softening tion is not used. If summation is needed, it is written out
behavior and the critical energy release rates, GIC explicitly.
and GIIC. In the work by Crisfield et al. [16] this coupling
exists. Critical energy release rates for mode I and mode
II opening are used in a damage based cohesive zone 2. Problem formulation
model to analyze mixed mode crack growth. Also
Jansson et al. [17] uses the critical energy release rate in As the delamination grows in an interface between
a damage and plasticity based model for the interface. two bodies it would be natural to implement the cohe-
Mixed mode II and III growth is studied by Wisnom sive zone model as a boundary condition between two
and Chang [18] using nonlinear springs to model the bodies. A less formal approach, taken in this work, is to
interface. neglect the presence of the interface when the FE for-
Foulk et al. [15] considers a cohesive zone model for a mulation is derived. This results in separate bodies that
general mixed mode case. In that work the energy does not interact with each other. The proper interac-
release rates are not used explicitly but the amount of tion is then enforced by calculating and adding contact
dissipated energy can be adjusted through other input forces based on the current geometry of the bodies and
parameters. Sprenger et al. [19] uses a similar approach the deformation history of the interface.
where a single critical energy release rate is used in a
failure criteria. Allix et al. [20] has a general mixed mode 2.1. Governing equations
damage formulation. It is not stated explicitly as a
cohesive zone model but has all of its characteristics. Several methods for implementing interfaces exists.
In Borg et al [21] a general cohesive zone model is Here a simple linear penalty formulation will form the
presented which accounts for different critical energy basis for the cohesive zone model. This type of penalty
release rates in all three crack opening modes. This is interface can be interpreted as tying coincident nodes
realized by postulating the existence of a maximum load with three orthogonal springs. However, no spring ele-
surface. This surface controls the reduction of the ments are added to the FE model, only the forces the
adhesive stresses in the interface such that the correct springs exert on it. A local corotational coordinate sys-
energy release rates are obtained. tem is placed in the plane of the delamination at the tip
The present work includes an extension of the model of the crack, see Fig. 1. All properties related to the
by Borg et al. [21], from here on denoted ‘‘the old delamination model are given with respect to this coor-
model’’. The main contribution in the present work is a dinate system.
damage formulation which leads to a simpler and more The initial response in the springs is linear, and the
robust implementation. The special case of mode I force displacement relations can be written
crack opening is also studied in more details.
The delamination model has been implemented in the Pi ¼ ki Di ð1Þ
FE code LS-DYNA [22] using explicit time integration.
The reason for choosing an explicit FE solver was two-
folded. Firstly, an explicit solver is well suited to simu- where Pi are penalty forces, Di are relative displace-
late short duration events, such as low velocity impacts, ments for the coincident nodes, and ki are the penalty
for which this delamination model is developed. Sec- stiffnesses (of the fictitious springs). A positive displace-
ondly, potential pitfalls such as snap-back and snap- ment Di represents opening of the crack, and a positive
through need no special treatment in a dynamic force strives to close the crack.
approach, as opposed to static approaches for which As the coincident nodes are forced apart the adhesive
path following algorithms will be needed. forces will eventually reach a maximum level and then
The governing equations are derived in Section 2 start to decrease. This reduction of the adhesive forces
leading to a set of equations similar to those found in can be accomplished by replacing Eq. (1) with a damage
classical plasticity. The equation governing the damage formulation
evolution is further discussed in Section 3. The imple-
mentation is described in Section 4 and some limitations Pi ¼ ð1 !i Þki Di ð2Þ
of the governing equations are discussed in Section 5.
