Sie sind auf Seite 1von 6

Composite Structures 71 (2005) 429–434

www.elsevier.com/locate/compstruct

Delamination modelling in a double cantilever beam


M. Meo *, E. Thieulot
Crashworthiness, Impact and Structural Mechanics Group, College of Aeronautics, School of Engineering, Cranfield University, Cranfield,
Bedfordshire MK43 0AL, UK

Abstract

The simulation of the delamination process in composite structures is quite complex, and requires advanced FE modelling techniques.
Failure analysis tools must be able to predict initiation, size and propagation of delamination process. The objective of the paper is to
present modelling techniques able to predict a delamination in composite structures. Four different ways of modelling delamination
growth of a double cantilever beam test (DCB) are proposed. The first two approaches were based on a cohesive zone model: the inter-
face being represented either by using delamination elements or non-linear springs. The idea of the third approach was to use a fracture
mechanics criterion, but to avoid the complex moving mesh techniques it often implies. The interface between the two layers was sim-
ulated with solid elements representing the matrix, which were eliminated when their energy release rate exceeded the critical value. The
energy release rate was computed using the virtual crack closure technique (VCCT). In the last approach, the interface behaviour was
modelled by a tiebreak contact. Coincident nodes were tied together with a constraint relation and remained joined, until when the max-
imum interlaminar stresses was reached. Once this value was exceeded, the nodes associated with that constraint were released to sim-
ulate the initiation of delamination. The comparison of the results of the first three modelling techniques with experimental data showed
that very good correlation was achieved. Poor results were obtained using tiebreak contact. It was due to the criterion used, since when
the critical interlaminar stress was reached, the delamination was experienced before the critical energy release rate was reached.
 2005 Elsevier Ltd. All rights reserved.

Keywords: Delamination; Virtual crack closure technique; Tiebreak contact

1. Introduction progressive debonding or delamination. Our attention


was focused on the simulation of delamination growth in
Delamination is one of the predominant forms of failure a double cantilever beam test. The different results
in laminated composites due to the lack of reinforcement in obtained with all four modelling techniques and the com-
the thickness direction. Delamination as a result of impact parison between them is presented below, showing the
or a manufacturing defect can cause a significant reduction capability of the proposed techniques to study both the
in the compressive load carrying capacity of a structure. delamination initiation and delamination propagation.
The fracture process of high performance composite lami-
nates is quite complex, involving not only interlaminar 2. Double cantilever beam test results
damage (delamination), but also intralaminar damage
mechanisms (e.g., matrix cracking, fiber fracture). For The problem analysed was a double cantilever beam
effective predictive capabilities, progressive failure analysis (DCB) test used to determine mode I toughness. The DCB
tools for all modes of failure are needed. The objective of test specimen was made of a unidirectional fiber-reinforced
this paper is to present modelling techniques to simulate laminate containing a thin insert at the mid-plane near the
loaded end. The specimen was 0.185-m long, 0.025 m wide,
*
Corresponding author. Tel.: +44 1234 750111x5220; fax: +44 1234
with two 2.5-mm thick plies, and with an initial crack length
752149. of 55 mm. The DCB test specimen was made (0)24, from a
E-mail address: m.meo@cranfield.ac.uk (M. Meo). carbon fiber-reinforced polyetheretherketone (C/PEEK)

0263-8223/$ - see front matter  2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compstruct.2005.09.026
430 M. Meo, E. Thieulot / Composite Structures 71 (2005) 429–434

