Sie sind auf Seite 1von 8

Meccanica (2010) 45: 297304

DOI 10.1007/s11012-009-9248-5

S I M U L AT I O N , O P T I M I Z AT I O N & I D E N T I F I C AT I O N

Computational models for mode I composite fracture


failure: the virtual crack closure technique
versus the two-step extension method
J. Bonhomme A. Argelles M.A. Castrillo
J. Via

Received: 24 March 2009 / Accepted: 11 September 2009 / Published online: 29 September 2009
Springer Science+Business Media B.V. 2009

Abstract This paper deals with the numerical de- results obtained from VCCT and Two-step extension
termination of the energy release rate under mode I models converge as element length decreases. Regard-
in carbon fibre reinforced composites (CFRC). Two ing the comparison between experimental and numer-
different models are reviewed: the virtual crack clo- ical results, the study showed that a correction for
sure technique (VCCT) and the Two-step extension testing devices compliance was needed to match both
method. The Two-step extension method needs two models.
computational steps in order to calculate the energy
release rate (G). The VCCT method is able to pro- Keywords Mode I fracture Composite
vide ERR value only from one computational step. delamination Virtual crack closure technique
Results were compared with empirical data obtained Two-step extension method
from double cantilever beam (DCB) tests carried out
on unidirectional AS4/8552 carbon/epoxy laminates.
This study showed that, in a pure mode I state, results
1 Introduction
obtained via the Two-step extension method were in
agreement with a straightforward calculation of the
Failure due to delamination is the most frequently ob-
elastic energy variation in the system. As expected,
served damage in carbon fibre reinforced plastic struc-
tures (CFRP) as the energy required to extend the
J. Bonhomme () A. Argelles M.A. Castrillo crack between plies is lower than that required for
Department of Construction and Manufacturing cracking across them.
Engineering, University of Oviedo, Edificio Over the last years, an extensive work has been
Departamental Oeste, 33203 Gijn, Asturias, Spain
e-mail: bonhomme@uniovi.es developed in order to predict the delamination onset
and crack growth by means of analytical methods, ex-
A. Argelles
e-mail: antonio@uniovi.es perimental tests and numerical procedures in differ-
ent laminate structures and crack configurations [1, 2].
M.A. Castrillo
e-mail: miguel@uniovi.es The understanding of fracture mechanisms could im-
prove the reliability of composite design and optimiza-
J. Via tion approaches [3]
Department of Materials Science and Metallurgical
There are several methods documented in the liter-
Engineering, University of Oviedo, Edificio
Departamental Este, 33203 Gijn, Asturias, Spain ature for computing the energy release rate by means
e-mail: jaure@uniovi.es of FE models. Among these methods, the virtual crack
298 Meccanica (2010) 45: 297304

closure technique (VCCT) has been the most widely


used to calculate G [47]. To date, only the ABAQUS
code has implemented this technique among the ma-
jor commercial general purpose finite element codes.
Nevertheless, it is possible to programme specific sub-
routines using APDL script in ANSYS and other FEM
commercial packages to perform the calculations.
This paper compares the VCCT model, the Two-
step extension method and a straightforward calcula-
tion of the elastic energy variation in the system for
the calculation of G in mode I. Along this work, em-
pirical data obtained from mode I tests of unidirec-
tional AS4/8552 CFRC have been used as a reference Fig. 1 Step 1 for the VCCT method. DOFs at coincident nodes
are coupled
to compare results.

2 The virtual crack closure technique

This method has been the most widely used for pre-
dicting failure in composite laminates. It has evolved
from the finite crack extension method and the virtual
crack extension method and is based on Irwins crack
closure integral [8]. Firstly, it was formulated by Ry-
bicki and Kanninen [9].
Giving a laminate with an initial crack length (a),
the following hypotheses are assumed in this method
when the crack extends a length a:
The released energy is identical to the energy re-
Fig. 2 Step 2 for the VCCT method. DOFs of the nodes at the
quired to close the crack.
crack front are released and the corresponding forces and dis-
Being a small enough, the stress state at the crack placements are computed
tip does not change significantly.
Figure 1 shows the FEM model. The crack path is
modelled using pairs of coincident nodes with identi- Fy2i : vertical forces at the crack tip
cal coordinates. Coincident nodes initially have all de- v1i : vertical displacements of the released nodes
grees of freedom (DOFs) coupled together in the un-
B: specimen width
loaded model. When the imposed load reaches a criti-
a: crack increment
cal value, the coupled DOFs of the nodes at the crack
n: number of nodes placed along the crack front
tip are released. The nodes are then allowed to move
and the crack extends one element length a (Fig. 2). (number of nodes in a row)
The critical energy release rate (GI c ) is then calcu- For elements with mid-side nodes, as described by
lated in the second step by means of the force at the
Krueger [5], two nodes must be released in order
crack tip and the displacement of the released nodes.
to avoid kinematic incompatibilities. This can be ex-
The above procedure can be analytically described
pressed as:
as follows:

1 
n
1 
n

GI = Fy2i (v1i v1 i ) (1) GI = Fy3i (v1i v1 i ) + Fy4i (v2i v2 i )
2Ba 2Ba
i=1 i=1

where: (2)
Meccanica (2010) 45: 297304 299

3 The two-step extension method placed on the midplane during lamination in order to
produce an artificial delamination to initiate the crack.
As in the VCCT, in this method the crack path is mod- The mechanical properties of the laminate are
elled using pairs of coincident nodes. Now, however, shown in Table 1.
the forces at the crack tip are calculated in the first
step. The imposed displacement in the sample is then
held and the coupled DOFs of the nodes at the crack 6 Experimental procedure
tip are released in the second step. Displacements are
then calculated in this second step. As a result, cal- Double cantilever beam (DCB) specimens were pre-
culation must be done with results from two different pared to carry out mode I fracture tests following
steps. In this case, forces and displacements are calcu- ASTM Standard D 5528-01. The DCB specimen,
lated in the same row of nodes (1 1 ). This procedure shown in Fig. 3, consists of a rectangular sample of
can be analytically described as follows: uniform thickness containing a non-adhesive insert on
the midplane that serves as a delamination initiator.
1 
n
Opening forces are applied to the specimen by means
GI = Fy1i (v1i v1 i ) (3)
2Ba of a pair of hinges bonded to one end of the specimen.
i=1
The applied force (P ) versus the opening displace-
And the corresponding equation for 8-node elements ment () is recorded on a computer. A critical value
is: of P (Pc ) is calculated from this curve as the onset
of the crack growth. This parameter may be calculated
1 
n

GI = Fy1i (v1i v1 i ) + Fy2i (v2i v2 i ) as the point of deviation from linearity (NL), the point
2Ba at which delamination is visually observed (VIS) and
i=1
(4) the point at which the compliance has increased by 5%
or the load has reached a maximum value (5%/max).
A travelling optical microscope (100) was used to
4 The elastic energy variation method measure the crack length during the test.

The delamination process is characterized by means of Table 1 Laminate mechanical properties


the energy release rate (G), which is a measurement of
Property MPa
the energy lost in the test specimen per unit of speci-
men width for an infinitesimal increase in delamina- E11 (Longitudinal elastic modulus) 144,000
tion length. In mathematical form: E22 (Transversal elastic modulus) 10,600
1 dU G12 (Shear elastic modulus) 5,360
G= (5) 11 (Longitudinal tensile strength) 1,703
B da
22 (Transversal tensile strength) 30,8
FEM software allows a direct calculation of the elastic
s (Shear strength) 67,7
energy of the system for a given load state. Therefore,
GI c can be calculated by means of (5) in two compu-
tational steps by computing the change in the elastic
energy of the system before and after the crack exten-
sion.

5 Material and specimen preparation

The material used as a reference in this study was


a Hexcel AS4/8552 6 mm thick unidirectional car-
bon fibre reinforced epoxy laminate. The structure of
the laminate was [0 ]16s . A non-adherent insert was Fig. 3 DCB specimen
300 Meccanica (2010) 45: 297304

Linear elastic behaviour is assumed in the calcula- In the CC method, n is the slope of the log C versus
tion of G used in this test method. log a least squares fit. In this case, the compliance and
Three data reduction methods were used to calcu- the delamination length follow the Berrys approach
late GI c [10]: [10, 11]:
(a) The Modified Beam Theory (MBT) Method: C = ka n (9)
3P On the other hand, A1 can be determined as the slope
GI c = (6)
2B (a + ||) of the least squares fit of a/ h versus C 1/3 [10].
In the DCB test, G value increases and then stabi-
(b) The Compliance Calibration (CC) Method:
lizes. The reason of this behaviour is the formation of
nP fibre bridging between the arms of the sample (Fig. 4).
GI c = (7) This test was carried out on an MTS testing ma-
2Ba
chine at a crosshead speed of 0.5 mm/min. Five sam-
(c) The Modified Compliance Calibration (MCC) ples were tested. Table 2 summarizes the results ob-
Method: tained in the tests.
2
3P 2 C 3
GI c = (8)
2A1 Bh 7 Numerical analysis
where:
A commercial FEM package ANSYS 8.0 was used to
P : load simulate mode I facture tests. Collapsed elements and
: load point displacement standard elements were studied at the crack tip. In or-
B: specimen width der to define the mesh around the crack tip for col-
h: specimen thickness lapsed elements, the ANSYS KSCON command has
a: delamination length
C: compliance
, n, A1 : calibration parameters
 corrects the beam theory for a perfectly built-in
double cantilever beam (rotation may occur at the de-
lamination front). It can be determined experimen-
tally by generating a least squares plot of the cube
root of compliance C 1/3 , as a function of delamina-
tion length a [10]. Fig. 4 Typical fibre bridging developed during mode I tests