Finally, in Section 6 simulation results obtained with where ki are the initial stiffnesses and where !i 2 ½0; 1
the current model are compared with experimental are damage variables. The value !i ¼ 0 corresponds to
results. undamaged adhesive properties in direction i, while
R. Borg et al. / Composites Science and Technology 62 (2002) 1299–1314 1301
In the present formulation it is not necessary to solution at time tnþ1 . Iterations within one time incre-
determine in what direction the crack will grow, since it ment are denoted with a left superscript r on the vari-
is known. It is controlled by the interface geometry. able at hand, rxn . The discrete counterparts of the
However, it is not known in what mode the crack governing equations read
growth will take place. Using the ideas put forward by
Wu [23] may have been a good way to determine crack Pinþ1 ¼ 1 !nþ1 i ki Dinþ1
: nþ1 :
growth mode but, as shown in the Appendix, this leads !i ¼ lnþ1
to an erroneous result.
F nþ1 ¼ f Pinþ1 1 g G nþ1 i ð12Þ
Instead the generality of the proposed model is : nþ1 :
nþ1 nþ1 nþ1
reduced and a simpler isotropic damage model is tacitly F 40 l 50 F l ¼0
assumed. Ð tnþ1 : 0 Pinþ1 Dinþ1
nþ1
Gi ¼ 0 Pi D i dt
2
: :
!i ¼ l ð10Þ
where
( Here 0 , 1 , 2 and 3 are positive material para-
@F X 3
@f meters. Note that only the positive part of the normal
: ¼ tki D nþ1 þ
nþ1 i
@l nþ1 @P force, P1C is accounted for. This is a simple way of
i¼1 i
: nþ1 2 ! ) treating compressive contact and it implies that the
@g tDi n : D elastic region is independent of P1 when P1 < 0. The
3Di þ 2tDi þ i
@Gi nþ1 6 2 forces P2 and P3 corresponds to modus II and III crack
: growth. They can take on both positive and negative
Dinþ1 ¼ Din þ tDi ð17Þ values and as i may be any positive number, the abso-
lute value of these forces are used. It is to be noted that
: the dependence of friction is not accounted for in the
Knowing lt and ignoring higher order terms : it is present setting. Friction can be accounted for by incor-
possible to obtain a better approximation for lnþ1 . The porating a Coulomb frictional law into f.
corrector step is repeated until a converged solution is The dissipated work in the contact is normalized
achieved similar to f
2
F nþ1 ¼ Fl:nþ1 ¼rþ1 l: ¼ ! 1 ! 2 ! 3 3 10
G 1 G 2 G 3
: @F
: g¼4 þ þ 5 ð20Þ
Fl:nþ1 ¼r l: þ rl : þO l2 ¼ 0 ð18Þ G 1C G 2C G 3C
: :
@lnþ1 l nþ1 ¼r l
where
: : : Box 2. Elastic predictor
r
l ¼rþ1 l r l 8 8
>
> < gnþ1 ¼1
<
(1) 1 gn < tholg ?
yes ! pnþ1 ¼0
: i
The fourth step in the algorithm is a ‘‘back door’’ >
> DONE
:
designed to force the iterations to stop when the damage t no ! 2:
parameter is getting close to unity. Thus, even before !i ¼ !ni
(2)
calculating the elastic trial state the outcome of the Pti ¼ ð1 !ti Þki Dnþ1
i
previous timestep is checked. If the adhesive forces are
close to failure they will be considered failed. This fail- (3) Ft ¼ FðPti ; Dnþ1
i Þ
ure is defined by gðG ni Þ > tholg , where tholg is close to 8 8 nþ1
>
> >
> g ¼ gn
unity. > >
>
> > !
< i
nþ1
¼ !ti
The complete algorithm is summarized in Boxes 2 and >
<
yes ! Pnþ1 ¼ Pti
3. Note that the force calculations in Boxes 2 and 3 are (4) Ft < 0 >
> :i
represented in their simplified version given in Eq. >
> > lnþ1
> ¼0
>
> :
(2) while Eq. (3) is used in the implemented algo- >
> DONE
:
rithm in order to obtain the proper contact inter- no ! BOX3
action.