Table 1 layers begin to separate. The damage accumulated at


Properties for APC-2/AS4-CFRP the interface is represented by a variable d, which is
E11 E22 = E33 G12 = G13 G23 m12 m13 m23 equal to zero when there is no damage and reaches 1
135 GPa 9 GPa 5.2 GPa 1.9 GPa 0.34 0.34 0.46 when the material is fully damaged: r = (1  d)k1bd;
• d > dF: decohesion part: decohesion of the two layers is
complete: there is no more bond between the two layers,
laminate (APC-2/AS4). The laminate material properties the traction across the interface is null. If crack closure is
are reported in Table 1. detected, interpenetration is prevented by reapplying
only the normal stiffness k1b.
3. The cohesive zone model
The area under the curve is the fracture energy GC for a
Delamination growth can be simulated by of placing particular mode. The delamination element implemented in
interfacial decohesion elements between composite layers. DYNA3D [1,2] consists of an 8-node element with zero-
A decohesion failure criterion that combines aspects of thickness. It connects two opposite shell elements, and its
strength-based analysis and fracture mechanics was used behaviour is based on the constitutive equations for the
to simulate debonding by softening the element. The inter- bilinear softening model presented previously. The element
face is characterized by constitutive equations which relate stiffness matrix is calculated using the constitutive equa-
the applied stress r to the relative displacement at the inter- tions of the softening bilinear model. Full details of the ele-
face d. The stress–relative displacement curve is divided ments and its implementation in the FE code can be found
into three main parts (Fig. 1) and the constitutive equations in [2].
are the following: The two layers of the DCB were modelled with shell ele-
ments, and the bond between these two layers was mod-
• d < d0: elastic part: the traction across the interface elled using delamination elements. The mesh is shown in
increases until it reaches a maximum, and the stress is Fig. 2. The specimen was clamped at one end, and loads
linked to the relative displacement via the interface stiff- were applied at the other end. In order to damp the vibra-
ness k1b: r = k1bd; tion, the dynamic relaxation option was used [1]. This
• d0 < d < dF: softening part: the traction across the inter- option allowed an explicit solver to conduct a static analy-
face decreases until it becomes equal to zero: the two sis by increasing the damping until the kinetic energy drops

Mode I Stiffness- k1b


2 k1b =E/h
σ
(1-d) k1b 5.5e+12 N/m

Initiation of damage- δo

7.E-6 m
δF δ
1 δo Final displacement - δF
3

Fig. 1. Bilinear softening model.

Fig. 2. (a) Deformed shape and (b) force–displacement curve of cohesive zone model.
M. Meo, E. Thieulot / Composite Structures 71 (2005) 429–434 431

to zero. In order to define the element constitutive equa- d = 4.6 mm, whereas the maximum load predicted is
tion, the penalty parameter and the interlaminar tensile 65.5 N for d = 4.94 m. The unloading response is well
strength had to must be determined. In particular, the reproduced by the numerical model, validating the pro-
properties required to define the bilinear interfacial posed interface modelling technique.
softening behaviour are the initial stiffness (penalty) k1b,
the fracture energies GI, and the corresponding nominal 5. Birth and death elements approach
interlaminar tensile strengths, S cr
1 . The accuracy of the anal-
ysis depends on the penalty stiffness k1b that is chosen. Too The idea of this approach was to use a fracture mech-
low of a value leads to an inaccurate representation of the anics criterion, but to avoid the complex moving mesh
mechanical behaviour of the interface, whereas high values techniques it often implies. The model was constituted of
will increase run-time and can promote numerical errors. three layers of solid elements. The upper and lower layer
Since the interface is a resin rich zone of small thickness, represented the two composite unidirectional layers, and
t, the penalty stiffness was defined as k1b = E33/t, where the mid layer was the interface as shown in Fig. 4. To sim-
E33 is the elastic modulus of the resin rich zone. The ulate delamination growth, the elements of the interface
parameters used to identify strain softening constitutive were eliminated when a user-defined criterion was reached.
models are reported in Fig. 1. The elements, representing the interface were deleted, when
In Fig. 2a, initiation and propagation of damage in the their energy release rate became greater than a critical
composite DCB is shown [2]. Fig. 2b shows that good value. To achieve this, the ‘‘Birth and Death’’ option pro-
agreement between the experimental data and the numeri- vided by ANSYS FE code was utilised. This option can be
cal predictions was obtained. The averaged maximum load used to deactivate or reactivate selected elements in cer-
obtained in the experiments is 65 N, whereas the maximum tain cases. The killed elements were not actually removed
load predicted is 68 N. The unloading response is well but they were deactivated by multiplying their stiffness by
reproduced by the numerical model, validating the unload- a severe reduction factor. The energy release rate was
ing behaviour of the constitutive equation proposed. computed using the virtual crack closure technique
(VCCT) [3]. The method evaluates the energy release rate
4. Non-linear springs G and compares it to the critical energy release rate GIC.
The virtual crack closure technique utilizes the product
This approach was similar to that used in delamination of nodal forces and the difference in nodal displacements
elements: the interface behaviour was still represented by a to calculate the energy release rates components for each
softening model. The two layers were modelled by shell ele- fracture mode. The method is based on two main
ments, and the interface between the layers was modelled assumptions:
using non-linear springs. This element is defined by two
node points and a generalized force-deflection curve. Same
mesh and bilinear softening model of the delamination ele-
ments were employed, in order to be able to compare the
results. For delamination elements, the behaviour of the
interface was modelled by stress versus displacement; in
this case there was the need to implement a force versus dis-
placement curve. The results in Fig. 3b show that the
results obtained with the FE simulation were in quite good
agreement with the experimental ones. The averaged max-
imum load obtained in the experiments is 65 N for Fig. 4. Description of the birth and death elements model.