Table 2 Mode I energy release rate. Experimental tests

Sample Displacement Max. Load GI c [J/m2 ] GI c [J/m2 ] GI c [J/m2 ]


at Max. Load [mm] [N] (MBT) (CC) (MCC)

1 1.851 138.0 298.2 310.5 292.5


2 2.085 130.7 303.4 317.5 309.8
3 1.865 167.0 347.0 383.9 350.3
4 1.834 123.3 265.5 276.2 253.2
5 1.873 134.1 296.5 309.6 284.7

Mean 1.902 138.6 302.1 319.5 298.1


S.d. 0.104 16.8 29.2 39.4 35.7

S.d.: standard deviation


Meccanica (2010) 45: 297304 301

been used. In this case, the mean length of the ele- Solid95: 20-node 3D solid element with three de-
ments at the crack tip was 0.6 mm (a/a = 0.012). grees of freedom at each node (translations in the
Nevertheless, different element lengths have been used nodal x, y and z directions)
at the crack tip to run the problem as showed in Table 4 Solid191: 20-node 3D layered solid element with
(from 0.1 to 1 mm). three degrees of freedom at each node (translations
Results showed that the energy release rate calcu- in the nodal x, y and z directions)
lation (GI c ) did not change significantly. Collapsed
Following the experimental procedure, the critical
elements seems to be more accurate than regular el-
load per unit width was applied to the model. A lin-
ements to obtain KI C values where results are related
ear elastic model was used to simulate the material
to stress concentrations, but regarding to energy calcu-
behaviour.
lations they do not seem to improve GI c calculations.
Table 3 shows a comparison between numerical and
Figures 5 and 6 show the mesh around the crack tip
experimental compliance obtained when the critical
for collapsed and regular elements.
load was applied to the ends of the sample.
The following element types were tested in differ-
As can be seen from these results, all the elements
ent runs:
used to model the loading step gave similar results.
Plane42: 4-node 2D solid element with two degrees The numerical models gave results that were 25%
of freedom at each node (translations in the nodal x stiffer than the experimental tests. Certain factors, in
and y directions) both the experimental and numerical models, can ex-
Plane82: 8-node 2D solid element with two degrees plain this mismatch. From the experimental point of
of freedom at each node (translations in the nodal x view the measurement of the displacement has been
and y directions) influenced by the compliance of the testing devices.

Fig. 5 2D model with collapsed elements around the crack tip (half model)

Fig. 6 2D model with regular elements at the crack tip


302 Meccanica (2010) 45: 297304
Table 3 FEM compliance error (experimental compliance: 0.0135 mm/N)

Geometry Element Material model Compliance (mm/N) Error

2D Plane42 Solid/orthotropic 0.0102 24%


2D Plane82 Solid/orthotropic 0.0103 24%
3D Solid95 Solid/orthotropic 0.0102 25%
3D Solid191 Layered/orthotropic 0.0103 24%

Fig. 7 Inconsistent solution due to kinematic incompatibility

On the other hand, a linear elastic model was chosen cided to use 2D models (Plane42 elements in plane
to perform the FE analysis whereas some degree of strain) to perform the energy release rate calculations.
non-linearity is always present in the real material near The critical load obtained in the experimental tests
the failure load. Nevertheless the load-displacement was applied to the model. GI c was then calculated
curves were highly linear up to failure, so the mis- through VCCT, Two-step extension method and by
match between experimental and numerical models computing the change in the elastic energy of the sys-
can be mainly attributed to the compliance of the test- tem before and after the crack extension. The study
ing devices. was performed for different element lengths. The ob-
Figure 7 shows the calculation of GI c for 8-node tained results are shown in Table 4.
elements (elements with midside nodes) in which the The MBT method was selected as the experimental
crack opening was modelled releasing the nodes one reference value because it provided the lowest stan-
by one. The kinematic incompatibility predicted by dard deviation (Table 2). As can be seen in Table 4,
Krueger [5] is shown as an inconsistent oscillatory FEM results were 30% lower than experimental results
solution that fluctuated depending on the number of for the analyzed sample (the difference oscillated be-
nodes released at each time. tween 20 to 30% error for other samples), similar to
As there were not significant differences between the difference observed between numerical and exper-
2D and 3D models in the preliminary runs, it was de- imental compliance in the preliminary calibration runs
Meccanica (2010) 45: 297304 303
Table 4 GI c numerical and experimental results

a (mm) (element length) 0.1 0.125 0.25 0.5 1

a/a 0.0022 0.0027 0.0055 0.0110 0.0220


GI c (VCCT) (J/m2 ) 225.4 225.2 223.8 221.0 215.3
GI c (Two-step extension) (J/m2 ) 226.3 226.3 226.0 225.3 223.8
GI c (U ) (J/m2 ) 226.2 226.2 226.1 225.4 223.8