1304 R. Borg et al. / Composites Science and Technology 62 (2002) 1299–1314
where 0 , 1 2 and 3 are positive material parameters. gram the this type of input most conveniently given in
Other possible expressions for g found in the literature area related units instead:
are summarized in Greenhalgh et al. [24], all of which
may not lend themselves to the normalized form nee- PiC !
iC , Critical Stresses
ded in the definition of the current delamination G iC ! GiC , Critical energy release rates in mode
model. I, II and mode III crack opening, respectively.
With the functions f and g chosen the model is com- ki ! kA i , Stiffnesses per unit area
pletely defined. The following user input is needed.
The choice of the functions f and g must be made with
0 ; i ; Material parameters for the function f caution. Not only must each of them reflect experi-
0 ; i , Material parameters for the function g mental results but, as will be shown in this section, they
PiC , Critical forces must also be consistent to each other.
G iC ; Amount of dissipated energy in mode I, II
and mode III crack opening, respectively 5.2. Single mode response
ki ; Spring stiffnesses
Considering the case of a prescribed monotonically
Note that some of the input to the model are in terms increasing displacement in mode I crack opening, the
of nodal related units. When implemented in a FE pro- response is initially elastic, but when P reaches PC
(1) r¼0
r
: :
(2) l ¼ ln
r
:
(3) !i ¼ !ni þ tr l
(4) r
Pi ¼ ð1 r !i Þki Dnþ1
i
(5) r
F ¼ Fðr Pi ; Dnþ1
i Þ
8 8 nþ1
>
> >
> g ¼ gðr Pi ; Dnþ1
i Þ
>
> >
> !nþ1
>
> < i ¼r !i
<
yes ! Pnþ1 ¼r P i
(6) jr Fj < tholF ? >
>
i
: nþ1 :
>
> >
> l ¼r l
>
> :
>
> DONE
:
no ! 7:
8 : r
>
>
r
l ¼ r F=ð@ð :FÞÞ
>
> @l :
>
< rþ1 !i ¼r !i þ tr l; rþ1 !i ¼ minð1; maxð0; rþ1 !i ÞÞ
(7) rþ1
Pi ¼ ð1 rþ1 !i Þki Dnþ1
>
> : r : r :
>
> rþ1
l ¼ l þ l
>
:
r¼rþ1
(8) ! 5:
R. Borg et al. / Composites Science and Technology 62 (2002) 1299–1314 1305
damage will start to grow. Damage growth continues Note how the choice 0 ¼ 1 to a great extent sim-
until a zero force is reached. Under these conditions plifies this expression.
damage is a function of the prescribed displacement, With the initial condition D(P=PC)= PkC the solution
! ¼ !ðDÞ, and consequently the adhesive force must of Eq. (23), obtained by the mathematical software
also be a function of the displacement, P ¼ PðDÞ. Maple [25], is
In order to obtain an analytical solution for the adhe- !
sive force, the parameters for the function g are chosen as 1 2G C 2G C 1
D¼ þ P P ð24Þ
0 ¼ 1 . The governing equations for mode I crack k 2 PC 2 2 P C
opening, excluding the initial elastic phase, are then
Fig. 2. The shape of the softening curve for different , analytical solution using k=2, PC=1, G C =2 and =1.
1306 R. Borg et al. / Composites Science and Technology 62 (2002) 1299–1314
Fig. 3. The shape of the softening curve for different values, numerical solution using k=2, PC=1, G C ¼ 2 and ¼ 1.
the performance of the numerical scheme and its imple- recorded. Now only D3 is set to zero while D1 and D2
mentation. A comparison of the results in Figs. 3 and 2 are monotonically increased.
shows no unexpected disagreements. Defining ¼ arctanðD2 =D1 Þ as a measure of mode
mixity, where ¼ 0 corresponds to pure mode I load-
5.3. Mixed mode response ing. Several simulations with constant but different
are performed for each of the setups. The force–dis-
Opposed to the single mode response there is no ana- placement relation P1 =D1 for setups 1 and 4 are recor-
lytical solution available for the mixed mode case. Thus ded and presented in Figs. 4 and 5. The linear response
it is not possible to give a mathematical proof regarding in Fig. 2 ( ¼ 1) remains linear for all used, see Fig. 4.