Fig. 3. (a) Deformed shape and (b) force–displacement curve for the non-linear springs model.
432 M. Meo, E. Thieulot / Composite Structures 71 (2005) 429–434

• The energy released (DE) when the crack is extended 1


G ¼ GI ¼ F w ð3Þ
from (a + Da) to (a + 2Da) is identical to the energy Da  b
required to close the crack between location i and k, In order to use the same properties of the interface as in the
and the forces required to close the crack are identical other modelling strategies, the Youngs modulus of the
to the forces acting on the upper and lower surfaces of interface was determined as proposed by [4]
the closed crack.
E3 ¼ K p t ð4Þ
• A crack extension from (a + Da) to (a + 2Da) does not
significantly alter the state at the crack tip. Therefore, where t is the thickness of the interface and Kp the stiffness
the displacements, behind the crack tip at node i, are used for the delamination elements. With the mesh used for
approximately equal to the displacements behind the the other methods, Da = 0.15/63 and b = 0.02/7. A non-
original crack tip at node ‘. linear analysis was performed and at the end of each load
step, the value of the energy release rate was calculated and
In a three-dimensional finite element model, the delam- compared with the critical energy release rate of the inter-
ination of length a is represented as a two-dimensional face. If the criterion was exceeded, the elements were elim-
discontinuity by two surfaces. The mode I, mode II and inated. Given the method used to calculate the energy
mode III components of the strain energy release rate are release rate, only one line of elements could be deleted at
calculated as follows: each analysis step. For this reason, the number of steps
1 had to be sufficient high to allow the delamination to grow:
GI ¼  Z Li  ðwL‘  wL‘ Þ; for example, if there are only five steps, then only five lines
2DA
1 of elements can be deleted and it may be insufficient to
GII ¼  X Li  ðuL‘  uL‘ Þ; completely simulate delamination growth in the specimen.
2DA
1 In our simulation, 40 steps were used. The force–displace-
GIII ¼  Y Li  ðvL‘  vL‘ Þ ð1Þ ment curve, shown in Fig. 6b, show that the results ob-
2DA
tained with the FE simulation were in quite good
where DA = Da Æ b, Da is the length of the elements at the agreement with the experimental ones. The adopted model-
delamination front and b is the width of the elements, ling approach was clearly capable of describing the first
XLi, YLi, ZLi, denote the forces at the delamination front linear part of the force displacement curve. However, the
in column L, row i as shown in Fig. 5. The corresponding peak of the load is higher compared to the experiment.
displacements behind the delamination at the top face node The maximum load obtained in the experiments is 65 N
row ‘ are denoted uL‘, vL‘, wL‘, and at the lower face row ‘* for d = 4.6 mm, whereas the maximum load predicted is
are denoted uL‘ , vL‘ , wL‘ . 71 N for d = 4.5 mm. The unloading response is well repro-
The calculation of the energy release rate of the elements duced by the numerical model, validating the proposed
of the interface for the double cantilever was performed interface modelling technique. The local loading and
according to the following equation: unloading in the softening zone of the force displacement
1 curve, are due to the crack growth (elements are elimi-
G ¼ GI ¼  Z Li  ðwL‘  wL‘ Þ ð2Þ nated). Sensitivity analysis of the material properties of
2Da  b
the interface showed that the slope of the linear part of
Due to the symmetry of the specimen wL‘ = wL‘* = w and
the force displacement curve is dependent on the Youngs
then, using the notations defined in Fig. 5:
modulus of the interface in the through-the-thickness direc-
tion, while the peak load is dependent on the longitudinal
Youngs modulus.
Local crack tip system z’, w’, Z’

6. Tiebreak contact

In this approach, the interface behaviour was modelled


b using a tiebreak contact [5]. The coincident nodes of the
wLl
vLl two layers are tied together with a constraint relation and
Zu,
uLl Yu x’, u’, X’ remained joined, until when the maximum interlaminar
wLl*
l vLl* Xu stresses was reached. Once this value was exceeded, all
uLl* nodes associated with that constraint were released to sim-
ulate the initiation of delamination. The failure of the inter-
l'
face was defined by the following equation:
z,w,ZX y,v,Y
 m  m
jrn j 1 jrs j 2
þ P1 ð5Þ
rn;fail rs;fail
a ∆a ∆a
Global system
x,u,XX
where rn and rs are the computed normal and shear stres-
Fig. 5. Virtual crack closure technique for solid elements [3]. ses, rn,fail and rs,fail are the tensile and shear failure stress.
M. Meo, E. Thieulot / Composite Structures 71 (2005) 429–434 433

Fig. 6. (a) Deformed shape and (b) force–displacement curve for the birth and death elements model.