GI c (exper.) (J/m2 ) (MBT) 302.1

GI c (VCCT): Virtual crack closure technique


GI c (Two-step extension): Two-step extension method
GI c (U ): GI c calculated through the elastic energy change in the system

Fig. 8 GI c versus element length

(Table 3). Another consequence of this stiffness mis- can also be observed in this figure that the VCCT
match, is that numerical results depend on weather we method is more sensitive to element size than the two-
select the experimental load (Pc ) or displacement (c ) step extension method. Furthermore, as expected, both
as critical values to input into the FEM model. models converge as the element length decreases.
On the other hand, it can be seen that the Two- The obtained results in this study, suggested the
step extension results are coincident with those de- need for testing machine compliance calibration to im-
rived from the elastic energy decrease in the system, prove the convergence between experimental and nu-
which can be easily calculated using the ANSYS code. merical models.
Figure 8 shows a plot of the calculated GI c values In order to determine the testing devices compli-
versus element length. As can be seen, VCCT gives ance, there were carried out four tests in mode I sam-
lower results than the two-step extension method. It ples without a crack in their mid-plane. The tests
304 Meccanica (2010) 45: 297304
Table 5 Stiffness of the testing devices (N/mm) the mismatch between both models from values next to
Sample Stiffness (N/mm) 30% to values less than 10%.

1 391.6
2 389.7 References
3 343.6
4 337.6 1. Carloni C, Piva A, Viola E (2004) Biaxial load effect
on crack initiation for orthotropic materials. Meccanica
Mean 365.6 39(4):331344. doi:10.1023/B:MECC.0000029363.67419.
S.d. 29.0 80
2. Krueger R (2006) Computational fracture mechanics for
S.d.: standard deviation composites. State of the art and challenges, NAFEMS
Nordic seminar: prediction and modelling of failure using
FEA, Copenhagen/Roskilde, Denmark
were carried out at the same speed as in mode I tests 3. Dehmous H, Welemane H, Karama M, Tahar K (2008) Re-
(0.5 mm/min) up to the maximum load found in GI c liability approach for fibre-reinforced composites design.
tests. Table 5 shows the stiffness of the testing devices. Int J Simul Multidiscip Des Optim 2:19. doi:10.1051/
smdo:2008001
Corresponding corrections were applied to both ex- 4. Leski A (2007) Implementation of the virtual crack clo-
perimental and numerical models, lowering the mis- sure technique in engineering FE calculations. Finite
match between them below 9%. Elem Anal Des 43(3):261268. doi:10.1016/j.finel.2006.
10.004
5. Krueger R (2004) Virtual crack closure technique. History,
approach and applications. Appl Mech Rev 57:109143.
8 Conclusions doi:10.1115/1.1595677
6. Mikulik Z, Prusty BG, Thomson RS, Kelly DW (2006)
Finite Element Modelling is a very useful technique Application of fracture mechanics-based methodologies for
for studying fracture problems from a theoretical point failure predictions in composite structures. Conference on
of view. Experimental and numerical models comple- damage in composite materials, Stuttgart
7. Hellen TK (1975) On the method of virtual crack extension.
ment each other in order to further understand the Int J Numer Meth Eng 9(1):187207
crack opening process in CFRC. 8. Irwin GR (1958) Fracture I. In: Flge S (ed) Handbuch der
In mode I crack opening, the VCCT method fur- Physik. Springer, New York
nishes slightly lower results than the two-step exten- 9. Rybicki EF, Kanninen MF (1977) Finite element calcu-
lation of stress intensity factors by a modified crack clo-
sion method. On the other hand the two-step extension
sure integral. Eng Fract Mech 9(4):931938. doi:10.1016/
method gives results that are coincident with the calcu- 0013-7944(77)90013-3
lation of the elastic energy change in the system. Both 10. Standard test method for mode I interlaminar fracture
VCCT and Two-step extension methods converge as toughness of unidirectional fiber-reinforced polymer matrix
element length decreases. composites (2001) ASTM D 5528-01, American Society
for Testing and Materials, Philadelphia, Pennsylvania
Finally, it was needed to apply a correction for test- 11. Berry JP (1963) Determination of fracture surface ener-
ing machine compliance in order to match the experi- gies by the cleavage technique. J Appl Phys 34(1):6268.
mental and numerical models. This correction lowered doi:10.1063/1.1729091

Das könnte Ihnen auch gefallen