the validity of the model behavior subjected to mixed The unphysical behavior using ¼ 4 presists in the
mode loading. Instead numerical simulations are used mixed mode response for all used, see Fig. 5. Clearly
to examine the model response. the findings for the single mode response appears to
In order to compare the mixed mode response with hold for the mixed mode response as well. Results for
the single mode response similar setups are studied, see the setups 2, 3, 5 and 6 are not presented but agree with
Table 1. Identical data is used for all three opening this conclusion.
modes. The only difference between the examined setups
are the 0 and i value. Setups 1–3 are comparable to 5.4. Conclusion
single mode loading with ¼ 1 while setups 4–6 are
comparable to single mode loading using ¼ 4. Based on the findings in the first part of this section a
The simulations of the mixed mode response are per- reliable choice of parameters for f and g in single mode
formed similar to the simulation of the single mode response is ¼ 1 and ¼ 1. Then the well known
response. Prescribed displacement histories are input to bilinear response is retained. Other shapes for the soft-
the implemented algorithm and the output forces are ening portion of the response can be obtained by using
< 1, but for really small values of the response
makes the model useless. The choice > 1 also gives an
unphysical response. Although not proved analytically
Table 1
Setup, mixed mode case
the same conclusions appears to hold for the mixed
mode response as well.
Setup 0 i 0 i ki PiC GiC The basic conclusion to be drawn is that certain
1 2 2 2 2 2 1 2 choices of parameters in f and g may lead to completely
2 3 3 2 2 2 1 2 unphysical responses! If one set of f and g has this flaw
3 6 6 2 2 2 1 2 it is reasonable to assume that others have it too. Thus,
4 0.5 2 2 2 2 1 2 if an analytical solution is unavailable for a specific set
5 0.75 3 2 2 2 1 2
6 1.5 6 2 2 2 1 2
of f and g, extensive numerical testing of the imple-
mented algorithm is mandatory.
R. Borg et al. / Composites Science and Technology 62 (2002) 1299–1314 1307
Fig. 6. Experimental setup; prescribed dosplacement I gives mode I (DCB), II gives mode II (ENF) and x gives mixed mode opening (MMB). a0 Is
the initial crack length.
A refined mesh was used in the region where the crack parameters 0 , i will effect the simulation but due to
was expected to propagate. The mesh was only refined the limited amount of experimental results they are
in the direction of crack growth, not in the width direc- chosen to be equal, i.e. 0 ¼ i .
tion. The FE model used in the simulation of the DCB Juntti et al. [28] reported the critical energy release
experiment is shown in Fig. 7. rate GC as a function of GII/G. These experimental data
The experiments are quasi static in nature with a are plotted in Fig. 8. A least square fit of Eq. (20) with
velocity of the prescribed displacements in the order of respect to the experimental data gave 0 ¼ i ¼ 1:23.
mm/min. If this load rate had been used in the simula- The delamination experiments were designed to study
tions, the timestep limitations inherent to the explicit and record mode I and mode II crack growth, although
FE formulation would have rendered unacceptable mode III crack growth may be present on a local scale.
long computing time. Instead the load rate was However, the amount of mode III crack growth in an
increased such that the simulations terminated within a ENF test specimen of similar design was reported in
reasonable time limits, while at the same time the Schön et al. [29], to be small. The effect of any mode III
amount of kinetic energy in the model was kept small crack growth, in the current simulation is therefore
with respect to the total energy. Mass scaling was not assumed to be small. No experimental data was avail-
used. able for mode III crack growth but by adopting the
In Borg [21] it was reported that the response of the assumption above, it was decided to use identical data
delamination model was insensitive to the critical stress. both for mode II and mode III input.
Thus the response should also be insensitive as to how The initial stiffness in the interface was selected based
the material parameters 0 , i are chosen and we set on the geometry and stiffness of the test specimen. Cri-
them all equal to 2. On the other hand, the material tical stresses for the interface were selected based on the
results in the old model, see Borg et al. [21]. Input to the
delamination model is summarized in Tables 3 and 4.