Fig. 7. (a) Deformed shape and (b) force–displacement curve for the tiebreak contact model.

In our case, there was only mode I crack opening (rs = 0). below its critical value. This explains why the maximum
The two layers were modelled using solid elements. The load was very low compared to the experiment.
mesh was the same as previously described. The results ob-
tained with this modelling strategy were not able to de- 7. Comparison of the different methods
scribe the experimental force displacement curve (Fig. 7).
This is mostly due to the fact that the criterion was based The comparison of the experimental data with the four
on maximum interlaminar shear stress and not critical en- modelling strategies is shown in Fig. 8. The force–displace-
ergy release rate, therefore, when the interlaminar stress ment curve obtained for delamination elements is quite
reached its maximum value, the energy release rate was still similar as the one obtained for non-linear springs, since

80 Delamination elements
Tiebreak contact
70 Springs
Experimental results
60 Birth and Death Elements
50
Force (N)

40

30

20

10

0
0 0.0025 0.005 0.0075 0.01 0.0125 0.015
Displacement (m)

Fig. 8. Comparison of the force–displacement curves obtained for each of the four methods.
434 M. Meo, E. Thieulot / Composite Structures 71 (2005) 429–434

the behaviour of the interface is the same. Using the third elements was quite similar as the one obtained for non-lin-
method, since delamination growth occurs when the energy ear springs, since the behaviour of the interface is the same.
release rate exceeds a critical value, similar results to the Using the third method, since delamination growth occurs
first two modelling techniques were achieved. In particular, when the energy release rate exceeds a critical value, similar
the birth and death elements methodology was capable of results to the first two modelling techniques were achieved.
capture the linear elastic behaviour of the force displace- The last modelling techniques simulate the interface with a
ment curve, while the use of non-linear springs was capable tiebreak contact. This method was not capable of accu-
of accurately predicted the softening part of the experimen- rately predicting the delamination growth. The main rea-
tal curve. The delamination element was capable of captur- son was due to the fact that the failure criterion was
ing the overall behaviour of the delamination failure; based on maximum interlaminar stress and not critical
however, bigger discrepancies were obtained with the energy release rate, therefore there was delamination
experimental curve. This was due to the fact that the ele- growth before the critical energy release rate was reached.
ment was implemented in an explicit transient FE code The comparison between the methods used shows that
(DYNA3D) and to achieve static solution dynamic relaxa- for the particular studied delamination phenomena, the
tion and damping were added to the model. Poor results proposed modelling techniques are capable of predicting
were obtained using tiebreak contact. It was due to the cri- the delamination growth in a double cantilever beam.
terion which has been used: once the critical interlaminar Future work will involve further validation of these model-
stress is reached, decohesion of the two layers occurs bru- ling techniques to model more complex structures and
tally, since there is delamination growth before the critical other delamination failure modes.
energy release rate has been reached.

8. Conclusions References

[1] DYNA3D user Manual. Lawrence Livermore National Laboratory,


The objective of the present work was to present model-
1999.
ling techniques able to predict a delamination failure mode [2] CU-Cast-EU project—Report on development of improved material
in composite structures. Four different ways of modelling model for composites—D4.1.6.
delamination growth of a double cantilever beam test [3] Ronald K. The virtual crack closure technique: history, approach
(DCB) are proposed. The first two approaches were based and applications, NASA/CR-2002-211628, ICASE Report No.
2002-10.
on a cohesive zone model, the third approach was based on
[4] Daudeville L, Allix O, Ladeveze P. Delamination analysis by damage
simulation of the interface with solid element and the last mechanics: some applications. Compos Eng 1995;5(1):17–24.
approach was based on the definition of a tiebreak contact. [5] LS-DYNA3D. Users Manual. Ver. 950 Livermore Software Techno-
The force–displacement curve obtained for delamination logy Corporation, Livermore, California, 1999.

Das könnte Ihnen auch gefallen