Table 2
Test specimen data
6.2. Simulations and results
Specification
Material: HTA/6376C The performance of the delamination model was
Layup: [012//(5/04)s]a
Geometry: 150203.1 (mm)
investigated by comparing results from the simulations
Initial crack length: 35 (mm) with experimental data. Two types of data were exam-
GIC: 259 (N/m) ined; the crack extensions as functions of prescribed
GIIC: 1008 (N/m) displacements, and the loading forces as functions of
Elastic constants prescribed displacements.
E11=146 GPa G12=5.25 GPa v12=0.3 The amount of crack growth in the experiments was
E12=10.5 GPa G13=5.25 GPa v13=0.3 determined by visual inspection. In the cohesive zone
E33=10.5 GPa G23=3.48 GPa v23=0.51
model used in the simulations there is no information
a
‘‘//’’ is the location in the stacking sequence where the initial crack on the exact location of the crack tip. Instead, the loca-
is located. tions where the normalized dissipated work g, see Eq.
R. Borg et al. / Composites Science and Technology 62 (2002) 1299–1314 1309
(20), takes on the values 0.25, 0.50 and 0.75 were used as ment relation, is filtered explicitly using an SAE 1000
indicators for the location of the delamination front. Hz high band filter included in the FE post processor
Computing crack extension along the test specimen LS-POST [30].
involves averaging data in the width direction. This Figs. 9 and 10 show the results for the DCB setup.
averaging acts as a filter and smooths this result com- The critical stresses
iC ¼ 20 106 Pa was used here. As
ponent. The other result component, the force displace- can be seen in Fig. 9 the model predicts the onset of
Fig. 8. Experimental results: critical energy release rate as a function of the degree of mixed mode.
1310 R. Borg et al. / Composites Science and Technology 62 (2002) 1299–1314
Fig. 9. Crack length as a function of applied displacement for the DCB setup.
be as stiff as in the experiments. This may also be the loading in the experiment has been shifted to the right in
explanation for the deviation found in Fig. 9. Fig. 12. This may be due to loosely fitted joints or some
The results for the ENF setup are shown in Figs. 11 similar problems in the experimental setup.
and 12. The critical stresses
iC ¼ 40 106 Pa was used The results for the MMB setup are shown in Figs. 13
here. With
iC ¼ 20 106 Pa the cohesive zone was too and 14. The critical stresses
iC ¼ 40 106 Pa was used
large. As can be seen from the crack growth recorded in here. In this experiment the crack initially grows a small
the experiment, this is an almost transient phenomena distance and is then arrested. This unstable behavior is
where the crack grows 15 mm in one single jump. Time probably due to some imperfection in the test specimen
scaling is used in the simulation causing the inertia of and should not be expected in a simulation. Similar to
the test specimen in the FE-model to increase, leading to the ENF setup the rapid crack growth in the simulation
a slower response than in reality, which is obvious both is smoother than in the experiment but the maximum
in Fig. 11 and in Fig. 12. However the onset of crack force and crack onset are well predicted.
growth is predicted quite well, cf Fig. 11. Also the initial The simulation results presented here are very similar
elastic loading as well as the maximum load is also pre- to the results presented in Borg et al. [21] which were
dicted with good agreement with the experiment, cf. based on a similar delamination model but with a dif-
Fig. 12. Note that the curve associated with the elastic ferent formulation.
Fig. 11. Crack length as a function of applied displacement for the ENF setup.
7. Discussion and conclusions in the way that the present model and the hypothesis of
maximum energy release rate are not quite compatible.
The aim of this work has been to develop a mathe- For simplicity, a simple isotropic damage growth law
matical model for simulating delamination under simple was instead suggested.
and combined loading. The main ideas in the old model, A vital part of the governing equations are the two
see Borg et al. [21] are retained in the present work, functions f and g. Numerical and analytical studies for
where in the latter the governing equations are based on single and mixed mode loading reveal that these func-
a damage formulation. tions must be carefully chosen. It is not sufficient to
In the damage formulation context, the governing study them individually, since their interaction may yet
equations take on a form similar to that of classical yield an unphysical behavior. Unless analytical solu-
plasticity. An attempt has been made to derive a tions are available, extensive numerical testing is
damage growth law by using the hypothesis of max- advised to validate the model.
imum energy release rate, but the so obtained formula- The simulations of DCB, ENF and MMB experi-
tion was found non-realistic. This should be interpreted ments all show good agreement with experimental
Fig. 13. Crack length as a function of applied displacement for MMB setup.
The total dissipation is the sum of the dissipation in Now, let us study a hypothetical two dimensional
each of the contact force components. situation. The consistency condition requires that the
:
projection of ! on the vector b to be constant, i.e. that
:
‘‘allowable !’’ are given by a line perpendicular to the
vector b, see Fig. A1. In general the vectors a and b are
:
not parallel. Thus, the further ‘‘away’’ ! is moved from
b towards and past a, the larger its projection will be on
a. As the damage growth rate must be positive in all its
Fig. A1. Determination of the damage growth rate. Fig. A2. Force response during load in 2D.
1314 R. Borg et al. / Composites Science and Technology 62 (2002) 1299–1314
:
components the optimal direction of ! is eventually [9] Wisnom MR. Modelling the effect of cracks on interlaminar
determined by one of the unit vectors. shear strength. Composites Part A 1996;27(1):17–24.
[10] Needleman A. A continuum model for void nucleation by
It is thus concluded that damage will only grow in one
inclusion debonding. Journal of Applied Mechanics 1987;54:
deformation mode at a time. When the first deformation 525–31.
mode has been exhausted damage will start to grow in a [11] Tvergaard V. Effect of fiber debonding in a whisker-reinforced
second deformation mode etc. This will continue until metal. Materials Science and Engineering, A125, A125 1990:203–
all springs have failed. Plotting the forces P1 and P2 15.
during such loading would give the qualitative response [12] Needleman A. Numerical modeling of crack growth under
dynamic loading conditions. Computational Mechanics 1997;
shown as Path A in Fig. A2. 19(6):463–9.
The exact material behavior at the tip of a crack is [13] Bolzon G. Hybride finite element approach to quasi-brittle frac-
complex. Nevertheless, it is reasonable to assume that ture. Computers and Structures 1996;60(5):733–41.
the same volume of matter transfers both normal and [14] Allen DH, Lo DC, Zocher MA. Modelling of damage evolution
tangential loads. Thus, if the material has failed com- in laminated viscoelastic composites. Internaltional Journal of
Damage Mechanics 1997;6(1):5–22.
pletely in say the normal direction, it should no longer [15] Foulk JW, Allen DH, Helms KLE. Formulation of a three-
be able to transfer any loads in the tangential direction. dimensional cohesive zone model for application to a finite ele-
Following this argumentation, the behavior of the pre- ments algorithm. Computer Methods in Applied Mechanics and
sent model, being derived so far, is regarded as unrea- Engineering 2000;183(1–2):51–67.
listic. [16] Crisfield M, Mi Y, Davies GAO, Hellweg HB. Finite element
methods and the progressive failure-modelling of composites
Instead the simplest possible assumption, isotropic structures. In: Owen DRJ, Onate E, Hinton E, editors. Compu-
damage, is adopted. Thus damage growth is equal in all tational plasticity. Barcelona; 1997. p. 239–54.
force components at all times: [17] Jansson N, Larsson R, Toll S. Finite element modelling of dela-
mination progression under mixed mode loadings. In: Fiber
: : reinforced composites 2000, composites for the millenium. New-
!i ¼ l ðA:6Þ
castle: 2000.
[18] Wisnom MR, Chang FK. Modelling of splitting and delamina-
For a proportional displacement driven loading, the tion in notched cross-ply laminates. Composites Science and
initial increase and subsequent decrease of adhesive Technology 2000;60(15):2849–56.
forces will then take place along the same path, see Path [19] Sprenger W, Gruttmann F, Wagner W. Delamination growth
analysis in laminated structures with continuum-based 3d-shell
B in Fig. A2. It is finally to be noted that, even though
elements and a viscoplastic softening model. Computer Methods
the dissipation maximization approach was abandoned, in Applied Mechanics and Engineering 2000;183(1–2):123–39.
Eq. (A.6) is directly seen to be in agreement with the [20] Allix O, Lévêque D, Perret L. Identification and forecast of
second principle of thermodynamics [see Eqs. (A.1) and delamination in composite laminates by an interlaminar interface
(A.2)]. model. Composites Science and Technology 1998;58(5):671–8.
[21] Borg R, Nilsson L, Simonsson K. Simulation of delamination in
fiber composites by the use of a discrete cohesive-failure model.
Composites Science and Technology 2001;61(5):667–77.
References [22] Hallquist JO. LS-DYNA theoreticla manual. Livermore: Liver-
more Software Technology Corporation; 1998.
[1] Borg R. An orthotropic material model with damage. Licentiate [23] Wu CH. Fracture under combined loads by maximum energy
thesis, LIU-TEK-LIC-1998: 38. Linköping: Linköping Uni- release rate criterion. Journal of Applied Mechanics 1978;45:553–
versity; 1998. 8.
[2] Dugdale DS. Yielding of steel sheets containing slits. Journal of [24] Greenhalgh G, Asp L, Singh S. Delamination resistance, failure
the Mechanics and Physics of Solids 1960;8:100–4. criteria and fracture morphology of 0 /0 , 0 /5 and 0 /90 ply
[3] Barenblatt GI. The mathematical theory of equillibrium cracks in interfaces in CFRP. In: 5th International Conference on Defor-
brittle fracture. Advances in Applied Mechanics 1962;7:55–129. mation & Fracture of Composites, London, 1999.
[4] Hillerborg A, Modéer M, Petersson P-E. Analysis of crack for- [25] Char BW, Geddes KO, Gonnet GH, Leong BL, Monogan MB,
mation and crack growth in concrete by means of fracture Watt SM. First leaves: a tutorial introduction to Maple V. New
mechanics and finite elements. Cement and Concrete Research York: Springer-Verlag; 1993.
1976;6:773–82. [26] Ottosen NS, Olsson KG. Hardening/softening plastic analysis of
[5] Andersson H, Bergkvist H. Analysis of a non-linear crack adhesive joints. Journal of Engineering Mechanics 1988;114(1):
model. Journal of the Mechanics and Physics of Solids 1970; 97–116.
18(1):1–28. [27] Asp LE. The effects of moisture and temperature on the inter-
[6] Nilsson, L., Oldenburg, M. Nonlinear wave propagation in plas- laminar delamination toughness of a carbon/epoxy composite.
tic fracturing materials—a constitutive modelling and finite ele- Composites Science and Technology 1998;58(6):967–77.
ment analysis. In: Nigual U, Engelbrecht J, editors. Nonlinear [28] Juntti M, Asp LE, Olsson R. Assessment of evaluation methods
deformation waves. Berlin: Springer; 1983. for the mixed-mode bending test. Journal of Composites Tech-
[7] Tin-Loi F, Li H. Numerical simulations of quasibrittle fracture nology and Research 1999;21(1):37–48.
processes using the discrete cohesive crack model. International [29] Schön J, Nyman T, Blom A, Ansell H. Numerical and experi-
Journal of Mechanical Sciences 2000;42(2):367–79. mental investigation of a composite enf-specimen. Engineering
[8] Cui W, Wisnom MR. A combined stress-based and fracture- Fracture Mechanics 2000;65:405–33.
mechanics-based model for predicting delamination in compo- [30] LS-POST, a new post-processor from LS-DYNA. Livermore:
sites. Composites 1993;24(6):467–74. Livermore Software Technology Corporation, 1999.